{
$(set_function_pointers c $ALL_ARCHS)
#if HAVE_DSPR2
+#if CONFIG_VP8
void dsputil_static_init();
dsputil_static_init();
#endif
+#if CONFIG_VP9
+void vp9_dsputil_static_init();
+vp9_dsputil_static_init();
+#endif
+#endif
}
#endif
$(common_bottom)
# this is used, it's used for two subsequent load instructions,
# where a hand-written version of it could merge two subsequent
# add and sub instructions.
- s/^(\s*)((ldr|str|pld)(ne)?)(\s+)(r\d+,)?\s*\[(\w+), -([^\]]+)\]/$1sub$4$5$7, $7, $8\n$1$2$5$6\[$7\]\n$1add$4$5$7, $7, $8/g;
+ s/^(\s*)((ldr|str|pld)(ne)?)(\s+)(r\d+,\s*)?\[(\w+), -([^\]]+)\]/$1sub$4$5$7, $7, $8\n$1$2$5$6\[$7\]\n$1add$4$5$7, $7, $8/g;
# Convert register post indexing to a separate add instruction.
# This converts "ldrneb r9, [r0], r2" into "ldrneb r9, [r0]",
INSTALL-LIBS-yes += include/vpx/vpx_codec.h
INSTALL-LIBS-yes += include/vpx/vpx_image.h
INSTALL-LIBS-yes += include/vpx/vpx_integer.h
-INSTALL-LIBS-yes += include/vpx/vpx_codec_impl_top.h
-INSTALL-LIBS-yes += include/vpx/vpx_codec_impl_bottom.h
INSTALL-LIBS-$(CONFIG_DECODERS) += include/vpx/vpx_decoder.h
INSTALL-LIBS-$(CONFIG_ENCODERS) += include/vpx/vpx_encoder.h
ifeq ($(CONFIG_EXTERNAL_BUILD),yes)
make_tuple(32, 64, &convolve8_neon),
make_tuple(64, 64, &convolve8_neon)));
#endif
+
+#if HAVE_DSPR2
+const ConvolveFunctions convolve8_dspr2(
+ vp9_convolve8_horiz_dspr2, vp9_convolve8_avg_horiz_dspr2,
+ vp9_convolve8_vert_dspr2, vp9_convolve8_avg_vert_dspr2,
+ vp9_convolve8_dspr2, vp9_convolve8_avg_dspr2);
+
+INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
+ make_tuple(4, 4, &convolve8_dspr2),
+ make_tuple(8, 4, &convolve8_dspr2),
+ make_tuple(4, 8, &convolve8_dspr2),
+ make_tuple(8, 8, &convolve8_dspr2),
+ make_tuple(16, 8, &convolve8_dspr2),
+ make_tuple(8, 16, &convolve8_dspr2),
+ make_tuple(16, 16, &convolve8_dspr2),
+ make_tuple(32, 16, &convolve8_dspr2),
+ make_tuple(16, 32, &convolve8_dspr2),
+ make_tuple(32, 32, &convolve8_dspr2),
+ make_tuple(64, 32, &convolve8_dspr2),
+ make_tuple(32, 64, &convolve8_dspr2),
+ make_tuple(64, 64, &convolve8_dspr2)));
+#endif
} // namespace
}
void idct4x4_add(int16_t* /*in*/, int16_t *out, uint8_t *dst,
int stride, int /*tx_type*/) {
- vp9_short_idct4x4_add_c(out, dst, stride >> 1);
+ vp9_idct4x4_16_add_c(out, dst, stride >> 1);
}
void fht4x4(int16_t *in, int16_t *out, uint8_t* /*dst*/,
int stride, int tx_type) {
65f45ec9a55537aac76104818278e0978f94a678 vp90-2-03-deltaq.webm.md5
4dbb87494c7f565ffc266c98d17d0d8c7a5c5aba vp90-2-05-resize.ivf
7f6d8879336239a43dbb6c9f13178cb11cf7ed09 vp90-2-05-resize.ivf.md5
+bf61ddc1f716eba58d4c9837d4e91031d9ce4ffe vp90-2-06-bilinear.webm
+f6235f937552e11d8eb331ec55da6b3aa596b9ac vp90-2-06-bilinear.webm.md5
495256cfd123fe777b2c0406862ed8468a1f4677 vp91-2-04-yv444.webm
65e3a7ffef61ab340d9140f335ecc49125970c2c vp91-2-04-yv444.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-deltaq.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-05-resize.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-05-resize.ivf.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-06-bilinear.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-06-bilinear.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp91-2-04-yv444.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp91-2-04-yv444.webm.md5
"vp90-2-03-size-226x202.webm", "vp90-2-03-size-226x208.webm",
"vp90-2-03-size-226x210.webm", "vp90-2-03-size-226x224.webm",
"vp90-2-03-size-226x226.webm", "vp90-2-03-deltaq.webm",
- "vp90-2-05-resize.ivf",
+ "vp90-2-05-resize.ivf", "vp90-2-06-bilinear.webm",
#if CONFIG_NON420
"vp91-2-04-yv444.webm"
#endif
*/
-#include <stdlib.h>
#include "filter.h"
-#include "vpx_ports/mem.h"
DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]) =
{
#ifndef FILTER_H
#define FILTER_H
+#include "vpx_ports/mem.h"
+
#define BLOCK_HEIGHT_WIDTH 4
#define VP8_FILTER_WEIGHT 128
#define VP8_FILTER_SHIFT 7
-extern const short vp8_bilinear_filters[8][2];
-extern const short vp8_sub_pel_filters[8][6];
+extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]);
+extern DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]);
#endif
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "vpx_ports/mem.h"
+#include "vp8/common/x86/filter_x86.h"
DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_4[8][8]) =
{
#ifndef FILTER_X86_H
#define FILTER_X86_H
+#include "vpx_ports/mem.h"
+
/* x86 assembly specific copy of vp8/common/filter.c:vp8_bilinear_filters with
* duplicated values */
-extern const short vp8_bilinear_filters_x86_4[8][8]; /* duplicated 4x */
-extern const short vp8_bilinear_filters_x86_8[8][16]; /* duplicated 8x */
+
+/* duplicated 4x */
+extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_4[8][8]);
+
+/* duplicated 8x */
+extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_8[8][16]);
#endif /* FILTER_X86_H */
int16_t skip_adding,
uint8_t *dest,
int dest_stride);
-extern void vp9_short_idct10_16x16_add_neon_pass1(int16_t *input,
+extern void vp9_short_idct16x16_10_add_neon_pass1(int16_t *input,
int16_t *output,
int output_stride);
-extern void vp9_short_idct10_16x16_add_neon_pass2(int16_t *src,
+extern void vp9_short_idct16x16_10_add_neon_pass2(int16_t *src,
int16_t *output,
int16_t *pass1Output,
int16_t skip_adding,
uint8_t *dest,
int dest_stride);
-extern void save_neon_registers();
-extern void restore_neon_registers();
+/* For ARM NEON, d8-d15 are callee-saved registers, and need to be saved. */
+extern void vp9_push_neon(int64_t *store);
+extern void vp9_pop_neon(int64_t *store);
void vp9_short_idct16x16_add_neon(int16_t *input,
uint8_t *dest, int dest_stride) {
+ int64_t store_reg[8];
int16_t pass1_output[16*16] = {0};
int16_t row_idct_output[16*16] = {0};
// save d8-d15 register values.
- save_neon_registers();
+ vp9_push_neon(store_reg);
/* Parallel idct on the upper 8 rows */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
dest_stride);
// restore d8-d15 register values.
- restore_neon_registers();
+ vp9_pop_neon(store_reg);
return;
}
-void vp9_short_idct10_16x16_add_neon(int16_t *input,
+void vp9_short_idct16x16_10_add_neon(int16_t *input,
uint8_t *dest, int dest_stride) {
+ int64_t store_reg[8];
int16_t pass1_output[16*16] = {0};
int16_t row_idct_output[16*16] = {0};
// save d8-d15 register values.
- save_neon_registers();
+ vp9_push_neon(store_reg);
/* Parallel idct on the upper 8 rows */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct10_16x16_add_neon_pass1(input, pass1_output, 8);
+ vp9_short_idct16x16_10_add_neon_pass1(input, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7
// which will be saved into row_idct_output.
- vp9_short_idct10_16x16_add_neon_pass2(input+1,
+ vp9_short_idct16x16_10_add_neon_pass2(input+1,
row_idct_output,
pass1_output,
0,
dest_stride);
// restore d8-d15 register values.
- restore_neon_registers();
+ vp9_pop_neon(store_reg);
return;
}
--- /dev/null
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+ EXPORT |vp9_push_neon|
+ EXPORT |vp9_pop_neon|
+
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+|vp9_push_neon| PROC
+ vst1.i64 {d8, d9, d10, d11}, [r0]!
+ vst1.i64 {d12, d13, d14, d15}, [r0]!
+ bx lr
+
+ ENDP
+
+|vp9_pop_neon| PROC
+ vld1.i64 {d8, d9, d10, d11}, [r0]!
+ vld1.i64 {d12, d13, d14, d15}, [r0]!
+ bx lr
+
+ ENDP
+
+ END
+
EXPORT |vp9_short_idct16x16_add_neon_pass1|
EXPORT |vp9_short_idct16x16_add_neon_pass2|
- EXPORT |vp9_short_idct10_16x16_add_neon_pass1|
- EXPORT |vp9_short_idct10_16x16_add_neon_pass2|
- EXPORT |save_neon_registers|
- EXPORT |restore_neon_registers|
+ EXPORT |vp9_short_idct16x16_10_add_neon_pass1|
+ EXPORT |vp9_short_idct16x16_10_add_neon_pass2|
ARM
REQUIRE8
PRESERVE8
bx lr
ENDP ; |vp9_short_idct16x16_add_neon_pass2|
-;void |vp9_short_idct10_16x16_add_neon_pass1|(int16_t *input,
+;void |vp9_short_idct16x16_10_add_neon_pass1|(int16_t *input,
; int16_t *output, int output_stride)
;
; r0 int16_t input
; idct16 stage1 - stage6 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct10_16x16_add_neon_pass1| PROC
+|vp9_short_idct16x16_10_add_neon_pass1| PROC
; TODO(hkuang): Find a better way to load the elements.
; load elements of 0, 2, 4, 6, 8, 10, 12, 14 into q8 - q15
vst1.64 {d31}, [r1], r2
bx lr
- ENDP ; |vp9_short_idct10_16x16_add_neon_pass1|
+ ENDP ; |vp9_short_idct16x16_10_add_neon_pass1|
-;void vp9_short_idct10_16x16_add_neon_pass2(int16_t *src,
+;void vp9_short_idct16x16_10_add_neon_pass2(int16_t *src,
; int16_t *output,
; int16_t *pass1Output,
; int16_t skip_adding,
; idct16 stage1 - stage7 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct10_16x16_add_neon_pass2| PROC
+|vp9_short_idct16x16_10_add_neon_pass2| PROC
push {r3-r9}
; TODO(hkuang): Find a better way to load the elements.
end_idct10_16x16_pass2
pop {r3-r9}
bx lr
- ENDP ; |vp9_short_idct10_16x16_add_neon_pass2|
-;void |save_neon_registers|()
-|save_neon_registers| PROC
- vpush {d8-d15}
- bx lr
- ENDP ; |save_registers|
-;void |restore_neon_registers|()
-|restore_neon_registers| PROC
- vpop {d8-d15}
- bx lr
- ENDP ; |restore_registers|
+ ENDP ; |vp9_short_idct16x16_10_add_neon_pass2|
END
;
- EXPORT |vp9_short_idct4x4_1_add_neon|
+ EXPORT |vp9_idct4x4_1_add_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp9_short_idct4x4_1_add_neon(int16_t *input, uint8_t *dest,
+;void vp9_idct4x4_1_add_neon(int16_t *input, uint8_t *dest,
; int dest_stride)
;
; r0 int16_t input
; r1 uint8_t *dest
; r2 int dest_stride)
-|vp9_short_idct4x4_1_add_neon| PROC
+|vp9_idct4x4_1_add_neon| PROC
ldrsh r0, [r0]
; generate cospi_16_64 = 11585
vst1.32 {d7[1]}, [r12]
bx lr
- ENDP ; |vp9_short_idct4x4_1_add_neon|
+ ENDP ; |vp9_idct4x4_1_add_neon|
END
; be found in the AUTHORS file in the root of the source tree.
;
- EXPORT |vp9_short_idct4x4_add_neon|
+ EXPORT |vp9_idct4x4_16_add_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
AREA Block, CODE, READONLY ; name this block of code
-;void vp9_short_idct4x4_add_neon(int16_t *input, uint8_t *dest, int dest_stride)
+;void vp9_idct4x4_16_add_neon(int16_t *input, uint8_t *dest, int dest_stride)
;
; r0 int16_t input
; r1 uint8_t *dest
; r2 int dest_stride)
-|vp9_short_idct4x4_add_neon| PROC
+|vp9_idct4x4_16_add_neon| PROC
; The 2D transform is done with two passes which are actually pretty
; similar. We first transform the rows. This is done by transposing
vst1.32 {d26[1]}, [r1], r2
vst1.32 {d26[0]}, [r1] ; no post-increment
bx lr
- ENDP ; |vp9_short_idct4x4_add_neon|
+ ENDP ; |vp9_idct4x4_16_add_neon|
END
;
EXPORT |vp9_short_idct8x8_add_neon|
- EXPORT |vp9_short_idct10_8x8_add_neon|
+ EXPORT |vp9_short_idct8x8_10_add_neon|
ARM
REQUIRE8
PRESERVE8
bx lr
ENDP ; |vp9_short_idct8x8_add_neon|
-;void vp9_short_idct10_8x8_add_neon(int16_t *input, uint8_t *dest, int dest_stride)
+;void vp9_short_idct8x8_10_add_neon(int16_t *input, uint8_t *dest, int dest_stride)
;
; r0 int16_t input
; r1 uint8_t *dest
; r2 int dest_stride)
-|vp9_short_idct10_8x8_add_neon| PROC
+|vp9_short_idct8x8_10_add_neon| PROC
push {r4-r9}
vpush {d8-d15}
vld1.s16 {q8,q9}, [r0]!
vpop {d8-d15}
pop {r4-r9}
bx lr
- ENDP ; |vp9_short_idct10_8x8_add_neon|
+ ENDP ; |vp9_short_idct8x8_10_add_neon|
END
#include "./vpx_config.h"
-#include "vp9_rtcd.h"
+#include "./vp9_rtcd.h"
#include "vp9/common/vp9_onyxc_int.h"
void vp9_machine_specific_config(VP9_COMMON *cm) {
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_COMMON_DSPR2_H_
+#define VP9_COMMON_VP9_COMMON_DSPR2_H_
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+
+#if HAVE_DSPR2
+#define CROP_WIDTH 512
+extern uint8_t *vp9_ff_cropTbl;
+
+#define DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input) ({ \
+ \
+ int32_t tmp, out; \
+ int dct_cost_rounding = DCT_CONST_ROUNDING; \
+ int in = input; \
+ \
+ __asm__ __volatile__ ( \
+ /* out = dct_const_round_shift(input_dc * cospi_16_64); */ \
+ "mtlo %[dct_cost_rounding], $ac1 \n\t"\
+ "mthi $zero, $ac1 \n\t"\
+ "madd $ac1, %[in], %[cospi_16_64] \n\t"\
+ "extp %[tmp], $ac1, 31 \n\t"\
+ \
+ /* out = dct_const_round_shift(out * cospi_16_64); */ \
+ "mtlo %[dct_cost_rounding], $ac2 \n\t"\
+ "mthi $zero, $ac2 \n\t"\
+ "madd $ac2, %[tmp], %[cospi_16_64] \n\t"\
+ "extp %[out], $ac2, 31 \n\t"\
+ \
+ : [tmp] "=&r" (tmp), [out] "=r" (out) \
+ : [in] "r" (in), \
+ [dct_cost_rounding] "r" (dct_cost_rounding), \
+ [cospi_16_64] "r" (cospi_16_64) \
+ ); \
+ out; })
+
+static INLINE void vp9_prefetch_load(const unsigned char *src) {
+ __asm__ __volatile__ (
+ "pref 0, 0(%[src]) \n\t"
+ :
+ : [src] "r" (src)
+ );
+}
+
+/* prefetch data for store */
+static INLINE void vp9_prefetch_store(unsigned char *dst) {
+ __asm__ __volatile__ (
+ "pref 1, 0(%[dst]) \n\t"
+ :
+ : [dst] "r" (dst)
+ );
+}
+
+static INLINE void vp9_prefetch_load_streamed(const unsigned char *src) {
+ __asm__ __volatile__ (
+ "pref 4, 0(%[src]) \n\t"
+ :
+ : [src] "r" (src)
+ );
+}
+
+/* prefetch data for store */
+static INLINE void vp9_prefetch_store_streamed(unsigned char *dst) {
+ __asm__ __volatile__ (
+ "pref 5, 0(%[dst]) \n\t"
+ :
+ : [dst] "r" (dst)
+ );
+}
+
+#endif // #if HAVE_DSPR2
+#endif // VP9_COMMON_VP9_COMMON_DSPR2_H_
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_avg_vert_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t w,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2, load3, load4;
+ uint32_t p1, p2;
+ uint32_t n1, n2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2;
+
+ vector1b = ((const int32_t *)filter_y)[0];
+ vector2b = ((const int32_t *)filter_y)[1];
+ vector3b = ((const int32_t *)filter_y)[2];
+ vector4b = ((const int32_t *)filter_y)[3];
+
+ src -= 3 * src_stride;
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+
+ for (x = 0; x < w; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+ "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
+ "extp %[Temp2], $ac3, 31 \n\t"
+ "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [load3] "=&r" (load3), [load4] "=&r" (load4),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [n1] "=&r" (n1), [n2] "=&r" (n2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride), [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_avg_vert_64_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2, load3, load4;
+ uint32_t p1, p2;
+ uint32_t n1, n2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2;
+
+ vector1b = ((const int32_t *)filter_y)[0];
+ vector2b = ((const int32_t *)filter_y)[1];
+ vector3b = ((const int32_t *)filter_y)[2];
+ vector4b = ((const int32_t *)filter_y)[3];
+
+ src -= 3 * src_stride;
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+ vp9_prefetch_store(dst + dst_stride + 32);
+
+ for (x = 0; x < 64; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+ "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
+ "extp %[Temp2], $ac3, 31 \n\t"
+ "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [load3] "=&r" (load3), [load4] "=&r" (load4),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [n1] "=&r" (n1), [n2] "=&r" (n2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride), [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vp9_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ if (((const int32_t *)filter_y)[1] == 0x800000) {
+ vp9_convolve_avg(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ } else {
+ if (16 == y_step_q4) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ case 8:
+ case 16:
+ case 32:
+ convolve_avg_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64:
+ vp9_prefetch_store(dst + 32);
+ convolve_avg_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, h);
+ break;
+ default:
+ vp9_convolve8_avg_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_avg_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+ }
+}
+
+void vp9_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ /* Fixed size intermediate buffer places limits on parameters. */
+ DECLARE_ALIGNED_ARRAY(32, uint8_t, temp, 64 * 135);
+ int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
+
+ assert(w <= 64);
+ assert(h <= 64);
+
+ if (intermediate_height < h)
+ intermediate_height = h;
+
+ if (x_step_q4 != 16 || y_step_q4 != 16)
+ return vp9_convolve8_avg_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+
+ vp9_convolve8_horiz(src - (src_stride * 3), src_stride,
+ temp, 64,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, intermediate_height);
+
+ vp9_convolve8_avg_vert(temp + (64*3), 64,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+}
+
+void vp9_convolve_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h) {
+ int x, y;
+ uint32_t tp1, tp2, tn1;
+ uint32_t tp3, tp4, tn2;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ /* 1 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 0(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "sw %[tn1], 0(%[dst]) \n\t" /* store */
+
+ : [tn1] "=&r" (tn1), [tp1] "=&r" (tp1),
+ [tp2] "=&r" (tp2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ case 8:
+ /* 2 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 0(%[dst]) \n\t"
+ "ulw %[tp3], 4(%[src]) \n\t"
+ "ulw %[tp4], 4(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "sw %[tn1], 0(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 4(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ case 16:
+ /* 4 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 0(%[dst]) \n\t"
+ "ulw %[tp3], 4(%[src]) \n\t"
+ "ulw %[tp4], 4(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 8(%[src]) \n\t"
+ "ulw %[tp2], 8(%[dst]) \n\t"
+ "sw %[tn1], 0(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 4(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 12(%[src]) \n\t"
+ "ulw %[tp4], 12(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "sw %[tn1], 8(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 12(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ case 32:
+ /* 8 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 0(%[dst]) \n\t"
+ "ulw %[tp3], 4(%[src]) \n\t"
+ "ulw %[tp4], 4(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 8(%[src]) \n\t"
+ "ulw %[tp2], 8(%[dst]) \n\t"
+ "sw %[tn1], 0(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 4(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 12(%[src]) \n\t"
+ "ulw %[tp4], 12(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 16(%[src]) \n\t"
+ "ulw %[tp2], 16(%[dst]) \n\t"
+ "sw %[tn1], 8(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 12(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 20(%[src]) \n\t"
+ "ulw %[tp4], 20(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 24(%[src]) \n\t"
+ "ulw %[tp2], 24(%[dst]) \n\t"
+ "sw %[tn1], 16(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 20(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 28(%[src]) \n\t"
+ "ulw %[tp4], 28(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "sw %[tn1], 24(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 28(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ case 64:
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ /* 16 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_load(src + src_stride + 64);
+ vp9_prefetch_store(dst + dst_stride);
+ vp9_prefetch_store(dst + dst_stride + 32);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 0(%[dst]) \n\t"
+ "ulw %[tp3], 4(%[src]) \n\t"
+ "ulw %[tp4], 4(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 8(%[src]) \n\t"
+ "ulw %[tp2], 8(%[dst]) \n\t"
+ "sw %[tn1], 0(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 4(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 12(%[src]) \n\t"
+ "ulw %[tp4], 12(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 16(%[src]) \n\t"
+ "ulw %[tp2], 16(%[dst]) \n\t"
+ "sw %[tn1], 8(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 12(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 20(%[src]) \n\t"
+ "ulw %[tp4], 20(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 24(%[src]) \n\t"
+ "ulw %[tp2], 24(%[dst]) \n\t"
+ "sw %[tn1], 16(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 20(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 28(%[src]) \n\t"
+ "ulw %[tp4], 28(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 32(%[src]) \n\t"
+ "ulw %[tp2], 32(%[dst]) \n\t"
+ "sw %[tn1], 24(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 28(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 36(%[src]) \n\t"
+ "ulw %[tp4], 36(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 40(%[src]) \n\t"
+ "ulw %[tp2], 40(%[dst]) \n\t"
+ "sw %[tn1], 32(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 36(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 44(%[src]) \n\t"
+ "ulw %[tp4], 44(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 48(%[src]) \n\t"
+ "ulw %[tp2], 48(%[dst]) \n\t"
+ "sw %[tn1], 40(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 44(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 52(%[src]) \n\t"
+ "ulw %[tp4], 52(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "ulw %[tp1], 56(%[src]) \n\t"
+ "ulw %[tp2], 56(%[dst]) \n\t"
+ "sw %[tn1], 48(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 52(%[dst]) \n\t" /* store */
+ "ulw %[tp3], 60(%[src]) \n\t"
+ "ulw %[tp4], 60(%[dst]) \n\t"
+ "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
+ "sw %[tn1], 56(%[dst]) \n\t" /* store */
+ "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
+ "sw %[tn2], 60(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ default:
+ for (y = h; y > 0; --y) {
+ for (x = 0; x < w; ++x) {
+ dst[x] = (dst[x] + src[x] + 1) >> 1;
+ }
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+}
+#endif
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_avg_horiz_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3, Temp4;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3, p4;
+ uint32_t n1, n2, n3, n4;
+ uint32_t tn1, tn2;
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "ulw %[tn2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn2] \n\t"
+ "balign %[tn1], %[tn2], 3 \n\t"
+ "balign %[tn2], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbu %[p2], 3(%[dst]) \n\t" /* load odd 2 */
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "lbu %[Temp1], 1(%[dst]) \n\t" /* load odd 1 */
+ "preceu.ph.qbr %[n1], %[tp2] \n\t"
+ "preceu.ph.qbl %[n2], %[tp2] \n\t"
+ "preceu.ph.qbr %[n3], %[tn2] \n\t"
+ "preceu.ph.qbl %[n4], %[tn2] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tn2], 0(%[dst]) \n\t" /* load even 1 */
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp3](%[cm]) \n\t" /* even 2 */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[n1], %[tn1] \n\t"
+ "lbux %[tn1], %[Temp2](%[cm]) \n\t" /* odd 1 */
+ "addqh_r.w %[tn2], %[tn2], %[tp1] \n\t" /* average even 1 */
+ "dpa.w.ph $ac2, %[n2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[n3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector4b] \n\t"
+ "extp %[Temp4], $ac2, 31 \n\t"
+
+ "lbu %[tp1], 2(%[dst]) \n\t" /* load even 2 */
+ "sb %[tn2], 0(%[dst]) \n\t" /* store even 1 */
+
+ /* clamp */
+ "addqh_r.w %[Temp1], %[Temp1], %[tn1] \n\t" /* average odd 1 */
+ "lbux %[n2], %[Temp4](%[cm]) \n\t" /* odd 2 */
+ "sb %[Temp1], 1(%[dst]) \n\t" /* store odd 1 */
+
+ "addqh_r.w %[tp1], %[tp1], %[tp2] \n\t" /* average even 2 */
+ "sb %[tp1], 2(%[dst]) \n\t" /* store even 2 */
+
+ "addqh_r.w %[p2], %[p2], %[n2] \n\t" /* average odd 2 */
+ "sb %[p2], 3(%[dst]) \n\t" /* store odd 2 */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1), [n2] "=&r" (n2), [n3] "=&r" (n3), [n4] "=&r" (n4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_avg_horiz_8_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3, p4, n1;
+ uint32_t tn1, tn2, tn3;
+ uint32_t st0, st1;
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "ulw %[tn2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+ "lbu %[Temp2], 0(%[dst]) \n\t"
+ "lbu %[tn3], 2(%[dst]) \n\t"
+
+ /* even 2. pixel */
+ "preceu.ph.qbr %[p1], %[tn2] \n\t"
+ "preceu.ph.qbl %[n1], %[tn2] \n\t"
+ "ulw %[tn1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[tn1] \n\t"
+ "lbux %[st1], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector4b] \n\t"
+ "extp %[Temp1], $ac1, 31 \n\t"
+
+ "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
+ "addqh_r.w %[tn3], %[tn3], %[st1] \n\t"
+ "sb %[Temp2], 0(%[dst]) \n\t"
+ "sb %[tn3], 2(%[dst]) \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "balign %[tn3], %[tn1], 3 \n\t"
+ "balign %[tn1], %[tn2], 3 \n\t"
+ "balign %[tn2], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "lbu %[Temp2], 4(%[dst]) \n\t"
+ "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "sb %[Temp2], 4(%[dst]) \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tn2] \n\t"
+ "preceu.ph.qbl %[p4], %[tn2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tp1], 6(%[dst]) \n\t"
+
+ /* odd 2. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn1] \n\t"
+ "preceu.ph.qbl %[n1], %[tn1] \n\t"
+ "lbux %[st0], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ "lbu %[tp2], 1(%[dst]) \n\t"
+ "lbu %[tn2], 3(%[dst]) \n\t"
+ "addqh_r.w %[tp1], %[tp1], %[st0] \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[st1], %[Temp2](%[cm]) \n\t"
+ "preceu.ph.qbr %[p2], %[tn3] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector4b] \n\t"
+ "addqh_r.w %[tp2], %[tp2], %[st1] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tn3], 5(%[dst]) \n\t"
+
+ /* odd 4. pixel */
+ "sb %[tp2], 1(%[dst]) \n\t"
+ "sb %[tp1], 6(%[dst]) \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbu %[tn1], 7(%[dst]) \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "addqh_r.w %[tn2], %[tn2], %[p4] \n\t"
+
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[tn3], %[tn3], %[p2] \n\t"
+
+ "lbux %[n1], %[Temp1](%[cm]) \n\t"
+ "addqh_r.w %[tn1], %[tn1], %[n1] \n\t"
+
+ /* store bytes */
+ "sb %[tn2], 3(%[dst]) \n\t"
+ "sb %[tn3], 5(%[dst]) \n\t"
+ "sb %[tn1], 7(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2), [tn3] "=&r" (tn3),
+ [st0] "=&r" (st0), [st1] "=&r" (st1),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_avg_horiz_16_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+ "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
+ "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
+ "ulw %[qload3], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
+ "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
+ "ulw %[qload3], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
+ "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
+
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
+
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
+
+ "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
+ "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
+ "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [qload3] "=&r" (qload3), [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+static void convolve_avg_horiz_64_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+ vp9_prefetch_store(dst_ptr + dst_stride + 32);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+ "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
+ "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
+ "ulw %[qload3], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
+ "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
+ "ulw %[qload3], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
+ "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
+
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
+
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
+
+ "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
+ "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
+ "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [qload3] "=&r" (qload3), [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+void vp9_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ if (((const int32_t *)filter_x)[1] == 0x800000) {
+ vp9_convolve_avg(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ } else {
+ if (16 == x_step_q4) {
+ uint32_t pos = 38;
+
+ src -= 3;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ convolve_avg_horiz_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 8:
+ convolve_avg_horiz_8_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 16:
+ convolve_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 1);
+ break;
+ case 32:
+ convolve_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 2);
+ break;
+ case 64:
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ convolve_avg_horiz_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ default:
+ vp9_convolve8_avg_horiz_c(src + 3, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_avg_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+ }
+}
+#endif
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+uint8_t vp9_ff_cropTbl_a[256 + 2 * CROP_WIDTH];
+uint8_t *vp9_ff_cropTbl;
+
+void vp9_dsputil_static_init(void) {
+ int i;
+
+ for (i = 0; i < 256; i++) vp9_ff_cropTbl_a[i + CROP_WIDTH] = i;
+
+ for (i = 0; i < CROP_WIDTH; i++) {
+ vp9_ff_cropTbl_a[i] = 0;
+ vp9_ff_cropTbl_a[i + CROP_WIDTH + 256] = 255;
+ }
+
+ vp9_ff_cropTbl = &vp9_ff_cropTbl_a[CROP_WIDTH];
+}
+
+static void convolve_horiz_4_transposed_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint8_t *dst_ptr;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3, Temp4;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3, p4;
+ uint32_t tn1, tn2;
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ dst_ptr = dst;
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "ulw %[tn2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn2] \n\t"
+ "balign %[tn1], %[tn2], 3 \n\t"
+ "balign %[tn2], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tn2] \n\t"
+ "preceu.ph.qbl %[p4], %[tn2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp3](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn1] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp4], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[tn1], %[Temp2](%[cm]) \n\t"
+ "lbux %[p2], %[Temp4](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[tp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[tn1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[tp2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[p2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), [tn1] "=&r" (tn1), [tn2] "=&r" (tn2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4),
+ [dst_ptr] "+r" (dst_ptr)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [src] "r" (src), [dst_stride] "r" (dst_stride)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+static void convolve_horiz_8_transposed_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint8_t *dst_ptr;
+ uint32_t vector4a = 64;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2, tp3;
+ uint32_t p1, p2, p3, p4, n1;
+ uint8_t *odd_dst;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+
+ dst_ptr = dst;
+ odd_dst = (dst_ptr + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp2], 0(%[src]) \n\t"
+ "ulw %[tp1], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tp1] \n\t"
+ "preceu.ph.qbl %[p4], %[tp1] \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "preceu.ph.qbr %[p1], %[tp3] \n\t"
+ "preceu.ph.qbl %[n1], %[tp3] \n\t"
+ "ulw %[tp2], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[tp2] \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector3b] \n\t"
+ "lbux %[tp3], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector4b] \n\t"
+ "extp %[p3], $ac1, 31 \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "sb %[Temp2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+ "sb %[tp3], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+
+ "ulw %[tp1], 1(%[src]) \n\t"
+ "ulw %[tp3], 5(%[src]) \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbux %[tp2], %[p3](%[cm]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp3] \n\t"
+ "preceu.ph.qbl %[p4], %[tp3] \n\t"
+ "sb %[tp2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+ "ulw %[tp2], 9(%[src]) \n\t"
+
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp1], %[Temp3](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[n1], %[tp2] \n\t"
+ "ulw %[Temp1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[vector1b] \n\t"
+ "sb %[tp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[tp3], %[Temp2](%[cm]) \n\t"
+ "preceu.ph.qbr %[p2], %[Temp1] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 4. pixel */
+ "sb %[tp3], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "lbux %[n1], %[Temp1](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[p4], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[p2], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[n1], 0(%[odd_dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), [tp3] "=&r" (tp3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst_ptr] "+r" (dst_ptr), [odd_dst] "+r" (odd_dst)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a), [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+static void convolve_horiz_16_transposed_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t c, y;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+ uint8_t *odd_dst;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+
+ src = src_ptr;
+ dst = dst_ptr;
+
+ odd_dst = (dst + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "ulw %[qload2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 2 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st3], 0(%[dst]) \n\t" /* even 3 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 4 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 5 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[st3], 0(%[dst]) \n\t" /* even 6 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 7 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 1 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 2 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 3 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 4 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 5 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 6 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 7 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [p5] "=&r" (p5),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst] "+r" (dst), [odd_dst] "+r" (odd_dst)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64), [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ src += 16;
+ dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
+ odd_dst = (dst + dst_stride);
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+
+ dst_ptr += 1;
+ }
+}
+
+static void convolve_horiz_64_transposed_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t c, y;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+ uint8_t *odd_dst;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+
+ src = src_ptr;
+ dst = dst_ptr;
+
+ odd_dst = (dst + dst_stride);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "ulw %[qload2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 2 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st3], 0(%[dst]) \n\t" /* even 3 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 4 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 5 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[st3], 0(%[dst]) \n\t" /* even 6 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 7 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 1 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 2 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 3 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 4 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 5 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 6 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 7 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [p5] "=&r" (p5),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst] "+r" (dst), [odd_dst] "+r" (odd_dst)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64), [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ src += 16;
+ dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
+ odd_dst = (dst + dst_stride);
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+
+ dst_ptr += 1;
+ }
+}
+
+void convolve_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter, int w, int h) {
+ int x, y, k;
+
+ for (y = 0; y < h; ++y) {
+ for (x = 0; x < w; ++x) {
+ int sum = 0;
+
+ for (k = 0; k < 8; ++k)
+ sum += src[x + k] * filter[k];
+
+ dst[x * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+ }
+
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+void vp9_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ DECLARE_ALIGNED_ARRAY(32, uint8_t, temp, 64 * 135);
+ int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ if (intermediate_height < h)
+ intermediate_height = h;
+
+ if (x_step_q4 != 16 || y_step_q4 != 16)
+ return vp9_convolve8_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+
+ if ((((const int32_t *)filter_x)[1] == 0x800000)
+ && (((const int32_t *)filter_y)[1] == 0x800000))
+ return vp9_convolve_copy(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+
+ /* copy the src to dst */
+ if (filter_x[3] == 0x80) {
+ int32_t y;
+ int32_t c;
+ const uint8_t *src_ptr = src - src_stride * 3;
+ uint8_t *dst_ptr = temp;
+
+ for (y = intermediate_height; y--;) {
+ for (c = 0; c < w; c++) {
+ dst_ptr[c * intermediate_height] = src_ptr[c];
+ }
+
+ /* next row... */
+ src_ptr += src_stride;
+ dst_ptr += 1;
+ }
+ } else {
+ src -= (src_stride * 3 + 3);
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+
+ switch (w) {
+ case 4:
+ convolve_horiz_4_transposed_dspr2(src, src_stride,
+ temp, intermediate_height,
+ filter_x, intermediate_height);
+ break;
+ case 8:
+ convolve_horiz_8_transposed_dspr2(src, src_stride,
+ temp, intermediate_height,
+ filter_x, intermediate_height);
+ break;
+ case 16:
+ case 32:
+ convolve_horiz_16_transposed_dspr2(src, src_stride,
+ temp, intermediate_height,
+ filter_x, intermediate_height,
+ (w/16));
+ break;
+ case 64:
+ vp9_prefetch_load(src + 32);
+ convolve_horiz_64_transposed_dspr2(src, src_stride,
+ temp, intermediate_height,
+ filter_x, intermediate_height);
+ break;
+ default:
+ convolve_horiz_transposed(src, src_stride,
+ temp, intermediate_height,
+ filter_x, w, intermediate_height);
+ break;
+ }
+ }
+
+ /* copy the src to dst */
+ if (filter_y[3] == 0x80) {
+ int32_t y;
+ int32_t c;
+ uint8_t *src_ptr = temp + 3;
+ uint8_t *dst_ptr = dst;
+
+ for (y = w; y--;) {
+ for (c = 0; c < h; c++) {
+ dst_ptr[c * dst_stride] = src_ptr[c];
+ }
+
+ /* next row... */
+ src_ptr += intermediate_height;
+ dst_ptr += 1;
+ }
+ } else {
+ switch (h) {
+ case 4:
+ convolve_horiz_4_transposed_dspr2(temp, intermediate_height,
+ dst, dst_stride,
+ filter_y, w);
+ break;
+ case 8:
+ convolve_horiz_8_transposed_dspr2(temp, intermediate_height,
+ dst, dst_stride,
+ filter_y, w);
+ break;
+ case 16:
+ case 32:
+ convolve_horiz_16_transposed_dspr2(temp, intermediate_height,
+ dst, dst_stride,
+ filter_y, w, (h/16));
+ break;
+ case 64:
+ convolve_horiz_64_transposed_dspr2(temp, intermediate_height,
+ dst, dst_stride,
+ filter_y, w);
+ break;
+ default:
+ convolve_horiz_transposed(temp, intermediate_height,
+ dst, dst_stride,
+ filter_y, h, w);
+ break;
+ }
+ }
+}
+
+void vp9_convolve_copy_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h) {
+ int x, y;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ {
+ uint32_t tp1;
+
+ /* 1 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], (%[src]) \n\t"
+ "sw %[tp1], (%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 8:
+ {
+ uint32_t tp1, tp2;
+
+ /* 2 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+ "sw %[tp1], 0(%[dst]) \n\t" /* store */
+ "sw %[tp2], 4(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 16:
+ {
+ uint32_t tp1, tp2, tp3, tp4;
+
+ /* 4 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "ulw %[tp4], 12(%[src]) \n\t"
+
+ "sw %[tp1], 0(%[dst]) \n\t" /* store */
+ "sw %[tp2], 4(%[dst]) \n\t" /* store */
+ "sw %[tp3], 8(%[dst]) \n\t" /* store */
+ "sw %[tp4], 12(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 32:
+ {
+ uint32_t tp1, tp2, tp3, tp4;
+ uint32_t tp5, tp6, tp7, tp8;
+
+ /* 8 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "ulw %[tp4], 12(%[src]) \n\t"
+ "ulw %[tp5], 16(%[src]) \n\t"
+ "ulw %[tp6], 20(%[src]) \n\t"
+ "ulw %[tp7], 24(%[src]) \n\t"
+ "ulw %[tp8], 28(%[src]) \n\t"
+
+ "sw %[tp1], 0(%[dst]) \n\t" /* store */
+ "sw %[tp2], 4(%[dst]) \n\t" /* store */
+ "sw %[tp3], 8(%[dst]) \n\t" /* store */
+ "sw %[tp4], 12(%[dst]) \n\t" /* store */
+ "sw %[tp5], 16(%[dst]) \n\t" /* store */
+ "sw %[tp6], 20(%[dst]) \n\t" /* store */
+ "sw %[tp7], 24(%[dst]) \n\t" /* store */
+ "sw %[tp8], 28(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tp5] "=&r" (tp5), [tp6] "=&r" (tp6),
+ [tp7] "=&r" (tp7), [tp8] "=&r" (tp8)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ break;
+ case 64:
+ {
+ uint32_t tp1, tp2, tp3, tp4;
+ uint32_t tp5, tp6, tp7, tp8;
+
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ /* 16 word storage */
+ for (y = h; y--; ) {
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_load(src + src_stride + 64);
+ vp9_prefetch_store(dst + dst_stride);
+ vp9_prefetch_store(dst + dst_stride + 32);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "ulw %[tp4], 12(%[src]) \n\t"
+ "ulw %[tp5], 16(%[src]) \n\t"
+ "ulw %[tp6], 20(%[src]) \n\t"
+ "ulw %[tp7], 24(%[src]) \n\t"
+ "ulw %[tp8], 28(%[src]) \n\t"
+
+ "sw %[tp1], 0(%[dst]) \n\t" /* store */
+ "sw %[tp2], 4(%[dst]) \n\t" /* store */
+ "sw %[tp3], 8(%[dst]) \n\t" /* store */
+ "sw %[tp4], 12(%[dst]) \n\t" /* store */
+ "sw %[tp5], 16(%[dst]) \n\t" /* store */
+ "sw %[tp6], 20(%[dst]) \n\t" /* store */
+ "sw %[tp7], 24(%[dst]) \n\t" /* store */
+ "sw %[tp8], 28(%[dst]) \n\t" /* store */
+
+ "ulw %[tp1], 32(%[src]) \n\t"
+ "ulw %[tp2], 36(%[src]) \n\t"
+ "ulw %[tp3], 40(%[src]) \n\t"
+ "ulw %[tp4], 44(%[src]) \n\t"
+ "ulw %[tp5], 48(%[src]) \n\t"
+ "ulw %[tp6], 52(%[src]) \n\t"
+ "ulw %[tp7], 56(%[src]) \n\t"
+ "ulw %[tp8], 60(%[src]) \n\t"
+
+ "sw %[tp1], 32(%[dst]) \n\t" /* store */
+ "sw %[tp2], 36(%[dst]) \n\t" /* store */
+ "sw %[tp3], 40(%[dst]) \n\t" /* store */
+ "sw %[tp4], 44(%[dst]) \n\t" /* store */
+ "sw %[tp5], 48(%[dst]) \n\t" /* store */
+ "sw %[tp6], 52(%[dst]) \n\t" /* store */
+ "sw %[tp7], 56(%[dst]) \n\t" /* store */
+ "sw %[tp8], 60(%[dst]) \n\t" /* store */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [tp5] "=&r" (tp5), [tp6] "=&r" (tp6),
+ [tp7] "=&r" (tp7), [tp8] "=&r" (tp8)
+ : [src] "r" (src), [dst] "r" (dst)
+ );
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ break;
+ default:
+ for (y = h; y--; ) {
+ for (x = 0; x < w; ++x) {
+ dst[x] = src[x];
+ }
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+}
+#endif
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_horiz_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3, Temp4;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3, p4;
+ uint32_t n1, n2, n3, n4;
+ uint32_t tn1, tn2;
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "ulw %[tn2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn2] \n\t"
+ "balign %[tn1], %[tn2], 3 \n\t"
+ "balign %[tn2], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[n1], %[tp2] \n\t"
+ "preceu.ph.qbl %[n2], %[tp2] \n\t"
+ "preceu.ph.qbr %[n3], %[tn2] \n\t"
+ "preceu.ph.qbl %[n4], %[tn2] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp3](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[n1], %[tn1] \n\t"
+ "dpa.w.ph $ac2, %[n2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[n3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector4b] \n\t"
+ "extp %[Temp4], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[tn1], %[Temp2](%[cm]) \n\t"
+ "lbux %[n2], %[Temp4](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[tp1], 0(%[dst]) \n\t"
+ "sb %[tn1], 1(%[dst]) \n\t"
+ "sb %[tp2], 2(%[dst]) \n\t"
+ "sb %[n2], 3(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1), [n2] "=&r" (n2), [n3] "=&r" (n3), [n4] "=&r" (n4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_horiz_8_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3, p4, n1;
+ uint32_t tn1, tn2, tn3;
+ uint32_t st0, st1;
+
+ vector1b = ((const int32_t *)filter_x0)[0];
+ vector2b = ((const int32_t *)filter_x0)[1];
+ vector3b = ((const int32_t *)filter_x0)[2];
+ vector4b = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "ulw %[tn2], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "preceu.ph.qbr %[p1], %[tn2] \n\t"
+ "preceu.ph.qbl %[n1], %[tn2] \n\t"
+ "ulw %[tn1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[tn1] \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector4b] \n\t"
+ "extp %[Temp1], $ac1, 31 \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st0], 0(%[dst]) \n\t"
+ "lbux %[st1], %[Temp3](%[cm]) \n\t"
+
+ "balign %[tn3], %[tn1], 3 \n\t"
+ "balign %[tn1], %[tn2], 3 \n\t"
+ "balign %[tn2], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 2(%[dst]) \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tn2] \n\t"
+ "preceu.ph.qbl %[p4], %[tn2] \n\t"
+ "sb %[st0], 4(%[dst]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tn1] \n\t"
+ "preceu.ph.qbl %[n1], %[tn1] \n\t"
+ "lbux %[st0], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[p3], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[vector4b] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[st1], %[Temp2](%[cm]) \n\t"
+ "preceu.ph.qbr %[p2], %[tn3] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 4. pixel */
+ "sb %[st1], 1(%[dst]) \n\t"
+ "sb %[st0], 6(%[dst]) \n\t"
+ "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
+ "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "lbux %[n1], %[Temp1](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[p4], 3(%[dst]) \n\t"
+ "sb %[p2], 5(%[dst]) \n\t"
+ "sb %[n1], 7(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2), [tn3] "=&r" (tn3),
+ [st0] "=&r" (st0), [st1] "=&r" (st1),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_horiz_16_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 2(%[dst]) \n\t" /* even 1 */
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st3], 4(%[dst]) \n\t" /* even 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 6(%[dst]) \n\t" /* even 4 */
+ "ulw %[qload3], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[st2], 8(%[dst]) \n\t" /* even 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[st3], 10(%[dst]) \n\t" /* even 6 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 12(%[dst]) \n\t" /* even 7 */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[st2], 14(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */
+ "ulw %[qload3], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */
+ "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */
+ "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst),
+ [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+static void convolve_horiz_64_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t filter12, filter34, filter56, filter78;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+
+ filter12 = ((const int32_t *)filter_x0)[0];
+ filter34 = ((const int32_t *)filter_x0)[1];
+ filter56 = ((const int32_t *)filter_x0)[2];
+ filter78 = ((const int32_t *)filter_x0)[3];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+ vp9_prefetch_store(dst_ptr + dst_stride + 32);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
+ "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
+ "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 2(%[dst]) \n\t" /* even 1 */
+ "ulw %[qload2], 16(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
+ "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st3], 4(%[dst]) \n\t" /* even 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
+ "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st1], 6(%[dst]) \n\t" /* even 4 */
+ "ulw %[qload3], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[st2], 8(%[dst]) \n\t" /* even 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
+ "sb %[st3], 10(%[dst]) \n\t" /* even 6 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 12(%[dst]) \n\t" /* even 7 */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
+ "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[st2], 14(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
+ "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
+ "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */
+ "ulw %[qload2], 17(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
+ "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p4], %[qload2] \n\t"
+ "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */
+ "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
+ "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbl %[p1], %[qload2] \n\t"
+ "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */
+ "ulw %[qload3], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
+ "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload3] \n\t"
+ "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */
+ "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
+ "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
+ "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */
+ "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */
+ "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter12] "r" (filter12), [filter34] "r" (filter34),
+ [filter56] "r" (filter56), [filter78] "r" (filter78),
+ [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst),
+ [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+void vp9_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ if (((const int32_t *)filter_x)[1] == 0x800000) {
+ vp9_convolve_copy(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ } else {
+ if (16 == x_step_q4) {
+ uint32_t pos = 38;
+
+ vp9_prefetch_load((const uint8_t *)filter_x);
+ src -= 3;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ convolve_horiz_4_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 8:
+ convolve_horiz_8_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 16:
+ convolve_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 1);
+ break;
+ case 32:
+ convolve_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 2);
+ break;
+ case 64:
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ convolve_horiz_64_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ default:
+ vp9_convolve8_horiz_c(src + 3, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+ }
+}
+#endif
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_vert_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t w,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2, load3, load4;
+ uint32_t p1, p2;
+ uint32_t n1, n2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2;
+
+ vector1b = ((const int32_t *)filter_y)[0];
+ vector2b = ((const int32_t *)filter_y)[1];
+ vector3b = ((const int32_t *)filter_y)[2];
+ vector4b = ((const int32_t *)filter_y)[3];
+
+ src -= 3 * src_stride;
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+
+ for (x = 0; x < w; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [load3] "=&r" (load3), [load4] "=&r" (load4),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [n1] "=&r" (n1), [n2] "=&r" (n2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a), [src_stride] "r" (src_stride),
+ [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_vert_64_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2, load3, load4;
+ uint32_t p1, p2;
+ uint32_t n1, n2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t vector1b, vector2b, vector3b, vector4b;
+ int32_t Temp1, Temp2;
+
+ vector1b = ((const int32_t *)filter_y)[0];
+ vector2b = ((const int32_t *)filter_y)[1];
+ vector3b = ((const int32_t *)filter_y)[2];
+ vector4b = ((const int32_t *)filter_y)[3];
+
+ src -= 3 * src_stride;
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+ vp9_prefetch_store(dst + dst_stride + 32);
+
+ for (x = 0; x < 64; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
+
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load3], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load4], 0(%[src_ptr]) \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbr %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbr %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+ "preceu.ph.qbl %[scratch2], %[load3] \n\t"
+ "preceu.ph.qbl %[p2], %[load4] \n\t"
+ "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
+ "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
+ "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
+ "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [load3] "=&r" (load3), [load4] "=&r" (load4),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [n1] "=&r" (n1), [n2] "=&r" (n2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
+ [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
+ [vector4a] "r" (vector4a), [src_stride] "r" (src_stride),
+ [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vp9_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {
+ if (((const int32_t *)filter_y)[1] == 0x800000) {
+ vp9_convolve_copy(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ } else {
+ if (16 == y_step_q4) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4 :
+ case 8 :
+ case 16 :
+ case 32 :
+ convolve_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64 :
+ vp9_prefetch_store(dst + 32);
+ convolve_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, h);
+ break;
+ default:
+ vp9_convolve8_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+ }
+}
+
+#endif
}
static void set_mb_mi(VP9_COMMON *cm, int aligned_width, int aligned_height) {
- cm->mb_cols = (aligned_width + 8) >> 4;
- cm->mb_rows = (aligned_height + 8) >> 4;
- cm->MBs = cm->mb_rows * cm->mb_cols;
-
cm->mi_cols = aligned_width >> MI_SIZE_LOG2;
cm->mi_rows = aligned_height >> MI_SIZE_LOG2;
cm->mode_info_stride = cm->mi_cols + MI_BLOCK_SIZE;
+
+ cm->mb_cols = (cm->mi_cols + 1) >> 1;
+ cm->mb_rows = (cm->mi_rows + 1) >> 1;
+ cm->MBs = cm->mb_rows * cm->mb_cols;
}
static void setup_mi(VP9_COMMON *cm) {
void vp9_create_common(VP9_COMMON *cm) {
vp9_machine_specific_config(cm);
- vp9_init_mbmode_probs(cm);
-
cm->tx_mode = ONLY_4X4;
cm->comp_pred_mode = HYBRID_PREDICTION;
-
- // Initialize reference frame sign bias structure to defaults
- vpx_memset(cm->ref_frame_sign_bias, 0, sizeof(cm->ref_frame_sign_bias));
}
void vp9_remove_common(VP9_COMMON *cm) {
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_common_data.h"
#include "vp9/common/vp9_enums.h"
+#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_mv.h"
#include "vp9/common/vp9_scale.h"
#include "vp9/common/vp9_seg_common.h"
NUM_FRAME_TYPES,
} FRAME_TYPE;
-typedef enum {
- EIGHTTAP = 0,
- EIGHTTAP_SMOOTH = 1,
- EIGHTTAP_SHARP = 2,
- BILINEAR = 3,
- SWITCHABLE = 4 /* should be the last one */
-} INTERPOLATIONFILTERTYPE;
-
typedef enum {
DC_PRED, // Average of above and left pixels
V_PRED, // Vertical
modes for the Y blocks to the left and above us; for interframes, there
is a single probability table. */
-union b_mode_info {
+typedef struct {
MB_PREDICTION_MODE as_mode;
int_mv as_mv[2]; // first, second inter predictor motion vectors
-};
+} b_mode_info;
typedef enum {
NONE = -1,
typedef struct {
MB_MODE_INFO mbmi;
- union b_mode_info bmi[4];
+ b_mode_info bmi[4];
} MODE_INFO;
static INLINE int is_inter_block(const MB_MODE_INFO *mbmi) {
unsigned char ab_index; // index of 4x4 block inside the 8x8 block
int q_index;
-
} MACROBLOCKD;
static INLINE uint8_t *get_sb_index(MACROBLOCKD *xd, BLOCK_SIZE subsize) {
extern const TX_SIZE max_uv_txsize_lookup[BLOCK_SIZES];
extern const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2];
-#endif // VP9_COMMON_VP9_COMMON_DATA_H
+#endif // VP9_COMMON_VP9_COMMON_DATA_H
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
-#ifndef VP9_COMMON_CONVOLVE_H_
-#define VP9_COMMON_CONVOLVE_H_
+#ifndef VP9_COMMON_VP9_CONVOLVE_H_
+#define VP9_COMMON_VP9_CONVOLVE_H_
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
const int16_t *filter_y, int y_step_q4,
int w, int h);
-struct subpix_fn_table {
- const int16_t (*filter_x)[8];
- const int16_t (*filter_y)[8];
-};
-
-#endif // VP9_COMMON_CONVOLVE_H_
+#endif // VP9_COMMON_VP9_CONVOLVE_H_
print_mi_data(cm, mvs, "Transform:", offsetof(MB_MODE_INFO, tx_size));
print_mi_data(cm, mvs, "UV Modes:", offsetof(MB_MODE_INFO, uv_mode));
- log_frame_info(cm, "Vectors ",mvs);
+ log_frame_info(cm, "Vectors ", mvs);
for (mi_row = 0; mi_row < rows; mi_row++) {
- fprintf(mvs,"V ");
+ fprintf(mvs, "V ");
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(mvs, "%4d:%4d ", mi_8x8[mi_index]->mbmi.mv[0].as_mv.row,
mi_8x8[mi_index]->mbmi.mv[0].as_mv.col);
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
+#ifndef VP9_COMMON_DEFAULT_COEF_PROBS_H_
+#define VP9_COMMON_DEFAULT_COEF_PROBS_H_
/*Generated file, included by vp9_entropy.c*/
static const vp9_coeff_probs_model default_coef_probs_4x4[BLOCK_TYPES] = {
}
};
+#endif // VP9_COMMON_DEFAULT_COEF_PROBS_H_
};
DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_16x16[256]) = {
- 0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80,
- 50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52,
- 98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69,
- 100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146,
- 55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25,
- 133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119,
- 26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194,
- 180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59,
- 12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13,
- 226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169,
- 242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108,
- 77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140,
- 230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141,
- 63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142,
- 219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159, 251,
- 190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239, 255,
+ 0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80,
+ 50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52,
+ 98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69,
+ 100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146,
+ 55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25,
+ 133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119,
+ 26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194,
+ 180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59,
+ 12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13,
+ 226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169,
+ 242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108,
+ 77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140,
+ 230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141,
+ 63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142,
+ 219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159,
+ 251,
+ 190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239,
+ 255,
};
DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_16x16[256]) = {
- 0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
- 34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4,
- 67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21,
- 146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85,
- 22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179,
- 225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24,
- 87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227,
- 88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167,
- 213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229,
- 74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59,
- 200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170,
- 60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202,
- 233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125,
- 62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79,
- 126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205, 236,
- 159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239, 255,
+ 0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
+ 34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4,
+ 67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21,
+ 146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85,
+ 22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179,
+ 225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24,
+ 87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227,
+ 88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167,
+ 213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229,
+ 74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59,
+ 200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170,
+ 60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202,
+ 233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125,
+ 62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79,
+ 126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205,
+ 236,
+ 159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239,
+ 255,
};
DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_16x16[256]) = {
- 0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20,
- 49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52,
- 23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69,
- 25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100,
- 13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102,
- 144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160,
- 89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176,
- 75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136,
- 165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166,
- 167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108,
- 197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170,
- 124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186,
- 156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110,
- 157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111, 158,
- 188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220, 175,
- 190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254, 255,
+ 0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20,
+ 49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52,
+ 23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69,
+ 25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100,
+ 13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102,
+ 144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160,
+ 89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176,
+ 75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136,
+ 165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166,
+ 167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108,
+ 197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170,
+ 124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186,
+ 156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110,
+ 157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111,
+ 158,
+ 188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220,
+ 175,
+ 190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254,
+ 255,
};
DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_32x32[1024]) = {
- 0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160, 129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193, 68, 131, 37, 100,
- 225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38, 258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321, 102, 352, 8, 197,
- 71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292, 135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293, 41, 417, 199, 136,
- 262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105, 419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169, 295, 420, 106, 451,
- 481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421, 75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391, 453, 139, 44, 234,
- 484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108, 546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577, 486, 77, 204, 362,
- 608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173, 610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17, 111, 238, 48, 143,
- 80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51, 83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424, 393, 300, 269, 176, 145,
- 52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301, 270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581, 550, 519, 488, 457, 426, 395,
- 364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737, 706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241, 210, 179, 117, 86, 55, 738, 707,
- 614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491, 367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676, 645, 552, 521, 428, 397, 304,
- 273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553, 522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26, 864, 833, 802, 771, 740, 709,
- 678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306, 275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741, 710, 679, 617, 586, 555, 493,
- 462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835, 742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867, 743, 619, 495, 371, 247, 123,
- 896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680, 649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929, 898, 836, 805, 774, 712, 681,
- 650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154, 92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682, 651, 620, 589, 558, 527,
- 496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124, 93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590, 559, 497, 466, 435, 373,
- 342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715, 622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623, 499, 375, 251, 127,
- 900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560, 529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716, 685, 654, 592, 561,
- 530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903, 872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469, 438, 407, 376, 345,
- 314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718, 687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998, 967, 874, 843, 750,
- 719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503, 379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657, 564, 533, 440, 409,
- 316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534, 472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783, 752, 721, 690, 659,
- 628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970, 939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381, 350, 319, 1002, 971,
- 878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631, 507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568, 537, 444, 413, 972,
- 941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414, 1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601, 570, 539, 508, 477,
- 446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571, 509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479, 1007, 883, 759, 635, 511,
- 912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945, 914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915, 884, 853, 822, 791,
- 760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823, 761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607, 1011, 887, 763, 639,
- 916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825, 794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733, 702, 671, 1013, 982,
- 951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015, 891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798, 1016, 985, 954, 923,
- 892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863, 1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021, 990, 959, 1022, 991, 1023,
+ 0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160,
+ 129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193,
+ 68, 131, 37, 100,
+ 225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38,
+ 258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321,
+ 102, 352, 8, 197,
+ 71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292,
+ 135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293,
+ 41, 417, 199, 136,
+ 262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105,
+ 419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169,
+ 295, 420, 106, 451,
+ 481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421,
+ 75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391,
+ 453, 139, 44, 234,
+ 484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108,
+ 546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577,
+ 486, 77, 204, 362,
+ 608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173,
+ 610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17,
+ 111, 238, 48, 143,
+ 80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51,
+ 83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424,
+ 393, 300, 269, 176, 145,
+ 52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301,
+ 270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581,
+ 550, 519, 488, 457, 426, 395,
+ 364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737,
+ 706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241,
+ 210, 179, 117, 86, 55, 738, 707,
+ 614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491,
+ 367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676,
+ 645, 552, 521, 428, 397, 304,
+ 273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553,
+ 522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26,
+ 864, 833, 802, 771, 740, 709,
+ 678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306,
+ 275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741,
+ 710, 679, 617, 586, 555, 493,
+ 462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835,
+ 742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867,
+ 743, 619, 495, 371, 247, 123,
+ 896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680,
+ 649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929,
+ 898, 836, 805, 774, 712, 681,
+ 650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154,
+ 92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682,
+ 651, 620, 589, 558, 527,
+ 496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124,
+ 93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590,
+ 559, 497, 466, 435, 373,
+ 342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715,
+ 622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623,
+ 499, 375, 251, 127,
+ 900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560,
+ 529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716,
+ 685, 654, 592, 561,
+ 530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903,
+ 872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469,
+ 438, 407, 376, 345,
+ 314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718,
+ 687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998,
+ 967, 874, 843, 750,
+ 719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503,
+ 379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657,
+ 564, 533, 440, 409,
+ 316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534,
+ 472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783,
+ 752, 721, 690, 659,
+ 628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970,
+ 939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381,
+ 350, 319, 1002, 971,
+ 878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631,
+ 507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568,
+ 537, 444, 413, 972,
+ 941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414,
+ 1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601,
+ 570, 539, 508, 477,
+ 446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571,
+ 509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479,
+ 1007, 883, 759, 635, 511,
+ 912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945,
+ 914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915,
+ 884, 853, 822, 791,
+ 760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823,
+ 761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607,
+ 1011, 887, 763, 639,
+ 916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825,
+ 794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733,
+ 702, 671, 1013, 982,
+ 951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015,
+ 891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798,
+ 1016, 985, 954, 923,
+ 892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863,
+ 1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021,
+ 990, 959, 1022, 991, 1023,
};
/* Array indices are identical to previously-existing CONTEXT_NODE indices */
-const vp9_tree_index vp9_coef_tree[ 22] = /* corresponding _CONTEXT_NODEs */
-{
+const vp9_tree_index vp9_coef_tree[ 22] = {
-DCT_EOB_TOKEN, 2, /* 0 = EOB */
-ZERO_TOKEN, 4, /* 1 = ZERO */
-ONE_TOKEN, 6, /* 2 = ONE */
vp9_default_scan_32x32_neighbors);
}
-const int16_t *vp9_get_coef_neighbors_handle(const int16_t *scan) {
- if (scan == vp9_default_scan_4x4) {
- return vp9_default_scan_4x4_neighbors;
- } else if (scan == vp9_row_scan_4x4) {
- return vp9_row_scan_4x4_neighbors;
- } else if (scan == vp9_col_scan_4x4) {
- return vp9_col_scan_4x4_neighbors;
- } else if (scan == vp9_default_scan_8x8) {
- return vp9_default_scan_8x8_neighbors;
- } else if (scan == vp9_row_scan_8x8) {
- return vp9_row_scan_8x8_neighbors;
- } else if (scan == vp9_col_scan_8x8) {
- return vp9_col_scan_8x8_neighbors;
- } else if (scan == vp9_default_scan_16x16) {
- return vp9_default_scan_16x16_neighbors;
- } else if (scan == vp9_row_scan_16x16) {
- return vp9_row_scan_16x16_neighbors;
- } else if (scan == vp9_col_scan_16x16) {
- return vp9_col_scan_16x16_neighbors;
- } else {
- assert(scan == vp9_default_scan_32x32);
- return vp9_default_scan_32x32_neighbors;
- }
-}
-
void vp9_coef_tree_initialize() {
vp9_init_neighbors();
init_bit_trees();
token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >> 1;
}
-const int16_t *vp9_get_coef_neighbors_handle(const int16_t *scan);
-
-
// 128 lists of probabilities are stored for the following ONE node probs:
// 1, 3, 5, 7, ..., 253, 255
// In between probabilities are interpolated linearly
typedef unsigned int vp9_coeff_count_model[REF_TYPES][COEF_BANDS]
[PREV_COEF_CONTEXTS]
[UNCONSTRAINED_NODES + 1];
-typedef unsigned int vp9_coeff_stats_model[REF_TYPES][COEF_BANDS]
- [PREV_COEF_CONTEXTS]
- [UNCONSTRAINED_NODES][2];
void vp9_model_to_full_probs(const vp9_prob *model, vp9_prob *full);
static void get_scan_and_band(const MACROBLOCKD *xd, TX_SIZE tx_size,
PLANE_TYPE type, int block_idx,
const int16_t **scan,
+ const int16_t **scan_nb,
const uint8_t **band_translate) {
switch (tx_size) {
case TX_4X4:
- *scan = get_scan_4x4(get_tx_type_4x4(type, xd, block_idx));
+ get_scan_nb_4x4(get_tx_type_4x4(type, xd, block_idx), scan, scan_nb);
*band_translate = vp9_coefband_trans_4x4;
break;
case TX_8X8:
- *scan = get_scan_8x8(get_tx_type_8x8(type, xd));
+ get_scan_nb_8x8(get_tx_type_8x8(type, xd), scan, scan_nb);
*band_translate = vp9_coefband_trans_8x8plus;
break;
case TX_16X16:
- *scan = get_scan_16x16(get_tx_type_16x16(type, xd));
+ get_scan_nb_16x16(get_tx_type_16x16(type, xd), scan, scan_nb);
*band_translate = vp9_coefband_trans_8x8plus;
break;
case TX_32X32:
*scan = vp9_default_scan_32x32;
+ *scan_nb = vp9_default_scan_32x32_neighbors;
*band_translate = vp9_coefband_trans_8x8plus;
break;
default:
{ 66 } }
};
-void tx_counts_to_branch_counts_32x32(unsigned int *tx_count_32x32p,
+void tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p,
unsigned int (*ct_32x32p)[2]) {
ct_32x32p[0][0] = tx_count_32x32p[TX_4X4];
ct_32x32p[0][1] = tx_count_32x32p[TX_8X8] +
ct_32x32p[2][1] = tx_count_32x32p[TX_32X32];
}
-void tx_counts_to_branch_counts_16x16(unsigned int *tx_count_16x16p,
+void tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p,
unsigned int (*ct_16x16p)[2]) {
ct_16x16p[0][0] = tx_count_16x16p[TX_4X4];
ct_16x16p[0][1] = tx_count_16x16p[TX_8X8] + tx_count_16x16p[TX_16X16];
ct_16x16p[1][1] = tx_count_16x16p[TX_16X16];
}
-void tx_counts_to_branch_counts_8x8(unsigned int *tx_count_8x8p,
+void tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p,
unsigned int (*ct_8x8p)[2]) {
ct_8x8p[0][0] = tx_count_8x8p[TX_4X4];
ct_8x8p[0][1] = tx_count_8x8p[TX_8X8];
#define COUNT_SAT 20
#define MAX_UPDATE_FACTOR 128
-static int update_ct(vp9_prob pre_prob, vp9_prob prob, unsigned int ct[2]) {
+static int update_ct(vp9_prob pre_prob, vp9_prob prob,
+ const unsigned int ct[2]) {
return merge_probs(pre_prob, prob, ct, COUNT_SAT, MAX_UPDATE_FACTOR);
}
-static int update_ct2(vp9_prob pre_prob, unsigned int ct[2]) {
+static int update_ct2(vp9_prob pre_prob, const unsigned int ct[2]) {
return merge_probs2(pre_prob, ct, COUNT_SAT, MAX_UPDATE_FACTOR);
}
static void update_mode_probs(int n_modes,
- const vp9_tree_index *tree, unsigned int *cnt,
- vp9_prob *pre_probs, vp9_prob *dst_probs,
+ const vp9_tree_index *tree,
+ const unsigned int *cnt,
+ const vp9_prob *pre_probs, vp9_prob *dst_probs,
unsigned int tok0_offset) {
#define MAX_PROBS 32
vp9_prob probs[MAX_PROBS];
void vp9_adapt_mode_probs(VP9_COMMON *cm) {
int i, j;
FRAME_CONTEXT *fc = &cm->fc;
- FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
- FRAME_COUNTS *counts = &cm->counts;
+ const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
+ const FRAME_COUNTS *counts = &cm->counts;
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
fc->intra_inter_prob[i] = update_ct2(pre_fc->intra_inter_prob[i],
void vp9_adapt_mode_probs(struct VP9Common *cm);
-void tx_counts_to_branch_counts_32x32(unsigned int *tx_count_32x32p,
+void tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p,
unsigned int (*ct_32x32p)[2]);
-void tx_counts_to_branch_counts_16x16(unsigned int *tx_count_16x16p,
+void tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p,
unsigned int (*ct_16x16p)[2]);
-void tx_counts_to_branch_counts_8x8(unsigned int *tx_count_8x8p,
+void tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p,
unsigned int (*ct_8x8p)[2]);
#endif // VP9_COMMON_VP9_ENTROPYMODE_H_
};
struct vp9_token vp9_mv_class_encodings[MV_CLASSES];
-const vp9_tree_index vp9_mv_class0_tree [2 * CLASS0_SIZE - 2] = {
+const vp9_tree_index vp9_mv_class0_tree[2 * CLASS0_SIZE - 2] = {
-0, -1,
};
struct vp9_token vp9_mv_class0_encodings[CLASS0_SIZE];
-const vp9_tree_index vp9_mv_fp_tree [2 * 4 - 2] = {
+const vp9_tree_index vp9_mv_fp_tree[2 * 4 - 2] = {
-0, 2,
-1, 4,
-2, -3
static const nmv_context default_nmv_context = {
{32, 64, 96},
- {
- { /* vert component */
+ { // NOLINT
+ { /* vert component */ // NOLINT
128, /* sign */
{224, 144, 192, 168, 192, 176, 192, 198, 198, 245}, /* class */
{216}, /* class0 */
160, /* class0_hp bit */
128, /* hp */
},
- { /* hor component */
+ { /* hor component */ // NOLINT
128, /* sign */
{216, 128, 176, 160, 176, 176, 192, 198, 198, 208}, /* class */
{208}, /* class0 */
static void inc_mv_component(int v, nmv_component_counts *comp_counts,
int incr, int usehp) {
int s, z, c, o, d, e, f;
- assert (v != 0); /* should not be zero */
+ assert(v != 0); /* should not be zero */
s = v < 0;
comp_counts->sign[s] += incr;
z = (s ? -v : v) - 1; /* magnitude - 1 */
vp9_prob this_probs[],
const vp9_prob last_probs[],
const unsigned int num_events[]) {
-
-
const unsigned int left = tree[i] <= 0
? num_events[-tree[i]]
: adapt_probs(tree[i], tree, this_probs, last_probs, num_events);
#define VP9_COMMON_VP9_ENTROPYMV_H_
#include "vp9/common/vp9_treecoder.h"
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/common/vp9_blockd.h"
struct VP9Common;
#define MV_MAX ((1 << MV_MAX_BITS) - 1)
#define MV_VALS ((MV_MAX << 1) + 1)
+#define MV_IN_USE_BITS 14
+#define MV_UPP ((1 << MV_IN_USE_BITS) - 1)
+#define MV_LOW (-(1 << MV_IN_USE_BITS))
+
extern const vp9_tree_index vp9_mv_class0_tree[2 * CLASS0_SIZE - 2];
extern struct vp9_token vp9_mv_class0_encodings[CLASS0_SIZE];
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
+
#include "vpx_ports/mem.h"
#include "vp9/common/vp9_filter.h"
-DECLARE_ALIGNED(256, const int16_t,
- vp9_bilinear_filters[SUBPEL_SHIFTS][SUBPEL_TAPS]) = {
+DECLARE_ALIGNED(256, const subpel_kernel,
+ vp9_bilinear_filters[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0 },
{ 0, 0, 0, 120, 8, 0, 0, 0 },
{ 0, 0, 0, 112, 16, 0, 0, 0 },
};
// Lagrangian interpolation filter
-DECLARE_ALIGNED(256, const int16_t,
- vp9_sub_pel_filters_8[SUBPEL_SHIFTS][SUBPEL_TAPS]) = {
+DECLARE_ALIGNED(256, const subpel_kernel,
+ vp9_sub_pel_filters_8[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{ 0, 1, -5, 126, 8, -3, 1, 0},
{ -1, 3, -10, 122, 18, -6, 2, 0},
};
// DCT based filter
-DECLARE_ALIGNED(256, const int16_t,
- vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][SUBPEL_TAPS]) = {
+DECLARE_ALIGNED(256, const subpel_kernel,
+ vp9_sub_pel_filters_8s[SUBPEL_SHIFTS]) = {
{0, 0, 0, 128, 0, 0, 0, 0},
{-1, 3, -7, 127, 8, -3, 1, 0},
{-2, 5, -13, 125, 17, -6, 3, -1},
};
// freqmultiplier = 0.5
-DECLARE_ALIGNED(256, const int16_t,
- vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][SUBPEL_TAPS]) = {
+DECLARE_ALIGNED(256, const subpel_kernel,
+ vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{-3, -1, 32, 64, 38, 1, -3, 0},
{-2, -2, 29, 63, 41, 2, -3, 0},
{ 0, -3, 2, 41, 63, 29, -2, -2},
{ 0, -3, 1, 38, 64, 32, -1, -3}
};
+
+const subpel_kernel *vp9_get_filter_kernel(INTERPOLATIONFILTERTYPE type) {
+ switch (type) {
+ case EIGHTTAP:
+ return vp9_sub_pel_filters_8;
+ case EIGHTTAP_SMOOTH:
+ return vp9_sub_pel_filters_8lp;
+ case EIGHTTAP_SHARP:
+ return vp9_sub_pel_filters_8s;
+ case BILINEAR:
+ return vp9_bilinear_filters;
+ default:
+ assert(!"Invalid filter type.");
+ return NULL;
+ }
+}
+
#ifndef VP9_COMMON_VP9_FILTER_H_
#define VP9_COMMON_VP9_FILTER_H_
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#define SUBPEL_BITS 4
#define SUBPEL_SHIFTS (1 << SUBPEL_BITS)
#define SUBPEL_TAPS 8
-extern const int16_t vp9_bilinear_filters[SUBPEL_SHIFTS][SUBPEL_TAPS];
-extern const int16_t vp9_sub_pel_filters_6[SUBPEL_SHIFTS][SUBPEL_TAPS];
-extern const int16_t vp9_sub_pel_filters_8[SUBPEL_SHIFTS][SUBPEL_TAPS];
-extern const int16_t vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][SUBPEL_TAPS];
-extern const int16_t vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][SUBPEL_TAPS];
+typedef enum {
+ EIGHTTAP = 0,
+ EIGHTTAP_SMOOTH = 1,
+ EIGHTTAP_SHARP = 2,
+ BILINEAR = 3,
+ SWITCHABLE = 4 /* should be the last one */
+} INTERPOLATIONFILTERTYPE;
+
+typedef const int16_t subpel_kernel[SUBPEL_TAPS];
+
+struct subpix_fn_table {
+ const subpel_kernel *filter_x;
+ const subpel_kernel *filter_y;
+};
+
+const subpel_kernel *vp9_get_filter_kernel(INTERPOLATIONFILTERTYPE type);
+
+extern const subpel_kernel vp9_bilinear_filters[SUBPEL_SHIFTS];
+extern const subpel_kernel vp9_sub_pel_filters_6[SUBPEL_SHIFTS];
+extern const subpel_kernel vp9_sub_pel_filters_8[SUBPEL_SHIFTS];
+extern const subpel_kernel vp9_sub_pel_filters_8s[SUBPEL_SHIFTS];
+extern const subpel_kernel vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS];
// The VP9_BILINEAR_FILTERS_2TAP macro returns a pointer to the bilinear
// filter kernel as a 2 tap filter.
vpx_memcpy(dst_list, mv_list, MAX_MV_REF_CANDIDATES * sizeof(int_mv));
} else if (block_idx == 1 || block_idx == 2) {
int dst = 0, n;
- union b_mode_info *bmi = mi->bmi;
+ b_mode_info *bmi = mi->bmi;
dst_list[dst++].as_int = bmi[0].as_mv[ref_idx].as_int;
for (n = 0; dst < MAX_MV_REF_CANDIDATES &&
dst_list[dst++].as_int = mv_list[n].as_int;
} else {
int dst = 0, n;
- union b_mode_info *bmi = mi->bmi;
+ b_mode_info *bmi = mi->bmi;
assert(block_idx == 3);
dst_list[dst++].as_int = bmi[2].as_mv[ref_idx].as_int;
if (!mi)
return DC_PRED;
- if (mi->mbmi.ref_frame[0] != INTRA_FRAME) {
+ if (is_inter_block(&mi->mbmi))
return DC_PRED;
- } else if (mi->mbmi.sb_type < BLOCK_8X8) {
- return ((mi->bmi + 1 + b)->as_mode);
- } else {
- return mi->mbmi.mode;
- }
+ else
+ return mi->mbmi.sb_type < BLOCK_8X8 ? (mi->bmi + 1 + b)->as_mode
+ : mi->mbmi.mode;
}
assert(b == 1 || b == 3);
return (mi->bmi + b - 1)->as_mode;
if (!mi)
return DC_PRED;
- if (mi->mbmi.ref_frame[0] != INTRA_FRAME) {
+ if (is_inter_block(&mi->mbmi))
return DC_PRED;
- } else if (mi->mbmi.sb_type < BLOCK_8X8) {
- return ((mi->bmi + 2 + b)->as_mode);
- } else {
- return mi->mbmi.mode;
- }
+ else
+ return mi->mbmi.sb_type < BLOCK_8X8 ? (mi->bmi + 2 + b)->as_mode
+ : mi->mbmi.mode;
}
return (mi->bmi + b - 2)->as_mode;
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_idct.h"
-void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
0.5 shifts per pixel. */
int i;
}
}
-void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
int i;
int a1, e1;
int16_t tmp[4];
output[3] = step[0] - step[3];
}
-void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
int16_t out[4 * 4];
int16_t *outptr = out;
int i, j;
}
}
-void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
int i;
int a1;
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+ dest[j * dest_stride + i]); }
}
-void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest,
+void vp9_short_idct8x8_10_add_c(int16_t *input, uint8_t *dest,
int dest_stride) {
int16_t out[8 * 8] = { 0 };
int16_t *outptr = out;
+ dest[j * dest_stride + i]); }
}
-void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest,
+void vp9_short_idct16x16_10_add_c(int16_t *input, uint8_t *dest,
int dest_stride) {
int16_t out[16 * 16] = { 0 };
int16_t *outptr = out;
}
}
-void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output) {
+void vp9_short_idct32x32_1_add_c(int16_t *input, uint8_t *dest,
+ int dest_stride) {
+ int i, j;
+ int a1;
+
int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
out = dct_const_round_shift(out * cospi_16_64);
- output[0] = ROUND_POWER_OF_TWO(out, 6);
+ a1 = ROUND_POWER_OF_TWO(out, 6);
+
+ for (j = 0; j < 32; ++j) {
+ for (i = 0; i < 32; ++i)
+ dest[i] = clip_pixel(dest[i] + a1);
+ dest += dest_stride;
+ }
+}
+
+// idct
+void vp9_idct4x4_add(int16_t *input, uint8_t *dest, int stride, int eob) {
+ if (eob > 1)
+ vp9_idct4x4_16_add(input, dest, stride);
+ else
+ vp9_idct4x4_1_add(input, dest, stride);
+}
+
+
+void vp9_iwht4x4_add(int16_t *input, uint8_t *dest, int stride, int eob) {
+ if (eob > 1)
+ vp9_iwht4x4_16_add(input, dest, stride);
+ else
+ vp9_iwht4x4_1_add(input, dest, stride);
+}
+
+void vp9_idct_add_8x8(int16_t *input, uint8_t *dest, int stride, int eob) {
+ // If dc is 1, then input[0] is the reconstructed value, do not need
+ // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
+
+ // The calculation can be simplified if there are not many non-zero dct
+ // coefficients. Use eobs to decide what to do.
+ // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
+ // Combine that with code here.
+ if (eob) {
+ if (eob == 1)
+ // DC only DCT coefficient
+ vp9_short_idct8x8_1_add(input, dest, stride);
+ else if (eob <= 10)
+ vp9_short_idct8x8_10_add(input, dest, stride);
+ else
+ vp9_short_idct8x8_add(input, dest, stride);
+ }
+}
+
+void vp9_idct_add_16x16(int16_t *input, uint8_t *dest, int stride, int eob) {
+ /* The calculation can be simplified if there are not many non-zero dct
+ * coefficients. Use eobs to separate different cases. */
+ if (eob) {
+ if (eob == 1)
+ /* DC only DCT coefficient. */
+ vp9_short_idct16x16_1_add(input, dest, stride);
+ else if (eob <= 10)
+ vp9_short_idct16x16_10_add(input, dest, stride);
+ else
+ vp9_short_idct16x16_add(input, dest, stride);
+ }
+}
+
+void vp9_idct_add_32x32(int16_t *input, uint8_t *dest, int stride, int eob) {
+ if (eob) {
+ if (eob == 1)
+ vp9_short_idct32x32_1_add(input, dest, stride);
+ else
+ vp9_short_idct32x32_add(input, dest, stride);
+ }
+}
+
+// iht
+void vp9_iht_add(TX_TYPE tx_type, int16_t *input, uint8_t *dest, int stride,
+ int eob) {
+ if (tx_type == DCT_DCT)
+ vp9_idct4x4_add(input, dest, stride, eob);
+ else
+ vp9_short_iht4x4_add(input, dest, stride, tx_type);
+}
+
+void vp9_iht_add_8x8(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+ int stride, int eob) {
+ if (tx_type == DCT_DCT) {
+ vp9_idct_add_8x8(input, dest, stride, eob);
+ } else {
+ if (eob > 0) {
+ vp9_short_iht8x8_add(input, dest, stride, tx_type);
+ }
+ }
+}
+
+void vp9_iht_add_16x16(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+ int stride, int eob) {
+ if (tx_type == DCT_DCT) {
+ vp9_idct_add_16x16(input, dest, stride, eob);
+ } else {
+ if (eob > 0) {
+ vp9_short_iht16x16_add(input, dest, stride, tx_type);
+ }
+ }
}
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_enums.h"
// Constants and Macros used by all idct/dct functions
transform_1d cols, rows; // vertical and horizontal
} transform_2d;
+
+void vp9_idct4x4_add(int16_t *input, uint8_t *dest, int stride, int eob);
+void vp9_iwht4x4_add(int16_t *input, uint8_t *dest, int stride, int eob);
+void vp9_idct_add_8x8(int16_t *input, uint8_t *dest, int stride, int eob);
+void vp9_idct_add_16x16(int16_t *input, uint8_t *dest, int stride, int eob);
+void vp9_idct_add_32x32(int16_t *input, uint8_t *dest, int stride, int eob);
+
+void vp9_iht_add(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+ int stride, int eob);
+
+void vp9_iht_add_8x8(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+ int stride, int eob);
+
+void vp9_iht_add_16x16(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+ int stride, int eob);
+
+
#endif // VP9_COMMON_VP9_IDCT_H_
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/common/vp9_loopfilter.h"
#include "vp9/common/vp9_onyxc_int.h"
#include "vp9/common/vp9_reconinter.h"
#define VP9_COMMON_VP9_LOOPFILTER_H_
#include "vpx_ports/mem.h"
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_seg_common.h"
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_loopfilter.h"
#include "vp9/common/vp9_onyxc_int.h"
};
// clamp_mv_ref
-#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units
+#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units
static void clamp_mv_ref(MV *mv, const MACROBLOCKD *xd) {
clamp_mv(mv, xd->mb_to_left_edge - MV_BORDER,
#ifdef __cplusplus
extern "C"
-{
+{ // NOLINT
#endif
#include "./vpx_config.h"
FOURFIVE = 1,
THREEFIVE = 2,
ONETWO = 3
-
} VPX_SCALING;
typedef enum {
// 3 - lowest quality/fastest decode
int width; // width of data passed to the compressor
int height; // height of data passed to the compressor
- double framerate; // set to passed in framerate
- int64_t target_bandwidth; // bandwidth to be used in kilobits per second
+ double framerate; // set to passed in framerate
+ int64_t target_bandwidth; // bandwidth to be used in kilobits per second
- int noise_sensitivity; // parameter used for applying pre processing blur: recommendation 0
- int Sharpness; // parameter used for sharpening output: recommendation 0:
+ int noise_sensitivity; // pre processing blur: recommendation 0
+ int Sharpness; // sharpening output: recommendation 0:
int cpu_used;
unsigned int rc_max_intra_bitrate_pct;
// mode ->
- // (0)=Realtime/Live Encoding. This mode is optimized for realtim encoding (for example, capturing
- // a television signal or feed from a live camera). ( speed setting controls how fast )
- // (1)=Good Quality Fast Encoding. The encoder balances quality with the amount of time it takes to
- // encode the output. ( speed setting controls how fast )
- // (2)=One Pass - Best Quality. The encoder places priority on the quality of the output over encoding
- // speed. The output is compressed at the highest possible quality. This option takes the longest
- // amount of time to encode. ( speed setting ignored )
- // (3)=Two Pass - First Pass. The encoder generates a file of statistics for use in the second encoding
- // pass. ( speed setting controls how fast )
- // (4)=Two Pass - Second Pass. The encoder uses the statistics that were generated in the first encoding
- // pass to create the compressed output. ( speed setting controls how fast )
- // (5)=Two Pass - Second Pass Best. The encoder uses the statistics that were generated in the first
- // encoding pass to create the compressed output using the highest possible quality, and taking a
+ // (0)=Realtime/Live Encoding. This mode is optimized for realtime
+ // encoding (for example, capturing a television signal or feed from
+ // a live camera). ( speed setting controls how fast )
+ // (1)=Good Quality Fast Encoding. The encoder balances quality with the
+ // amount of time it takes to encode the output. ( speed setting
+ // controls how fast )
+ // (2)=One Pass - Best Quality. The encoder places priority on the
+ // quality of the output over encoding speed. The output is compressed
+ // at the highest possible quality. This option takes the longest
+ // amount of time to encode. ( speed setting ignored )
+ // (3)=Two Pass - First Pass. The encoder generates a file of statistics
+ // for use in the second encoding pass. ( speed setting controls how
+ // fast )
+ // (4)=Two Pass - Second Pass. The encoder uses the statistics that were
+ // generated in the first encoding pass to create the compressed
+ // output. ( speed setting controls how fast )
+ // (5)=Two Pass - Second Pass Best. The encoder uses the statistics that
+ // were generated in the first encoding pass to create the compressed
+ // output using the highest possible quality, and taking a
// longer amount of time to encode.. ( speed setting ignored )
- int Mode; //
+ int Mode;
// Key Framing Operations
- int auto_key; // automatically detect cut scenes and set the keyframes
- int key_freq; // maximum distance to key frame.
+ int auto_key; // autodetect cut scenes and set the keyframes
+ int key_freq; // maximum distance to key frame.
- int allow_lag; // allow lagged compression (if 0 lagin frames is ignored)
- int lag_in_frames; // how many frames lag before we start encoding
+ int allow_lag; // allow lagged compression (if 0 lagin frames is ignored)
+ int lag_in_frames; // how many frames lag before we start encoding
// ----------------------------------------------------------------
// DATARATE CONTROL OPTIONS
- int end_usage; // vbr or cbr
+ int end_usage; // vbr or cbr
// buffer targeting aggressiveness
int under_shoot_pct;
int play_alternate;
int alt_freq;
- int encode_breakout; // early breakout encode threshold : for video conf recommend 800
+ int encode_breakout; // early breakout : for video conf recommend 800
/* Bitfield defining the error resiliency features to enable.
* Can provide decodable frames after losses in previous
void vp9_change_config(VP9_PTR onyx, VP9_CONFIG *oxcf);
-// receive a frames worth of data caller can assume that a copy of this frame is made
-// and not just a copy of the pointer..
+ // receive a frames worth of data. caller can assume that a copy of this
+ // frame is made and not just a copy of the pointer..
int vp9_receive_raw_frame(VP9_PTR comp, unsigned int frame_flags,
YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
int64_t end_time_stamp);
#ifndef VP9_COMMON_VP9_ONYXC_INT_H_
#define VP9_COMMON_VP9_ONYXC_INT_H_
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vpx/internal/vpx_codec_internal.h"
-#include "vp9_rtcd.h"
+#include "./vp9_rtcd.h"
#include "vp9/common/vp9_loopfilter.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_entropy.h"
YV12_BUFFER_CONFIG post_proc_buffer;
- FRAME_TYPE last_frame_type; /* Save last frame's frame type for motion search. */
+ FRAME_TYPE last_frame_type; /* last frame's frame type for motion search.*/
FRAME_TYPE frame_type;
int show_frame;
xd->right_available = (mi_col + bw < cm->cur_tile_mi_col_end);
}
-static int get_token_alloc(int mb_rows, int mb_cols) {
- return mb_rows * mb_cols * (48 * 16 + 4);
-}
-
static void set_prev_mi(VP9_COMMON *cm) {
const int use_prev_in_find_mv_refs = cm->width == cm->last_width &&
cm->height == cm->last_height &&
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <math.h>
+#include <stdlib.h>
+#include <stdio.h>
#include "./vpx_config.h"
#include "vpx_scale/yv12config.h"
#include "./vp9_rtcd.h"
#include "./vpx_scale_rtcd.h"
-
-#include <math.h>
-#include <stdlib.h>
-#include <stdio.h>
-
#define RGB_TO_YUV(t) \
( (0.257*(float)(t >> 16)) + (0.504*(float)(t >> 8 & 0xff)) + \
(0.098*(float)(t & 0xff)) + 16), \
p_dst = dst_ptr;
for (col = 0; col < cols; col++) {
-
int kernel = 4;
int v = p_src[col];
void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
int rows, int cols, int flimit) {
int r, c, i;
- const short *rv3 = &vp9_rv[63 & rand()];
+ const short *rv3 = &vp9_rv[63 & rand()]; // NOLINT
for (c = 0; c < cols; c++) {
uint8_t *s = &dst[c];
next = next + j;
}
-
}
for (; next < 256; next++)
}
for (i = 0; i < 3072; i++) {
- state->noise[i] = char_dist[rand() & 0xff];
+ state->noise[i] = char_dist[rand() & 0xff]; // NOLINT
}
for (i = 0; i < 16; i++) {
#if 0 && CONFIG_POSTPROC_VISUALIZER
if (flags & VP9D_DEBUG_TXT_FRAME_INFO) {
char message[512];
- sprintf(message, "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
- (cm->frame_type == KEY_FRAME),
- cm->refresh_golden_frame,
- cm->base_qindex,
- cm->filter_level,
- flags,
- cm->mb_cols, cm->mb_rows);
+ snprintf(message, sizeof(message) -1,
+ "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
+ (cm->frame_type == KEY_FRAME),
+ cm->refresh_golden_frame,
+ cm->base_qindex,
+ cm->filter_level,
+ flags,
+ cm->mb_cols, cm->mb_rows);
vp9_blit_text(message, cm->post_proc_buffer.y_buffer,
cm->post_proc_buffer.y_stride);
}
for (j = 0; j < mb_cols; j++) {
char zz[4];
- sprintf(zz, "%c", mi[mb_index].mbmi.mode + 'a');
+ snprintf(zz, sizeof(zz) - 1, "%c", mi[mb_index].mbmi.mode + 'a');
vp9_blit_text(zz, y_ptr, post->y_stride);
mb_index++;
mb_index++; /* border */
y_ptr += post->y_stride * 16 - post->y_width;
-
}
}
mi[mb_index].mbmi.skip_coeff);
if (cm->frame_type == KEY_FRAME)
- sprintf(zz, "a");
+ snprintf(zz, sizeof(zz) - 1, "a");
else
- sprintf(zz, "%c", dc_diff + '0');
+ snprintf(zz, sizeof(zz) - 1, "%c", dc_diff + '0');
vp9_blit_text(zz, y_ptr, post->y_stride);
mb_index++;
mb_index++; /* border */
y_ptr += post->y_stride * 16 - post->y_width;
-
}
}
constrain_line(lx0, &x1, ly0 + 1, &y1, width, height);
vp9_blit_line(lx0, x1, ly0 + 1, y1, y_buffer, y_stride);
- } else
+ } else {
vp9_blit_line(lx0, x1, ly0, y1, y_buffer, y_stride);
+ }
}
mi++;
xd->this_mi->mbmi.seg_id_predicted = pred_flag;
}
-void vp9_set_pred_flag_mbskip(MACROBLOCKD *xd, BLOCK_SIZE bsize,
- uint8_t pred_flag) {
- xd->this_mi->mbmi.skip_coeff = pred_flag;
-}
-
int vp9_get_segment_id(VP9_COMMON *cm, const uint8_t *segment_ids,
BLOCK_SIZE bsize, int mi_row, int mi_col) {
const int mi_offset = mi_row * cm->mi_cols + mi_col;
return xd->this_mi->mbmi.skip_coeff;
}
-void vp9_set_pred_flag_mbskip(MACROBLOCKD *xd, BLOCK_SIZE bsize,
- uint8_t pred_flag);
-
unsigned char vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd);
unsigned char vp9_get_pred_context_intra_inter(const MACROBLOCKD *xd);
const MACROBLOCKD *xd);
-static INLINE vp9_prob vp9_get_pred_prob_comp_inter_inter(const VP9_COMMON *cm,
- const MACROBLOCKD *xd) {
+static INLINE
+vp9_prob vp9_get_pred_prob_comp_inter_inter(const VP9_COMMON *cm,
+ const MACROBLOCKD *xd) {
const int pred_context = vp9_get_pred_context_comp_inter_inter(cm, xd);
return cm->fc.comp_inter_prob[pred_context];
}
#if 1
static const int16_t dc_qlookup[QINDEX_RANGE] = {
- 4, 8, 8, 9, 10, 11, 12, 12,
- 13, 14, 15, 16, 17, 18, 19, 19,
- 20, 21, 22, 23, 24, 25, 26, 26,
- 27, 28, 29, 30, 31, 32, 32, 33,
- 34, 35, 36, 37, 38, 38, 39, 40,
- 41, 42, 43, 43, 44, 45, 46, 47,
- 48, 48, 49, 50, 51, 52, 53, 53,
- 54, 55, 56, 57, 57, 58, 59, 60,
- 61, 62, 62, 63, 64, 65, 66, 66,
- 67, 68, 69, 70, 70, 71, 72, 73,
- 74, 74, 75, 76, 77, 78, 78, 79,
- 80, 81, 81, 82, 83, 84, 85, 85,
- 87, 88, 90, 92, 93, 95, 96, 98,
- 99, 101, 102, 104, 105, 107, 108, 110,
- 111, 113, 114, 116, 117, 118, 120, 121,
- 123, 125, 127, 129, 131, 134, 136, 138,
- 140, 142, 144, 146, 148, 150, 152, 154,
- 156, 158, 161, 164, 166, 169, 172, 174,
- 177, 180, 182, 185, 187, 190, 192, 195,
- 199, 202, 205, 208, 211, 214, 217, 220,
- 223, 226, 230, 233, 237, 240, 243, 247,
- 250, 253, 257, 261, 265, 269, 272, 276,
- 280, 284, 288, 292, 296, 300, 304, 309,
- 313, 317, 322, 326, 330, 335, 340, 344,
- 349, 354, 359, 364, 369, 374, 379, 384,
- 389, 395, 400, 406, 411, 417, 423, 429,
- 435, 441, 447, 454, 461, 467, 475, 482,
- 489, 497, 505, 513, 522, 530, 539, 549,
- 559, 569, 579, 590, 602, 614, 626, 640,
- 654, 668, 684, 700, 717, 736, 755, 775,
- 796, 819, 843, 869, 896, 925, 955, 988,
+ 4, 8, 8, 9, 10, 11, 12, 12,
+ 13, 14, 15, 16, 17, 18, 19, 19,
+ 20, 21, 22, 23, 24, 25, 26, 26,
+ 27, 28, 29, 30, 31, 32, 32, 33,
+ 34, 35, 36, 37, 38, 38, 39, 40,
+ 41, 42, 43, 43, 44, 45, 46, 47,
+ 48, 48, 49, 50, 51, 52, 53, 53,
+ 54, 55, 56, 57, 57, 58, 59, 60,
+ 61, 62, 62, 63, 64, 65, 66, 66,
+ 67, 68, 69, 70, 70, 71, 72, 73,
+ 74, 74, 75, 76, 77, 78, 78, 79,
+ 80, 81, 81, 82, 83, 84, 85, 85,
+ 87, 88, 90, 92, 93, 95, 96, 98,
+ 99, 101, 102, 104, 105, 107, 108, 110,
+ 111, 113, 114, 116, 117, 118, 120, 121,
+ 123, 125, 127, 129, 131, 134, 136, 138,
+ 140, 142, 144, 146, 148, 150, 152, 154,
+ 156, 158, 161, 164, 166, 169, 172, 174,
+ 177, 180, 182, 185, 187, 190, 192, 195,
+ 199, 202, 205, 208, 211, 214, 217, 220,
+ 223, 226, 230, 233, 237, 240, 243, 247,
+ 250, 253, 257, 261, 265, 269, 272, 276,
+ 280, 284, 288, 292, 296, 300, 304, 309,
+ 313, 317, 322, 326, 330, 335, 340, 344,
+ 349, 354, 359, 364, 369, 374, 379, 384,
+ 389, 395, 400, 406, 411, 417, 423, 429,
+ 435, 441, 447, 454, 461, 467, 475, 482,
+ 489, 497, 505, 513, 522, 530, 539, 549,
+ 559, 569, 579, 590, 602, 614, 626, 640,
+ 654, 668, 684, 700, 717, 736, 755, 775,
+ 796, 819, 843, 869, 896, 925, 955, 988,
1022, 1058, 1098, 1139, 1184, 1232, 1282, 1336,
};
static const int16_t ac_qlookup[QINDEX_RANGE] = {
- 4, 8, 9, 10, 11, 12, 13, 14,
- 15, 16, 17, 18, 19, 20, 21, 22,
- 23, 24, 25, 26, 27, 28, 29, 30,
- 31, 32, 33, 34, 35, 36, 37, 38,
- 39, 40, 41, 42, 43, 44, 45, 46,
- 47, 48, 49, 50, 51, 52, 53, 54,
- 55, 56, 57, 58, 59, 60, 61, 62,
- 63, 64, 65, 66, 67, 68, 69, 70,
- 71, 72, 73, 74, 75, 76, 77, 78,
- 79, 80, 81, 82, 83, 84, 85, 86,
- 87, 88, 89, 90, 91, 92, 93, 94,
- 95, 96, 97, 98, 99, 100, 101, 102,
- 104, 106, 108, 110, 112, 114, 116, 118,
- 120, 122, 124, 126, 128, 130, 132, 134,
- 136, 138, 140, 142, 144, 146, 148, 150,
- 152, 155, 158, 161, 164, 167, 170, 173,
- 176, 179, 182, 185, 188, 191, 194, 197,
- 200, 203, 207, 211, 215, 219, 223, 227,
- 231, 235, 239, 243, 247, 251, 255, 260,
- 265, 270, 275, 280, 285, 290, 295, 300,
- 305, 311, 317, 323, 329, 335, 341, 347,
- 353, 359, 366, 373, 380, 387, 394, 401,
- 408, 416, 424, 432, 440, 448, 456, 465,
- 474, 483, 492, 501, 510, 520, 530, 540,
- 550, 560, 571, 582, 593, 604, 615, 627,
- 639, 651, 663, 676, 689, 702, 715, 729,
- 743, 757, 771, 786, 801, 816, 832, 848,
- 864, 881, 898, 915, 933, 951, 969, 988,
+ 4, 8, 9, 10, 11, 12, 13, 14,
+ 15, 16, 17, 18, 19, 20, 21, 22,
+ 23, 24, 25, 26, 27, 28, 29, 30,
+ 31, 32, 33, 34, 35, 36, 37, 38,
+ 39, 40, 41, 42, 43, 44, 45, 46,
+ 47, 48, 49, 50, 51, 52, 53, 54,
+ 55, 56, 57, 58, 59, 60, 61, 62,
+ 63, 64, 65, 66, 67, 68, 69, 70,
+ 71, 72, 73, 74, 75, 76, 77, 78,
+ 79, 80, 81, 82, 83, 84, 85, 86,
+ 87, 88, 89, 90, 91, 92, 93, 94,
+ 95, 96, 97, 98, 99, 100, 101, 102,
+ 104, 106, 108, 110, 112, 114, 116, 118,
+ 120, 122, 124, 126, 128, 130, 132, 134,
+ 136, 138, 140, 142, 144, 146, 148, 150,
+ 152, 155, 158, 161, 164, 167, 170, 173,
+ 176, 179, 182, 185, 188, 191, 194, 197,
+ 200, 203, 207, 211, 215, 219, 223, 227,
+ 231, 235, 239, 243, 247, 251, 255, 260,
+ 265, 270, 275, 280, 285, 290, 295, 300,
+ 305, 311, 317, 323, 329, 335, 341, 347,
+ 353, 359, 366, 373, 380, 387, 394, 401,
+ 408, 416, 424, 432, 440, 448, 456, 465,
+ 474, 483, 492, 501, 510, 520, 530, 540,
+ 550, 560, 571, 582, 593, 604, 615, 627,
+ 639, 651, 663, 676, 689, 702, 715, 729,
+ 743, 757, 771, 786, 801, 816, 832, 848,
+ 864, 881, 898, 915, 933, 951, 969, 988,
1007, 1026, 1046, 1066, 1087, 1108, 1129, 1151,
1173, 1196, 1219, 1243, 1267, 1292, 1317, 1343,
1369, 1396, 1423, 1451, 1479, 1508, 1537, 1567,
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
-
void vp9_setup_interp_filters(MACROBLOCKD *xd,
INTERPOLATIONFILTERTYPE mcomp_filter_type,
VP9_COMMON *cm) {
if (xd->mi_8x8 && xd->this_mi) {
- MB_MODE_INFO * mbmi = &xd->this_mi->mbmi;
+ MB_MODE_INFO *const mbmi = &xd->this_mi->mbmi;
- set_scale_factors(xd, mbmi->ref_frame[0] - 1, mbmi->ref_frame[1] - 1,
- cm->active_ref_scale);
+ set_scale_factors(xd, mbmi->ref_frame[0] - LAST_FRAME,
+ mbmi->ref_frame[1] - LAST_FRAME,
+ cm->active_ref_scale);
} else {
set_scale_factors(xd, -1, -1, cm->active_ref_scale);
}
- switch (mcomp_filter_type) {
- case EIGHTTAP:
- case SWITCHABLE:
- xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8;
- break;
- case EIGHTTAP_SMOOTH:
- xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8lp;
- break;
- case EIGHTTAP_SHARP:
- xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8s;
- break;
- case BILINEAR:
- xd->subpix.filter_x = xd->subpix.filter_y = vp9_bilinear_filters;
- break;
- }
+ xd->subpix.filter_x = xd->subpix.filter_y =
+ vp9_get_filter_kernel(mcomp_filter_type == SWITCHABLE ?
+ EIGHTTAP : mcomp_filter_type);
+
assert(((intptr_t)xd->subpix.filter_x & 0xff) == 0);
}
const int x = 4 * (block & ((1 << bwl) - 1));
const int y = 4 * (block >> bwl);
const MODE_INFO *mi = xd->this_mi;
- const int use_second_ref = mi->mbmi.ref_frame[1] > 0;
+ const int is_compound = has_second_ref(&mi->mbmi);
int ref;
assert(x < bw);
assert(mi->mbmi.sb_type < BLOCK_8X8 || 4 << pred_w == bw);
assert(mi->mbmi.sb_type < BLOCK_8X8 || 4 << pred_h == bh);
- for (ref = 0; ref < 1 + use_second_ref; ++ref) {
+ for (ref = 0; ref < 1 + is_compound; ++ref) {
struct scale_factors *const scale = &xd->scale_factor[ref];
struct buf_2d *const pre_buf = &pd->pre[ref];
struct buf_2d *const dst_buf = &pd->dst;
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/vpx_once.h"
-#include "vp9_rtcd.h"
+#include "./vp9_rtcd.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_onyxc_int.h"
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "vpx_config.h"
+#include "./vpx_config.h"
#define RTCD_C
-#include "vp9_rtcd.h"
+#include "./vp9_rtcd.h"
#include "vpx_ports/vpx_once.h"
void vpx_scale_rtcd(void);
# this variable is for functions that are 64 bit only.
[ $arch = "x86_64" ] && mmx_x86_64=mmx && sse2_x86_64=sse2 && ssse3_x86_64=ssse3
-#
-# Dequant
-#
-
-prototype void vp9_idct_add_16x16 "int16_t *input, uint8_t *dest, int stride, int eob"
-specialize vp9_idct_add_16x16
-
-prototype void vp9_idct_add_8x8 "int16_t *input, uint8_t *dest, int stride, int eob"
-specialize vp9_idct_add_8x8
-
-prototype void vp9_idct_add "int16_t *input, uint8_t *dest, int stride, int eob"
-specialize vp9_idct_add
-
-prototype void vp9_idct_add_32x32 "int16_t *q, uint8_t *dst, int stride, int eob"
-specialize vp9_idct_add_32x32
-
#
# RECON
#
prototype void vp9_dc_128_predictor_32x32 "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"
specialize vp9_dc_128_predictor_32x32
-if [ "$CONFIG_VP9_DECODER" = "yes" ]; then
-prototype void vp9_add_constant_residual_8x8 "const int16_t diff, uint8_t *dest, int stride"
-specialize vp9_add_constant_residual_8x8 sse2 neon
-
-prototype void vp9_add_constant_residual_16x16 "const int16_t diff, uint8_t *dest, int stride"
-specialize vp9_add_constant_residual_16x16 sse2 neon
-
-prototype void vp9_add_constant_residual_32x32 "const int16_t diff, uint8_t *dest, int stride"
-specialize vp9_add_constant_residual_32x32 sse2 neon
-fi
-
#
# Loopfilter
#
# Sub Pixel Filters
#
prototype void vp9_convolve_copy "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve_copy $sse2_x86inc neon
+specialize vp9_convolve_copy $sse2_x86inc neon dspr2
prototype void vp9_convolve_avg "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve_avg $sse2_x86inc neon
+specialize vp9_convolve_avg $sse2_x86inc neon dspr2
prototype void vp9_convolve8 "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8 ssse3 neon
+specialize vp9_convolve8 ssse3 neon dspr2
prototype void vp9_convolve8_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_horiz ssse3 neon
+specialize vp9_convolve8_horiz ssse3 neon dspr2
prototype void vp9_convolve8_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_vert ssse3 neon
+specialize vp9_convolve8_vert ssse3 neon dspr2
prototype void vp9_convolve8_avg "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg ssse3 neon
+specialize vp9_convolve8_avg ssse3 neon dspr2
prototype void vp9_convolve8_avg_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg_horiz ssse3 neon
+specialize vp9_convolve8_avg_horiz ssse3 neon dspr2
prototype void vp9_convolve8_avg_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg_vert ssse3 neon
+specialize vp9_convolve8_avg_vert ssse3 neon dspr2
#
# dct
#
-prototype void vp9_short_idct4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct4x4_1_add sse2 neon
+prototype void vp9_idct4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_idct4x4_1_add sse2 neon
-prototype void vp9_short_idct4x4_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct4x4_add sse2 neon
+prototype void vp9_idct4x4_16_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_idct4x4_16_add sse2 neon
prototype void vp9_short_idct8x8_1_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct8x8_1_add sse2 neon
prototype void vp9_short_idct8x8_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct8x8_add sse2 neon
-prototype void vp9_short_idct10_8x8_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct10_8x8_add sse2 neon
+prototype void vp9_short_idct8x8_10_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct8x8_10_add sse2 neon
prototype void vp9_short_idct16x16_1_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct16x16_1_add sse2 neon
prototype void vp9_short_idct16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct16x16_add sse2 neon
-prototype void vp9_short_idct10_16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct10_16x16_add sse2 neon
+prototype void vp9_short_idct16x16_10_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct16x16_10_add sse2 neon
prototype void vp9_short_idct32x32_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct32x32_add sse2 neon
-prototype void vp9_short_idct1_32x32 "int16_t *input, int16_t *output"
-specialize vp9_short_idct1_32x32
+prototype void vp9_short_idct32x32_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct32x32_1_add sse2
prototype void vp9_short_iht4x4_add "int16_t *input, uint8_t *dest, int dest_stride, int tx_type"
specialize vp9_short_iht4x4_add sse2 neon
specialize vp9_idct4_1d sse2
# dct and add
-prototype void vp9_short_iwalsh4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_iwalsh4x4_1_add
+prototype void vp9_iwht4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_iwht4x4_1_add
-prototype void vp9_short_iwalsh4x4_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_iwalsh4x4_add
+prototype void vp9_iwht4x4_16_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_iwht4x4_16_add
#
# Encoder functions below this point.
sf->y_scale_fp != REF_NO_SCALE;
}
-#endif // VP9_COMMON_VP9_SCALE_H_
+#endif // VP9_COMMON_VP9_SCALE_H_
+++ /dev/null
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifndef VP9_COMMON_VP9_SUBPELVAR_H_
-#define VP9_COMMON_VP9_SUBPELVAR_H_
-
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_convolve.h"
-
-static void variance(const uint8_t *src_ptr,
- int source_stride,
- const uint8_t *ref_ptr,
- int recon_stride,
- int w,
- int h,
- unsigned int *sse,
- int *sum) {
- int i, j;
- int diff;
-
- *sum = 0;
- *sse = 0;
-
- for (i = 0; i < h; i++) {
- for (j = 0; j < w; j++) {
- diff = src_ptr[j] - ref_ptr[j];
- *sum += diff;
- *sse += diff * diff;
- }
-
- src_ptr += source_stride;
- ref_ptr += recon_stride;
- }
-}
-
-/****************************************************************************
- *
- * ROUTINE : filter_block2d_bil_first_pass
- *
- * INPUTS : uint8_t *src_ptr : Pointer to source block.
- * uint32_t src_pixels_per_line : Stride of input block.
- * uint32_t pixel_step : Offset between filter input samples (see notes).
- * uint32_t output_height : Input block height.
- * uint32_t output_width : Input block width.
- * int32_t *vp9_filter : Array of 2 bi-linear filter taps.
- *
- * OUTPUTS : int32_t *output_ptr : Pointer to filtered block.
- *
- * RETURNS : void
- *
- * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
- * either horizontal or vertical direction to produce the
- * filtered output block. Used to implement first-pass
- * of 2-D separable filter.
- *
- * SPECIAL NOTES : Produces int32_t output to retain precision for next pass.
- * Two filter taps should sum to VP9_FILTER_WEIGHT.
- * pixel_step defines whether the filter is applied
- * horizontally (pixel_step=1) or vertically (pixel_step=stride).
- * It defines the offset required to move from one input
- * to the next.
- *
- ****************************************************************************/
-static void var_filter_block2d_bil_first_pass(const uint8_t *src_ptr,
- uint16_t *output_ptr,
- unsigned int src_pixels_per_line,
- int pixel_step,
- unsigned int output_height,
- unsigned int output_width,
- const int16_t *vp9_filter) {
- unsigned int i, j;
-
- for (i = 0; i < output_height; i++) {
- for (j = 0; j < output_width; j++) {
- output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
- (int)src_ptr[pixel_step] * vp9_filter[1],
- FILTER_BITS);
-
- src_ptr++;
- }
-
- // Next row...
- src_ptr += src_pixels_per_line - output_width;
- output_ptr += output_width;
- }
-}
-
-/****************************************************************************
- *
- * ROUTINE : filter_block2d_bil_second_pass
- *
- * INPUTS : int32_t *src_ptr : Pointer to source block.
- * uint32_t src_pixels_per_line : Stride of input block.
- * uint32_t pixel_step : Offset between filter input samples (see notes).
- * uint32_t output_height : Input block height.
- * uint32_t output_width : Input block width.
- * int32_t *vp9_filter : Array of 2 bi-linear filter taps.
- *
- * OUTPUTS : uint16_t *output_ptr : Pointer to filtered block.
- *
- * RETURNS : void
- *
- * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
- * either horizontal or vertical direction to produce the
- * filtered output block. Used to implement second-pass
- * of 2-D separable filter.
- *
- * SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass.
- * Two filter taps should sum to VP9_FILTER_WEIGHT.
- * pixel_step defines whether the filter is applied
- * horizontally (pixel_step=1) or vertically (pixel_step=stride).
- * It defines the offset required to move from one input
- * to the next.
- *
- ****************************************************************************/
-static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
- uint8_t *output_ptr,
- unsigned int src_pixels_per_line,
- unsigned int pixel_step,
- unsigned int output_height,
- unsigned int output_width,
- const int16_t *vp9_filter) {
- unsigned int i, j;
-
- for (i = 0; i < output_height; i++) {
- for (j = 0; j < output_width; j++) {
- output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
- (int)src_ptr[pixel_step] * vp9_filter[1],
- FILTER_BITS);
- src_ptr++;
- }
-
- src_ptr += src_pixels_per_line - output_width;
- output_ptr += output_width;
- }
-}
-
-#endif // VP9_COMMON_VP9_SUBPELVAR_H_
if (j <= 0) {
p[-j].value = v;
p[-j].len = l;
- } else
+ } else {
tree2tok(p, t, j, v, l);
+ }
} while (++v & 1);
}
return left + right;
}
-void vp9_tree_probs_from_distribution(
- vp9_tree tree,
- vp9_prob probs [ /* n-1 */ ],
- unsigned int branch_ct [ /* n-1 */ ] [2],
- const unsigned int num_events[ /* n */ ],
- unsigned int tok0_offset) {
+void vp9_tree_probs_from_distribution(vp9_tree tree, vp9_prob probs[/* n-1 */],
+ unsigned int branch_ct[/* n-1 */][2],
+ const unsigned int num_events[/* n */],
+ unsigned int tok0_offset) {
convert_distribution(0, tree, probs, branch_ct, num_events, tok0_offset);
}
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_idct.h"
-void vp9_short_idct4x4_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct4x4_16_add_sse2(int16_t *input, uint8_t *dest, int stride) {
const __m128i zero = _mm_setzero_si128();
const __m128i eight = _mm_set1_epi16(8);
const __m128i cst = _mm_setr_epi16((int16_t)cospi_16_64, (int16_t)cospi_16_64,
RECON_AND_STORE4X4(dest, input3);
}
-void vp9_short_idct4x4_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct4x4_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
__m128i dc_value;
const __m128i zero = _mm_setzero_si128();
int a;
RECON_AND_STORE(dest, in[7]);
}
-void vp9_short_idct10_8x8_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+void vp9_short_idct8x8_10_add_sse2(int16_t *input, uint8_t *dest, int stride) {
const __m128i zero = _mm_setzero_si128();
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<4);
TRANSPOSE_8X4(in0, in1, in2, in3, in0, in1, in2, in3)
// Stage1
- {
+ { //NOLINT
const __m128i lo_17 = _mm_unpackhi_epi16(in0, in3);
const __m128i lo_35 = _mm_unpackhi_epi16(in1, in2);
}
// Stage2
- {
+ { //NOLINT
const __m128i lo_04 = _mm_unpacklo_epi16(in0, in2);
const __m128i lo_26 = _mm_unpacklo_epi16(in1, in3);
}
// Stage3
- {
+ { //NOLINT
const __m128i lo_56 = _mm_unpacklo_epi16(stp2_5, stp2_6);
stp1_0 = _mm_adds_epi16(stp2_0, stp2_3);
stp1_1 = _mm_adds_epi16(stp2_1, stp2_2);
write_buffer_8x16(dest, in1, stride);
}
-void vp9_short_idct10_16x16_add_sse2(int16_t *input, uint8_t *dest,
+void vp9_short_idct16x16_10_add_sse2(int16_t *input, uint8_t *dest,
int stride) {
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<5);
dest += 8 - (stride * 32);
}
}
+} //NOLINT
+
+void vp9_short_idct32x32_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+ __m128i dc_value;
+ const __m128i zero = _mm_setzero_si128();
+ int a, i;
+
+ a = dct_const_round_shift(input[0] * cospi_16_64);
+ a = dct_const_round_shift(a * cospi_16_64);
+ a = ROUND_POWER_OF_TWO(a, 6);
+
+ dc_value = _mm_set1_epi16(a);
+
+ for (i = 0; i < 4; ++i) {
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ RECON_AND_STORE(dest, dc_value);
+ dest += 8 - (stride * 32);
+ }
}
#endif
#endif
-#endif
+#endif // VP9_COMMON_X86_VP9_POSTPROC_X86_H_
ret
;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%macro HORIZx4_ROW 2
+ movdqa %2, %1
+ pshufb %1, [GLOBAL(shuf_t0t1)]
+ pshufb %2, [GLOBAL(shuf_t2t3)]
+ pmaddubsw %1, xmm6
+ pmaddubsw %2, xmm7
+
+ paddsw %1, %2
+ movdqa %2, %1
+ psrldq %2, 8
+ paddsw %1, %2
+ paddsw %1, xmm5
+ psraw %1, 7
+ packuswb %1, %1
+%endm
%macro HORIZx4 1
mov rdx, arg(5) ;filter ptr
movdqa xmm4, [rdx] ;load filters
movq xmm5, rcx
packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
+ pshuflw xmm6, xmm4, 0b ;k0_k1
+ pshufhw xmm6, xmm6, 10101010b ;k0_k1_k4_k5
+ pshuflw xmm7, xmm4, 01010101b ;k2_k3
+ pshufhw xmm7, xmm7, 11111111b ;k2_k3_k6_k7
+ pshufd xmm5, xmm5, 0 ;rounding
movsxd rax, dword ptr arg(1) ;src_pixels_per_line
movsxd rdx, dword ptr arg(3) ;output_pitch
movsxd rcx, dword ptr arg(4) ;output_height
-
+ shr rcx, 1
.loop:
- movq xmm0, [rsi - 3] ; -3 -2 -1 0 1 2 3 4
-
- movq xmm3, [rsi + 5] ; 5 6 7 8 9 10 11 12
- punpcklqdq xmm0, xmm3
+ ;Do two rows once
+ movq xmm0, [rsi - 3] ;load src
+ movq xmm1, [rsi + 5]
+ movq xmm2, [rsi + rax - 3]
+ movq xmm3, [rsi + rax + 5]
+ punpcklqdq xmm0, xmm1
+ punpcklqdq xmm2, xmm3
+
+ HORIZx4_ROW xmm0, xmm1
+ HORIZx4_ROW xmm2, xmm3
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+ movd xmm3, [rdi + rdx]
+ pavgb xmm2, xmm3
+%endif
+ movd [rdi], xmm0
+ movd [rdi +rdx], xmm2
- movdqa xmm1, xmm0
- pshufb xmm0, [GLOBAL(shuf_t0t1)]
- pmaddubsw xmm0, k0k1
+ lea rsi, [rsi + rax]
+ prefetcht0 [rsi + 4 * rax - 3]
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + 2 * rdx]
+ prefetcht0 [rsi + 2 * rax - 3]
- movdqa xmm2, xmm1
- pshufb xmm1, [GLOBAL(shuf_t2t3)]
- pmaddubsw xmm1, k2k3
+ dec rcx
+ jnz .loop
- movdqa xmm4, xmm2
- pshufb xmm2, [GLOBAL(shuf_t4t5)]
- pmaddubsw xmm2, k4k5
+ ; Do last row if output_height is odd
+ movsxd rcx, dword ptr arg(4) ;output_height
+ and rcx, 1
+ je .done
- pshufb xmm4, [GLOBAL(shuf_t6t7)]
- pmaddubsw xmm4, k6k7
+ movq xmm0, [rsi - 3] ; load src
+ movq xmm1, [rsi + 5]
+ punpcklqdq xmm0, xmm1
- paddsw xmm0, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
+ HORIZx4_ROW xmm0, xmm1
%if %1
movd xmm1, [rdi]
pavgb xmm0, xmm1
%endif
- lea rsi, [rsi + rax]
movd [rdi], xmm0
+.done
+%endm
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
+%macro HORIZx8_ROW 4
+ movdqa %2, %1
+ movdqa %3, %1
+ movdqa %4, %1
+
+ pshufb %1, [GLOBAL(shuf_t0t1)]
+ pshufb %2, [GLOBAL(shuf_t2t3)]
+ pshufb %3, [GLOBAL(shuf_t4t5)]
+ pshufb %4, [GLOBAL(shuf_t6t7)]
+
+ pmaddubsw %1, k0k1
+ pmaddubsw %2, k2k3
+ pmaddubsw %3, k4k5
+ pmaddubsw %4, k6k7
+
+ paddsw %1, %2
+ paddsw %1, %4
+ paddsw %1, %3
+ paddsw %1, krd
+ psraw %1, 7
+ packuswb %1, %1
%endm
%macro HORIZx8 1
movsxd rax, dword ptr arg(1) ;src_pixels_per_line
movsxd rdx, dword ptr arg(3) ;output_pitch
movsxd rcx, dword ptr arg(4) ;output_height
+ shr rcx, 1
.loop:
- movq xmm0, [rsi - 3] ; -3 -2 -1 0 1 2 3 4
-
- movq xmm3, [rsi + 5] ; 5 6 7 8 9 10 11 12
+ movq xmm0, [rsi - 3] ;load src
+ movq xmm3, [rsi + 5]
+ movq xmm4, [rsi + rax - 3]
+ movq xmm7, [rsi + rax + 5]
punpcklqdq xmm0, xmm3
+ punpcklqdq xmm4, xmm7
- movdqa xmm1, xmm0
- pshufb xmm0, [GLOBAL(shuf_t0t1)]
- pmaddubsw xmm0, k0k1
+ HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
+ HORIZx8_ROW xmm4, xmm5, xmm6, xmm7
+%if %1
+ movq xmm1, [rdi]
+ movq xmm2, [rdi + rdx]
+ pavgb xmm0, xmm1
+ pavgb xmm4, xmm2
+%endif
+ movq [rdi], xmm0
+ movq [rdi + rdx], xmm4
- movdqa xmm2, xmm1
- pshufb xmm1, [GLOBAL(shuf_t2t3)]
- pmaddubsw xmm1, k2k3
+ lea rsi, [rsi + rax]
+ prefetcht0 [rsi + 4 * rax - 3]
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + 2 * rdx]
+ prefetcht0 [rsi + 2 * rax - 3]
+ dec rcx
+ jnz .loop
- movdqa xmm4, xmm2
- pshufb xmm2, [GLOBAL(shuf_t4t5)]
- pmaddubsw xmm2, k4k5
+ ;Do last row if output_height is odd
+ movsxd rcx, dword ptr arg(4) ;output_height
+ and rcx, 1
+ je .done
- pshufb xmm4, [GLOBAL(shuf_t6t7)]
- pmaddubsw xmm4, k6k7
+ movq xmm0, [rsi - 3]
+ movq xmm3, [rsi + 5]
+ punpcklqdq xmm0, xmm3
- paddsw xmm0, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
+ HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
%if %1
movq xmm1, [rdi]
pavgb xmm0, xmm1
%endif
-
- lea rsi, [rsi + rax]
movq [rdi], xmm0
-
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
+.done
%endm
%macro HORIZx16 1
movsxd rcx, dword ptr arg(4) ;output_height
.loop:
- movq xmm0, [rsi - 3] ; -3 -2 -1 0 1 2 3 4
+ prefetcht0 [rsi + 2 * rax -3]
- movq xmm3, [rsi + 5] ; 5 6 7 8 9 10 11 12
- punpcklqdq xmm0, xmm3
+ movq xmm0, [rsi - 3] ;load src data
+ movq xmm4, [rsi + 5]
+ movq xmm7, [rsi + 13]
+ punpcklqdq xmm0, xmm4
+ punpcklqdq xmm4, xmm7
movdqa xmm1, xmm0
- pshufb xmm0, [GLOBAL(shuf_t0t1)]
- pmaddubsw xmm0, k0k1
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm5, xmm4
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm4
- movdqa xmm2, xmm1
+ pshufb xmm0, [GLOBAL(shuf_t0t1)]
pshufb xmm1, [GLOBAL(shuf_t2t3)]
- pmaddubsw xmm1, k2k3
-
- movdqa xmm4, xmm2
pshufb xmm2, [GLOBAL(shuf_t4t5)]
- pmaddubsw xmm2, k4k5
+ pshufb xmm3, [GLOBAL(shuf_t6t7)]
+ pshufb xmm4, [GLOBAL(shuf_t0t1)]
+ pshufb xmm5, [GLOBAL(shuf_t2t3)]
+ pshufb xmm6, [GLOBAL(shuf_t4t5)]
+ pshufb xmm7, [GLOBAL(shuf_t6t7)]
- pshufb xmm4, [GLOBAL(shuf_t6t7)]
- pmaddubsw xmm4, k6k7
+ pmaddubsw xmm0, k0k1
+ pmaddubsw xmm1, k2k3
+ pmaddubsw xmm2, k4k5
+ pmaddubsw xmm3, k6k7
+ pmaddubsw xmm4, k0k1
+ pmaddubsw xmm5, k2k3
+ pmaddubsw xmm6, k4k5
+ pmaddubsw xmm7, k6k7
paddsw xmm0, xmm1
- paddsw xmm0, xmm4
+ paddsw xmm0, xmm3
paddsw xmm0, xmm2
+ paddsw xmm4, xmm5
+ paddsw xmm4, xmm7
+ paddsw xmm4, xmm6
+
paddsw xmm0, krd
+ paddsw xmm4, krd
psraw xmm0, 7
+ psraw xmm4, 7
packuswb xmm0, xmm0
-
-
- movq xmm3, [rsi + 5]
- movq xmm7, [rsi + 13]
- punpcklqdq xmm3, xmm7
-
- movdqa xmm1, xmm3
- pshufb xmm3, [GLOBAL(shuf_t0t1)]
- pmaddubsw xmm3, k0k1
-
- movdqa xmm2, xmm1
- pshufb xmm1, [GLOBAL(shuf_t2t3)]
- pmaddubsw xmm1, k2k3
-
- movdqa xmm4, xmm2
- pshufb xmm2, [GLOBAL(shuf_t4t5)]
- pmaddubsw xmm2, k4k5
-
- pshufb xmm4, [GLOBAL(shuf_t6t7)]
- pmaddubsw xmm4, k6k7
-
- paddsw xmm3, xmm1
- paddsw xmm3, xmm4
- paddsw xmm3, xmm2
- paddsw xmm3, krd
- psraw xmm3, 7
- packuswb xmm3, xmm3
- punpcklqdq xmm0, xmm3
+ packuswb xmm4, xmm4
+ punpcklqdq xmm0, xmm4
%if %1
movdqa xmm1, [rdi]
pavgb xmm0, xmm1
push rdi
; end prolog
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
HORIZx4 0
- add rsp, 16*5
- pop rsp
-
; begin epilog
pop rdi
pop rsi
push rdi
; end prolog
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
HORIZx4 1
- add rsp, 16*5
- pop rsp
-
; begin epilog
pop rdi
pop rsi
+++ /dev/null
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
- EXPORT |vp9_add_constant_residual_8x8_neon|
- EXPORT |vp9_add_constant_residual_16x16_neon|
- EXPORT |vp9_add_constant_residual_32x32_neon|
- ARM
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
- MACRO
- LD_16x8 $src, $stride
- vld1.8 {q8}, [$src], $stride
- vld1.8 {q9}, [$src], $stride
- vld1.8 {q10}, [$src], $stride
- vld1.8 {q11}, [$src], $stride
- vld1.8 {q12}, [$src], $stride
- vld1.8 {q13}, [$src], $stride
- vld1.8 {q14}, [$src], $stride
- vld1.8 {q15}, [$src], $stride
- MEND
-
- MACRO
- ADD_DIFF_16x8 $diff
- vqadd.u8 q8, q8, $diff
- vqadd.u8 q9, q9, $diff
- vqadd.u8 q10, q10, $diff
- vqadd.u8 q11, q11, $diff
- vqadd.u8 q12, q12, $diff
- vqadd.u8 q13, q13, $diff
- vqadd.u8 q14, q14, $diff
- vqadd.u8 q15, q15, $diff
- MEND
-
- MACRO
- SUB_DIFF_16x8 $diff
- vqsub.u8 q8, q8, $diff
- vqsub.u8 q9, q9, $diff
- vqsub.u8 q10, q10, $diff
- vqsub.u8 q11, q11, $diff
- vqsub.u8 q12, q12, $diff
- vqsub.u8 q13, q13, $diff
- vqsub.u8 q14, q14, $diff
- vqsub.u8 q15, q15, $diff
- MEND
-
- MACRO
- ST_16x8 $dst, $stride
- vst1.8 {q8}, [$dst], $stride
- vst1.8 {q9}, [$dst], $stride
- vst1.8 {q10}, [$dst], $stride
- vst1.8 {q11}, [$dst], $stride
- vst1.8 {q12}, [$dst], $stride
- vst1.8 {q13}, [$dst], $stride
- vst1.8 {q14}, [$dst], $stride
- vst1.8 {q15}, [$dst], $stride
- MEND
-
-; void add_constant_residual(const int16_t diff, uint8_t *dest, int stride,
-; int width, int height) {
-; int r, c;
-;
-; for (r = 0; r < height; r++) {
-; for (c = 0; c < width; c++)
-; dest[c] = clip_pixel(diff + dest[c]);
-;
-; dest += stride;
-; }
-;}
-;void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest,
-; int stride) {
-; add_constant_residual(diff, dest, stride, 8, 8);
-;}
-; r0 : const int16_t diff
-; r1 : const uint8_t *dest
-; r2 : int stride
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-|vp9_add_constant_residual_8x8_neon| PROC
- mov r3, r1 ; r3: save dest to r3
- vld1.8 {d0}, [r1], r2
- vld1.8 {d1}, [r1], r2
- vld1.8 {d2}, [r1], r2
- vld1.8 {d3}, [r1], r2
- vld1.8 {d4}, [r1], r2
- vld1.8 {d5}, [r1], r2
- vld1.8 {d6}, [r1], r2
- vld1.8 {d7}, [r1], r2
- cmp r0, #0
- bge DIFF_POSITIVE_8x8
-
-DIFF_NEGATIVE_8x8 ; diff < 0
- neg r0, r0
- usat r0, #8, r0
- vdup.u8 q8, r0
-
- vqsub.u8 q0, q0, q8
- vqsub.u8 q1, q1, q8
- vqsub.u8 q2, q2, q8
- vqsub.u8 q3, q3, q8
- b DIFF_SAVE_8x8
-
-DIFF_POSITIVE_8x8 ; diff >= 0
- usat r0, #8, r0
- vdup.u8 q8, r0
-
- vqadd.u8 q0, q0, q8
- vqadd.u8 q1, q1, q8
- vqadd.u8 q2, q2, q8
- vqadd.u8 q3, q3, q8
-
-DIFF_SAVE_8x8
- vst1.8 {d0}, [r3], r2
- vst1.8 {d1}, [r3], r2
- vst1.8 {d2}, [r3], r2
- vst1.8 {d3}, [r3], r2
- vst1.8 {d4}, [r3], r2
- vst1.8 {d5}, [r3], r2
- vst1.8 {d6}, [r3], r2
- vst1.8 {d7}, [r3], r2
-
- bx lr
- ENDP
-
-;void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest,
-; int stride) {
-; add_constant_residual(diff, dest, stride, 16, 16);
-;}
-; r0 : const int16_t diff
-; r1 : const uint8_t *dest
-; r2 : int stride
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-|vp9_add_constant_residual_16x16_neon| PROC
- mov r3, r1
- LD_16x8 r1, r2
- cmp r0, #0
- bge DIFF_POSITIVE_16x16
-
-|DIFF_NEGATIVE_16x16|
- neg r0, r0
- usat r0, #8, r0
- vdup.u8 q0, r0
-
- SUB_DIFF_16x8 q0
- ST_16x8 r3, r2
- LD_16x8 r1, r2
- SUB_DIFF_16x8 q0
- b DIFF_SAVE_16x16
-
-|DIFF_POSITIVE_16x16|
- usat r0, #8, r0
- vdup.u8 q0, r0
-
- ADD_DIFF_16x8 q0
- ST_16x8 r3, r2
- LD_16x8 r1, r2
- ADD_DIFF_16x8 q0
-
-|DIFF_SAVE_16x16|
- ST_16x8 r3, r2
- bx lr
- ENDP
-
-;void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest,
-; int stride) {
-; add_constant_residual(diff, dest, stride, 32, 32);
-;}
-; r0 : const int16_t diff
-; r1 : const uint8_t *dest
-; r2 : int stride
-;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-|vp9_add_constant_residual_32x32_neon| PROC
- push {r4,lr}
- pld [r1]
- mov r3, r1
- add r4, r1, #16 ; r4 dest + 16 for second loop
- cmp r0, #0
- bge DIFF_POSITIVE_32x32
-
-|DIFF_NEGATIVE_32x32|
- neg r0, r0
- usat r0, #8, r0
- vdup.u8 q0, r0
- mov r0, #4
-
-|DIFF_NEGATIVE_32x32_LOOP|
- sub r0, #1
- LD_16x8 r1, r2
- SUB_DIFF_16x8 q0
- ST_16x8 r3, r2
-
- LD_16x8 r1, r2
- SUB_DIFF_16x8 q0
- ST_16x8 r3, r2
- cmp r0, #2
- moveq r1, r4
- moveq r3, r4
- cmp r0, #0
- bne DIFF_NEGATIVE_32x32_LOOP
- pop {r4,pc}
-
-|DIFF_POSITIVE_32x32|
- usat r0, #8, r0
- vdup.u8 q0, r0
- mov r0, #4
-
-|DIFF_POSITIVE_32x32_LOOP|
- sub r0, #1
- LD_16x8 r1, r2
- ADD_DIFF_16x8 q0
- ST_16x8 r3, r2
-
- LD_16x8 r1, r2
- ADD_DIFF_16x8 q0
- ST_16x8 r3, r2
- cmp r0, #2
- moveq r1, r4
- moveq r3, r4
- cmp r0, #0
- bne DIFF_POSITIVE_32x32_LOOP
- pop {r4,pc}
- ENDP
-
- END
return (MB_PREDICTION_MODE)treed_read(r, vp9_intra_mode_tree, p);
}
+static MB_PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, vp9_reader *r,
+ int size_group) {
+ const MB_PREDICTION_MODE y_mode = read_intra_mode(r,
+ cm->fc.y_mode_prob[size_group]);
+ ++cm->counts.y_mode[size_group][y_mode];
+ return y_mode;
+}
+
+static MB_PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, vp9_reader *r,
+ MB_PREDICTION_MODE y_mode) {
+ const MB_PREDICTION_MODE uv_mode = read_intra_mode(r,
+ cm->fc.uv_mode_prob[y_mode]);
+ ++cm->counts.uv_mode[y_mode][uv_mode];
+ return uv_mode;
+}
+
static MB_PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, vp9_reader *r,
uint8_t context) {
- MB_PREDICTION_MODE mode = treed_read(r, vp9_inter_mode_tree,
- cm->fc.inter_mode_probs[context]);
+ const MB_PREDICTION_MODE mode = treed_read(r, vp9_inter_mode_tree,
+ cm->fc.inter_mode_probs[context]);
++cm->counts.inter_mode[context][inter_mode_offset(mode)];
return mode;
}
static int read_mv_component(vp9_reader *r,
const nmv_component *mvcomp, int usehp) {
-
int mag, d, fr, hp;
const int sign = vp9_read(r, mvcomp->sign);
const int mv_class = treed_read(r, vp9_mv_class_tree, mvcomp->classes);
int i, j;
for (j = 0; j < SWITCHABLE_FILTERS + 1; ++j)
for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB,
+ &fc->switchable_interp_prob[j][i]);
}
static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
int i, j;
for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
for (j = 0; j < INTER_MODES - 1; ++j)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &fc->inter_mode_probs[i][j]);
}
static INLINE COMPPREDMODE_TYPE read_comp_pred_mode(vp9_reader *r) {
mbmi->ref_frame[1] = NONE;
if (bsize >= BLOCK_8X8) {
- const int size_group = size_group_lookup[bsize];
- mbmi->mode = read_intra_mode(r, cm->fc.y_mode_prob[size_group]);
- cm->counts.y_mode[size_group][mbmi->mode]++;
+ mbmi->mode = read_intra_mode_y(cm, r, size_group_lookup[bsize]);
} else {
// Only 4x4, 4x8, 8x4 blocks
const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize]; // 1 or 2
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
const int ib = idy * 2 + idx;
- const int b_mode = read_intra_mode(r, cm->fc.y_mode_prob[0]);
+ const int b_mode = read_intra_mode_y(cm, r, 0);
mi->bmi[ib].as_mode = b_mode;
- cm->counts.y_mode[0][b_mode]++;
-
if (num_4x4_h == 2)
mi->bmi[ib + 2].as_mode = b_mode;
if (num_4x4_w == 2)
mbmi->mode = mi->bmi[3].as_mode;
}
- mbmi->uv_mode = read_intra_mode(r, cm->fc.uv_mode_prob[mbmi->mode]);
- cm->counts.uv_mode[mbmi->mode][mbmi->uv_mode]++;
+ mbmi->uv_mode = read_intra_mode_uv(cm, r, mbmi->mode);
+}
+
+static INLINE void assign_mv(VP9_COMMON *cm, MB_PREDICTION_MODE mode,
+ int_mv mv[2], int_mv best_mv[2],
+ int_mv nearest_mv[2], int_mv near_mv[2],
+ int is_compound, int allow_hp, vp9_reader *r) {
+ int i;
+
+ switch (mode) {
+ case NEWMV:
+ read_mv(r, &mv[0].as_mv, &best_mv[0].as_mv,
+ &cm->fc.nmvc, &cm->counts.mv, allow_hp);
+ if (is_compound)
+ read_mv(r, &mv[1].as_mv, &best_mv[1].as_mv,
+ &cm->fc.nmvc, &cm->counts.mv, allow_hp);
+ break;
+ case NEARESTMV:
+ mv[0].as_int = nearest_mv[0].as_int;
+ if (is_compound)
+ mv[1].as_int = nearest_mv[1].as_int;
+ break;
+ case NEARMV:
+ mv[0].as_int = near_mv[0].as_int;
+ if (is_compound)
+ mv[1].as_int = near_mv[1].as_int;
+ break;
+ case ZEROMV:
+ mv[0].as_int = 0;
+ if (is_compound)
+ mv[1].as_int = 0;
+ break;
+ default:
+ assert(!"Invalid inter mode value.");
+ }
+
+ for (i = 0; i < 1 + is_compound; ++i) {
+ assert(mv[i].as_mv.row < MV_UPP && mv[i].as_mv.row > MV_LOW);
+ assert(mv[i].as_mv.col < MV_UPP && mv[i].as_mv.col > MV_LOW);
+ }
}
static int read_is_inter_block(VP9D_COMP *pbi, int segment_id, vp9_reader *r) {
int mi_row, int mi_col, vp9_reader *r) {
VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
- nmv_context *const nmvc = &cm->fc.nmvc;
MB_MODE_INFO *const mbmi = &mi->mbmi;
- int_mv *const mv0 = &mbmi->mv[0];
- int_mv *const mv1 = &mbmi->mv[1];
const BLOCK_SIZE bsize = mbmi->sb_type;
const int allow_hp = xd->allow_high_precision_mv;
- int_mv nearest, nearby, best_mv;
- int_mv nearest_second, nearby_second, best_mv_second;
+ int_mv nearest[2], nearmv[2], best[2];
uint8_t inter_mode_ctx;
MV_REFERENCE_FRAME ref0;
int is_compound;
// nearest, nearby
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {
- vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref0], &nearest, &nearby);
- best_mv.as_int = mbmi->ref_mvs[ref0][0].as_int;
+ vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref0], &nearest[0], &nearmv[0]);
+ best[0].as_int = nearest[0].as_int;
}
if (is_compound) {
ref1, mbmi->ref_mvs[ref1], mi_row, mi_col);
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {
- vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref1],
- &nearest_second, &nearby_second);
- best_mv_second.as_int = mbmi->ref_mvs[ref1][0].as_int;
+ vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref1], &nearest[1], &nearmv[1]);
+ best[1].as_int = nearest[1].as_int;
}
}
int b_mode;
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
- int_mv blockmv, secondmv;
+ int_mv block[2];
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV) {
- vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest, &nearby, j, 0,
+ vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest[0],
+ &nearmv[0], j, 0,
mi_row, mi_col);
if (is_compound)
- vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest_second,
- &nearby_second, j, 1,
- mi_row, mi_col);
+ vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest[1],
+ &nearmv[1], j, 1,
+ mi_row, mi_col);
}
- switch (b_mode) {
- case NEWMV:
- read_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
- &cm->counts.mv, allow_hp);
-
- if (is_compound)
- read_mv(r, &secondmv.as_mv, &best_mv_second.as_mv, nmvc,
- &cm->counts.mv, allow_hp);
- break;
- case NEARESTMV:
- blockmv.as_int = nearest.as_int;
- if (is_compound)
- secondmv.as_int = nearest_second.as_int;
- break;
- case NEARMV:
- blockmv.as_int = nearby.as_int;
- if (is_compound)
- secondmv.as_int = nearby_second.as_int;
- break;
- case ZEROMV:
- blockmv.as_int = 0;
- if (is_compound)
- secondmv.as_int = 0;
- break;
- default:
- assert(!"Invalid inter mode value");
- }
- mi->bmi[j].as_mv[0].as_int = blockmv.as_int;
+ assign_mv(cm, b_mode, block, best, nearest, nearmv,
+ is_compound, allow_hp, r);
+
+ mi->bmi[j].as_mv[0].as_int = block[0].as_int;
if (is_compound)
- mi->bmi[j].as_mv[1].as_int = secondmv.as_int;
+ mi->bmi[j].as_mv[1].as_int = block[1].as_int;
if (num_4x4_h == 2)
mi->bmi[j + 2] = mi->bmi[j];
}
mi->mbmi.mode = b_mode;
- mv0->as_int = mi->bmi[3].as_mv[0].as_int;
- mv1->as_int = mi->bmi[3].as_mv[1].as_int;
- } else {
- switch (mbmi->mode) {
- case NEARMV:
- mv0->as_int = nearby.as_int;
- if (is_compound)
- mv1->as_int = nearby_second.as_int;
- break;
-
- case NEARESTMV:
- mv0->as_int = nearest.as_int;
- if (is_compound)
- mv1->as_int = nearest_second.as_int;
- break;
-
- case ZEROMV:
- mv0->as_int = 0;
- if (is_compound)
- mv1->as_int = 0;
- break;
- case NEWMV:
- read_mv(r, &mv0->as_mv, &best_mv.as_mv, nmvc, &cm->counts.mv, allow_hp);
- if (is_compound)
- read_mv(r, &mv1->as_mv, &best_mv_second.as_mv, nmvc, &cm->counts.mv,
- allow_hp);
- break;
- default:
- assert(!"Invalid inter mode value");
- }
+ mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
+ mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
+ } else {
+ assign_mv(cm, mbmi->mode, mbmi->mv, best, nearest, nearmv,
+ is_compound, allow_hp, r);
}
}
if (cm->comp_pred_mode == HYBRID_PREDICTION)
for (i = 0; i < COMP_INTER_CONTEXTS; i++)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.comp_inter_prob[i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.comp_inter_prob[i]);
if (cm->comp_pred_mode != COMP_PREDICTION_ONLY)
for (i = 0; i < REF_CONTEXTS; i++) {
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][0]);
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][1]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.single_ref_prob[i][0]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.single_ref_prob[i][1]);
}
if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
for (i = 0; i < REF_CONTEXTS; i++)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.comp_ref_prob[i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.comp_ref_prob[i]);
}
void vp9_prepare_read_mode_info(VP9D_COMP* pbi, vp9_reader *r) {
// TODO(jkoleszar): does this clear more than MBSKIP_CONTEXTS? Maybe remove.
// vpx_memset(cm->fc.mbskip_probs, 0, sizeof(cm->fc.mbskip_probs));
for (k = 0; k < MBSKIP_CONTEXTS; ++k)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.mbskip_probs[k]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.mbskip_probs[k]);
if (cm->frame_type != KEY_FRAME && !cm->intra_only) {
nmv_context *const nmvc = &pbi->common.fc.nmvc;
read_switchable_interp_probs(&cm->fc, r);
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.intra_inter_prob[i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.intra_inter_prob[i]);
read_comp_pred(cm, r);
for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
for (i = 0; i < INTRA_MODES - 1; ++i)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.y_mode_prob[j][i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &cm->fc.y_mode_prob[j][i]);
for (j = 0; j < NUM_PARTITION_CONTEXTS; ++j)
for (i = 0; i < PARTITION_TYPES - 1; ++i)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &cm->fc.partition_prob[INTER_FRAME][j][i]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB,
+ &cm->fc.partition_prob[INTER_FRAME][j][i]);
read_mv_probs(r, nmvc, xd->allow_high_precision_mv);
}
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_extend.h"
+#include "vp9/common/vp9_idct.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_quant_common.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/decoder/vp9_detokenize.h"
#include "vp9/decoder/vp9_decodemv.h"
#include "vp9/decoder/vp9_dsubexp.h"
-#include "vp9/decoder/vp9_idct_blk.h"
#include "vp9/decoder/vp9_onyxd_int.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
#include "vp9/decoder/vp9_thread.h"
for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
for (j = 0; j < TX_SIZES - 3; ++j)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &tx_probs->p8x8[i][j]);
for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
for (j = 0; j < TX_SIZES - 2; ++j)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &tx_probs->p16x16[i][j]);
for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
for (j = 0; j < TX_SIZES - 1; ++j)
- if (vp9_read(r, MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
+ vp9_diff_update_prob(r, MODE_UPDATE_PROB, &tx_probs->p32x32[i][j]);
}
static void setup_plane_dequants(VP9_COMMON *cm, MACROBLOCKD *xd, int q_index) {
int16_t* const qcoeff = BLOCK_OFFSET(pd->qcoeff, block);
const int stride = pd->dst.stride;
const int eob = pd->eobs[block];
- const int raster_block = txfrm_block_to_raster_block(plane_bsize, tx_size,
- block);
- uint8_t* const dst = raster_block_offset_uint8(plane_bsize, raster_block,
- pd->dst.buf, stride);
- switch (tx_size) {
- case TX_4X4: {
- const TX_TYPE tx_type = get_tx_type_4x4(pd->plane_type, xd, raster_block);
- if (tx_type == DCT_DCT)
- xd->itxm_add(qcoeff, dst, stride, eob);
+ if (eob > 0) {
+ TX_TYPE tx_type;
+ const int raster_block = txfrm_block_to_raster_block(plane_bsize, tx_size,
+ block);
+ uint8_t* const dst = raster_block_offset_uint8(plane_bsize, raster_block,
+ pd->dst.buf, stride);
+ switch (tx_size) {
+ case TX_4X4:
+ tx_type = get_tx_type_4x4(pd->plane_type, xd, raster_block);
+ if (tx_type == DCT_DCT)
+ xd->itxm_add(qcoeff, dst, stride, eob);
+ else
+ vp9_iht_add(tx_type, qcoeff, dst, stride, eob);
+ break;
+ case TX_8X8:
+ tx_type = get_tx_type_8x8(pd->plane_type, xd);
+ vp9_iht_add_8x8(tx_type, qcoeff, dst, stride, eob);
+ break;
+ case TX_16X16:
+ tx_type = get_tx_type_16x16(pd->plane_type, xd);
+ vp9_iht_add_16x16(tx_type, qcoeff, dst, stride, eob);
+ break;
+ case TX_32X32:
+ tx_type = DCT_DCT;
+ vp9_idct_add_32x32(qcoeff, dst, stride, eob);
+ break;
+ default:
+ assert(!"Invalid transform size");
+ }
+
+ if (eob == 1) {
+ *((int32_t *)qcoeff) = 0;
+ } else {
+ if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
+ vpx_memset(qcoeff, 0, 4 * (4 << tx_size) * sizeof(qcoeff[0]));
else
- vp9_iht_add_c(tx_type, qcoeff, dst, stride, eob);
- break;
+ vpx_memset(qcoeff, 0, (16 << (tx_size << 1)) * sizeof(qcoeff[0]));
}
- case TX_8X8:
- vp9_iht_add_8x8_c(get_tx_type_8x8(pd->plane_type, xd), qcoeff, dst,
- stride, eob);
- break;
- case TX_16X16:
- vp9_iht_add_16x16_c(get_tx_type_16x16(pd->plane_type, xd), qcoeff, dst,
- stride, eob);
- break;
- case TX_32X32:
- vp9_idct_add_32x32(qcoeff, dst, stride, eob);
- break;
- default:
- assert(!"Invalid transform size");
}
}
static void decode_modes_b(VP9D_COMP *pbi, int mi_row, int mi_col,
vp9_reader *r, BLOCK_SIZE bsize) {
- VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
const int less8x8 = bsize < BLOCK_8X8;
MB_MODE_INFO *mbmi;
+ int eobtotal;
if (less8x8)
if (xd->ab_index > 0)
// Has to be called after set_offsets
mbmi = &xd->this_mi->mbmi;
+ eobtotal = decode_tokens(pbi, bsize, r);
if (!is_inter_block(mbmi)) {
// Intra reconstruction
- decode_tokens(pbi, bsize, r);
foreach_transformed_block(xd, bsize, decode_block_intra, xd);
} else {
// Inter reconstruction
- int eobtotal;
+ const int decode_blocks = (eobtotal > 0);
+
+ if (!less8x8) {
+ assert(mbmi->sb_type == bsize);
+ if (eobtotal == 0)
+ mbmi->skip_coeff = 1; // skip loopfilter
+ }
set_ref(pbi, 0, mi_row, mi_col);
if (has_second_ref(mbmi))
set_ref(pbi, 1, mi_row, mi_col);
- vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
+ xd->subpix.filter_x = xd->subpix.filter_y =
+ vp9_get_filter_kernel(mbmi->interp_filter);
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
- eobtotal = decode_tokens(pbi, bsize, r);
- if (less8x8) {
- if (eobtotal >= 0)
- foreach_transformed_block(xd, bsize, decode_block, xd);
- } else {
- assert(mbmi->sb_type == bsize);
- if (eobtotal == 0)
- // skip loopfilter
- vp9_set_pred_flag_mbskip(xd, bsize, 1);
- else if (eobtotal > 0)
- foreach_transformed_block(xd, bsize, decode_block, xd);
- }
+
+ if (decode_blocks)
+ foreach_transformed_block(xd, bsize, decode_block, xd);
}
xd->corrupted |= vp9_reader_has_error(r);
}
for (l = 0; l < PREV_COEF_CONTEXTS; l++)
if (k > 0 || l < 3)
for (m = 0; m < UNCONSTRAINED_NODES; m++)
- if (vp9_read(r, VP9_COEF_UPDATE_PROB))
- vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
+ vp9_diff_update_prob(r, VP9_COEF_UPDATE_PROB,
+ &coef_probs[i][j][k][l][m]);
}
static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode,
static void setup_loopfilter(struct loopfilter *lf,
struct vp9_read_bit_buffer *rb) {
-
lf->filter_level = vp9_rb_read_literal(rb, 6);
lf->sharpness_level = vp9_rb_read_literal(rb, 3);
cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
- xd->itxm_add = xd->lossless ? vp9_idct_add_lossless_c
- : vp9_idct_add;
+ xd->itxm_add = xd->lossless ? vp9_iwht4x4_add : vp9_idct4x4_add;
}
static INTERPOLATIONFILTERTYPE read_interp_filter_type(
struct vp9_read_bit_buffer *rb) {
const INTERPOLATIONFILTERTYPE literal_to_type[] = { EIGHTTAP_SMOOTH,
EIGHTTAP,
- EIGHTTAP_SHARP };
+ EIGHTTAP_SHARP,
+ BILINEAR };
return vp9_rb_read_bit(rb) ? SWITCHABLE
: literal_to_type[vp9_rb_read_literal(rb, 2)];
}
const uint8_t *band_translate;
uint8_t token_cache[1024];
int pt = get_entropy_context(tx_size, A, L);
- get_scan_and_band(xd, tx_size, type, block_idx, &scan, &band_translate);
- nb = vp9_get_coef_neighbors_handle(scan);
+ get_scan_and_band(xd, tx_size, type, block_idx, &scan, &nb, &band_translate);
while (1) {
int val;
if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
break;
-SKIP_START:
+ SKIP_START:
if (c >= seg_eob)
break;
if (c)
206, 207, 208, 209, 210, 211, 212, 213, 215, 216, 217, 218, 219, 220, 221,
222, 223, 224, 225, 226, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237,
238, 239, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252,
-
};
// v = merge_index(v, MAX_PROBS - 1, MODULUS_PARAM);
v = inv_map_table[v];
return word;
}
-void vp9_diff_update_prob(vp9_reader *r, vp9_prob* p) {
- int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
- *p = (vp9_prob)inv_remap_prob(delp, *p);
+void vp9_diff_update_prob(vp9_reader *r, int update_prob, vp9_prob* p) {
+ if (vp9_read(r, update_prob)) {
+ const int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
+ *p = (vp9_prob)inv_remap_prob(delp, *p);
+ }
}
#include "vp9/decoder/vp9_dboolhuff.h"
-void vp9_diff_update_prob(vp9_reader *r, vp9_prob* p);
+void vp9_diff_update_prob(vp9_reader *r, int update_prob, vp9_prob* p);
#endif // VP9_DECODER_VP9_DSUBEXP_H_
+++ /dev/null
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "vp9_rtcd.h"
-#include "vp9/common/vp9_blockd.h"
-#include "vp9/decoder/vp9_idct_blk.h"
-
-static void add_constant_residual(const int16_t diff, uint8_t *dest, int stride,
- int width, int height) {
- int r, c;
-
- for (r = 0; r < height; r++) {
- for (c = 0; c < width; c++)
- dest[c] = clip_pixel(diff + dest[c]);
-
- dest += stride;
- }
-}
-
-void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest,
- int stride) {
- add_constant_residual(diff, dest, stride, 8, 8);
-}
-
-void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest,
- int stride) {
- add_constant_residual(diff, dest, stride, 16, 16);
-}
-
-void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest,
- int stride) {
- add_constant_residual(diff, dest, stride, 32, 32);
-}
-
-void vp9_iht_add_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest, int stride,
- int eob) {
- if (tx_type == DCT_DCT) {
- vp9_idct_add(input, dest, stride, eob);
- } else {
- vp9_short_iht4x4_add(input, dest, stride, tx_type);
- vpx_memset(input, 0, 32);
- }
-}
-
-void vp9_iht_add_8x8_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
- int stride, int eob) {
- if (tx_type == DCT_DCT) {
- vp9_idct_add_8x8(input, dest, stride, eob);
- } else {
- if (eob > 0) {
- vp9_short_iht8x8_add(input, dest, stride, tx_type);
- vpx_memset(input, 0, 128);
- }
- }
-}
-
-void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob) {
- if (eob > 1) {
- vp9_short_idct4x4_add(input, dest, stride);
- vpx_memset(input, 0, 32);
- } else {
- vp9_short_idct4x4_1_add(input, dest, stride);
- ((int *)input)[0] = 0;
- }
-}
-
-void vp9_idct_add_lossless_c(int16_t *input, uint8_t *dest, int stride,
- int eob) {
- if (eob > 1) {
- vp9_short_iwalsh4x4_add(input, dest, stride);
- vpx_memset(input, 0, 32);
- } else {
- vp9_short_iwalsh4x4_1_add_c(input, dest, stride);
- ((int *)input)[0] = 0;
- }
-}
-
-void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob) {
- // If dc is 1, then input[0] is the reconstructed value, do not need
- // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
-
- // The calculation can be simplified if there are not many non-zero dct
- // coefficients. Use eobs to decide what to do.
- // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
- // Combine that with code here.
- if (eob) {
- if (eob == 1) {
- // DC only DCT coefficient
- vp9_short_idct8x8_1_add(input, dest, stride);
- input[0] = 0;
- } else if (eob <= 10) {
- vp9_short_idct10_8x8_add(input, dest, stride);
- vpx_memset(input, 0, 128);
- } else {
- vp9_short_idct8x8_add(input, dest, stride);
- vpx_memset(input, 0, 128);
- }
- }
-}
-
-void vp9_iht_add_16x16_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
- int stride, int eob) {
- if (tx_type == DCT_DCT) {
- vp9_idct_add_16x16(input, dest, stride, eob);
- } else {
- if (eob > 0) {
- vp9_short_iht16x16_add(input, dest, stride, tx_type);
- vpx_memset(input, 0, 512);
- }
- }
-}
-
-void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob) {
- /* The calculation can be simplified if there are not many non-zero dct
- * coefficients. Use eobs to separate different cases. */
- if (eob) {
- if (eob == 1) {
- /* DC only DCT coefficient. */
- vp9_short_idct16x16_1_add(input, dest, stride);
- input[0] = 0;
- } else if (eob <= 10) {
- vp9_short_idct10_16x16_add(input, dest, stride);
- vpx_memset(input, 0, 512);
- } else {
- vp9_short_idct16x16_add(input, dest, stride);
- vpx_memset(input, 0, 512);
- }
- }
-}
-
-void vp9_idct_add_32x32_c(int16_t *input, uint8_t *dest, int stride, int eob) {
- DECLARE_ALIGNED_ARRAY(16, int16_t, output, 1024);
-
- if (eob) {
- if (eob == 1) {
- vp9_short_idct1_32x32(input, output);
- vp9_add_constant_residual_32x32(output[0], dest, stride);
- input[0] = 0;
- } else {
- vp9_short_idct32x32_add(input, dest, stride);
- vpx_memset(input, 0, 2048);
- }
- }
-}
-
+++ /dev/null
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-
-#ifndef VP9_DECODER_VP9_IDCT_BLK_H_
-#define VP9_DECODER_VP9_IDCT_BLK_H_
-
-#include "vp9/common/vp9_blockd.h"
-
-void vp9_idct_add_lossless_c(int16_t *input, uint8_t *dest,
- int stride, int eob);
-
-void vp9_iht_add_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
- int stride, int eob);
-
-void vp9_iht_add_8x8_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
- int stride, int eob);
-
-void vp9_iht_add_16x16_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
- int stride, int eob);
-
-#endif // VP9_DECODER_VP9_IDCT_BLK_H_
* be found in the AUTHORS file in the root of the source tree.
*/
-#ifndef VP9_COMMON_VP9_ONYXD_H_
-#define VP9_COMMON_VP9_ONYXD_H_
+#ifndef VP9_DECODER_VP9_ONYXD_H_
+#define VP9_DECODER_VP9_ONYXD_H_
#ifdef __cplusplus
extern "C" {
void vp9_initialize_dec();
int vp9_receive_compressed_data(VP9D_PTR comp,
- uint64_t size, const uint8_t **dest,
+ size_t size, const uint8_t **dest,
int64_t time_stamp);
int vp9_get_raw_frame(VP9D_PTR comp, YV12_BUFFER_CONFIG *sd,
}
#endif
-#endif // VP9_COMMON_VP9_ONYXD_H_
+#endif // VP9_DECODER_VP9_ONYXD_H_
#endif
#if WRITE_RECON_BUFFER == 2
void write_dx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame) {
-
// write the frame
FILE *yframe;
int i;
char filename[255];
- sprintf(filename, "dx\\y%04d.raw", this_frame);
+ snprintf(filename, sizeof(filename)-1, "dx\\y%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->y_height; i++)
frame->y_width, 1, yframe);
fclose(yframe);
- sprintf(filename, "dx\\u%04d.raw", this_frame);
+ snprintf(filename, sizeof(filename)-1, "dx\\u%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->uv_height; i++)
frame->uv_width, 1, yframe);
fclose(yframe);
- sprintf(filename, "dx\\v%04d.raw", this_frame);
+ snprintf(filename, sizeof(filename)-1, "dx\\v%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->uv_height; i++)
* vpxenc --test-decode functionality working, and will be replaced in a
* later commit that adds VP9-specific controls for this functionality.
*/
- if (ref_frame_flag == VP9_LAST_FLAG)
+ if (ref_frame_flag == VP9_LAST_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[0];
- else if (ref_frame_flag == VP9_GOLD_FLAG)
+ } else if (ref_frame_flag == VP9_GOLD_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[1];
- else if (ref_frame_flag == VP9_ALT_FLAG)
+ } else if (ref_frame_flag == VP9_ALT_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[2];
- else {
+ } else {
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Invalid reference frame");
return pbi->common.error.error_code;
}
int vp9_receive_compressed_data(VP9D_PTR ptr,
- uint64_t size, const uint8_t **psource,
+ size_t size, const uint8_t **psource,
int64_t time_stamp) {
VP9D_COMP *pbi = (VP9D_COMP *) ptr;
VP9_COMMON *cm = &pbi->common;
VP9D_CONFIG oxcf;
const uint8_t *source;
- uint32_t source_sz;
+ size_t source_sz;
int64_t last_time_stamp;
int ready_for_new_data;
VP9Worker lf_worker;
} VP9D_COMP;
-#endif // VP9_DECODER_VP9_TREEREADER_H_
+#endif // VP9_DECODER_VP9_ONYXD_INT_H_
* be found in the AUTHORS file in the root of the source tree.
*/
-#ifndef VP9_READ_BIT_BUFFER_
-#define VP9_READ_BIT_BUFFER_
+#ifndef VP9_DECODER_VP9_READ_BIT_BUFFER_H_
+#define VP9_DECODER_VP9_READ_BIT_BUFFER_H_
#include <limits.h>
return vp9_rb_read_bit(rb) ? -value : value;
}
-#endif // VP9_READ_BIT_BUFFER_
+#endif // VP9_DECODER_VP9_READ_BIT_BUFFER_H_
//------------------------------------------------------------------------------
// simplistic pthread emulation layer
-#include <process.h>
+#include <process.h> // NOLINT
// _beginthreadex requires __stdcall
#define THREADFN unsigned int __stdcall
#ifndef VP9_DECODER_VP9_THREAD_H_
#define VP9_DECODER_VP9_THREAD_H_
-#include "vpx_config.h"
+#include "./vpx_config.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#if defined(_WIN32)
-#include <windows.h>
+#include <windows.h> // NOLINT
typedef HANDLE pthread_t;
typedef CRITICAL_SECTION pthread_mutex_t;
typedef struct {
#else
-#include <pthread.h>
+#include <pthread.h> // NOLINT
#endif /* _WIN32 */
#endif /* CONFIG_MULTITHREAD */
} // extern "C"
#endif
-#endif /* VP9_DECODER_VP9_THREAD_H_ */
+#endif // VP9_DECODER_VP9_THREAD_H_
const vp9_prob *const p) {
register vp9_tree_index i = 0;
- while ((i = t[ i + vp9_read(r, p[i >> 1])]) > 0);
+ while ((i = t[ i + vp9_read(r, p[i >> 1])]) > 0)
+ continue;
return -i;
}
+++ /dev/null
-/*
- * Copyright (c) 2012 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <assert.h>
-#include <emmintrin.h> // SSE2
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_idct.h"
-
-void vp9_add_constant_residual_8x8_sse2(const int16_t diff, uint8_t *dest,
- int stride) {
- uint8_t abs_diff;
- __m128i d;
-
- // Prediction data.
- __m128i p0 = _mm_loadl_epi64((const __m128i *)(dest + 0 * stride));
- __m128i p1 = _mm_loadl_epi64((const __m128i *)(dest + 1 * stride));
- __m128i p2 = _mm_loadl_epi64((const __m128i *)(dest + 2 * stride));
- __m128i p3 = _mm_loadl_epi64((const __m128i *)(dest + 3 * stride));
- __m128i p4 = _mm_loadl_epi64((const __m128i *)(dest + 4 * stride));
- __m128i p5 = _mm_loadl_epi64((const __m128i *)(dest + 5 * stride));
- __m128i p6 = _mm_loadl_epi64((const __m128i *)(dest + 6 * stride));
- __m128i p7 = _mm_loadl_epi64((const __m128i *)(dest + 7 * stride));
-
- p0 = _mm_unpacklo_epi64(p0, p1);
- p2 = _mm_unpacklo_epi64(p2, p3);
- p4 = _mm_unpacklo_epi64(p4, p5);
- p6 = _mm_unpacklo_epi64(p6, p7);
-
- // Clip diff value to [0, 255] range. Then, do addition or subtraction
- // according to its sign.
- if (diff >= 0) {
- abs_diff = (diff > 255) ? 255 : diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
-
- p0 = _mm_adds_epu8(p0, d);
- p2 = _mm_adds_epu8(p2, d);
- p4 = _mm_adds_epu8(p4, d);
- p6 = _mm_adds_epu8(p6, d);
- } else {
- abs_diff = (diff < -255) ? 255 : -diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
-
- p0 = _mm_subs_epu8(p0, d);
- p2 = _mm_subs_epu8(p2, d);
- p4 = _mm_subs_epu8(p4, d);
- p6 = _mm_subs_epu8(p6, d);
- }
-
- _mm_storel_epi64((__m128i *)(dest + 0 * stride), p0);
- p0 = _mm_srli_si128(p0, 8);
- _mm_storel_epi64((__m128i *)(dest + 1 * stride), p0);
-
- _mm_storel_epi64((__m128i *)(dest + 2 * stride), p2);
- p2 = _mm_srli_si128(p2, 8);
- _mm_storel_epi64((__m128i *)(dest + 3 * stride), p2);
-
- _mm_storel_epi64((__m128i *)(dest + 4 * stride), p4);
- p4 = _mm_srli_si128(p4, 8);
- _mm_storel_epi64((__m128i *)(dest + 5 * stride), p4);
-
- _mm_storel_epi64((__m128i *)(dest + 6 * stride), p6);
- p6 = _mm_srli_si128(p6, 8);
- _mm_storel_epi64((__m128i *)(dest + 7 * stride), p6);
-}
-
-void vp9_add_constant_residual_16x16_sse2(const int16_t diff, uint8_t *dest,
- int stride) {
- uint8_t abs_diff;
- __m128i d;
-
- // Prediction data.
- __m128i p0 = _mm_load_si128((const __m128i *)(dest + 0 * stride));
- __m128i p1 = _mm_load_si128((const __m128i *)(dest + 1 * stride));
- __m128i p2 = _mm_load_si128((const __m128i *)(dest + 2 * stride));
- __m128i p3 = _mm_load_si128((const __m128i *)(dest + 3 * stride));
- __m128i p4 = _mm_load_si128((const __m128i *)(dest + 4 * stride));
- __m128i p5 = _mm_load_si128((const __m128i *)(dest + 5 * stride));
- __m128i p6 = _mm_load_si128((const __m128i *)(dest + 6 * stride));
- __m128i p7 = _mm_load_si128((const __m128i *)(dest + 7 * stride));
- __m128i p8 = _mm_load_si128((const __m128i *)(dest + 8 * stride));
- __m128i p9 = _mm_load_si128((const __m128i *)(dest + 9 * stride));
- __m128i p10 = _mm_load_si128((const __m128i *)(dest + 10 * stride));
- __m128i p11 = _mm_load_si128((const __m128i *)(dest + 11 * stride));
- __m128i p12 = _mm_load_si128((const __m128i *)(dest + 12 * stride));
- __m128i p13 = _mm_load_si128((const __m128i *)(dest + 13 * stride));
- __m128i p14 = _mm_load_si128((const __m128i *)(dest + 14 * stride));
- __m128i p15 = _mm_load_si128((const __m128i *)(dest + 15 * stride));
-
- // Clip diff value to [0, 255] range. Then, do addition or subtraction
- // according to its sign.
- if (diff >= 0) {
- abs_diff = (diff > 255) ? 255 : diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
-
- p0 = _mm_adds_epu8(p0, d);
- p1 = _mm_adds_epu8(p1, d);
- p2 = _mm_adds_epu8(p2, d);
- p3 = _mm_adds_epu8(p3, d);
- p4 = _mm_adds_epu8(p4, d);
- p5 = _mm_adds_epu8(p5, d);
- p6 = _mm_adds_epu8(p6, d);
- p7 = _mm_adds_epu8(p7, d);
- p8 = _mm_adds_epu8(p8, d);
- p9 = _mm_adds_epu8(p9, d);
- p10 = _mm_adds_epu8(p10, d);
- p11 = _mm_adds_epu8(p11, d);
- p12 = _mm_adds_epu8(p12, d);
- p13 = _mm_adds_epu8(p13, d);
- p14 = _mm_adds_epu8(p14, d);
- p15 = _mm_adds_epu8(p15, d);
- } else {
- abs_diff = (diff < -255) ? 255 : -diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
-
- p0 = _mm_subs_epu8(p0, d);
- p1 = _mm_subs_epu8(p1, d);
- p2 = _mm_subs_epu8(p2, d);
- p3 = _mm_subs_epu8(p3, d);
- p4 = _mm_subs_epu8(p4, d);
- p5 = _mm_subs_epu8(p5, d);
- p6 = _mm_subs_epu8(p6, d);
- p7 = _mm_subs_epu8(p7, d);
- p8 = _mm_subs_epu8(p8, d);
- p9 = _mm_subs_epu8(p9, d);
- p10 = _mm_subs_epu8(p10, d);
- p11 = _mm_subs_epu8(p11, d);
- p12 = _mm_subs_epu8(p12, d);
- p13 = _mm_subs_epu8(p13, d);
- p14 = _mm_subs_epu8(p14, d);
- p15 = _mm_subs_epu8(p15, d);
- }
-
- // Store results
- _mm_store_si128((__m128i *)(dest + 0 * stride), p0);
- _mm_store_si128((__m128i *)(dest + 1 * stride), p1);
- _mm_store_si128((__m128i *)(dest + 2 * stride), p2);
- _mm_store_si128((__m128i *)(dest + 3 * stride), p3);
- _mm_store_si128((__m128i *)(dest + 4 * stride), p4);
- _mm_store_si128((__m128i *)(dest + 5 * stride), p5);
- _mm_store_si128((__m128i *)(dest + 6 * stride), p6);
- _mm_store_si128((__m128i *)(dest + 7 * stride), p7);
- _mm_store_si128((__m128i *)(dest + 8 * stride), p8);
- _mm_store_si128((__m128i *)(dest + 9 * stride), p9);
- _mm_store_si128((__m128i *)(dest + 10 * stride), p10);
- _mm_store_si128((__m128i *)(dest + 11 * stride), p11);
- _mm_store_si128((__m128i *)(dest + 12 * stride), p12);
- _mm_store_si128((__m128i *)(dest + 13 * stride), p13);
- _mm_store_si128((__m128i *)(dest + 14 * stride), p14);
- _mm_store_si128((__m128i *)(dest + 15 * stride), p15);
-}
-
-void vp9_add_constant_residual_32x32_sse2(const int16_t diff, uint8_t *dest,
- int stride) {
- uint8_t abs_diff;
- __m128i d;
- int i = 8;
-
- if (diff >= 0) {
- abs_diff = (diff > 255) ? 255 : diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
- } else {
- abs_diff = (diff < -255) ? 255 : -diff;
- d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
- }
-
- do {
- // Prediction data.
- __m128i p0 = _mm_load_si128((const __m128i *)(dest + 0 * stride));
- __m128i p1 = _mm_load_si128((const __m128i *)(dest + 0 * stride + 16));
- __m128i p2 = _mm_load_si128((const __m128i *)(dest + 1 * stride));
- __m128i p3 = _mm_load_si128((const __m128i *)(dest + 1 * stride + 16));
- __m128i p4 = _mm_load_si128((const __m128i *)(dest + 2 * stride));
- __m128i p5 = _mm_load_si128((const __m128i *)(dest + 2 * stride + 16));
- __m128i p6 = _mm_load_si128((const __m128i *)(dest + 3 * stride));
- __m128i p7 = _mm_load_si128((const __m128i *)(dest + 3 * stride + 16));
-
- // Clip diff value to [0, 255] range. Then, do addition or subtraction
- // according to its sign.
- if (diff >= 0) {
- p0 = _mm_adds_epu8(p0, d);
- p1 = _mm_adds_epu8(p1, d);
- p2 = _mm_adds_epu8(p2, d);
- p3 = _mm_adds_epu8(p3, d);
- p4 = _mm_adds_epu8(p4, d);
- p5 = _mm_adds_epu8(p5, d);
- p6 = _mm_adds_epu8(p6, d);
- p7 = _mm_adds_epu8(p7, d);
- } else {
- p0 = _mm_subs_epu8(p0, d);
- p1 = _mm_subs_epu8(p1, d);
- p2 = _mm_subs_epu8(p2, d);
- p3 = _mm_subs_epu8(p3, d);
- p4 = _mm_subs_epu8(p4, d);
- p5 = _mm_subs_epu8(p5, d);
- p6 = _mm_subs_epu8(p6, d);
- p7 = _mm_subs_epu8(p7, d);
- }
-
- // Store results
- _mm_store_si128((__m128i *)(dest + 0 * stride), p0);
- _mm_store_si128((__m128i *)(dest + 0 * stride + 16), p1);
- _mm_store_si128((__m128i *)(dest + 1 * stride), p2);
- _mm_store_si128((__m128i *)(dest + 1 * stride + 16), p3);
- _mm_store_si128((__m128i *)(dest + 2 * stride), p4);
- _mm_store_si128((__m128i *)(dest + 2 * stride + 16), p5);
- _mm_store_si128((__m128i *)(dest + 3 * stride), p6);
- _mm_store_si128((__m128i *)(dest + 3 * stride + 16), p7);
-
- dest += 4 * stride;
- } while (--i);
-}
static void write_interp_filter_type(INTERPOLATIONFILTERTYPE type,
struct vp9_write_bit_buffer *wb) {
- const int type_to_literal[] = { 1, 0, 2 };
+ const int type_to_literal[] = { 1, 0, 2, 3 };
vp9_wb_write_bit(wb, type == SWITCHABLE);
if (type != SWITCHABLE)
} PARTITION_INFO;
// Structure to hold snapshot of coding context during the mode picking process
-// TODO Do we need all of these?
typedef struct {
MODE_INFO mic;
PARTITION_INFO partition_info;
#endif
const unsigned int vp9_prob_cost[256] = {
- 2047, 2047, 1791, 1641, 1535, 1452, 1385, 1328, 1279, 1235, 1196, 1161, 1129, 1099, 1072, 1046,
- 1023, 1000, 979, 959, 940, 922, 905, 889, 873, 858, 843, 829, 816, 803, 790, 778,
- 767, 755, 744, 733, 723, 713, 703, 693, 684, 675, 666, 657, 649, 641, 633, 625,
- 617, 609, 602, 594, 587, 580, 573, 567, 560, 553, 547, 541, 534, 528, 522, 516,
- 511, 505, 499, 494, 488, 483, 477, 472, 467, 462, 457, 452, 447, 442, 437, 433,
- 428, 424, 419, 415, 410, 406, 401, 397, 393, 389, 385, 381, 377, 373, 369, 365,
- 361, 357, 353, 349, 346, 342, 338, 335, 331, 328, 324, 321, 317, 314, 311, 307,
- 304, 301, 297, 294, 291, 288, 285, 281, 278, 275, 272, 269, 266, 263, 260, 257,
- 255, 252, 249, 246, 243, 240, 238, 235, 232, 229, 227, 224, 221, 219, 216, 214,
- 211, 208, 206, 203, 201, 198, 196, 194, 191, 189, 186, 184, 181, 179, 177, 174,
- 172, 170, 168, 165, 163, 161, 159, 156, 154, 152, 150, 148, 145, 143, 141, 139,
- 137, 135, 133, 131, 129, 127, 125, 123, 121, 119, 117, 115, 113, 111, 109, 107,
- 105, 103, 101, 99, 97, 95, 93, 92, 90, 88, 86, 84, 82, 81, 79, 77,
- 75, 73, 72, 70, 68, 66, 65, 63, 61, 60, 58, 56, 55, 53, 51, 50,
- 48, 46, 45, 43, 41, 40, 38, 37, 35, 33, 32, 30, 29, 27, 25, 24,
- 22, 21, 19, 18, 16, 15, 13, 12, 10, 9, 7, 6, 4, 3, 1, 1
-};
+ 2047, 2047, 1791, 1641, 1535, 1452, 1385, 1328, 1279, 1235, 1196, 1161,
+ 1129, 1099, 1072, 1046, 1023, 1000, 979, 959, 940, 922, 905, 889,
+ 873, 858, 843, 829, 816, 803, 790, 778, 767, 755, 744, 733,
+ 723, 713, 703, 693, 684, 675, 666, 657, 649, 641, 633, 625,
+ 617, 609, 602, 594, 587, 580, 573, 567, 560, 553, 547, 541,
+ 534, 528, 522, 516, 511, 505, 499, 494, 488, 483, 477, 472,
+ 467, 462, 457, 452, 447, 442, 437, 433, 428, 424, 419, 415,
+ 410, 406, 401, 397, 393, 389, 385, 381, 377, 373, 369, 365,
+ 361, 357, 353, 349, 346, 342, 338, 335, 331, 328, 324, 321,
+ 317, 314, 311, 307, 304, 301, 297, 294, 291, 288, 285, 281,
+ 278, 275, 272, 269, 266, 263, 260, 257, 255, 252, 249, 246,
+ 243, 240, 238, 235, 232, 229, 227, 224, 221, 219, 216, 214,
+ 211, 208, 206, 203, 201, 198, 196, 194, 191, 189, 186, 184,
+ 181, 179, 177, 174, 172, 170, 168, 165, 163, 161, 159, 156,
+ 154, 152, 150, 148, 145, 143, 141, 139, 137, 135, 133, 131,
+ 129, 127, 125, 123, 121, 119, 117, 115, 113, 111, 109, 107,
+ 105, 103, 101, 99, 97, 95, 93, 92, 90, 88, 86, 84,
+ 82, 81, 79, 77, 75, 73, 72, 70, 68, 66, 65, 63,
+ 61, 60, 58, 56, 55, 53, 51, 50, 48, 46, 45, 43,
+ 41, 40, 38, 37, 35, 33, 32, 30, 29, 27, 25, 24,
+ 22, 21, 19, 18, 16, 15, 13, 12, 10, 9, 7, 6,
+ 4, 3, 1, 1};
void vp9_start_encode(vp9_writer *br, uint8_t *source) {
br->lowvalue = 0;
/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
pixel. */
-void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
+void vp9_short_walsh4x4_c(int16_t *input, int16_t *output, int pitch) {
int i;
int a1, b1, c1, d1, e1;
- short *ip = input;
- short *op = output;
+ int16_t *ip = input;
+ int16_t *op = output;
int pitch_short = pitch >> 1;
for (i = 0; i < 4; i++) {
}
}
-void vp9_short_walsh8x4_c(short *input, short *output, int pitch) {
+void vp9_short_walsh8x4_c(int16_t *input, int16_t *output, int pitch) {
vp9_short_walsh4x4_c(input, output, pitch);
vp9_short_walsh4x4_c(input + 4, output + 16, pitch);
}
tmp = sortlist[j - 1];
sortlist[j - 1] = sortlist[j];
sortlist[j] = tmp;
- } else
- break;
+ } else {
+ break;
+ }
}
}
#if OUTPUT_NORM_ACT_STATS
fprintf(f, "\n");
#endif
-
}
#if OUTPUT_NORM_ACT_STATS
fclose(f);
#endif
-
}
#endif // USE_ACT_INDEX
// Calculate an activity index number of each mb
calc_activity_index(cpi, x);
#endif
-
}
// Macroblock activity masking
const int mi_height = num_8x8_blocks_high_lookup[bsize];
assert(mi->mbmi.mode < MB_MODE_COUNT);
- assert(mb_mode_index < MAX_MODES);
assert(mi->mbmi.ref_frame[0] < MAX_REF_FRAMES);
assert(mi->mbmi.ref_frame[1] < MAX_REF_FRAMES);
assert(mi->mbmi.sb_type == bsize);
THR_D207_PRED /*D207_PRED*/,
THR_D63_PRED /*D63_PRED*/,
THR_TM /*TM_PRED*/,
- THR_B_PRED /*I4X4_PRED*/,
};
cpi->mode_chosen_counts[kf_mode_index[mi->mbmi.mode]]++;
#endif
}
void vp9_setup_src_planes(MACROBLOCK *x, const YV12_BUFFER_CONFIG *src,
- int mb_row, int mb_col) {
- uint8_t *buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer, src
- ->alpha_buffer};
- int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride, src
- ->alpha_stride};
+ int mi_row, int mi_col) {
+ uint8_t *const buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+ src->alpha_buffer};
+ const int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+ src->alpha_stride};
int i;
- for (i = 0; i < MAX_MB_PLANE; i++) {
- setup_pred_plane(&x->plane[i].src, buffers[i], strides[i], mb_row, mb_col,
+ for (i = 0; i < MAX_MB_PLANE; i++)
+ setup_pred_plane(&x->plane[i].src, buffers[i], strides[i], mi_row, mi_col,
NULL, x->e_mbd.plane[i].subsampling_x,
x->e_mbd.plane[i].subsampling_y);
- }
}
static void set_offsets(VP9_COMP *cpi, int mi_row, int mi_col,
// cannot be used.
xd->last_mi = cm->prev_mi ? xd->prev_mi_8x8[0] : NULL;
- xd->this_mi =
- xd->mi_8x8[0] = cm->mi + idx_str;
+ xd->this_mi = xd->mi_8x8[0] = cm->mi + idx_str;
mbmi = &xd->this_mi->mbmi;
// Find best coding mode & reconstruct the MB so it is available
// as a predictor for MBs that follow in the SB
- if (cm->frame_type == KEY_FRAME)
+ if (cm->frame_type == KEY_FRAME) {
vp9_rd_pick_intra_mode_sb(cpi, x, totalrate, totaldist, bsize, ctx,
best_rd);
- else
- vp9_rd_pick_inter_mode_sb(cpi, x, mi_row, mi_col, totalrate, totaldist,
- bsize, ctx, best_rd);
+ } else {
+ if (bsize >= BLOCK_8X8)
+ vp9_rd_pick_inter_mode_sb(cpi, x, mi_row, mi_col, totalrate, totaldist,
+ bsize, ctx, best_rd);
+ else
+ vp9_rd_pick_inter_mode_sub8x8(cpi, x, mi_row, mi_col, totalrate,
+ totaldist, bsize, ctx, best_rd);
+ }
}
static void update_stats(VP9_COMP *cpi) {
return &x->ab4x4_context[xd->sb_index][xd->mb_index][xd->b_index];
default:
assert(0);
- return NULL ;
+ return NULL;
}
}
return &x->b_partitioning[xd->sb_index][xd->mb_index][xd->b_index];
default:
assert(0);
- return NULL ;
+ return NULL;
}
}
}
}
-static void set_block_size(VP9_COMMON * const cm, MODE_INFO **mi_8x8,
- BLOCK_SIZE bsize, int mis, int mi_row,
- int mi_col) {
- int r, c;
- const int bs = MAX(num_8x8_blocks_wide_lookup[bsize],
- num_8x8_blocks_high_lookup[bsize]);
- const int idx_str = mis * mi_row + mi_col;
- MODE_INFO **const mi2 = &mi_8x8[idx_str];
-
- mi2[0] = cm->mi + idx_str;
- mi2[0]->mbmi.sb_type = bsize;
-
- for (r = 0; r < bs; r++)
- for (c = 0; c < bs; c++)
- if (mi_row + r < cm->mi_rows && mi_col + c < cm->mi_cols)
- mi2[r * mis + c] = mi2[0];
-}
-
-typedef struct {
- int64_t sum_square_error;
- int64_t sum_error;
- int count;
- int variance;
-} var;
-
-typedef struct {
- var none;
- var horz[2];
- var vert[2];
-} partition_variance;
-
-#define VT(TYPE, BLOCKSIZE) \
- typedef struct { \
- partition_variance vt; \
- BLOCKSIZE split[4]; } TYPE;
-
-VT(v8x8, var)
-VT(v16x16, v8x8)
-VT(v32x32, v16x16)
-VT(v64x64, v32x32)
-
-typedef struct {
- partition_variance *vt;
- var *split[4];
-} vt_node;
-
-typedef enum {
- V16X16,
- V32X32,
- V64X64,
-} TREE_LEVEL;
-
-static void tree_to_node(void *data, BLOCK_SIZE bsize, vt_node *node) {
- int i;
- switch (bsize) {
- case BLOCK_64X64: {
- v64x64 *vt = (v64x64 *) data;
- node->vt = &vt->vt;
- for (i = 0; i < 4; i++)
- node->split[i] = &vt->split[i].vt.none;
- break;
- }
- case BLOCK_32X32: {
- v32x32 *vt = (v32x32 *) data;
- node->vt = &vt->vt;
- for (i = 0; i < 4; i++)
- node->split[i] = &vt->split[i].vt.none;
- break;
- }
- case BLOCK_16X16: {
- v16x16 *vt = (v16x16 *) data;
- node->vt = &vt->vt;
- for (i = 0; i < 4; i++)
- node->split[i] = &vt->split[i].vt.none;
- break;
- }
- case BLOCK_8X8: {
- v8x8 *vt = (v8x8 *) data;
- node->vt = &vt->vt;
- for (i = 0; i < 4; i++)
- node->split[i] = &vt->split[i];
- break;
- }
- default:
- node->vt = 0;
- for (i = 0; i < 4; i++)
- node->split[i] = 0;
- assert(-1);
- }
-}
-
-// Set variance values given sum square error, sum error, count.
-static void fill_variance(var *v, int64_t s2, int64_t s, int c) {
- v->sum_square_error = s2;
- v->sum_error = s;
- v->count = c;
- if (c > 0)
- v->variance = (int)(256
- * (v->sum_square_error - v->sum_error * v->sum_error / v->count)
- / v->count);
- else
- v->variance = 0;
-}
-
-// Combine 2 variance structures by summing the sum_error, sum_square_error,
-// and counts and then calculating the new variance.
-void sum_2_variances(var *r, var *a, var*b) {
- fill_variance(r, a->sum_square_error + b->sum_square_error,
- a->sum_error + b->sum_error, a->count + b->count);
-}
-
-static void fill_variance_tree(void *data, BLOCK_SIZE bsize) {
- vt_node node;
- tree_to_node(data, bsize, &node);
- sum_2_variances(&node.vt->horz[0], node.split[0], node.split[1]);
- sum_2_variances(&node.vt->horz[1], node.split[2], node.split[3]);
- sum_2_variances(&node.vt->vert[0], node.split[0], node.split[2]);
- sum_2_variances(&node.vt->vert[1], node.split[1], node.split[3]);
- sum_2_variances(&node.vt->none, &node.vt->vert[0], &node.vt->vert[1]);
-}
-
-#if PERFORM_RANDOM_PARTITIONING
-static int set_vt_partitioning(VP9_COMP *cpi, void *data, MODE_INFO *m,
- BLOCK_SIZE block_size, int mi_row,
- int mi_col, int mi_size) {
- VP9_COMMON * const cm = &cpi->common;
- vt_node vt;
- const int mis = cm->mode_info_stride;
- int64_t threshold = 4 * cpi->common.base_qindex * cpi->common.base_qindex;
-
- tree_to_node(data, block_size, &vt);
-
- // split none is available only if we have more than half a block size
- // in width and height inside the visible image
- if (mi_col + mi_size < cm->mi_cols && mi_row + mi_size < cm->mi_rows &&
- (rand() & 3) < 1) {
- set_block_size(cm, m, block_size, mis, mi_row, mi_col);
- return 1;
- }
-
- // vertical split is available on all but the bottom border
- if (mi_row + mi_size < cm->mi_rows && vt.vt->vert[0].variance < threshold
- && (rand() & 3) < 1) {
- set_block_size(cm, m, get_subsize(block_size, PARTITION_VERT), mis, mi_row,
- mi_col);
- return 1;
- }
-
- // horizontal split is available on all but the right border
- if (mi_col + mi_size < cm->mi_cols && vt.vt->horz[0].variance < threshold
- && (rand() & 3) < 1) {
- set_block_size(cm, m, get_subsize(block_size, PARTITION_HORZ), mis, mi_row,
- mi_col);
- return 1;
- }
-
- return 0;
-}
-
-#else // !PERFORM_RANDOM_PARTITIONING
-
-static int set_vt_partitioning(VP9_COMP *cpi, void *data, MODE_INFO **m,
- BLOCK_SIZE bsize, int mi_row,
- int mi_col, int mi_size) {
- VP9_COMMON * const cm = &cpi->common;
- vt_node vt;
- const int mis = cm->mode_info_stride;
- int64_t threshold = 50 * cpi->common.base_qindex;
-
- tree_to_node(data, bsize, &vt);
-
- // split none is available only if we have more than half a block size
- // in width and height inside the visible image
- if (mi_col + mi_size < cm->mi_cols && mi_row + mi_size < cm->mi_rows
- && vt.vt->none.variance < threshold) {
- set_block_size(cm, m, bsize, mis, mi_row, mi_col);
- return 1;
- }
-
- // vertical split is available on all but the bottom border
- if (mi_row + mi_size < cm->mi_rows && vt.vt->vert[0].variance < threshold
- && vt.vt->vert[1].variance < threshold) {
- set_block_size(cm, m, get_subsize(bsize, PARTITION_VERT), mis, mi_row,
- mi_col);
- return 1;
- }
-
- // horizontal split is available on all but the right border
- if (mi_col + mi_size < cm->mi_cols && vt.vt->horz[0].variance < threshold
- && vt.vt->horz[1].variance < threshold) {
- set_block_size(cm, m, get_subsize(bsize, PARTITION_HORZ), mis, mi_row,
- mi_col);
- return 1;
- }
-
- return 0;
-}
-#endif // PERFORM_RANDOM_PARTITIONING
-
-static void choose_partitioning(VP9_COMP *cpi, MODE_INFO **mi_8x8,
- int mi_row, int mi_col) {
- VP9_COMMON * const cm = &cpi->common;
- MACROBLOCK *x = &cpi->mb;
- MACROBLOCKD *xd = &cpi->mb.e_mbd;
+static int sb_has_motion(VP9_COMP *cpi, MODE_INFO **prev_mi_8x8) {
+ VP9_COMMON *const cm = &cpi->common;
const int mis = cm->mode_info_stride;
- // TODO(JBB): More experimentation or testing of this threshold;
- int64_t threshold = 4;
- int i, j, k;
- v64x64 vt;
- unsigned char * s;
- int sp;
- const unsigned char * d;
- int dp;
- int pixels_wide = 64, pixels_high = 64;
-
- vp9_zero(vt);
- set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
-
- if (xd->mb_to_right_edge < 0)
- pixels_wide += (xd->mb_to_right_edge >> 3);
-
- if (xd->mb_to_bottom_edge < 0)
- pixels_high += (xd->mb_to_bottom_edge >> 3);
-
- s = x->plane[0].src.buf;
- sp = x->plane[0].src.stride;
-
- // TODO(JBB): Clearly the higher the quantizer the fewer partitions we want
- // but this needs more experimentation.
- threshold = threshold * cpi->common.base_qindex * cpi->common.base_qindex;
-
- d = vp9_64x64_zeros;
- dp = 64;
- if (cm->frame_type != KEY_FRAME) {
- int_mv nearest_mv, near_mv;
- const int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, LAST_FRAME)];
- YV12_BUFFER_CONFIG *ref_fb = &cm->yv12_fb[idx];
- YV12_BUFFER_CONFIG *second_ref_fb = NULL;
-
- setup_pre_planes(xd, 0, ref_fb, mi_row, mi_col,
- &xd->scale_factor[0]);
- setup_pre_planes(xd, 1, second_ref_fb, mi_row, mi_col,
- &xd->scale_factor[1]);
+ int block_row, block_col;
- xd->this_mi->mbmi.ref_frame[0] = LAST_FRAME;
- xd->this_mi->mbmi.sb_type = BLOCK_64X64;
- vp9_find_best_ref_mvs(xd,
- mi_8x8[0]->mbmi.ref_mvs[mi_8x8[0]->mbmi.ref_frame[0]],
- &nearest_mv, &near_mv);
-
- xd->this_mi->mbmi.mv[0] = nearest_mv;
- vp9_build_inter_predictors_sby(xd, mi_row, mi_col, BLOCK_64X64);
-
- d = xd->plane[0].dst.buf;
- dp = xd->plane[0].dst.stride;
- }
-
- // Fill in the entire tree of 8x8 variances for splits.
- for (i = 0; i < 4; i++) {
- const int x32_idx = ((i & 1) << 5);
- const int y32_idx = ((i >> 1) << 5);
- for (j = 0; j < 4; j++) {
- const int x16_idx = x32_idx + ((j & 1) << 4);
- const int y16_idx = y32_idx + ((j >> 1) << 4);
- v16x16 *vst = &vt.split[i].split[j];
- for (k = 0; k < 4; k++) {
- int x_idx = x16_idx + ((k & 1) << 3);
- int y_idx = y16_idx + ((k >> 1) << 3);
- unsigned int sse = 0;
- int sum = 0;
- if (x_idx < pixels_wide && y_idx < pixels_high)
- vp9_get_sse_sum_8x8(s + y_idx * sp + x_idx, sp,
- d + y_idx * dp + x_idx, dp, &sse, &sum);
- fill_variance(&vst->split[k].vt.none, sse, sum, 64);
- }
- }
- }
- // Fill the rest of the variance tree by summing the split partition
- // values.
- for (i = 0; i < 4; i++) {
- for (j = 0; j < 4; j++) {
- fill_variance_tree(&vt.split[i].split[j], BLOCK_16X16);
- }
- fill_variance_tree(&vt.split[i], BLOCK_32X32);
- }
- fill_variance_tree(&vt, BLOCK_64X64);
- // Now go through the entire structure, splitting every block size until
- // we get to one that's got a variance lower than our threshold, or we
- // hit 8x8.
- if (!set_vt_partitioning(cpi, &vt, mi_8x8, BLOCK_64X64, mi_row, mi_col,
- 4)) {
- for (i = 0; i < 4; ++i) {
- const int x32_idx = ((i & 1) << 2);
- const int y32_idx = ((i >> 1) << 2);
- if (!set_vt_partitioning(cpi, &vt.split[i], mi_8x8, BLOCK_32X32,
- (mi_row + y32_idx), (mi_col + x32_idx), 2)) {
- for (j = 0; j < 4; ++j) {
- const int x16_idx = ((j & 1) << 1);
- const int y16_idx = ((j >> 1) << 1);
- if (!set_vt_partitioning(cpi, &vt.split[i].split[j], mi_8x8,
- BLOCK_16X16,
- (mi_row + y32_idx + y16_idx),
- (mi_col + x32_idx + x16_idx), 1)) {
- for (k = 0; k < 4; ++k) {
- const int x8_idx = (k & 1);
- const int y8_idx = (k >> 1);
- set_block_size(cm, mi_8x8, BLOCK_8X8, mis,
- (mi_row + y32_idx + y16_idx + y8_idx),
- (mi_col + x32_idx + x16_idx + x8_idx));
- }
- }
+ if (cm->prev_mi) {
+ for (block_row = 0; block_row < 8; ++block_row) {
+ for (block_col = 0; block_col < 8; ++block_col) {
+ MODE_INFO * prev_mi = prev_mi_8x8[block_row * mis + block_col];
+ if (prev_mi) {
+ if (abs(prev_mi->mbmi.mv[0].as_mv.row) >= 8 ||
+ abs(prev_mi->mbmi.mv[0].as_mv.col) >= 8)
+ return 1;
}
}
}
}
+ return 0;
}
static void rd_use_partition(VP9_COMP *cpi, MODE_INFO **mi_8x8,
static void rd_auto_partition_range(VP9_COMP *cpi, int row, int col,
BLOCK_SIZE *min_block_size,
BLOCK_SIZE *max_block_size) {
+ VP9_COMMON * const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
MODE_INFO ** mi_8x8 = xd->mi_8x8;
+ MODE_INFO ** prev_mi_8x8 = xd->prev_mi_8x8;
+
const int left_in_image = xd->left_available && mi_8x8[-1];
const int above_in_image = xd->up_available &&
mi_8x8[-xd->mode_info_stride];
MODE_INFO ** above_sb64_mi_8x8;
MODE_INFO ** left_sb64_mi_8x8;
- // Frequency check
- if (cpi->sf.auto_min_max_partition_count <= 0) {
- cpi->sf.auto_min_max_partition_count =
- cpi->sf.auto_min_max_partition_interval;
+ int row8x8_remaining = cm->cur_tile_mi_row_end - row;
+ int col8x8_remaining = cm->cur_tile_mi_col_end - col;
+ int bh, bw;
+
+ // Trap case where we do not have a prediction.
+ if (!left_in_image && !above_in_image &&
+ ((cm->frame_type == KEY_FRAME) || !cm->prev_mi)) {
*min_block_size = BLOCK_4X4;
*max_block_size = BLOCK_64X64;
} else {
- --cpi->sf.auto_min_max_partition_count;
-
- // Set default values if no left or above neighbour
- if (!left_in_image && !above_in_image) {
- *min_block_size = BLOCK_4X4;
- *max_block_size = BLOCK_64X64;
- } else {
- VP9_COMMON *const cm = &cpi->common;
- int row8x8_remaining = cm->cur_tile_mi_row_end - row;
- int col8x8_remaining = cm->cur_tile_mi_col_end - col;
- int bh, bw;
-
- // Default "min to max" and "max to min"
- *min_block_size = BLOCK_64X64;
- *max_block_size = BLOCK_4X4;
-
- // Find the min and max partition sizes used in the left SB64
- if (left_in_image) {
- left_sb64_mi_8x8 = &mi_8x8[-MI_BLOCK_SIZE];
- get_sb_partition_size_range(cpi, left_sb64_mi_8x8,
- min_block_size, max_block_size);
- }
-
- // Find the min and max partition sizes used in the above SB64 taking
- // the values found for left as a starting point.
- if (above_in_image) {
- above_sb64_mi_8x8 = &mi_8x8[-xd->mode_info_stride * MI_BLOCK_SIZE];
- get_sb_partition_size_range(cpi, above_sb64_mi_8x8,
- min_block_size, max_block_size);
- }
+ // Default "min to max" and "max to min"
+ *min_block_size = BLOCK_64X64;
+ *max_block_size = BLOCK_4X4;
+
+ // NOTE: each call to get_sb_partition_size_range() uses the previous
+ // passed in values for min and max as a starting point.
+ //
+ // Find the min and max partition used in previous frame at this location
+ if (cm->prev_mi && (cm->frame_type != KEY_FRAME)) {
+ get_sb_partition_size_range(cpi, prev_mi_8x8,
+ min_block_size, max_block_size);
+ }
- // Give a bit of leaway either side of the observed min and max
- *min_block_size = min_partition_size[*min_block_size];
- *max_block_size = max_partition_size[*max_block_size];
+ // Find the min and max partition sizes used in the left SB64
+ if (left_in_image) {
+ left_sb64_mi_8x8 = &mi_8x8[-MI_BLOCK_SIZE];
+ get_sb_partition_size_range(cpi, left_sb64_mi_8x8,
+ min_block_size, max_block_size);
+ }
- // Check border cases where max and min from neighbours may not be legal.
- *max_block_size = find_partition_size(*max_block_size,
- row8x8_remaining, col8x8_remaining,
- &bh, &bw);
- *min_block_size = MIN(*min_block_size, *max_block_size);
+ // Find the min and max partition sizes used in the above SB64.
+ if (above_in_image) {
+ above_sb64_mi_8x8 = &mi_8x8[-xd->mode_info_stride * MI_BLOCK_SIZE];
+ get_sb_partition_size_range(cpi, above_sb64_mi_8x8,
+ min_block_size, max_block_size);
}
}
+
+ // Give a bit of leaway either side of the observed min and max
+ *min_block_size = min_partition_size[*min_block_size];
+ *max_block_size = max_partition_size[*max_block_size];
+
+ // Check border cases where max and min from neighbours may not be legal.
+ *max_block_size = find_partition_size(*max_block_size,
+ row8x8_remaining, col8x8_remaining,
+ &bh, &bw);
+ *min_block_size = MIN(*min_block_size, *max_block_size);
}
static void compute_fast_motion_search_level(VP9_COMP *cpi, BLOCK_SIZE bsize) {
best_dist = sum_dist;
best_rd = sum_rd;
*(get_sb_partitioning(x, bsize)) = subsize;
- } else {
- // skip rectangular partition test when larger block size
- // gives better rd cost
- if (cpi->sf.less_rectangular_check)
- do_rect &= !partition_none_allowed;
}
+ } else {
+ // skip rectangular partition test when larger block size
+ // gives better rd cost
+ if (cpi->sf.less_rectangular_check)
+ do_rect &= !partition_none_allowed;
}
partition_split_done = 1;
restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
int dummy_rate;
int64_t dummy_dist;
- vpx_memset(cpi->mb.pred_mv, 0, sizeof(cpi->mb.pred_mv));
+ vp9_zero(cpi->mb.pred_mv);
if (cpi->sf.reference_masking)
rd_pick_reference_frame(cpi, mi_row, mi_col);
- if (cpi->sf.partition_by_variance || cpi->sf.use_lastframe_partitioning ||
+ if (cpi->sf.use_lastframe_partitioning ||
cpi->sf.use_one_partition_size_always ) {
const int idx_str = cm->mode_info_stride * mi_row + mi_col;
MODE_INFO **mi_8x8 = cm->mi_grid_visible + idx_str;
set_partitioning(cpi, mi_8x8, mi_row, mi_col);
rd_use_partition(cpi, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
&dummy_rate, &dummy_dist, 1);
- } else if (cpi->sf.partition_by_variance) {
- choose_partitioning(cpi, cm->mi_grid_visible, mi_row, mi_col);
- rd_use_partition(cpi, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
- &dummy_rate, &dummy_dist, 1);
} else {
if ((cpi->common.current_video_frame
% cpi->sf.last_partitioning_redo_frequency) == 0
|| cm->prev_mi == 0
|| cpi->common.show_frame == 0
|| cpi->common.frame_type == KEY_FRAME
- || cpi->is_src_frame_alt_ref) {
+ || cpi->is_src_frame_alt_ref
+ || ((cpi->sf.use_lastframe_partitioning ==
+ LAST_FRAME_PARTITION_LOW_MOTION) &&
+ sb_has_motion(cpi, prev_mi_8x8))) {
// If required set upper and lower partition size limits
if (cpi->sf.auto_min_max_partition_size) {
set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
// printf("Switching to lossless\n");
cpi->mb.fwd_txm8x4 = vp9_short_walsh8x4;
cpi->mb.fwd_txm4x4 = vp9_short_walsh4x4;
- cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_short_iwalsh4x4_1_add;
- cpi->mb.e_mbd.inv_txm4x4_add = vp9_short_iwalsh4x4_add;
+ cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_iwht4x4_1_add;
+ cpi->mb.e_mbd.inv_txm4x4_add = vp9_iwht4x4_16_add;
cpi->mb.optimize = 0;
cpi->common.lf.filter_level = 0;
cpi->zbin_mode_boost_enabled = 0;
// printf("Not lossless\n");
cpi->mb.fwd_txm8x4 = vp9_short_fdct8x4;
cpi->mb.fwd_txm4x4 = vp9_short_fdct4x4;
- cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_short_idct4x4_1_add;
- cpi->mb.e_mbd.inv_txm4x4_add = vp9_short_idct4x4_add;
+ cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_idct4x4_1_add;
+ cpi->mb.e_mbd.inv_txm4x4_add = vp9_idct4x4_16_add;
}
}
xd->mi_8x8 = cm->mi_grid_visible;
// required for vp9_frame_init_quantizer
- xd->this_mi =
- xd->mi_8x8[0] = cm->mi;
+ xd->this_mi = xd->mi_8x8[0] = cm->mi;
xd->mic_stream_ptr = cm->mi;
xd->last_mi = cm->prev_mi;
-
vp9_zero(cpi->NMVcount);
vp9_zero(cpi->coef_counts);
vp9_zero(cm->counts.eob_branch);
// Keep record of the total distortion this time around for future use
cpi->last_frame_distortion = cpi->frame_distortion;
#endif
-
}
static int check_dual_ref_flags(VP9_COMP *cpi) {
int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, mbmi->ref_frame[0])];
YV12_BUFFER_CONFIG *ref_fb = &cm->yv12_fb[idx];
YV12_BUFFER_CONFIG *second_ref_fb = NULL;
- if (mbmi->ref_frame[1] > 0) {
+ if (has_second_ref(mbmi)) {
idx = cm->ref_frame_map[get_ref_frame_idx(cpi, mbmi->ref_frame[1])];
second_ref_fb = &cm->yv12_fb[idx];
}
void vp9_setup_src_planes(struct macroblock *x,
const struct yv12_buffer_config *src,
- int mb_row, int mb_col);
+ int mi_row, int mi_col);
#endif // VP9_ENCODER_VP9_ENCODEFRAME_H_
*/
#include "./vpx_config.h"
-#include "vp9_rtcd.h"
+#include "./vp9_rtcd.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/encoder/vp9_encodemb.h"
* be found in the AUTHORS file in the root of the source tree.
*/
+
+#include "./vp9_rtcd.h"
#include "./vpx_config.h"
-#include "vp9/encoder/vp9_encodemb.h"
+
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/common/vp9_idct.h"
#include "vp9/common/vp9_reconinter.h"
-#include "vp9/encoder/vp9_quantize.h"
-#include "vp9/encoder/vp9_tokenize.h"
#include "vp9/common/vp9_reconintra.h"
-#include "vpx_mem/vpx_mem.h"
-#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/common/vp9_systemdependent.h"
-#include "vp9_rtcd.h"
+
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/encoder/vp9_tokenize.h"
DECLARE_ALIGNED(16, extern const uint8_t,
vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
xd->inv_txm4x4_add(dqcoeff, dest, stride);
}
-static void inverse_transform_b_8x8_add(int eob,
- int16_t *dqcoeff, uint8_t *dest,
- int stride) {
- if (eob <= 1)
- vp9_short_idct8x8_1_add(dqcoeff, dest, stride);
- else if (eob <= 10)
- vp9_short_idct10_8x8_add(dqcoeff, dest, stride);
- else
- vp9_short_idct8x8_add(dqcoeff, dest, stride);
-}
-
-static void inverse_transform_b_16x16_add(int eob,
- int16_t *dqcoeff, uint8_t *dest,
- int stride) {
- if (eob <= 1)
- vp9_short_idct16x16_1_add(dqcoeff, dest, stride);
- else if (eob <= 10)
- vp9_short_idct10_16x16_add(dqcoeff, dest, stride);
- else
- vp9_short_idct16x16_add(dqcoeff, dest, stride);
-}
-
static void subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
struct macroblock_plane *const p = &x->plane[plane];
const MACROBLOCKD *const xd = &x->e_mbd;
vp9_subtract_sbuv(x, bsize);
}
-
-#define RDTRUNC(RM,DM,R,D) ( (128+(R)*(RM)) & 0xFF )
+#define RDTRUNC(RM, DM, R, D) ((128 + (R) * (RM)) & 0xFF)
typedef struct vp9_token_state vp9_token_state;
struct vp9_token_state {
short qc;
};
-// TODO: experiments to find optimal multiple numbers
+// TODO(jimbankoski): experiment to find optimal RD numbers.
#define Y1_RD_MULT 4
#define UV_RD_MULT 2
assert((!type && !plane) || (type && plane));
dqcoeff_ptr = BLOCK_OFFSET(pd->dqcoeff, block);
qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
- get_scan_and_band(xd, tx_size, type, ib, &scan, &band_translate);
+ get_scan_and_band(xd, tx_size, type, ib, &scan, &nb, &band_translate);
assert(eob <= default_eob);
/* Now set up a Viterbi trellis to evaluate alternative roundings. */
for (i = 0; i < eob; i++)
token_cache[scan[i]] = vp9_pt_energy_class[vp9_dct_value_tokens_ptr[
qcoeff_ptr[scan[i]]].token];
- nb = vp9_get_coef_neighbors_handle(scan);
for (i = eob; i-- > i0;) {
int base_bits, d2, dx;
best_index[i][1] = best;
/* Finally, make this the new head of the trellis. */
next = i;
- }
- /* There's no choice to make for a zero coefficient, so we don't
- * add a new trellis node, but we do need to update the costs.
- */
- else {
+ } else {
+ /* There's no choice to make for a zero coefficient, so we don't
+ * add a new trellis node, but we do need to update the costs.
+ */
band = get_coef_band(band_translate, i + 1);
t0 = tokens[next][0].token;
t1 = tokens[next][1].token;
vp9_short_idct32x32_add(dqcoeff, dst, pd->dst.stride);
break;
case TX_16X16:
- inverse_transform_b_16x16_add(pd->eobs[block], dqcoeff, dst,
- pd->dst.stride);
+ vp9_idct_add_16x16(dqcoeff, dst, pd->dst.stride, pd->eobs[block]);
break;
case TX_8X8:
- inverse_transform_b_8x8_add(pd->eobs[block], dqcoeff, dst,
- pd->dst.stride);
+ vp9_idct_add_8x8(dqcoeff, dst, pd->dst.stride, pd->eobs[block]);
break;
case TX_4X4:
// this is like vp9_short_idct4x4 but has a special case around eob<=1
vp9_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan, iscan);
- if (!x->skip_encode && *eob) {
- if (tx_type == DCT_DCT)
- inverse_transform_b_16x16_add(*eob, dqcoeff, dst, pd->dst.stride);
- else
- vp9_short_iht16x16_add(dqcoeff, dst, pd->dst.stride, tx_type);
- }
+ if (!x->skip_encode && *eob)
+ vp9_iht_add_16x16(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
break;
case TX_8X8:
tx_type = get_tx_type_8x8(pd->plane_type, xd);
vp9_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan, iscan);
- if (!x->skip_encode && *eob) {
- if (tx_type == DCT_DCT)
- inverse_transform_b_8x8_add(*eob, dqcoeff, dst, pd->dst.stride);
- else
- vp9_short_iht8x8_add(dqcoeff, dst, pd->dst.stride, tx_type);
- }
+ if (!x->skip_encode && *eob)
+ vp9_iht_add_8x8(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
break;
case TX_4X4:
tx_type = get_tx_type_4x4(pd->plane_type, xd, block);
#include "vp9/encoder/vp9_onyx_int.h"
#include "vp9/common/vp9_onyxc_int.h"
-typedef enum {
- RD_DC_PRED = DC_PRED,
- RD_V_PRED = V_PRED,
- RD_H_PRED = H_PRED,
- RD_D45_PRED = D45_PRED,
- RD_D135_PRED = D135_PRED,
- RD_D117_PRED = D117_PRED,
- RD_D153_PRED = D153_PRED,
- RD_D207_PRED = D207_PRED,
- RD_D63_PRED = D63_PRED,
- RD_TM_PRED = TM_PRED,
- RD_NEARESTMV = NEARESTMV,
- RD_NEARMV = NEARMV,
- RD_ZEROMV = ZEROMV,
- RD_NEWMV = NEWMV,
- RD_I4X4_PRED,
- RD_SPLITMV,
- RD_MODE_COUNT
-} RD_PREDICTION_MODE;
-
typedef struct {
- RD_PREDICTION_MODE mode;
+ MB_PREDICTION_MODE mode;
MV_REFERENCE_FRAME ref_frame;
MV_REFERENCE_FRAME second_ref_frame;
} MODE_DEFINITION;
+typedef struct {
+ MV_REFERENCE_FRAME ref_frame;
+ MV_REFERENCE_FRAME second_ref_frame;
+} REF_DEFINITION;
+
struct optimize_ctx {
ENTROPY_CONTEXT ta[MAX_MB_PLANE][16];
ENTROPY_CONTEXT tl[MAX_MB_PLANE][16];
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <math.h>
#include "vp9/common/vp9_common.h"
-#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_encodemv.h"
-#include <math.h>
#ifdef ENTROPY_STATS
extern unsigned int active_section;
mb_row << 1,
1 << mi_height_log2(xd->this_mi->mbmi.sb_type),
mb_col << 1,
- 1 << mi_height_log2(xd->this_mi->mbmi.sb_type));
+ 1 << mi_width_log2(xd->this_mi->mbmi.sb_type));
// do intra 16x16 prediction
this_error = vp9_encode_intra(x, use_dc_pred);
#include <assert.h>
#include <stdlib.h>
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/common/vp9_common.h"
#include "vp9/encoder/vp9_lookahead.h"
#include "vp9/common/vp9_extend.h"
goto bail;
}
return ctx;
-bail:
+ bail:
vp9_lookahead_destroy(ctx);
return NULL;
}
ref_full.as_mv.row = ref_mv->as_mv.row >> 3;
/*cpi->sf.search_method == HEX*/
- best_err = vp9_hex_search(x, &ref_full, step_param, x->errorperbit,
+ best_err = vp9_hex_search(x, &ref_full.as_mv, step_param, x->errorperbit,
0, &v_fn_ptr,
- 0, ref_mv, dst_mv);
+ 0, &ref_mv->as_mv, &dst_mv->as_mv);
// Try sub-pixel MC
// if (bestsme > error_thresh && bestsme < INT_MAX)
dst_mv->as_int = tmp_mv.as_int;
}
- // If the current best reference mv is not centred on 0,0 then do a 0,0 based search as well
+ // If the current best reference mv is not centered on 0,0 then do a 0,0
+ // based search as well.
if (ref_mv->as_int) {
unsigned int tmp_err;
int_mv zero_ref_mv, tmp_mv;
stats->ref[GOLDEN_FRAME].m.mv.as_int = 0;
}
- // Alt-ref frame MV search, if it exists and is different than last/golden frame
+ // Do an Alt-ref frame MV search, if it exists and is different than
+ // last/golden frame.
if (alt_ref) {
int a_motion_error;
xd->plane[0].pre[0].buf = alt_ref->y_buffer + mb_y_offset;
int_mv arf_top_mv, gld_top_mv;
MODE_INFO mi_local = { { 0 } };
- // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+ // Set up limit values for motion vectors to prevent them extending outside
+ // the UMV borders.
arf_top_mv.as_int = 0;
gld_top_mv.as_int = 0;
x->mv_row_min = -BORDER_MV_PIXELS_B16;
int arf_y_in_offset = arf_y_offset;
int gld_y_in_offset = gld_y_offset;
- // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+ // Set up limit values for motion vectors to prevent them extending outside
+ // the UMV borders.
arf_left_mv.as_int = arf_top_mv.as_int;
gld_left_mv.as_int = gld_top_mv.as_int;
x->mv_col_min = -BORDER_MV_PIXELS_B16;
for (i = 0; i < n_frames; i++) {
MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
vpx_memset(frame_stats->mb_stats, 0,
- cm->mb_rows * cm->mb_cols * sizeof(*cpi->mbgraph_stats[i].mb_stats));
+ cm->mb_rows * cm->mb_cols *
+ sizeof(*cpi->mbgraph_stats[i].mb_stats));
}
// do motion search to find contribution of each reference to data
#define CHECK_POINT \
{\
- if (this_mv.as_mv.col < x->mv_col_min) continue;\
- if (this_mv.as_mv.col > x->mv_col_max) continue;\
- if (this_mv.as_mv.row < x->mv_row_min) continue;\
- if (this_mv.as_mv.row > x->mv_row_max) continue;\
+ if (this_mv.col < x->mv_col_min) continue;\
+ if (this_mv.col > x->mv_col_max) continue;\
+ if (this_mv.row < x->mv_row_min) continue;\
+ if (this_mv.row > x->mv_row_max) continue;\
}
#define CHECK_BETTER \
if (thissad < bestsad)\
{\
if (use_mvcost) \
- thissad += mvsad_err_cost(&this_mv.as_mv, &fcenter_mv.as_mv, \
+ thissad += mvsad_err_cost(&this_mv, &fcenter_mv.as_mv, \
mvjsadcost, mvsadcost, \
sad_per_bit);\
if (thissad < bestsad)\
// candidates as indicated in the num_candidates and candidates arrays
// passed into this function
static int vp9_pattern_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int sad_per_bit,
int do_init_search,
int do_refine,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
- int_mv *center_mv, int_mv *best_mv,
+ const MV *center_mv, MV *best_mv,
const int num_candidates[MAX_PATTERN_SCALES],
const MV candidates[MAX_PATTERN_SCALES]
[MAX_PATTERN_CANDIDATES]) {
int what_stride = x->plane[0].src.stride;
int in_what_stride = xd->plane[0].pre[0].stride;
int br, bc;
- int_mv this_mv;
+ MV this_mv;
int bestsad = INT_MAX;
int thissad;
uint8_t *base_offset;
int *mvjsadcost = x->nmvjointsadcost;
int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
- fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
- fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+ fcenter_mv.as_mv.row = center_mv->row >> 3;
+ fcenter_mv.as_mv.col = center_mv->col >> 3;
// adjust ref_mv to make sure it is within MV range
- clamp_mv(&ref_mv->as_mv,
- x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
- br = ref_mv->as_mv.row;
- bc = ref_mv->as_mv.col;
+ clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+ br = ref_mv->row;
+ bc = ref_mv->col;
// Work out the start point for the search
base_offset = (uint8_t *)(xd->plane[0].pre[0].buf);
this_offset = base_offset + (br * in_what_stride) + bc;
- this_mv.as_mv.row = br;
- this_mv.as_mv.col = bc;
+ this_mv.row = br;
+ this_mv.col = bc;
bestsad = vfp->sdf(what, what_stride, this_offset, in_what_stride, 0x7fffffff)
- + mvsad_err_cost(&this_mv.as_mv, &fcenter_mv.as_mv,
+ + mvsad_err_cost(&this_mv, &fcenter_mv.as_mv,
mvjsadcost, mvsadcost, sad_per_bit);
// Search all possible scales upto the search param around the center point
CHECK_BOUNDS((1 << t))
if (all_in) {
for (i = 0; i < num_candidates[t]; i++) {
- this_mv.as_mv.row = br + candidates[t][i].row;
- this_mv.as_mv.col = bc + candidates[t][i].col;
- this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) +
- this_mv.as_mv.col;
+ this_mv.row = br + candidates[t][i].row;
+ this_mv.col = bc + candidates[t][i].col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
}
} else {
for (i = 0; i < num_candidates[t]; i++) {
- this_mv.as_mv.row = br + candidates[t][i].row;
- this_mv.as_mv.col = bc + candidates[t][i].col;
+ this_mv.row = br + candidates[t][i].row;
+ this_mv.col = bc + candidates[t][i].col;
CHECK_POINT
- this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) +
- this_mv.as_mv.col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
CHECK_BOUNDS((1 << s))
if (all_in) {
for (i = 0; i < num_candidates[s]; i++) {
- this_mv.as_mv.row = br + candidates[s][i].row;
- this_mv.as_mv.col = bc + candidates[s][i].col;
- this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) +
- this_mv.as_mv.col;
+ this_mv.row = br + candidates[s][i].row;
+ this_mv.col = bc + candidates[s][i].col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
}
} else {
for (i = 0; i < num_candidates[s]; i++) {
- this_mv.as_mv.row = br + candidates[s][i].row;
- this_mv.as_mv.col = bc + candidates[s][i].col;
+ this_mv.row = br + candidates[s][i].row;
+ this_mv.col = bc + candidates[s][i].col;
CHECK_POINT
- this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) +
- this_mv.as_mv.col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
get_next_chkpts(next_chkpts_indices, k, num_candidates[s]);
if (all_in) {
for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
- this_mv.as_mv.row = br +
- candidates[s][next_chkpts_indices[i]].row;
- this_mv.as_mv.col = bc +
- candidates[s][next_chkpts_indices[i]].col;
- this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) +
- this_mv.as_mv.col;
+ this_mv.row = br + candidates[s][next_chkpts_indices[i]].row;
+ this_mv.col = bc + candidates[s][next_chkpts_indices[i]].col;
+ this_offset = base_offset + (this_mv.row * (in_what_stride)) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
}
} else {
for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
- this_mv.as_mv.row = br +
- candidates[s][next_chkpts_indices[i]].row;
- this_mv.as_mv.col = bc +
- candidates[s][next_chkpts_indices[i]].col;
+ this_mv.row = br + candidates[s][next_chkpts_indices[i]].row;
+ this_mv.col = bc + candidates[s][next_chkpts_indices[i]].col;
CHECK_POINT
- this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) +
- this_mv.as_mv.col;
+ this_offset = base_offset + (this_mv.row * (in_what_stride)) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
CHECK_BOUNDS(1)
if (all_in) {
for (i = 0; i < 4; i++) {
- this_mv.as_mv.row = br + neighbors[i].row;
- this_mv.as_mv.col = bc + neighbors[i].col;
- this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) +
- this_mv.as_mv.col;
+ this_mv.row = br + neighbors[i].row;
+ this_mv.col = bc + neighbors[i].col;
+ this_offset = base_offset + (this_mv.row * (in_what_stride)) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
}
} else {
for (i = 0; i < 4; i++) {
- this_mv.as_mv.row = br + neighbors[i].row;
- this_mv.as_mv.col = bc + neighbors[i].col;
+ this_mv.row = br + neighbors[i].row;
+ this_mv.col = bc + neighbors[i].col;
CHECK_POINT
- this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) +
- this_mv.as_mv.col;
+ this_offset = base_offset + (this_mv.row * (in_what_stride)) +
+ this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
bestsad);
CHECK_BETTER
}
}
- best_mv->as_mv.row = br;
- best_mv->as_mv.col = bc;
+ best_mv->row = br;
+ best_mv->col = bc;
- this_offset = base_offset + (best_mv->as_mv.row * (in_what_stride)) +
- best_mv->as_mv.col;
- this_mv.as_mv.row = best_mv->as_mv.row * 8;
- this_mv.as_mv.col = best_mv->as_mv.col * 8;
+ this_offset = base_offset + (best_mv->row * in_what_stride) +
+ best_mv->col;
+ this_mv.row = best_mv->row * 8;
+ this_mv.col = best_mv->col * 8;
if (bestsad == INT_MAX)
return INT_MAX;
return vfp->vf(what, what_stride, this_offset, in_what_stride,
(unsigned int *)&bestsad) +
- use_mvcost ? mv_err_cost(&this_mv.as_mv, ¢er_mv->as_mv,
+ use_mvcost ? mv_err_cost(&this_mv, center_mv,
x->nmvjointcost, x->mvcost, x->errorperbit)
: 0;
}
int vp9_hex_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int sad_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
- int_mv *center_mv, int_mv *best_mv) {
+ const MV *center_mv, MV *best_mv) {
// First scale has 8-closest points, the rest have 6 points in hex shape
// at increasing scales
static const int hex_num_candidates[MAX_PATTERN_SCALES] = {
}
int vp9_bigdia_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int sad_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
- int_mv *center_mv,
- int_mv *best_mv) {
+ const MV *center_mv,
+ MV *best_mv) {
// First scale has 4-closest points, the rest have 8 points in diamond
// shape at increasing scales
static const int bigdia_num_candidates[MAX_PATTERN_SCALES] = {
{{-512, -512}, {0, -1024}, {512, -512}, {1024, 0}, {512, 512}, {0, 1024},
{-512, 512}, {-1024, 0}},
};
- return
- vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
- do_init_search, 0, vfp, use_mvcost,
- center_mv, best_mv,
- bigdia_num_candidates, bigdia_candidates);
+ return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
+ do_init_search, 0, vfp, use_mvcost,
+ center_mv, best_mv,
+ bigdia_num_candidates, bigdia_candidates);
}
int vp9_square_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int sad_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
- int_mv *center_mv,
- int_mv *best_mv) {
+ const MV *center_mv,
+ MV *best_mv) {
// All scales have 8 closest points in square shape
static const int square_num_candidates[MAX_PATTERN_SCALES] = {
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
{{-1024, -1024}, {0, -1024}, {1024, -1024}, {1024, 0}, {1024, 1024},
{0, 1024}, {-1024, 1024}, {-1024, 0}},
};
- return
- vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
- do_init_search, 0, vfp, use_mvcost,
- center_mv, best_mv,
- square_num_candidates, square_candidates);
+ return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
+ do_init_search, 0, vfp, use_mvcost,
+ center_mv, best_mv,
+ square_num_candidates, square_candidates);
};
#undef CHECK_BOUNDS
int_mv *ref_mv, int_mv *dst_mv);
int vp9_hex_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int error_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vf,
int use_mvcost,
- int_mv *center_mv,
- int_mv *best_mv);
+ const MV *center_mv,
+ MV *best_mv);
int vp9_bigdia_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int error_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vf,
int use_mvcost,
- int_mv *center_mv,
- int_mv *best_mv);
+ const MV *center_mv,
+ MV *best_mv);
int vp9_square_search(MACROBLOCK *x,
- int_mv *ref_mv,
+ MV *ref_mv,
int search_param,
int error_per_bit,
int do_init_search,
const vp9_variance_fn_ptr_t *vf,
int use_mvcost,
- int_mv *center_mv,
- int_mv *best_mv);
+ const MV *center_mv,
+ MV *best_mv);
typedef int (fractional_mv_step_fp) (
MACROBLOCK *x,
now so that HIGH_PRECISION is always
chosen */
+// Masks for partially or completely disabling split mode
+#define DISABLE_ALL_SPLIT 0x3F
+#define DISABLE_ALL_INTER_SPLIT 0x1F
+#define DISABLE_COMPOUND_SPLIT 0x18
+#define LAST_AND_INTRA_SPLIT_ONLY 0x1E
+
#if CONFIG_INTERNAL_STATS
#include "math.h"
gf_low_motion_minq[i] = calculate_minq_index(maxq,
0.0000015,
-0.0009,
- 0.33,
+ 0.32,
0.0);
gf_high_motion_minq[i] = calculate_minq_index(maxq,
0.0000021,
-0.00125,
- 0.45,
+ 0.50,
0.0);
inter_minq[i] = calculate_minq_index(maxq,
0.00000271,
-0.00113,
- 0.697,
+ 0.75,
0.0);
afq_low_motion_minq[i] = calculate_minq_index(maxq,
0.0000015,
}
}
+static int get_active_quality(int q,
+ int gfu_boost,
+ int low,
+ int high,
+ int *low_motion_minq,
+ int *high_motion_minq) {
+ int active_best_quality;
+ if (gfu_boost > high) {
+ active_best_quality = low_motion_minq[q];
+ } else if (gfu_boost < low) {
+ active_best_quality = high_motion_minq[q];
+ } else {
+ const int gap = high - low;
+ const int offset = high - gfu_boost;
+ const int qdiff = high_motion_minq[q] - low_motion_minq[q];
+ const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
+ active_best_quality = low_motion_minq[q] + adjustment;
+ }
+ return active_best_quality;
+}
+
static void set_mvcost(MACROBLOCK *mb) {
if (mb->e_mbd.allow_high_precision_mv) {
mb->mvcost = mb->nmvcost_hp;
sf->thresh_mult[THR_COMP_NEARGA] += 1500;
sf->thresh_mult[THR_COMP_NEWGA] += 2000;
- sf->thresh_mult[THR_SPLITMV] += 2500;
- sf->thresh_mult[THR_SPLITG] += 2500;
- sf->thresh_mult[THR_SPLITA] += 2500;
- sf->thresh_mult[THR_COMP_SPLITLA] += 4500;
- sf->thresh_mult[THR_COMP_SPLITGA] += 4500;
-
sf->thresh_mult[THR_ZEROMV] += 2000;
sf->thresh_mult[THR_ZEROG] += 2000;
sf->thresh_mult[THR_ZEROA] += 2000;
sf->thresh_mult[THR_COMP_ZEROLA] += 2500;
sf->thresh_mult[THR_COMP_ZEROGA] += 2500;
- sf->thresh_mult[THR_B_PRED] += 2500;
sf->thresh_mult[THR_H_PRED] += 2000;
sf->thresh_mult[THR_V_PRED] += 2000;
sf->thresh_mult[THR_D45_PRED ] += 2500;
sf->thresh_mult[THR_NEARESTMV] = INT_MAX;
sf->thresh_mult[THR_ZEROMV ] = INT_MAX;
sf->thresh_mult[THR_NEARMV ] = INT_MAX;
- sf->thresh_mult[THR_SPLITMV ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
sf->thresh_mult[THR_NEARESTG ] = INT_MAX;
sf->thresh_mult[THR_ZEROG ] = INT_MAX;
sf->thresh_mult[THR_NEARG ] = INT_MAX;
sf->thresh_mult[THR_NEWG ] = INT_MAX;
- sf->thresh_mult[THR_SPLITG ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
sf->thresh_mult[THR_NEARESTA ] = INT_MAX;
sf->thresh_mult[THR_ZEROA ] = INT_MAX;
sf->thresh_mult[THR_NEARA ] = INT_MAX;
sf->thresh_mult[THR_NEWA ] = INT_MAX;
- sf->thresh_mult[THR_SPLITA ] = INT_MAX;
}
if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
sf->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARLA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWLA ] = INT_MAX;
- sf->thresh_mult[THR_COMP_SPLITLA ] = INT_MAX;
}
if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
(VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
sf->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARGA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWGA ] = INT_MAX;
- sf->thresh_mult[THR_COMP_SPLITGA ] = INT_MAX;
}
+}
- if (sf->disable_splitmv == 1) {
- sf->thresh_mult[THR_SPLITMV ] = INT_MAX;
- sf->thresh_mult[THR_SPLITG ] = INT_MAX;
- sf->thresh_mult[THR_SPLITA ] = INT_MAX;
+static void set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi, int mode) {
+ SPEED_FEATURES *sf = &cpi->sf;
+ int i;
- sf->thresh_mult[THR_COMP_SPLITLA ] = INT_MAX;
- sf->thresh_mult[THR_COMP_SPLITGA ] = INT_MAX;
+ for (i = 0; i < MAX_REFS; ++i)
+ sf->thresh_mult_sub8x8[i] = mode == 0 ? -500 : 0;
+
+ sf->thresh_mult_sub8x8[THR_LAST] += 2500;
+ sf->thresh_mult_sub8x8[THR_GOLD] += 2500;
+ sf->thresh_mult_sub8x8[THR_ALTR] += 2500;
+ sf->thresh_mult_sub8x8[THR_INTRA] += 2500;
+ sf->thresh_mult_sub8x8[THR_COMP_LA] += 4500;
+ sf->thresh_mult_sub8x8[THR_COMP_GA] += 4500;
+
+ // Check for masked out split cases.
+ for (i = 0; i < MAX_REFS; i++) {
+ if (sf->disable_split_mask & (1 << i))
+ sf->thresh_mult_sub8x8[i] = INT_MAX;
}
+
+ // disable mode test if frame flag is not set
+ if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
+ sf->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
+ if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
+ sf->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
+ if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
+ sf->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
+ if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_LAST_FLAG | VP9_ALT_FLAG))
+ sf->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
+ if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_GOLD_FLAG | VP9_ALT_FLAG))
+ sf->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
}
void vp9_set_speed_features(VP9_COMP *cpi) {
if (mode > 1)
mode = 1;
- // Initialise default mode frequency sampling variables
- for (i = 0; i < MAX_MODES; i ++) {
- cpi->mode_check_freq[i] = 0;
- cpi->mode_test_hit_counts[i] = 0;
+ for (i = 0; i < MAX_MODES; ++i)
cpi->mode_chosen_counts[i] = 0;
- }
// best quality defaults
sf->RD = 1;
sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
sf->comp_inter_joint_search_thresh = BLOCK_4X4;
sf->adaptive_rd_thresh = 0;
- sf->use_lastframe_partitioning = 0;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_OFF;
sf->tx_size_search_method = USE_FULL_RD;
sf->use_lp32x32fdct = 0;
sf->adaptive_motion_search = 0;
sf->use_avoid_tested_higherror = 0;
sf->reference_masking = 0;
- sf->partition_by_variance = 0;
sf->use_one_partition_size_always = 0;
sf->less_rectangular_check = 0;
sf->use_square_partition_only = 0;
sf->auto_min_max_partition_size = 0;
- sf->auto_min_max_partition_interval = 0;
- sf->auto_min_max_partition_count = 0;
sf->max_partition_size = BLOCK_64X64;
sf->min_partition_size = BLOCK_4X4;
sf->adjust_partitioning_from_last_frame = 0;
sf->last_partitioning_redo_frequency = 4;
- sf->disable_splitmv = 0;
+ sf->disable_split_mask = 0;
sf->mode_search_skip_flags = 0;
sf->disable_split_var_thresh = 0;
sf->disable_filter_search_var_thresh = 0;
sf->static_segmentation = 0;
#endif
sf->use_avoid_tested_higherror = 1;
- sf->adaptive_rd_thresh = MIN((speed + 1), 4);
+ sf->adaptive_rd_thresh = 1;
if (speed == 1) {
- sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->less_rectangular_check = 1;
+ sf->use_square_partition_only = !(cpi->common.frame_type == KEY_FRAME ||
+ cpi->common.intra_only);
sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0) ?
- USE_FULL_RD :
- USE_LARGESTALL);
+ cpi->common.intra_only)
+ ? USE_FULL_RD : USE_LARGESTALL);
+
+ if (MIN(cpi->common.width, cpi->common.height) >= 720)
+ sf->disable_split_mask = DISABLE_ALL_SPLIT;
+ else
+ sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
+
+ sf->use_rd_breakout = 1;
+ sf->adaptive_motion_search = 1;
+ sf->auto_mv_step_size = 1;
+ sf->adaptive_rd_thresh = 2;
+ }
+ if (speed == 2) {
sf->use_square_partition_only = !(cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0);
- sf->disable_splitmv =
- (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
+ cpi->common.intra_only);
+ sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
+ cpi->common.intra_only)
+ ? USE_FULL_RD : USE_LARGESTALL);
+
+ if (MIN(cpi->common.width, cpi->common.height) >= 720)
+ sf->disable_split_mask = DISABLE_ALL_SPLIT;
+ else
+ sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
- sf->use_uv_intra_rd_estimate = 1;
+
sf->use_rd_breakout = 1;
- sf->skip_encode_sb = 1;
- sf->use_lp32x32fdct = 1;
sf->adaptive_motion_search = 1;
sf->auto_mv_step_size = 1;
- sf->auto_min_max_partition_size = 1;
- sf->auto_min_max_partition_interval = 1;
- // FIXME(jingning): temporarily turn off disable_split_var_thresh
- // during refactoring process. will get this back after finishing
- // the main framework of partition search type.
- sf->disable_split_var_thresh = 0;
sf->disable_filter_search_var_thresh = 16;
+ sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->intra_y_mode_mask = INTRA_DC_TM_H_V;
- sf->intra_uv_mode_mask = INTRA_DC_TM_H_V;
- sf->use_fast_coef_updates = 1;
+ sf->auto_min_max_partition_size = 1;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
+ sf->adjust_partitioning_from_last_frame = 1;
+ sf->last_partitioning_redo_frequency = 3;
+
+ sf->adaptive_rd_thresh = 2;
sf->mode_skip_start = 11;
}
- if (speed == 2) {
- sf->less_rectangular_check = 1;
+ if (speed == 3) {
sf->use_square_partition_only = 1;
- sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->use_lastframe_partitioning = 1;
- sf->adjust_partitioning_from_last_frame = 1;
- sf->last_partitioning_redo_frequency = 3;
- sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0) ?
- USE_FULL_RD :
- USE_LARGESTALL);
+ sf->tx_size_search_method = USE_LARGESTALL;
+
+ if (MIN(cpi->common.width, cpi->common.height) >= 720)
+ sf->disable_split_mask = DISABLE_ALL_SPLIT;
+ else
+ sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
- FLAG_SKIP_COMP_REFMISMATCH |
- FLAG_SKIP_INTRA_LOWVAR |
- FLAG_EARLY_TERMINATE;
- sf->intra_y_mode_mask = INTRA_DC_TM;
- sf->intra_uv_mode_mask = INTRA_DC_TM;
- sf->use_uv_intra_rd_estimate = 1;
+ FLAG_SKIP_INTRA_LOWVAR;
+
sf->use_rd_breakout = 1;
- sf->skip_encode_sb = 1;
- sf->use_lp32x32fdct = 1;
sf->adaptive_motion_search = 1;
- sf->using_small_partition_info = 0;
- sf->disable_splitmv =
- (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
sf->auto_mv_step_size = 1;
- sf->search_method = SQUARE;
- sf->subpel_iters_per_step = 1;
- sf->use_fast_lpf_pick = 1;
+
+ sf->disable_filter_search_var_thresh = 16;
+ sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
+
sf->auto_min_max_partition_size = 1;
- sf->auto_min_max_partition_interval = 2;
- sf->disable_split_var_thresh = 32;
- sf->disable_filter_search_var_thresh = 32;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
+ sf->adjust_partitioning_from_last_frame = 1;
+ sf->last_partitioning_redo_frequency = 3;
+
+ sf->use_uv_intra_rd_estimate = 1;
+ sf->skip_encode_sb = 1;
+ sf->use_lp32x32fdct = 1;
+ sf->subpel_iters_per_step = 1;
sf->use_fast_coef_updates = 2;
+
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
- if (speed == 3) {
- sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->partition_by_variance = 1;
- sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0) ?
- USE_FULL_RD :
- USE_LARGESTALL);
+ if (speed == 4) {
+ sf->use_square_partition_only = 1;
+ sf->tx_size_search_method = USE_LARGESTALL;
+ sf->disable_split_mask = DISABLE_ALL_SPLIT;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_COMP_REFMISMATCH |
FLAG_SKIP_INTRA_LOWVAR |
FLAG_EARLY_TERMINATE;
+
sf->use_rd_breakout = 1;
+ sf->adaptive_motion_search = 1;
+ sf->auto_mv_step_size = 1;
+
+ sf->disable_filter_search_var_thresh = 16;
+ sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
+
+ sf->auto_min_max_partition_size = 1;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
+ sf->adjust_partitioning_from_last_frame = 1;
+ sf->last_partitioning_redo_frequency = 3;
+
+ sf->use_uv_intra_rd_estimate = 1;
sf->skip_encode_sb = 1;
sf->use_lp32x32fdct = 1;
- sf->disable_splitmv = 1;
- sf->auto_mv_step_size = 1;
- sf->search_method = BIGDIA;
sf->subpel_iters_per_step = 1;
- sf->disable_split_var_thresh = 64;
- sf->disable_filter_search_var_thresh = 64;
- sf->intra_y_mode_mask = INTRA_DC_ONLY;
- sf->intra_uv_mode_mask = INTRA_DC_ONLY;
sf->use_fast_coef_updates = 2;
+
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
+
+ /* sf->intra_y_mode_mask = INTRA_DC_ONLY;
+ sf->intra_uv_mode_mask = INTRA_DC_ONLY;
+ sf->search_method = BIGDIA;
+ sf->disable_split_var_thresh = 64;
+ sf->disable_filter_search_var_thresh = 64; */
}
- if (speed == 4) {
+ if (speed == 5) {
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->use_one_partition_size_always = 1;
sf->always_this_block_size = BLOCK_16X16;
// sf->reduce_first_step_size = 1;
// sf->reference_masking = 1;
- sf->disable_splitmv = 1;
+ sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->search_method = HEX;
sf->subpel_iters_per_step = 1;
sf->disable_split_var_thresh = 64;
sf->intra_y_mode_mask = INTRA_DC_ONLY;
sf->intra_uv_mode_mask = INTRA_DC_ONLY;
sf->use_fast_coef_updates = 2;
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
break;
// Set rd thresholds based on mode and speed setting
set_rd_speed_thresholds(cpi, mode);
+ set_rd_speed_thresholds_sub8x8(cpi, mode);
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
cpi->oxcf.lossless = oxcf->lossless;
if (cpi->oxcf.lossless) {
- cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_short_iwalsh4x4_1_add;
- cpi->mb.e_mbd.inv_txm4x4_add = vp9_short_iwalsh4x4_add;
+ cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_iwht4x4_1_add;
+ cpi->mb.e_mbd.inv_txm4x4_add = vp9_iwht4x4_16_add;
} else {
- cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_short_idct4x4_1_add;
- cpi->mb.e_mbd.inv_txm4x4_add = vp9_short_idct4x4_add;
+ cpi->mb.e_mbd.inv_txm4x4_1_add = vp9_idct4x4_1_add;
+ cpi->mb.e_mbd.inv_txm4x4_add = vp9_idct4x4_16_add;
}
cpi->baseline_gf_interval = DEFAULT_GF_INTERVAL;
vp9_set_speed_features(cpi);
// Default rd threshold factors for mode selection
- for (i = 0; i < BLOCK_SIZES; ++i)
+ for (i = 0; i < BLOCK_SIZES; ++i) {
for (j = 0; j < MAX_MODES; ++j)
cpi->rd_thresh_freq_fact[i][j] = 32;
+ for (j = 0; j < MAX_REFS; ++j)
+ cpi->rd_thresh_freq_sub8x8[i][j] = 32;
+ }
#define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SVFHH, SVFHV, SVFHHV, \
SDX3F, SDX8F, SDX4DF)\
double q_val;
// Baseline value derived from cpi->active_worst_quality and kf boost
- if (cpi->kf_boost > high) {
- cpi->active_best_quality = kf_low_motion_minq[q];
- } else if (cpi->kf_boost < low) {
- cpi->active_best_quality = kf_high_motion_minq[q];
- } else {
- const int gap = high - low;
- const int offset = high - cpi->kf_boost;
- const int qdiff = kf_high_motion_minq[q] - kf_low_motion_minq[q];
- const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
-
- cpi->active_best_quality = kf_low_motion_minq[q] + adjustment;
- }
+ cpi->active_best_quality = get_active_quality(q, cpi->kf_boost,
+ low, high,
+ kf_low_motion_minq,
+ kf_high_motion_minq);
// Allow somewhat lower kf minq with small image formats.
if ((cm->width * cm->height) <= (352 * 288)) {
q = cpi->avg_frame_qindex;
}
// For constrained quality dont allow Q less than the cq level
- if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY &&
- q < cpi->cq_target_quality) {
- q = cpi->cq_target_quality;
- }
- if (cpi->gfu_boost > high) {
- cpi->active_best_quality = gf_low_motion_minq[q];
- } else if (cpi->gfu_boost < low) {
- cpi->active_best_quality = gf_high_motion_minq[q];
- } else {
- const int gap = high - low;
- const int offset = high - cpi->gfu_boost;
- const int qdiff = gf_high_motion_minq[q] - gf_low_motion_minq[q];
- const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
-
- cpi->active_best_quality = gf_low_motion_minq[q] + adjustment;
- }
-
- // Constrained quality use slightly lower active best.
- if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY)
+ if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
+ if (q < cpi->cq_target_quality)
+ q = cpi->cq_target_quality;
+ if (cpi->frames_since_key > 1) {
+ cpi->active_best_quality = get_active_quality(q, cpi->gfu_boost,
+ low, high,
+ afq_low_motion_minq,
+ afq_high_motion_minq);
+ } else {
+ cpi->active_best_quality = get_active_quality(q, cpi->gfu_boost,
+ low, high,
+ gf_low_motion_minq,
+ gf_high_motion_minq);
+ }
+ // Constrained quality use slightly lower active best.
cpi->active_best_quality = cpi->active_best_quality * 15 / 16;
- // TODO(debargha): Refine the logic below
- if (cpi->oxcf.end_usage == USAGE_CONSTANT_QUALITY) {
+ } else if (cpi->oxcf.end_usage == USAGE_CONSTANT_QUALITY) {
if (!cpi->refresh_alt_ref_frame) {
cpi->active_best_quality = cpi->cq_target_quality;
} else {
if (cpi->frames_since_key > 1) {
- if (cpi->gfu_boost > high) {
- cpi->active_best_quality = afq_low_motion_minq[q];
- } else if (cpi->gfu_boost < low) {
- cpi->active_best_quality = afq_high_motion_minq[q];
- } else {
- const int gap = high - low;
- const int offset = high - cpi->gfu_boost;
- const int qdiff = afq_high_motion_minq[q] - afq_low_motion_minq[q];
- const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
-
- cpi->active_best_quality = afq_low_motion_minq[q] + adjustment;
- }
+ cpi->active_best_quality = get_active_quality(q, cpi->gfu_boost,
+ low, high,
+ afq_low_motion_minq,
+ afq_high_motion_minq);
+ } else {
+ cpi->active_best_quality = get_active_quality(q, cpi->gfu_boost,
+ low, high,
+ gf_low_motion_minq,
+ gf_high_motion_minq);
}
}
+ } else {
+ if (!cpi->refresh_alt_ref_frame) {
+ cpi->active_best_quality = inter_minq[q];
+ } else {
+ cpi->active_best_quality = get_active_quality(q, cpi->gfu_boost,
+ low, high,
+ gf_low_motion_minq,
+ gf_high_motion_minq);
+ }
}
} else {
if (cpi->oxcf.end_usage == USAGE_CONSTANT_QUALITY) {
cm->frame_type, cpi->refresh_golden_frame,
cpi->refresh_alt_ref_frame);
- for (i = 0; i < MAX_MODES; i++)
+ for (i = 0; i < MAX_MODES; ++i)
fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
+ for (i = 0; i < MAX_REFS; ++i)
+ fprintf(fmodes, "%5d ", cpi->sub8x8_mode_chosen_counts[i]);
fprintf(fmodes, "\n");
#define DISABLE_RC_LONG_TERM_MEM 0
#endif
-// #define MODE_TEST_HIT_STATS
+#define MODE_TEST_HIT_STATS
// #define SPEEDSTATS 1
#if CONFIG_MULTIPLE_ARF
#define KEY_FRAME_CONTEXT 5
-#define MAX_MODES 36
+#define MAX_MODES 30
+#define MAX_REFS 6
#define MIN_THRESHMULT 32
#define MAX_THRESHMULT 512
#define INTRA_ZBIN_BOOST 0
typedef struct {
- nmv_context nmvc;
int nmvjointcost[MV_JOINTS];
int nmvcosts[2][MV_VALS];
int nmvcosts_hp[2][MV_VALS];
vp9_prob segment_pred_probs[PREDICTION_PROBS];
- vp9_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
- vp9_prob comp_inter_prob[COMP_INTER_CONTEXTS];
- vp9_prob single_ref_prob[REF_CONTEXTS][2];
- vp9_prob comp_ref_prob[REF_CONTEXTS];
unsigned char *last_frame_seg_map_copy;
// 0 = ZERO_MV, MV
signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
- vp9_coeff_probs_model coef_probs[TX_SIZES][BLOCK_TYPES];
-
- vp9_prob y_mode_prob[4][INTRA_MODES - 1];
- vp9_prob uv_mode_prob[INTRA_MODES][INTRA_MODES - 1];
- vp9_prob partition_prob[2][NUM_PARTITION_CONTEXTS][PARTITION_TYPES - 1];
-
- vp9_prob switchable_interp_prob[SWITCHABLE_FILTERS + 1]
- [SWITCHABLE_FILTERS - 1];
-
int inter_mode_counts[INTER_MODE_CONTEXTS][INTER_MODES - 1][2];
- vp9_prob inter_mode_probs[INTER_MODE_CONTEXTS][INTER_MODES - 1];
-
- struct tx_probs tx_probs;
- vp9_prob mbskip_probs[MBSKIP_CONTEXTS];
+ FRAME_CONTEXT fc;
} CODING_CONTEXT;
typedef struct {
THR_COMP_NEARGA,
THR_COMP_NEWGA,
- THR_SPLITMV,
- THR_SPLITG,
- THR_SPLITA,
- THR_COMP_SPLITLA,
- THR_COMP_SPLITGA,
-
THR_ZEROMV,
THR_ZEROG,
THR_ZEROA,
THR_COMP_ZEROLA,
THR_COMP_ZEROGA,
- THR_B_PRED,
THR_H_PRED,
THR_V_PRED,
THR_D135_PRED,
THR_D45_PRED,
} THR_MODES;
+typedef enum {
+ THR_LAST,
+ THR_GOLD,
+ THR_ALTR,
+ THR_COMP_LA,
+ THR_COMP_GA,
+ THR_INTRA,
+} THR_MODES_SUB8X8;
+
typedef enum {
DIAMOND = 0,
NSTEP = 1,
#define INTRA_DC_TM ((1 << TM_PRED) | (1 << DC_PRED))
#define INTRA_DC_TM_H_V (INTRA_DC_TM | (1 << V_PRED) | (1 << H_PRED))
+typedef enum {
+ LAST_FRAME_PARTITION_OFF = 0,
+ LAST_FRAME_PARTITION_LOW_MOTION = 1,
+ LAST_FRAME_PARTITION_ALL = 2
+} LAST_FRAME_PARTITION_METHOD;
+
typedef struct {
int RD;
SEARCH_METHODS search_method;
SUBPEL_SEARCH_METHODS subpel_search_method;
int subpel_iters_per_step;
int thresh_mult[MAX_MODES];
+ int thresh_mult_sub8x8[MAX_REFS];
int max_step_search_steps;
int reduce_first_step_size;
int auto_mv_step_size;
int adaptive_rd_thresh;
int skip_encode_sb;
int skip_encode_frame;
- int use_lastframe_partitioning;
+ LAST_FRAME_PARTITION_METHOD use_lastframe_partitioning;
TX_SIZE_SEARCH_METHOD tx_size_search_method;
int use_lp32x32fdct;
int use_avoid_tested_higherror;
- int partition_by_variance;
int use_one_partition_size_always;
int less_rectangular_check;
int use_square_partition_only;
int reference_masking;
BLOCK_SIZE always_this_block_size;
int auto_min_max_partition_size;
- int auto_min_max_partition_interval;
- int auto_min_max_partition_count;
BLOCK_SIZE min_partition_size;
BLOCK_SIZE max_partition_size;
int adjust_partitioning_from_last_frame;
int last_partitioning_redo_frequency;
- int disable_splitmv;
+ int disable_split_mask;
int using_small_partition_info;
// TODO(jingning): combine the related motion search speed features
int adaptive_motion_search;
YV12_BUFFER_CONFIG scaled_source;
unsigned int frames_till_alt_ref_frame;
- int source_alt_ref_pending; // frame in src_buffers has been identified to be encoded as an alt ref
- int source_alt_ref_active; // an alt ref frame has been encoded and is usable
+ int source_alt_ref_pending;
+ int source_alt_ref_active;
- int is_src_frame_alt_ref; // source of frame to encode is an exact copy of an alt ref frame
+ int is_src_frame_alt_ref;
- int gold_is_last; // golden frame same as last frame ( short circuit gold searches)
- int alt_is_last; // Alt reference frame same as last ( short circuit altref search)
+ int gold_is_last; // gold same as last frame ( short circuit gold searches)
+ int alt_is_last; // Alt same as last ( short circuit altref search)
int gold_is_alt; // don't do both alt and gold search ( just do gold).
int scaled_ref_idx[3];
// Ambient reconstruction err target for force key frames
int ambient_err;
- unsigned int mode_check_freq[MAX_MODES];
- unsigned int mode_test_hit_counts[MAX_MODES];
unsigned int mode_chosen_counts[MAX_MODES];
+ unsigned int sub8x8_mode_chosen_counts[MAX_REFS];
int64_t mode_skip_mask;
int ref_frame_mask;
int set_ref_frame_mask;
int rd_threshes[BLOCK_SIZES][MAX_MODES];
int rd_thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
+ int rd_thresh_sub8x8[BLOCK_SIZES][MAX_REFS];
+ int rd_thresh_freq_sub8x8[BLOCK_SIZES][MAX_REFS];
int64_t rd_comp_pred_diff[NB_PREDICTION_TYPES];
int64_t rd_prediction_type_threshes[4][NB_PREDICTION_TYPES];
double gf_rate_correction_factor;
unsigned int frames_since_golden;
- int frames_till_gf_update_due; // Count down till next GF
+ int frames_till_gf_update_due; // Count down till next GF
- int gf_overspend_bits; // Total bits overspent becasue of GF boost (cumulative)
+ int gf_overspend_bits; // cumulative bits overspent because of GF boost
- int non_gf_bitrate_adjustment; // Used in the few frames following a GF to recover the extra bits spent in that GF
+ int non_gf_bitrate_adjustment; // Following GF to recover extra bits spent
- int kf_overspend_bits; // Extra bits spent on key frames that need to be recovered on inter frames
- int kf_bitrate_adjustment; // Current number of bit s to try and recover on each inter frame.
+ int kf_overspend_bits; // Bits spent on key frames to be recovered on inters
+ int kf_bitrate_adjustment; // number of bits to recover on each inter frame.
int max_gf_interval;
int baseline_gf_interval;
int active_arnr_frames; // <= cpi->oxcf.arnr_max_frames
int64_t key_frame_count;
int prior_key_frame_distance[KEY_FRAME_CONTEXT];
- int per_frame_bandwidth; // Current section per frame bandwidth target
- int av_per_frame_bandwidth; // Average frame size target for clip
- int min_frame_bandwidth; // Minimum allocation that should be used for any frame
+ int per_frame_bandwidth; // Current section per frame bandwidth target
+ int av_per_frame_bandwidth; // Average frame size target for clip
+ int min_frame_bandwidth; // Minimum allocation used for any frame
int inter_frame_target;
double output_framerate;
int64_t last_time_stamp_seen;
unsigned int max_mv_magnitude;
int mv_step_param;
- // Data used for real time conferencing mode to help determine if it would be good to update the gf
+ // Data used for real time conferencing mode to help determine if it
+ // would be good to update the gf
int inter_zz_count;
int gf_bad_count;
int gf_update_recommended;
void vp9_alloc_compressor_data(VP9_COMP *cpi);
+static int get_token_alloc(int mb_rows, int mb_cols) {
+ return mb_rows * mb_cols * (48 * 16 + 4);
+}
+
#endif // VP9_ENCODER_VP9_ONYX_INT_H_
src += srcoffset;
dst += dstoffset;
- // Loop through the Y plane raw and reconstruction data summing (square differences)
+ // Loop through the raw Y plane and reconstruction data summing the square
+ // differences.
for (i = 0; i < linestocopy; i += 16) {
for (j = 0; j < source->y_width; j += 16) {
unsigned int sse;
// Enforce a minimum filter level based upon baseline Q
static int get_min_filter_level(VP9_COMP *cpi, int base_qindex) {
int min_filter_level;
- /*int q = (int) vp9_convert_qindex_to_q(base_qindex);
-
- if (cpi->source_alt_ref_active && cpi->common.refresh_golden_frame && !cpi->common.refresh_alt_ref_frame)
- min_filter_level = 0;
- else
- {
- if (q <= 10)
- min_filter_level = 0;
- else if (q <= 64)
- min_filter_level = 1;
- else
- min_filter_level = (q >> 6);
- }
- */
min_filter_level = 0;
return min_filter_level;
// Enforce a maximum filter level based upon baseline Q
static int get_max_filter_level(VP9_COMP *cpi, int base_qindex) {
- // PGW August 2006: Highest filter values almost always a bad idea
-
- // jbb chg: 20100118 - not so any more with this overquant stuff allow high values
- // with lots of intra coming in.
- int max_filter_level = MAX_LOOP_FILTER;// * 3 / 4;
+ int max_filter_level = MAX_LOOP_FILTER;
(void)base_qindex;
if (cpi->twopass.section_intra_rating > 8)
int filt_best;
int filt_direction = 0;
- int Bias = 0; // Bias against raising loop filter and in favour of lowering it
+ int Bias = 0; // Bias against raising loop filter in favor of lowering it.
// Make a copy of the unfiltered / processed recon buffer
vpx_yv12_copy_y(cm->frame_to_show, &cpi->last_frame_uf);
lf->sharpness_level = cm->frame_type == KEY_FRAME ? 0
: cpi->oxcf.Sharpness;
- // Start the search at the previous frame filter level unless it is now out of range.
+ // Start the search at the previous frame filter level unless it is now out of
+ // range.
filt_mid = clamp(lf->filter_level, min_filter_level, max_filter_level);
// Define the initial step size
vpx_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
while (filter_step > 0) {
- Bias = (best_err >> (15 - (filt_mid / 8))) * filter_step; // PGW change 12/12/06 for small images
+ Bias = (best_err >> (15 - (filt_mid / 8))) * filter_step;
- // jbb chg: 20100118 - in sections with lots of new material coming in don't bias as much to a low filter value
if (cpi->twopass.section_intra_rating < 20)
Bias = Bias * cpi->twopass.section_intra_rating / 20;
if (cpi->common.tx_mode != ONLY_4X4)
Bias >>= 1;
- filt_high = ((filt_mid + filter_step) > max_filter_level) ? max_filter_level : (filt_mid + filter_step);
- filt_low = ((filt_mid - filter_step) < min_filter_level) ? min_filter_level : (filt_mid - filter_step);
+ filt_high = ((filt_mid + filter_step) > max_filter_level)
+ ? max_filter_level
+ : (filt_mid + filter_step);
+ filt_low = ((filt_mid - filter_step) < min_filter_level)
+ ? min_filter_level
+ : (filt_mid - filter_step);
if ((filt_direction <= 0) && (filt_low != filt_mid)) {
// Get Low filter error score
// Re-instate the unfiltered frame
vpx_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
- // If value is close to the best so far then bias towards a lower loop filter value.
+ // If value is close to the best so far then bias towards a lower loop
+ // filter value.
if ((filt_err - Bias) < best_err) {
// Was it actually better than the previous best?
if (filt_err < best_err)
lf->filter_level = filt_best;
}
-
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <math.h>
#include "vpx_scale/yv12config.h"
-#include "math.h"
#define MAX_PSNR 100
// restored with a call to vp9_restore_coding_context. These functions are
// intended for use in a re-code loop in vp9_compress_frame where the
// quantizer value is adjusted between loop iterations.
-
- cc->nmvc = cm->fc.nmvc;
vp9_copy(cc->nmvjointcost, cpi->mb.nmvjointcost);
vp9_copy(cc->nmvcosts, cpi->mb.nmvcosts);
vp9_copy(cc->nmvcosts_hp, cpi->mb.nmvcosts_hp);
- vp9_copy(cc->inter_mode_probs, cm->fc.inter_mode_probs);
-
- vp9_copy(cc->y_mode_prob, cm->fc.y_mode_prob);
- vp9_copy(cc->uv_mode_prob, cm->fc.uv_mode_prob);
- vp9_copy(cc->partition_prob, cm->fc.partition_prob);
-
vp9_copy(cc->segment_pred_probs, cm->seg.pred_probs);
- vp9_copy(cc->intra_inter_prob, cm->fc.intra_inter_prob);
- vp9_copy(cc->comp_inter_prob, cm->fc.comp_inter_prob);
- vp9_copy(cc->single_ref_prob, cm->fc.single_ref_prob);
- vp9_copy(cc->comp_ref_prob, cm->fc.comp_ref_prob);
-
vpx_memcpy(cpi->coding_context.last_frame_seg_map_copy,
cm->last_frame_seg_map, (cm->mi_rows * cm->mi_cols));
vp9_copy(cc->last_ref_lf_deltas, cm->lf.last_ref_deltas);
vp9_copy(cc->last_mode_lf_deltas, cm->lf.last_mode_deltas);
- vp9_copy(cc->coef_probs, cm->fc.coef_probs);
- vp9_copy(cc->switchable_interp_prob, cm->fc.switchable_interp_prob);
- cc->tx_probs = cm->fc.tx_probs;
- vp9_copy(cc->mbskip_probs, cm->fc.mbskip_probs);
+ cc->fc = cm->fc;
}
void vp9_restore_coding_context(VP9_COMP *cpi) {
// Restore key state variables to the snapshot state stored in the
// previous call to vp9_save_coding_context.
-
- cm->fc.nmvc = cc->nmvc;
vp9_copy(cpi->mb.nmvjointcost, cc->nmvjointcost);
vp9_copy(cpi->mb.nmvcosts, cc->nmvcosts);
vp9_copy(cpi->mb.nmvcosts_hp, cc->nmvcosts_hp);
- vp9_copy(cm->fc.inter_mode_probs, cc->inter_mode_probs);
-
- vp9_copy(cm->fc.y_mode_prob, cc->y_mode_prob);
- vp9_copy(cm->fc.uv_mode_prob, cc->uv_mode_prob);
- vp9_copy(cm->fc.partition_prob, cc->partition_prob);
-
vp9_copy(cm->seg.pred_probs, cc->segment_pred_probs);
- vp9_copy(cm->fc.intra_inter_prob, cc->intra_inter_prob);
- vp9_copy(cm->fc.comp_inter_prob, cc->comp_inter_prob);
- vp9_copy(cm->fc.single_ref_prob, cc->single_ref_prob);
- vp9_copy(cm->fc.comp_ref_prob, cc->comp_ref_prob);
-
vpx_memcpy(cm->last_frame_seg_map,
cpi->coding_context.last_frame_seg_map_copy,
(cm->mi_rows * cm->mi_cols));
vp9_copy(cm->lf.last_ref_deltas, cc->last_ref_lf_deltas);
vp9_copy(cm->lf.last_mode_deltas, cc->last_mode_lf_deltas);
- vp9_copy(cm->fc.coef_probs, cc->coef_probs);
- vp9_copy(cm->fc.switchable_interp_prob, cc->switchable_interp_prob);
- cm->fc.tx_probs = cc->tx_probs;
- vp9_copy(cm->fc.mbskip_probs, cc->mbskip_probs);
+ cm->fc = cc->fc;
}
void vp9_setup_key_frame(VP9_COMP *cpi) {
DECLARE_ALIGNED(16, extern const uint8_t,
vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
-#define LAST_FRAME_MODE_MASK 0xFFDADCD60
-#define GOLDEN_FRAME_MODE_MASK 0xFFB5A3BB0
-#define ALT_REF_MODE_MASK 0xFF8C648D0
+#define LAST_FRAME_MODE_MASK 0xFFEDCD60
+#define GOLDEN_FRAME_MODE_MASK 0xFFDA3BB0
+#define ALT_REF_MODE_MASK 0xFFC648D0
+
+#define MIN_EARLY_TERM_INDEX 3
const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
- {RD_NEARESTMV, LAST_FRAME, NONE},
- {RD_NEARESTMV, ALTREF_FRAME, NONE},
- {RD_NEARESTMV, GOLDEN_FRAME, NONE},
-
- {RD_DC_PRED, INTRA_FRAME, NONE},
-
- {RD_NEWMV, LAST_FRAME, NONE},
- {RD_NEWMV, ALTREF_FRAME, NONE},
- {RD_NEWMV, GOLDEN_FRAME, NONE},
-
- {RD_NEARMV, LAST_FRAME, NONE},
- {RD_NEARMV, ALTREF_FRAME, NONE},
- {RD_NEARESTMV, LAST_FRAME, ALTREF_FRAME},
- {RD_NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
-
- {RD_TM_PRED, INTRA_FRAME, NONE},
-
- {RD_NEARMV, LAST_FRAME, ALTREF_FRAME},
- {RD_NEWMV, LAST_FRAME, ALTREF_FRAME},
- {RD_NEARMV, GOLDEN_FRAME, NONE},
- {RD_NEARMV, GOLDEN_FRAME, ALTREF_FRAME},
- {RD_NEWMV, GOLDEN_FRAME, ALTREF_FRAME},
-
- {RD_SPLITMV, LAST_FRAME, NONE},
- {RD_SPLITMV, GOLDEN_FRAME, NONE},
- {RD_SPLITMV, ALTREF_FRAME, NONE},
- {RD_SPLITMV, LAST_FRAME, ALTREF_FRAME},
- {RD_SPLITMV, GOLDEN_FRAME, ALTREF_FRAME},
-
- {RD_ZEROMV, LAST_FRAME, NONE},
- {RD_ZEROMV, GOLDEN_FRAME, NONE},
- {RD_ZEROMV, ALTREF_FRAME, NONE},
- {RD_ZEROMV, LAST_FRAME, ALTREF_FRAME},
- {RD_ZEROMV, GOLDEN_FRAME, ALTREF_FRAME},
-
- {RD_I4X4_PRED, INTRA_FRAME, NONE},
- {RD_H_PRED, INTRA_FRAME, NONE},
- {RD_V_PRED, INTRA_FRAME, NONE},
- {RD_D135_PRED, INTRA_FRAME, NONE},
- {RD_D207_PRED, INTRA_FRAME, NONE},
- {RD_D153_PRED, INTRA_FRAME, NONE},
- {RD_D63_PRED, INTRA_FRAME, NONE},
- {RD_D117_PRED, INTRA_FRAME, NONE},
- {RD_D45_PRED, INTRA_FRAME, NONE},
+ {NEARESTMV, LAST_FRAME, NONE},
+ {NEARESTMV, ALTREF_FRAME, NONE},
+ {NEARESTMV, GOLDEN_FRAME, NONE},
+
+ {DC_PRED, INTRA_FRAME, NONE},
+
+ {NEWMV, LAST_FRAME, NONE},
+ {NEWMV, ALTREF_FRAME, NONE},
+ {NEWMV, GOLDEN_FRAME, NONE},
+
+ {NEARMV, LAST_FRAME, NONE},
+ {NEARMV, ALTREF_FRAME, NONE},
+ {NEARESTMV, LAST_FRAME, ALTREF_FRAME},
+ {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
+
+ {TM_PRED, INTRA_FRAME, NONE},
+
+ {NEARMV, LAST_FRAME, ALTREF_FRAME},
+ {NEWMV, LAST_FRAME, ALTREF_FRAME},
+ {NEARMV, GOLDEN_FRAME, NONE},
+ {NEARMV, GOLDEN_FRAME, ALTREF_FRAME},
+ {NEWMV, GOLDEN_FRAME, ALTREF_FRAME},
+
+ {ZEROMV, LAST_FRAME, NONE},
+ {ZEROMV, GOLDEN_FRAME, NONE},
+ {ZEROMV, ALTREF_FRAME, NONE},
+ {ZEROMV, LAST_FRAME, ALTREF_FRAME},
+ {ZEROMV, GOLDEN_FRAME, ALTREF_FRAME},
+
+ {H_PRED, INTRA_FRAME, NONE},
+ {V_PRED, INTRA_FRAME, NONE},
+ {D135_PRED, INTRA_FRAME, NONE},
+ {D207_PRED, INTRA_FRAME, NONE},
+ {D153_PRED, INTRA_FRAME, NONE},
+ {D63_PRED, INTRA_FRAME, NONE},
+ {D117_PRED, INTRA_FRAME, NONE},
+ {D45_PRED, INTRA_FRAME, NONE},
+};
+
+const REF_DEFINITION vp9_ref_order[MAX_REFS] = {
+ {LAST_FRAME, NONE},
+ {GOLDEN_FRAME, NONE},
+ {ALTREF_FRAME, NONE},
+ {LAST_FRAME, ALTREF_FRAME},
+ {GOLDEN_FRAME, ALTREF_FRAME},
+ {INTRA_FRAME, NONE},
};
// The baseline rd thresholds for breaking out of the rd loop for
return (11 * q * q) >> 2;
}
-static MB_PREDICTION_MODE rd_mode_to_mode(RD_PREDICTION_MODE rd_mode) {
- if (rd_mode == RD_SPLITMV || rd_mode == RD_I4X4_PRED) {
- assert(!"Invalid rd_mode");
- return MB_MODE_COUNT;
- }
- assert((int)rd_mode < (int)MB_MODE_COUNT);
- return (MB_PREDICTION_MODE)rd_mode;
-}
-
void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex) {
cpi->mb.sadperbit16 = sad_per_bit16lut[qindex];
cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
}
-
void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
int q, i, bsize;
// cpi->common.refresh_alt_ref_frame)
qindex = clamp(qindex, 0, MAXQ);
+ cpi->RDDIV = 100;
cpi->RDMULT = compute_rd_mult(qindex);
if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
if (cpi->twopass.next_iiratio > 31)
if (q < 8)
q = 8;
- if (cpi->RDMULT > 1000) {
- cpi->RDDIV = 1;
- cpi->RDMULT /= 100;
-
- for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
- for (i = 0; i < MAX_MODES; ++i) {
- // Threshold here seem unecessarily harsh but fine given actual
- // range of values used for cpi->sf.thresh_mult[]
- int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
+ for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
+ for (i = 0; i < MAX_MODES; ++i) {
+ // Threshold here seem unecessarily harsh but fine given actual
+ // range of values used for cpi->sf.thresh_mult[]
+ int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
- // *4 relates to the scaling of rd_thresh_block_size_factor[]
- if ((int64_t)cpi->sf.thresh_mult[i] < thresh_max) {
- cpi->rd_threshes[bsize][i] =
+ if (cpi->sf.thresh_mult[i] < thresh_max) {
+ cpi->rd_threshes[bsize][i] =
cpi->sf.thresh_mult[i] * q *
- rd_thresh_block_size_factor[bsize] / (4 * 100);
- } else {
- cpi->rd_threshes[bsize][i] = INT_MAX;
- }
+ rd_thresh_block_size_factor[bsize] / 4;
+ } else {
+ cpi->rd_threshes[bsize][i] = INT_MAX;
}
}
- } else {
- cpi->RDDIV = 100;
- for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
- for (i = 0; i < MAX_MODES; i++) {
- // Threshold here seem unecessarily harsh but fine given actual
- // range of values used for cpi->sf.thresh_mult[]
- int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
+ for (i = 0; i < MAX_REFS; ++i) {
+ int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
- if (cpi->sf.thresh_mult[i] < thresh_max) {
- cpi->rd_threshes[bsize][i] =
- cpi->sf.thresh_mult[i] * q *
+ if (cpi->sf.thresh_mult_sub8x8[i] < thresh_max) {
+ cpi->rd_thresh_sub8x8[bsize][i] =
+ cpi->sf.thresh_mult_sub8x8[i] * q *
rd_thresh_block_size_factor[bsize] / 4;
- } else {
- cpi->rd_threshes[bsize][i] = INT_MAX;
- }
+ } else {
+ cpi->rd_thresh_sub8x8[bsize][i] = INT_MAX;
}
}
}
const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
const int num_4x4_w = num_4x4_blocks_wide_lookup[bs];
const int num_4x4_h = num_4x4_blocks_high_lookup[bs];
+ const uint8_t *band_translate; // just for the get_scan_and_band call
struct rdcost_block_args args = { x, { 0 }, { 0 }, tx_size,
num_4x4_w, num_4x4_h,
vp9_get_entropy_contexts(tx_size, args.t_above, args.t_left,
pd->above_context, pd->left_context,
num_4x4_w, num_4x4_h);
- switch (tx_size) {
- case TX_4X4:
- get_scan_nb_4x4(get_tx_type_4x4(pd->plane_type, xd, 0),
- &args.scan, &args.nb);
- break;
- case TX_8X8:
- get_scan_nb_8x8(get_tx_type_8x8(pd->plane_type, xd),
- &args.scan, &args.nb);
- break;
- case TX_16X16:
- get_scan_nb_16x16(get_tx_type_16x16(pd->plane_type, xd),
- &args.scan, &args.nb);
- break;
- case TX_32X32:
- args.scan = vp9_default_scan_32x32;
- args.nb = vp9_default_scan_32x32_neighbors;
- break;
- default:
- assert(0);
- }
+
+ get_scan_and_band(xd, tx_size, pd->plane_type, 0, &args.scan, &args.nb,
+ &band_translate);
foreach_transformed_block_in_plane(xd, bsize, plane, block_yrd_txfm, &args);
if (args.skip) {
for (idx = 0; idx < num_4x4_blocks_wide; ++idx) {
int64_t ssz;
const int16_t *scan;
+ const int16_t *nb;
uint8_t *src = src_init + idx * 4 + idy * 4 * src_stride;
uint8_t *dst = dst_init + idx * 4 + idy * 4 * dst_stride;
x->quantize_b_4x4(x, block, tx_type, 16);
}
- scan = get_scan_4x4(get_tx_type_4x4(PLANE_TYPE_Y_WITH_DC, xd, block));
+ get_scan_nb_4x4(get_tx_type_4x4(PLANE_TYPE_Y_WITH_DC, xd, block),
+ &scan, &nb);
ratey += cost_coeffs(x, 0, block,
- tempa + idx, templ + idy, TX_4X4, scan,
- vp9_get_coef_neighbors_handle(scan));
+ tempa + idx, templ + idy, TX_4X4, scan, nb);
distortion += vp9_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, block),
16, &ssz) >> 2;
if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
// adjust src pointer for this block
mi_buf_shift(x, i);
if (cpi->sf.search_method == HEX) {
- bestsme = vp9_hex_search(x, &mvp_full,
+ bestsme = vp9_hex_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1, v_fn_ptr, 1,
- bsi->ref_mv, &mode_mv[NEWMV]);
+ &bsi->ref_mv->as_mv,
+ &mode_mv[NEWMV].as_mv);
} else if (cpi->sf.search_method == SQUARE) {
- bestsme = vp9_square_search(x, &mvp_full,
+ bestsme = vp9_square_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1, v_fn_ptr, 1,
- bsi->ref_mv, &mode_mv[NEWMV]);
+ &bsi->ref_mv->as_mv,
+ &mode_mv[NEWMV].as_mv);
} else if (cpi->sf.search_method == BIGDIA) {
- bestsme = vp9_bigdia_search(x, &mvp_full,
+ bestsme = vp9_bigdia_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1, v_fn_ptr, 1,
- bsi->ref_mv, &mode_mv[NEWMV]);
+ &bsi->ref_mv->as_mv,
+ &mode_mv[NEWMV].as_mv);
} else {
bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
sadpb, further_steps, 0, v_fn_ptr,
further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
if (cpi->sf.search_method == HEX) {
- bestsme = vp9_hex_search(x, &mvp_full,
+ bestsme = vp9_hex_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1,
&cpi->fn_ptr[block_size], 1,
- &ref_mv, tmp_mv);
+ &ref_mv.as_mv, &tmp_mv->as_mv);
} else if (cpi->sf.search_method == SQUARE) {
- bestsme = vp9_square_search(x, &mvp_full,
+ bestsme = vp9_square_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1,
&cpi->fn_ptr[block_size], 1,
- &ref_mv, tmp_mv);
+ &ref_mv.as_mv, &tmp_mv->as_mv);
} else if (cpi->sf.search_method == BIGDIA) {
- bestsme = vp9_bigdia_search(x, &mvp_full,
+ bestsme = vp9_bigdia_search(x, &mvp_full.as_mv,
step_param,
sadpb, 1,
&cpi->fn_ptr[block_size], 1,
- &ref_mv, tmp_mv);
+ &ref_mv.as_mv, &tmp_mv->as_mv);
} else {
bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
sadpb, further_steps, 1,
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->this_mi->mbmi;
- const int is_comp_pred = (mbmi->ref_frame[1] > 0);
+ const int is_comp_pred = has_second_ref(mbmi);
const int num_refs = is_comp_pred ? 2 : 1;
const int this_mode = mbmi->mode;
int_mv *frame_mv = mode_mv[this_mode];
MB_MODE_INFO *mbmi = &xd->this_mi->mbmi;
const struct segmentation *seg = &cm->seg;
const BLOCK_SIZE block_size = get_plane_block_size(bsize, &xd->plane[0]);
- RD_PREDICTION_MODE this_mode;
+ MB_PREDICTION_MODE this_mode;
MV_REFERENCE_FRAME ref_frame, second_ref_frame;
unsigned char segment_id = mbmi->segment_id;
int comp_pred, i;
cpi->gld_fb_idx,
cpi->alt_fb_idx};
int64_t best_rd = best_rd_so_far;
- int64_t best_yrd = best_rd_so_far; // FIXME(rbultje) more precise
int64_t best_tx_rd[TX_MODES];
int64_t best_tx_diff[TX_MODES];
int64_t best_pred_diff[NB_PREDICTION_TYPES];
int64_t best_intra_rd = INT64_MAX;
int64_t best_inter_rd = INT64_MAX;
MB_PREDICTION_MODE best_intra_mode = DC_PRED;
- // MB_PREDICTION_MODE best_inter_mode = ZEROMV;
MV_REFERENCE_FRAME best_inter_ref_frame = LAST_FRAME;
INTERPOLATIONFILTERTYPE tmp_best_filter = SWITCHABLE;
int rate_uv_intra[TX_SIZES], rate_uv_tokenonly[TX_SIZES];
int64_t frame_distortions[MAX_REF_FRAMES] = {-1};
int intra_cost_penalty = 20 * vp9_dc_quant(cpi->common.base_qindex,
cpi->common.y_dc_delta_q);
- int_mv seg_mvs[4][MAX_REF_FRAMES];
- union b_mode_info best_bmodes[4];
- PARTITION_INFO best_partition;
const int bws = num_8x8_blocks_wide_lookup[bsize] / 2;
const int bhs = num_8x8_blocks_high_lookup[bsize] / 2;
int best_skip2 = 0;
unsigned char best_zcoeff_blk[256] = { 0 };
x->skip_encode = cpi->sf.skip_encode_frame && xd->q_index < QIDX_SKIP_THRESH;
- vpx_memset(x->zcoeff_blk, 0, sizeof(x->zcoeff_blk));
- vpx_memset(ctx->zcoeff_blk, 0, sizeof(ctx->zcoeff_blk));
+ vp9_zero(x->zcoeff_blk);
+ vp9_zero(ctx->zcoeff_blk);
- for (i = 0; i < 4; i++) {
- int j;
- for (j = 0; j < MAX_REF_FRAMES; j++)
- seg_mvs[i][j].as_int = INVALID_MV;
- }
// Everywhere the flag is set the error is much higher than its neighbors.
ctx->frames_with_high_error = 0;
ctx->modes_with_high_error = 0;
second_ref_frame != best_inter_ref_frame)
continue;
}
- // TODO(jingning, jkoleszar): scaling reference frame not supported for
- // SPLITMV.
- if (ref_frame > 0 &&
- vp9_is_scaled(&scale_factor[ref_frame]) &&
- this_mode == RD_SPLITMV)
- continue;
-
- if (second_ref_frame > 0 &&
- vp9_is_scaled(&scale_factor[second_ref_frame]) &&
- this_mode == RD_SPLITMV)
- continue;
-
- if (bsize >= BLOCK_8X8 &&
- (this_mode == RD_I4X4_PRED || this_mode == RD_SPLITMV))
- continue;
-
- if (bsize < BLOCK_8X8 &&
- !(this_mode == RD_I4X4_PRED || this_mode == RD_SPLITMV))
- continue;
set_scale_factors(xd, ref_frame, second_ref_frame, scale_factor);
mbmi->uv_mode = DC_PRED;
// If the segment skip feature is enabled....
// then do nothing if the current mode is not allowed..
} else if (vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP) &&
- (this_mode != RD_ZEROMV && ref_frame != INTRA_FRAME)) {
+ (this_mode != ZEROMV && ref_frame != INTRA_FRAME)) {
continue;
// Disable this drop out case if the ref frame
// segment level feature is enabled for this segment. This is to
// an unfiltered alternative. We allow near/nearest as well
// because they may result in zero-zero MVs but be cheaper.
if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
- if ((this_mode != RD_ZEROMV &&
- !(this_mode == RD_NEARMV &&
- frame_mv[RD_NEARMV][ALTREF_FRAME].as_int == 0) &&
- !(this_mode == RD_NEARESTMV &&
- frame_mv[RD_NEARESTMV][ALTREF_FRAME].as_int == 0)) ||
+ if ((this_mode != ZEROMV &&
+ !(this_mode == NEARMV &&
+ frame_mv[NEARMV][ALTREF_FRAME].as_int == 0) &&
+ !(this_mode == NEARESTMV &&
+ frame_mv[NEARESTMV][ALTREF_FRAME].as_int == 0)) ||
ref_frame != ALTREF_FRAME) {
continue;
}
// a representative block in the boundary ( first ) and then implement a
// function that does sads when inside the border..
if (((mi_row + bhs) > cm->mi_rows || (mi_col + bws) > cm->mi_cols) &&
- this_mode == RD_NEWMV) {
+ this_mode == NEWMV) {
continue;
}
cpi->mode_test_hits[bsize]++;
#endif
- if (this_mode == RD_I4X4_PRED) {
- int rate;
-
- /*
- if ((cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_BESTINTER) &&
- (vp9_mode_order[best_mode_index].ref_frame > INTRA_FRAME))
- continue;
- */
-
- // RD_I4X4_PRED is only considered for block sizes less than 8x8.
- mbmi->tx_size = TX_4X4;
- if (rd_pick_intra_sub_8x8_y_mode(cpi, x, &rate, &rate_y,
- &distortion_y, best_rd) >= best_rd)
- continue;
- rate2 += rate;
- rate2 += intra_cost_penalty;
- distortion2 += distortion_y;
- if (rate_uv_intra[TX_4X4] == INT_MAX) {
- choose_intra_uv_mode(cpi, bsize, &rate_uv_intra[TX_4X4],
- &rate_uv_tokenonly[TX_4X4],
- &dist_uv[TX_4X4], &skip_uv[TX_4X4],
- &mode_uv[TX_4X4]);
- }
- rate2 += rate_uv_intra[TX_4X4];
- rate_uv = rate_uv_tokenonly[TX_4X4];
- distortion2 += dist_uv[TX_4X4];
- distortion_uv = dist_uv[TX_4X4];
- mbmi->uv_mode = mode_uv[TX_4X4];
- tx_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
- for (i = 0; i < TX_MODES; ++i)
- tx_cache[i] = tx_cache[ONLY_4X4];
- } else if (ref_frame == INTRA_FRAME) {
+ if (ref_frame == INTRA_FRAME) {
TX_SIZE uv_tx;
// Disable intra modes other than DC_PRED for blocks with low variance
// Threshold for intra skipping based on source variance
64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
};
if ((cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_LOWVAR) &&
- this_mode != RD_DC_PRED &&
+ this_mode != DC_PRED &&
x->source_variance < skip_intra_var_thresh[mbmi->sb_type])
continue;
// Only search the oblique modes if the best so far is
// one of the neighboring directional modes
if ((cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_BESTINTER) &&
- (this_mode >= RD_D45_PRED && this_mode <= RD_TM_PRED)) {
+ (this_mode >= D45_PRED && this_mode <= TM_PRED)) {
if (vp9_mode_order[best_mode_index].ref_frame > INTRA_FRAME)
continue;
}
- mbmi->mode = rd_mode_to_mode(this_mode);
+ mbmi->mode = this_mode;
if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_DIRMISMATCH) {
if (conditional_skipintra(mbmi->mode, best_intra_mode))
continue;
mbmi->uv_mode = mode_uv[uv_tx];
rate2 = rate_y + x->mbmode_cost[mbmi->mode] + rate_uv_intra[uv_tx];
- if (this_mode != RD_DC_PRED && this_mode != RD_TM_PRED)
+ if (this_mode != DC_PRED && this_mode != TM_PRED)
rate2 += intra_cost_penalty;
distortion2 = distortion_y + distortion_uv;
- } else if (this_mode == RD_SPLITMV) {
- const int is_comp_pred = second_ref_frame > 0;
- int rate;
- int64_t distortion;
- int64_t this_rd_thresh;
- int64_t tmp_rd, tmp_best_rd = INT64_MAX, tmp_best_rdu = INT64_MAX;
- int tmp_best_rate = INT_MAX, tmp_best_ratey = INT_MAX;
- int64_t tmp_best_distortion = INT_MAX, tmp_best_sse, uv_sse;
- int tmp_best_skippable = 0;
- int switchable_filter_index;
- int_mv *second_ref = is_comp_pred ?
- &mbmi->ref_mvs[second_ref_frame][0] : NULL;
- union b_mode_info tmp_best_bmodes[16];
- MB_MODE_INFO tmp_best_mbmode;
- PARTITION_INFO tmp_best_partition;
- BEST_SEG_INFO bsi[SWITCHABLE_FILTERS];
- int pred_exists = 0;
- int uv_skippable;
- if (is_comp_pred) {
- if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA)
- if (vp9_mode_order[best_mode_index].ref_frame == INTRA_FRAME)
- continue;
- if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_REFMISMATCH)
- if (ref_frame != best_inter_ref_frame &&
- second_ref_frame != best_inter_ref_frame)
- continue;
- }
-
- this_rd_thresh = (ref_frame == LAST_FRAME) ?
- cpi->rd_threshes[bsize][THR_NEWMV] :
- cpi->rd_threshes[bsize][THR_NEWA];
- this_rd_thresh = (ref_frame == GOLDEN_FRAME) ?
- cpi->rd_threshes[bsize][THR_NEWG] : this_rd_thresh;
- xd->this_mi->mbmi.tx_size = TX_4X4;
-
- cpi->rd_filter_cache[SWITCHABLE_FILTERS] = INT64_MAX;
- if (cm->mcomp_filter_type != BILINEAR) {
- tmp_best_filter = EIGHTTAP;
- if (x->source_variance <
- cpi->sf.disable_filter_search_var_thresh) {
- tmp_best_filter = EIGHTTAP;
- vp9_zero(cpi->rd_filter_cache);
- } else {
- for (switchable_filter_index = 0;
- switchable_filter_index < SWITCHABLE_FILTERS;
- ++switchable_filter_index) {
- int newbest, rs;
- int64_t rs_rd;
- mbmi->interp_filter = switchable_filter_index;
- vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
-
- tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
- &mbmi->ref_mvs[ref_frame][0],
- second_ref,
- best_yrd,
- &rate, &rate_y, &distortion,
- &skippable, &total_sse,
- (int)this_rd_thresh, seg_mvs,
- bsi, switchable_filter_index,
- mi_row, mi_col);
-
- if (tmp_rd == INT64_MAX)
- continue;
- cpi->rd_filter_cache[switchable_filter_index] = tmp_rd;
- rs = get_switchable_rate(x);
- rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
- cpi->rd_filter_cache[SWITCHABLE_FILTERS] =
- MIN(cpi->rd_filter_cache[SWITCHABLE_FILTERS],
- tmp_rd + rs_rd);
- if (cm->mcomp_filter_type == SWITCHABLE)
- tmp_rd += rs_rd;
-
- newbest = (tmp_rd < tmp_best_rd);
- if (newbest) {
- tmp_best_filter = mbmi->interp_filter;
- tmp_best_rd = tmp_rd;
- }
- if ((newbest && cm->mcomp_filter_type == SWITCHABLE) ||
- (mbmi->interp_filter == cm->mcomp_filter_type &&
- cm->mcomp_filter_type != SWITCHABLE)) {
- tmp_best_rdu = tmp_rd;
- tmp_best_rate = rate;
- tmp_best_ratey = rate_y;
- tmp_best_distortion = distortion;
- tmp_best_sse = total_sse;
- tmp_best_skippable = skippable;
- tmp_best_mbmode = *mbmi;
- tmp_best_partition = *x->partition_info;
- for (i = 0; i < 4; i++)
- tmp_best_bmodes[i] = xd->this_mi->bmi[i];
- pred_exists = 1;
- if (switchable_filter_index == 0 &&
- cpi->sf.use_rd_breakout &&
- best_rd < INT64_MAX) {
- if (tmp_best_rdu / 2 > best_rd) {
- // skip searching the other filters if the first is
- // already substantially larger than the best so far
- tmp_best_filter = mbmi->interp_filter;
- tmp_best_rdu = INT64_MAX;
- break;
- }
- }
- }
- } // switchable_filter_index loop
- }
- }
-
- if (tmp_best_rdu == INT64_MAX)
- continue;
-
- mbmi->interp_filter = (cm->mcomp_filter_type == SWITCHABLE ?
- tmp_best_filter : cm->mcomp_filter_type);
- vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
- if (!pred_exists) {
- // Handles the special case when a filter that is not in the
- // switchable list (bilinear, 6-tap) is indicated at the frame level
- tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
- &mbmi->ref_mvs[ref_frame][0],
- second_ref,
- best_yrd,
- &rate, &rate_y, &distortion,
- &skippable, &total_sse,
- (int)this_rd_thresh, seg_mvs,
- bsi, 0,
- mi_row, mi_col);
- if (tmp_rd == INT64_MAX)
- continue;
- } else {
- if (cpi->common.mcomp_filter_type == SWITCHABLE) {
- int rs = get_switchable_rate(x);
- tmp_best_rdu -= RDCOST(x->rdmult, x->rddiv, rs, 0);
- }
- tmp_rd = tmp_best_rdu;
- total_sse = tmp_best_sse;
- rate = tmp_best_rate;
- rate_y = tmp_best_ratey;
- distortion = tmp_best_distortion;
- skippable = tmp_best_skippable;
- *mbmi = tmp_best_mbmode;
- *x->partition_info = tmp_best_partition;
- for (i = 0; i < 4; i++)
- xd->this_mi->bmi[i] = tmp_best_bmodes[i];
- }
-
- rate2 += rate;
- distortion2 += distortion;
-
- if (cpi->common.mcomp_filter_type == SWITCHABLE)
- rate2 += get_switchable_rate(x);
-
- if (!mode_excluded) {
- if (is_comp_pred)
- mode_excluded = cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY;
- else
- mode_excluded = cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY;
- }
- compmode_cost = vp9_cost_bit(comp_mode_p, is_comp_pred);
-
- tmp_best_rdu = best_rd -
- MIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
- RDCOST(x->rdmult, x->rddiv, 0, total_sse));
-
- if (tmp_best_rdu > 0) {
- // If even the 'Y' rd value of split is higher than best so far
- // then dont bother looking at UV
- vp9_build_inter_predictors_sbuv(&x->e_mbd, mi_row, mi_col,
- BLOCK_8X8);
- super_block_uvrd(cm, x, &rate_uv, &distortion_uv, &uv_skippable,
- &uv_sse, BLOCK_8X8, tmp_best_rdu);
- if (rate_uv == INT_MAX)
- continue;
- rate2 += rate_uv;
- distortion2 += distortion_uv;
- skippable = skippable && uv_skippable;
- total_sse += uv_sse;
-
- tx_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
- for (i = 0; i < TX_MODES; ++i)
- tx_cache[i] = tx_cache[ONLY_4X4];
- }
} else {
- mbmi->mode = rd_mode_to_mode(this_mode);
+ mbmi->mode = this_mode;
compmode_cost = vp9_cost_bit(comp_mode_p, second_ref_frame > INTRA_FRAME);
this_rd = handle_inter_mode(cpi, x, bsize,
tx_cache,
const int mb_skip_allowed = !vp9_segfeature_active(seg, segment_id,
SEG_LVL_SKIP);
- if (skippable && bsize >= BLOCK_8X8) {
+ if (skippable) {
// Back out the coefficient coding costs
rate2 -= (rate_y + rate_uv);
// for best yrd calculation
this_rd < best_inter_rd) {
best_inter_rd = this_rd;
best_inter_ref_frame = ref_frame;
- // best_inter_mode = xd->this_mi->mbmi.mode;
}
if (!disable_skip && ref_frame == INTRA_FRAME) {
best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
}
- if (this_mode != RD_I4X4_PRED && this_mode != RD_SPLITMV) {
- // Store the respective mode distortions for later use.
- if (mode_distortions[this_mode] == -1
- || distortion2 < mode_distortions[this_mode]) {
- mode_distortions[this_mode] = distortion2;
- }
- if (frame_distortions[ref_frame] == -1
- || distortion2 < frame_distortions[ref_frame]) {
- frame_distortions[ref_frame] = distortion2;
- }
+ // Store the respective mode distortions for later use.
+ if (mode_distortions[this_mode] == -1
+ || distortion2 < mode_distortions[this_mode]) {
+ mode_distortions[this_mode] = distortion2;
+ }
+ if (frame_distortions[ref_frame] == -1
+ || distortion2 < frame_distortions[ref_frame]) {
+ frame_distortions[ref_frame] = distortion2;
}
// Did this mode help.. i.e. is it the new best mode
*returnrate = rate2;
*returndistortion = distortion2;
best_rd = this_rd;
- best_yrd = best_rd -
- RDCOST(x->rdmult, x->rddiv, rate_uv, distortion_uv);
best_mbmode = *mbmi;
best_skip2 = this_skip2;
- best_partition = *x->partition_info;
vpx_memcpy(best_zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
sizeof(best_zcoeff_blk));
- if (this_mode == RD_I4X4_PRED || this_mode == RD_SPLITMV)
- for (i = 0; i < 4; i++)
- best_bmodes[i] = xd->this_mi->bmi[i];
-
// TODO(debargha): enhance this test with a better distortion prediction
// based on qp, activity mask and history
- if (cpi->sf.mode_search_skip_flags & FLAG_EARLY_TERMINATE) {
+ if ((cpi->sf.mode_search_skip_flags & FLAG_EARLY_TERMINATE) &&
+ (mode_index > MIN_EARLY_TERM_INDEX)) {
const int qstep = xd->plane[0].dequant[1];
// TODO(debargha): Enhance this by specializing for each mode_index
int scale = 4;
/* keep record of best txfm size */
if (bsize < BLOCK_32X32) {
- if (bsize < BLOCK_16X16) {
- if (this_mode == RD_SPLITMV || this_mode == RD_I4X4_PRED)
- tx_cache[ALLOW_8X8] = tx_cache[ONLY_4X4];
+ if (bsize < BLOCK_16X16)
tx_cache[ALLOW_16X16] = tx_cache[ALLOW_8X8];
- }
+
tx_cache[ALLOW_32X32] = tx_cache[ALLOW_16X16];
}
if (!mode_excluded && this_rd != INT64_MAX) {
for (i = 0; i < TX_MODES && tx_cache[i] < INT64_MAX; i++) {
int64_t adj_rd = INT64_MAX;
- if (this_mode != RD_I4X4_PRED) {
- adj_rd = this_rd + tx_cache[i] - tx_cache[cm->tx_mode];
- } else {
- adj_rd = this_rd;
- }
+ adj_rd = this_rd + tx_cache[i] - tx_cache[cm->tx_mode];
if (adj_rd < best_tx_rd[i])
best_tx_rd[i] = adj_rd;
}
}
- if (best_rd == INT64_MAX && bsize < BLOCK_8X8) {
- *returnrate = INT_MAX;
- *returndistortion = INT_MAX;
- return best_rd;
+ assert((cm->mcomp_filter_type == SWITCHABLE) ||
+ (cm->mcomp_filter_type == best_mbmode.interp_filter) ||
+ (best_mbmode.ref_frame[0] == INTRA_FRAME));
+
+ // Updating rd_thresh_freq_fact[] here means that the different
+ // partition/block sizes are handled independently based on the best
+ // choice for the current partition. It may well be better to keep a scaled
+ // best rd so far value and update rd_thresh_freq_fact based on the mode/size
+ // combination that wins out.
+ if (cpi->sf.adaptive_rd_thresh) {
+ for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
+ if (mode_index == best_mode_index) {
+ cpi->rd_thresh_freq_fact[bsize][mode_index] -=
+ (cpi->rd_thresh_freq_fact[bsize][mode_index] >> 3);
+ } else {
+ cpi->rd_thresh_freq_fact[bsize][mode_index] += RD_THRESH_INC;
+ if (cpi->rd_thresh_freq_fact[bsize][mode_index] >
+ (cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT)) {
+ cpi->rd_thresh_freq_fact[bsize][mode_index] =
+ cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT;
+ }
+ }
+ }
+ }
+
+ // macroblock modes
+ *mbmi = best_mbmode;
+ x->skip |= best_skip2;
+
+ vpx_memcpy(x->zcoeff_blk[mbmi->tx_size], best_zcoeff_blk,
+ sizeof(best_zcoeff_blk));
+
+ for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
+ if (best_pred_rd[i] == INT64_MAX)
+ best_pred_diff[i] = INT_MIN;
+ else
+ best_pred_diff[i] = best_rd - best_pred_rd[i];
+ }
+
+ if (!x->skip) {
+ for (i = 0; i <= SWITCHABLE_FILTERS; i++) {
+ if (best_filter_rd[i] == INT64_MAX)
+ best_filter_diff[i] = 0;
+ else
+ best_filter_diff[i] = best_rd - best_filter_rd[i];
+ }
+ if (cm->mcomp_filter_type == SWITCHABLE)
+ assert(best_filter_diff[SWITCHABLE_FILTERS] == 0);
+ } else {
+ vp9_zero(best_filter_diff);
+ }
+
+ if (!x->skip) {
+ for (i = 0; i < TX_MODES; i++) {
+ if (best_tx_rd[i] == INT64_MAX)
+ best_tx_diff[i] = 0;
+ else
+ best_tx_diff[i] = best_rd - best_tx_rd[i];
+ }
+ } else {
+ vp9_zero(best_tx_diff);
+ }
+
+ set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
+ scale_factor);
+ store_coding_context(x, ctx, best_mode_index,
+ NULL,
+ &mbmi->ref_mvs[mbmi->ref_frame[0]][0],
+ &mbmi->ref_mvs[mbmi->ref_frame[1] < 0 ? 0 :
+ mbmi->ref_frame[1]][0],
+ best_pred_diff, best_tx_diff, best_filter_diff);
+
+ return best_rd;
+}
+
+
+int64_t vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
+ int mi_row, int mi_col,
+ int *returnrate,
+ int64_t *returndistortion,
+ BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx,
+ int64_t best_rd_so_far) {
+ VP9_COMMON *cm = &cpi->common;
+ MACROBLOCKD *xd = &x->e_mbd;
+ MB_MODE_INFO *mbmi = &xd->this_mi->mbmi;
+ const struct segmentation *seg = &cm->seg;
+ const BLOCK_SIZE block_size = get_plane_block_size(bsize, &xd->plane[0]);
+ MV_REFERENCE_FRAME ref_frame, second_ref_frame;
+ unsigned char segment_id = mbmi->segment_id;
+ int comp_pred, i;
+ int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
+ struct buf_2d yv12_mb[4][MAX_MB_PLANE];
+ static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
+ VP9_ALT_FLAG };
+ int idx_list[4] = {0,
+ cpi->lst_fb_idx,
+ cpi->gld_fb_idx,
+ cpi->alt_fb_idx};
+ int64_t best_rd = best_rd_so_far;
+ int64_t best_yrd = best_rd_so_far; // FIXME(rbultje) more precise
+ int64_t best_tx_rd[TX_MODES];
+ int64_t best_tx_diff[TX_MODES];
+ int64_t best_pred_diff[NB_PREDICTION_TYPES];
+ int64_t best_pred_rd[NB_PREDICTION_TYPES];
+ int64_t best_filter_rd[SWITCHABLE_FILTERS + 1];
+ int64_t best_filter_diff[SWITCHABLE_FILTERS + 1];
+ MB_MODE_INFO best_mbmode = { 0 };
+ int mode_index, best_mode_index = 0;
+ unsigned int ref_costs_single[MAX_REF_FRAMES], ref_costs_comp[MAX_REF_FRAMES];
+ vp9_prob comp_mode_p;
+ int64_t best_inter_rd = INT64_MAX;
+ MV_REFERENCE_FRAME best_inter_ref_frame = LAST_FRAME;
+ INTERPOLATIONFILTERTYPE tmp_best_filter = SWITCHABLE;
+ int rate_uv_intra[TX_SIZES], rate_uv_tokenonly[TX_SIZES];
+ int64_t dist_uv[TX_SIZES];
+ int skip_uv[TX_SIZES];
+ MB_PREDICTION_MODE mode_uv[TX_SIZES] = { 0 };
+ struct scale_factors scale_factor[4];
+ unsigned int ref_frame_mask = 0;
+ unsigned int mode_mask = 0;
+ int intra_cost_penalty = 20 * vp9_dc_quant(cpi->common.base_qindex,
+ cpi->common.y_dc_delta_q);
+ int_mv seg_mvs[4][MAX_REF_FRAMES];
+ b_mode_info best_bmodes[4];
+ PARTITION_INFO best_partition;
+ int best_skip2 = 0;
+ unsigned char best_zcoeff_blk[256] = { 0 };
+
+ x->skip_encode = cpi->sf.skip_encode_frame && xd->q_index < QIDX_SKIP_THRESH;
+ vp9_zero(x->zcoeff_blk);
+ vp9_zero(ctx->zcoeff_blk);
+
+ for (i = 0; i < 4; i++) {
+ int j;
+ for (j = 0; j < MAX_REF_FRAMES; j++)
+ seg_mvs[i][j].as_int = INVALID_MV;
+ }
+
+ estimate_ref_frame_costs(cpi, segment_id, ref_costs_single, ref_costs_comp,
+ &comp_mode_p);
+
+ for (i = 0; i < NB_PREDICTION_TYPES; ++i)
+ best_pred_rd[i] = INT64_MAX;
+ for (i = 0; i < TX_MODES; i++)
+ best_tx_rd[i] = INT64_MAX;
+ for (i = 0; i <= SWITCHABLE_FILTERS; i++)
+ best_filter_rd[i] = INT64_MAX;
+ for (i = 0; i < TX_SIZES; i++)
+ rate_uv_intra[i] = INT_MAX;
+
+ *returnrate = INT_MAX;
+
+ // Create a mask set to 1 for each reference frame used by a smaller
+ // resolution.
+ if (cpi->sf.use_avoid_tested_higherror) {
+ ref_frame_mask = 0;
+ mode_mask = 0;
+ ref_frame_mask = ~ref_frame_mask;
+ mode_mask = ~mode_mask;
+ }
+
+ for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
+ if (cpi->ref_frame_flags & flag_list[ref_frame]) {
+ setup_buffer_inter(cpi, x, idx_list[ref_frame], ref_frame, block_size,
+ mi_row, mi_col, frame_mv[NEARESTMV], frame_mv[NEARMV],
+ yv12_mb, scale_factor);
+ }
+ frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
+ frame_mv[ZEROMV][ref_frame].as_int = 0;
+ }
+
+ for (mode_index = 0; mode_index < MAX_REFS; ++mode_index) {
+ int mode_excluded = 0;
+ int64_t this_rd = INT64_MAX;
+ int disable_skip = 0;
+ int compmode_cost = 0;
+ int rate2 = 0, rate_y = 0, rate_uv = 0;
+ int64_t distortion2 = 0, distortion_y = 0, distortion_uv = 0;
+ int skippable = 0;
+ int64_t tx_cache[TX_MODES];
+ int i;
+ int this_skip2 = 0;
+ int64_t total_sse = INT_MAX;
+ int early_term = 0;
+
+ for (i = 0; i < TX_MODES; ++i)
+ tx_cache[i] = INT64_MAX;
+
+ x->skip = 0;
+ ref_frame = vp9_ref_order[mode_index].ref_frame;
+ second_ref_frame = vp9_ref_order[mode_index].second_ref_frame;
+
+ // Skip if the current reference frame has been masked off
+ if (cpi->sf.reference_masking && !cpi->set_ref_frame_mask &&
+ (cpi->ref_frame_mask & (1 << ref_frame)))
+ continue;
+
+ // Test best rd so far against threshold for trying this mode.
+ if ((best_rd < ((int64_t)cpi->rd_thresh_sub8x8[bsize][mode_index] *
+ cpi->rd_thresh_freq_sub8x8[bsize][mode_index] >> 5)) ||
+ cpi->rd_thresh_sub8x8[bsize][mode_index] == INT_MAX)
+ continue;
+
+ // Do not allow compound prediction if the segment level reference
+ // frame feature is in use as in this case there can only be one reference.
+ if ((second_ref_frame > INTRA_FRAME) &&
+ vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
+ continue;
+
+ mbmi->ref_frame[0] = ref_frame;
+ mbmi->ref_frame[1] = second_ref_frame;
+
+ if (!(ref_frame == INTRA_FRAME
+ || (cpi->ref_frame_flags & flag_list[ref_frame]))) {
+ continue;
+ }
+ if (!(second_ref_frame == NONE
+ || (cpi->ref_frame_flags & flag_list[second_ref_frame]))) {
+ continue;
+ }
+
+ comp_pred = second_ref_frame > INTRA_FRAME;
+ if (comp_pred) {
+ if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA)
+ if (vp9_ref_order[best_mode_index].ref_frame == INTRA_FRAME)
+ continue;
+ if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_REFMISMATCH)
+ if (ref_frame != best_inter_ref_frame &&
+ second_ref_frame != best_inter_ref_frame)
+ continue;
+ }
+
+ // TODO(jingning, jkoleszar): scaling reference frame not supported for
+ // sub8x8 blocks.
+ if (ref_frame > 0 &&
+ vp9_is_scaled(&scale_factor[ref_frame]))
+ continue;
+
+ if (second_ref_frame > 0 &&
+ vp9_is_scaled(&scale_factor[second_ref_frame]))
+ continue;
+
+ set_scale_factors(xd, ref_frame, second_ref_frame, scale_factor);
+ mbmi->uv_mode = DC_PRED;
+
+ // Evaluate all sub-pel filters irrespective of whether we can use
+ // them for this frame.
+ mbmi->interp_filter = cm->mcomp_filter_type;
+ vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+
+ if (comp_pred) {
+ if (!(cpi->ref_frame_flags & flag_list[second_ref_frame]))
+ continue;
+ set_scale_factors(xd, ref_frame, second_ref_frame, scale_factor);
+
+ mode_excluded = mode_excluded
+ ? mode_excluded
+ : cm->comp_pred_mode == SINGLE_PREDICTION_ONLY;
+ } else {
+ if (ref_frame != INTRA_FRAME && second_ref_frame != INTRA_FRAME) {
+ mode_excluded =
+ mode_excluded ?
+ mode_excluded : cm->comp_pred_mode == COMP_PREDICTION_ONLY;
+ }
+ }
+
+ // Select prediction reference frames.
+ for (i = 0; i < MAX_MB_PLANE; i++) {
+ xd->plane[i].pre[0] = yv12_mb[ref_frame][i];
+ if (comp_pred)
+ xd->plane[i].pre[1] = yv12_mb[second_ref_frame][i];
+ }
+
+ // If the segment reference frame feature is enabled....
+ // then do nothing if the current ref frame is not allowed..
+ if (vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
+ vp9_get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) !=
+ (int)ref_frame) {
+ continue;
+ // If the segment skip feature is enabled....
+ // then do nothing if the current mode is not allowed..
+ } else if (vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP) &&
+ ref_frame != INTRA_FRAME) {
+ continue;
+ // Disable this drop out case if the ref frame
+ // segment level feature is enabled for this segment. This is to
+ // prevent the possibility that we end up unable to pick any mode.
+ } else if (!vp9_segfeature_active(seg, segment_id,
+ SEG_LVL_REF_FRAME)) {
+ // Only consider ZEROMV/ALTREF_FRAME for alt ref frame,
+ // unless ARNR filtering is enabled in which case we want
+ // an unfiltered alternative. We allow near/nearest as well
+ // because they may result in zero-zero MVs but be cheaper.
+ if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0))
+ continue;
+ }
+
+#ifdef MODE_TEST_HIT_STATS
+ // TEST/DEBUG CODE
+ // Keep a rcord of the number of test hits at each size
+ cpi->mode_test_hits[bsize]++;
+#endif
+
+ if (ref_frame == INTRA_FRAME) {
+ int rate;
+ mbmi->tx_size = TX_4X4;
+ if (rd_pick_intra_sub_8x8_y_mode(cpi, x, &rate, &rate_y,
+ &distortion_y, best_rd) >= best_rd)
+ continue;
+ rate2 += rate;
+ rate2 += intra_cost_penalty;
+ distortion2 += distortion_y;
+
+ if (rate_uv_intra[TX_4X4] == INT_MAX) {
+ choose_intra_uv_mode(cpi, bsize, &rate_uv_intra[TX_4X4],
+ &rate_uv_tokenonly[TX_4X4],
+ &dist_uv[TX_4X4], &skip_uv[TX_4X4],
+ &mode_uv[TX_4X4]);
+ }
+ rate2 += rate_uv_intra[TX_4X4];
+ rate_uv = rate_uv_tokenonly[TX_4X4];
+ distortion2 += dist_uv[TX_4X4];
+ distortion_uv = dist_uv[TX_4X4];
+ mbmi->uv_mode = mode_uv[TX_4X4];
+ tx_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+ for (i = 0; i < TX_MODES; ++i)
+ tx_cache[i] = tx_cache[ONLY_4X4];
+ } else {
+ int rate;
+ int64_t distortion;
+ int64_t this_rd_thresh;
+ int64_t tmp_rd, tmp_best_rd = INT64_MAX, tmp_best_rdu = INT64_MAX;
+ int tmp_best_rate = INT_MAX, tmp_best_ratey = INT_MAX;
+ int64_t tmp_best_distortion = INT_MAX, tmp_best_sse, uv_sse;
+ int tmp_best_skippable = 0;
+ int switchable_filter_index;
+ int_mv *second_ref = comp_pred ?
+ &mbmi->ref_mvs[second_ref_frame][0] : NULL;
+ b_mode_info tmp_best_bmodes[16];
+ MB_MODE_INFO tmp_best_mbmode;
+ PARTITION_INFO tmp_best_partition;
+ BEST_SEG_INFO bsi[SWITCHABLE_FILTERS];
+ int pred_exists = 0;
+ int uv_skippable;
+
+ this_rd_thresh = (ref_frame == LAST_FRAME) ?
+ cpi->rd_thresh_sub8x8[bsize][THR_LAST] :
+ cpi->rd_thresh_sub8x8[bsize][THR_ALTR];
+ this_rd_thresh = (ref_frame == GOLDEN_FRAME) ?
+ cpi->rd_thresh_sub8x8[bsize][THR_GOLD] : this_rd_thresh;
+ xd->this_mi->mbmi.tx_size = TX_4X4;
+
+ cpi->rd_filter_cache[SWITCHABLE_FILTERS] = INT64_MAX;
+ if (cm->mcomp_filter_type != BILINEAR) {
+ tmp_best_filter = EIGHTTAP;
+ if (x->source_variance <
+ cpi->sf.disable_filter_search_var_thresh) {
+ tmp_best_filter = EIGHTTAP;
+ vp9_zero(cpi->rd_filter_cache);
+ } else {
+ for (switchable_filter_index = 0;
+ switchable_filter_index < SWITCHABLE_FILTERS;
+ ++switchable_filter_index) {
+ int newbest, rs;
+ int64_t rs_rd;
+ mbmi->interp_filter = switchable_filter_index;
+ vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+
+ tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+ &mbmi->ref_mvs[ref_frame][0],
+ second_ref,
+ best_yrd,
+ &rate, &rate_y, &distortion,
+ &skippable, &total_sse,
+ (int)this_rd_thresh, seg_mvs,
+ bsi, switchable_filter_index,
+ mi_row, mi_col);
+
+ if (tmp_rd == INT64_MAX)
+ continue;
+ cpi->rd_filter_cache[switchable_filter_index] = tmp_rd;
+ rs = get_switchable_rate(x);
+ rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
+ cpi->rd_filter_cache[SWITCHABLE_FILTERS] =
+ MIN(cpi->rd_filter_cache[SWITCHABLE_FILTERS],
+ tmp_rd + rs_rd);
+ if (cm->mcomp_filter_type == SWITCHABLE)
+ tmp_rd += rs_rd;
+
+ newbest = (tmp_rd < tmp_best_rd);
+ if (newbest) {
+ tmp_best_filter = mbmi->interp_filter;
+ tmp_best_rd = tmp_rd;
+ }
+ if ((newbest && cm->mcomp_filter_type == SWITCHABLE) ||
+ (mbmi->interp_filter == cm->mcomp_filter_type &&
+ cm->mcomp_filter_type != SWITCHABLE)) {
+ tmp_best_rdu = tmp_rd;
+ tmp_best_rate = rate;
+ tmp_best_ratey = rate_y;
+ tmp_best_distortion = distortion;
+ tmp_best_sse = total_sse;
+ tmp_best_skippable = skippable;
+ tmp_best_mbmode = *mbmi;
+ tmp_best_partition = *x->partition_info;
+ for (i = 0; i < 4; i++)
+ tmp_best_bmodes[i] = xd->this_mi->bmi[i];
+ pred_exists = 1;
+ if (switchable_filter_index == 0 &&
+ cpi->sf.use_rd_breakout &&
+ best_rd < INT64_MAX) {
+ if (tmp_best_rdu / 2 > best_rd) {
+ // skip searching the other filters if the first is
+ // already substantially larger than the best so far
+ tmp_best_filter = mbmi->interp_filter;
+ tmp_best_rdu = INT64_MAX;
+ break;
+ }
+ }
+ }
+ } // switchable_filter_index loop
+ }
+ }
+
+ if (tmp_best_rdu == INT64_MAX)
+ continue;
+
+ mbmi->interp_filter = (cm->mcomp_filter_type == SWITCHABLE ?
+ tmp_best_filter : cm->mcomp_filter_type);
+ vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+ if (!pred_exists) {
+ // Handles the special case when a filter that is not in the
+ // switchable list (bilinear, 6-tap) is indicated at the frame level
+ tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+ &mbmi->ref_mvs[ref_frame][0],
+ second_ref,
+ best_yrd,
+ &rate, &rate_y, &distortion,
+ &skippable, &total_sse,
+ (int)this_rd_thresh, seg_mvs,
+ bsi, 0,
+ mi_row, mi_col);
+ if (tmp_rd == INT64_MAX)
+ continue;
+ } else {
+ if (cpi->common.mcomp_filter_type == SWITCHABLE) {
+ int rs = get_switchable_rate(x);
+ tmp_best_rdu -= RDCOST(x->rdmult, x->rddiv, rs, 0);
+ }
+ tmp_rd = tmp_best_rdu;
+ total_sse = tmp_best_sse;
+ rate = tmp_best_rate;
+ rate_y = tmp_best_ratey;
+ distortion = tmp_best_distortion;
+ skippable = tmp_best_skippable;
+ *mbmi = tmp_best_mbmode;
+ *x->partition_info = tmp_best_partition;
+ for (i = 0; i < 4; i++)
+ xd->this_mi->bmi[i] = tmp_best_bmodes[i];
+ }
+
+ rate2 += rate;
+ distortion2 += distortion;
+
+ if (cpi->common.mcomp_filter_type == SWITCHABLE)
+ rate2 += get_switchable_rate(x);
+
+ if (!mode_excluded) {
+ if (comp_pred)
+ mode_excluded = cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY;
+ else
+ mode_excluded = cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY;
+ }
+ compmode_cost = vp9_cost_bit(comp_mode_p, comp_pred);
+
+ tmp_best_rdu = best_rd -
+ MIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
+ RDCOST(x->rdmult, x->rddiv, 0, total_sse));
+
+ if (tmp_best_rdu > 0) {
+ // If even the 'Y' rd value of split is higher than best so far
+ // then dont bother looking at UV
+ vp9_build_inter_predictors_sbuv(&x->e_mbd, mi_row, mi_col,
+ BLOCK_8X8);
+ super_block_uvrd(cm, x, &rate_uv, &distortion_uv, &uv_skippable,
+ &uv_sse, BLOCK_8X8, tmp_best_rdu);
+ if (rate_uv == INT_MAX)
+ continue;
+ rate2 += rate_uv;
+ distortion2 += distortion_uv;
+ skippable = skippable && uv_skippable;
+ total_sse += uv_sse;
+
+ tx_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+ for (i = 0; i < TX_MODES; ++i)
+ tx_cache[i] = tx_cache[ONLY_4X4];
+ }
+ }
+
+ if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+ rate2 += compmode_cost;
+ }
+
+ // Estimate the reference frame signaling cost and add it
+ // to the rolling cost variable.
+ if (second_ref_frame > INTRA_FRAME) {
+ rate2 += ref_costs_comp[ref_frame];
+ } else {
+ rate2 += ref_costs_single[ref_frame];
+ }
+
+ if (!disable_skip) {
+ // Test for the condition where skip block will be activated
+ // because there are no non zero coefficients and make any
+ // necessary adjustment for rate. Ignore if skip is coded at
+ // segment level as the cost wont have been added in.
+ // Is Mb level skip allowed (i.e. not coded at segment level).
+ const int mb_skip_allowed = !vp9_segfeature_active(seg, segment_id,
+ SEG_LVL_SKIP);
+
+ if (mb_skip_allowed && ref_frame != INTRA_FRAME && !xd->lossless) {
+ if (RDCOST(x->rdmult, x->rddiv, rate_y + rate_uv, distortion2) <
+ RDCOST(x->rdmult, x->rddiv, 0, total_sse)) {
+ // Add in the cost of the no skip flag.
+ int prob_skip_cost = vp9_cost_bit(vp9_get_pred_prob_mbskip(cm, xd),
+ 0);
+ rate2 += prob_skip_cost;
+ } else {
+ // FIXME(rbultje) make this work for splitmv also
+ int prob_skip_cost = vp9_cost_bit(vp9_get_pred_prob_mbskip(cm, xd),
+ 1);
+ rate2 += prob_skip_cost;
+ distortion2 = total_sse;
+ assert(total_sse >= 0);
+ rate2 -= (rate_y + rate_uv);
+ rate_y = 0;
+ rate_uv = 0;
+ this_skip2 = 1;
+ }
+ } else if (mb_skip_allowed) {
+ // Add in the cost of the no skip flag.
+ int prob_skip_cost = vp9_cost_bit(vp9_get_pred_prob_mbskip(cm, xd),
+ 0);
+ rate2 += prob_skip_cost;
+ }
+
+ // Calculate the final RD estimate for this mode.
+ this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+ }
+
+ // Keep record of best inter rd with single reference
+ if (xd->this_mi->mbmi.ref_frame[0] > INTRA_FRAME &&
+ xd->this_mi->mbmi.ref_frame[1] == NONE &&
+ !mode_excluded &&
+ this_rd < best_inter_rd) {
+ best_inter_rd = this_rd;
+ best_inter_ref_frame = ref_frame;
+ }
+
+ if (!disable_skip && ref_frame == INTRA_FRAME) {
+ for (i = 0; i < NB_PREDICTION_TYPES; ++i)
+ best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+ for (i = 0; i <= SWITCHABLE_FILTERS; i++)
+ best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
+ }
+
+ // Did this mode help.. i.e. is it the new best mode
+ if (this_rd < best_rd || x->skip) {
+ if (!mode_excluded) {
+ // Note index of best mode so far
+ best_mode_index = mode_index;
+
+ if (ref_frame == INTRA_FRAME) {
+ /* required for left and above block mv */
+ mbmi->mv[0].as_int = 0;
+ }
+
+ *returnrate = rate2;
+ *returndistortion = distortion2;
+ best_rd = this_rd;
+ best_yrd = best_rd -
+ RDCOST(x->rdmult, x->rddiv, rate_uv, distortion_uv);
+ best_mbmode = *mbmi;
+ best_skip2 = this_skip2;
+ best_partition = *x->partition_info;
+ vpx_memcpy(best_zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
+ sizeof(best_zcoeff_blk));
+
+ for (i = 0; i < 4; i++)
+ best_bmodes[i] = xd->this_mi->bmi[i];
+
+ // TODO(debargha): enhance this test with a better distortion prediction
+ // based on qp, activity mask and history
+ if ((cpi->sf.mode_search_skip_flags & FLAG_EARLY_TERMINATE) &&
+ (mode_index > MIN_EARLY_TERM_INDEX)) {
+ const int qstep = xd->plane[0].dequant[1];
+ // TODO(debargha): Enhance this by specializing for each mode_index
+ int scale = 4;
+ if (x->source_variance < UINT_MAX) {
+ const int var_adjust = (x->source_variance < 16);
+ scale -= var_adjust;
+ }
+ if (ref_frame > INTRA_FRAME &&
+ distortion2 * scale < qstep * qstep) {
+ early_term = 1;
+ }
+ }
+ }
+ }
+
+ /* keep record of best compound/single-only prediction */
+ if (!disable_skip && ref_frame != INTRA_FRAME) {
+ int single_rd, hybrid_rd, single_rate, hybrid_rate;
+
+ if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+ single_rate = rate2 - compmode_cost;
+ hybrid_rate = rate2;
+ } else {
+ single_rate = rate2;
+ hybrid_rate = rate2 + compmode_cost;
+ }
+
+ single_rd = RDCOST(x->rdmult, x->rddiv, single_rate, distortion2);
+ hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
+
+ if (second_ref_frame <= INTRA_FRAME &&
+ single_rd < best_pred_rd[SINGLE_PREDICTION_ONLY]) {
+ best_pred_rd[SINGLE_PREDICTION_ONLY] = single_rd;
+ } else if (second_ref_frame > INTRA_FRAME &&
+ single_rd < best_pred_rd[COMP_PREDICTION_ONLY]) {
+ best_pred_rd[COMP_PREDICTION_ONLY] = single_rd;
+ }
+ if (hybrid_rd < best_pred_rd[HYBRID_PREDICTION])
+ best_pred_rd[HYBRID_PREDICTION] = hybrid_rd;
+ }
+
+ /* keep record of best filter type */
+ if (!mode_excluded && !disable_skip && ref_frame != INTRA_FRAME &&
+ cm->mcomp_filter_type != BILINEAR) {
+ int64_t ref = cpi->rd_filter_cache[cm->mcomp_filter_type == SWITCHABLE ?
+ SWITCHABLE_FILTERS : cm->mcomp_filter_type];
+ for (i = 0; i <= SWITCHABLE_FILTERS; i++) {
+ int64_t adj_rd;
+ // In cases of poor prediction, filter_cache[] can contain really big
+ // values, which actually are bigger than this_rd itself. This can
+ // cause negative best_filter_rd[] values, which is obviously silly.
+ // Therefore, if filter_cache < ref, we do an adjusted calculation.
+ if (cpi->rd_filter_cache[i] >= ref)
+ adj_rd = this_rd + cpi->rd_filter_cache[i] - ref;
+ else // FIXME(rbultje) do this for comppred also
+ adj_rd = this_rd - (ref - cpi->rd_filter_cache[i]) * this_rd / ref;
+ best_filter_rd[i] = MIN(best_filter_rd[i], adj_rd);
+ }
+ }
+
+ /* keep record of best txfm size */
+ if (bsize < BLOCK_32X32) {
+ if (bsize < BLOCK_16X16) {
+ tx_cache[ALLOW_8X8] = tx_cache[ONLY_4X4];
+ tx_cache[ALLOW_16X16] = tx_cache[ALLOW_8X8];
+ }
+ tx_cache[ALLOW_32X32] = tx_cache[ALLOW_16X16];
+ }
+ if (!mode_excluded && this_rd != INT64_MAX) {
+ for (i = 0; i < TX_MODES && tx_cache[i] < INT64_MAX; i++) {
+ int64_t adj_rd = INT64_MAX;
+ if (ref_frame > INTRA_FRAME)
+ adj_rd = this_rd + tx_cache[i] - tx_cache[cm->tx_mode];
+ else
+ adj_rd = this_rd;
+
+ if (adj_rd < best_tx_rd[i])
+ best_tx_rd[i] = adj_rd;
+ }
+ }
+
+ if (early_term)
+ break;
+
+ if (x->skip && !comp_pred)
+ break;
+ }
+
+ if (best_rd >= best_rd_so_far)
+ return INT64_MAX;
+
+ // If we used an estimate for the uv intra rd in the loop above...
+ if (cpi->sf.use_uv_intra_rd_estimate) {
+ // Do Intra UV best rd mode selection if best mode choice above was intra.
+ if (vp9_ref_order[best_mode_index].ref_frame == INTRA_FRAME) {
+ TX_SIZE uv_tx_size = get_uv_tx_size(mbmi);
+ rd_pick_intra_sbuv_mode(cpi, x, &rate_uv_intra[uv_tx_size],
+ &rate_uv_tokenonly[uv_tx_size],
+ &dist_uv[uv_tx_size],
+ &skip_uv[uv_tx_size],
+ BLOCK_8X8);
+ }
+ }
+
+ // If we are using reference masking and the set mask flag is set then
+ // create the reference frame mask.
+ if (cpi->sf.reference_masking && cpi->set_ref_frame_mask)
+ cpi->ref_frame_mask = ~(1 << vp9_ref_order[best_mode_index].ref_frame);
+
+ if (best_rd == INT64_MAX && bsize < BLOCK_8X8) {
+ *returnrate = INT_MAX;
+ *returndistortion = INT_MAX;
+ return best_rd;
}
assert((cm->mcomp_filter_type == SWITCHABLE) ||
// best rd so far value and update rd_thresh_freq_fact based on the mode/size
// combination that wins out.
if (cpi->sf.adaptive_rd_thresh) {
- for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
+ for (mode_index = 0; mode_index < MAX_REFS; ++mode_index) {
if (mode_index == best_mode_index) {
- cpi->rd_thresh_freq_fact[bsize][mode_index] -=
- (cpi->rd_thresh_freq_fact[bsize][mode_index] >> 3);
+ cpi->rd_thresh_freq_sub8x8[bsize][mode_index] -=
+ (cpi->rd_thresh_freq_sub8x8[bsize][mode_index] >> 3);
} else {
- cpi->rd_thresh_freq_fact[bsize][mode_index] += RD_THRESH_INC;
- if (cpi->rd_thresh_freq_fact[bsize][mode_index] >
+ cpi->rd_thresh_freq_sub8x8[bsize][mode_index] += RD_THRESH_INC;
+ if (cpi->rd_thresh_freq_sub8x8[bsize][mode_index] >
(cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT)) {
- cpi->rd_thresh_freq_fact[bsize][mode_index] =
+ cpi->rd_thresh_freq_sub8x8[bsize][mode_index] =
cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT;
}
}
// macroblock modes
*mbmi = best_mbmode;
x->skip |= best_skip2;
- if (best_mbmode.ref_frame[0] == INTRA_FRAME &&
- best_mbmode.sb_type < BLOCK_8X8) {
+ if (best_mbmode.ref_frame[0] == INTRA_FRAME) {
for (i = 0; i < 4; i++)
xd->this_mi->bmi[i].as_mode = best_bmodes[i].as_mode;
- }
-
- if (best_mbmode.ref_frame[0] != INTRA_FRAME &&
- best_mbmode.sb_type < BLOCK_8X8) {
+ } else {
for (i = 0; i < 4; i++)
xd->this_mi->bmi[i].as_mv[0].as_int =
best_bmodes[i].as_mv[0].as_int;
- if (mbmi->ref_frame[1] > 0)
+ if (has_second_ref(mbmi))
for (i = 0; i < 4; i++)
- xd->this_mi->bmi[i].as_mv[1].as_int =
- best_bmodes[i].as_mv[1].as_int;
+ xd->this_mi->bmi[i].as_mv[1].as_int = best_bmodes[i].as_mv[1].as_int;
*x->partition_info = best_partition;
if (cm->mcomp_filter_type == SWITCHABLE)
assert(best_filter_diff[SWITCHABLE_FILTERS] == 0);
} else {
- vpx_memset(best_filter_diff, 0, sizeof(best_filter_diff));
+ vp9_zero(best_filter_diff);
}
if (!x->skip) {
best_tx_diff[i] = best_rd - best_tx_rd[i];
}
} else {
- vpx_memset(best_tx_diff, 0, sizeof(best_tx_diff));
+ vp9_zero(best_tx_diff);
}
set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
return best_rd;
}
+
#ifndef VP9_ENCODER_VP9_RDOPT_H_
#define VP9_ENCODER_VP9_RDOPT_H_
-#define RDCOST(RM,DM,R,D) ( ((128+((int64_t)R)*(RM)) >> 8) + ((int64_t)DM)*(D) )
+#define RDCOST(RM, DM, R, D) \
+ (((128 + ((int64_t)R) * (RM)) >> 8) + ((int64_t)DM) * (D))
#define QIDX_SKIP_THRESH 115
void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex);
int *r, int64_t *d, BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx, int64_t best_rd);
+int64_t vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
+ int mi_row, int mi_col,
+ int *r, int64_t *d, BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx, int64_t best_rd);
+
void vp9_init_me_luts();
void vp9_set_mbmode_and_mvs(MACROBLOCK *x,
}
}
-const static int64_t cc1 = 26634; // (64^2*(.01*255)^2
-const static int64_t cc2 = 239708; // (64^2*(.03*255)^2
+static const int64_t cc1 = 26634; // (64^2*(.01*255)^2
+static const int64_t cc2 = 239708; // (64^2*(.03*255)^2
static double similarity(unsigned long sum_s, unsigned long sum_r,
unsigned long sum_sq_s, unsigned long sum_sq_r,
#include "vpx_ports/vpx_timer.h"
#define ALT_REF_MC_ENABLED 1 // dis/enable MC in AltRef filtering
-#define ALT_REF_SUBPEL_ENABLED 1 // dis/enable subpel in MC AltRef filtering
+#define ALT_REF_SUBPEL_ENABLED 1 // dis/enable subpel in MC AltRef filtering
static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
uint8_t *y_mb_ptr,
for (i = 0, k = 0; i < block_size; i++) {
for (j = 0; j < block_size; j++, k++) {
-
int src_byte = frame1[byte];
int pixel_value = *frame2++;
step_param = MIN(step_param, (cpi->sf.max_step_search_steps - 2));
/*cpi->sf.search_method == HEX*/
- // TODO Check that the 16x16 vf & sdf are selected here
// Ignore mv costing by sending NULL pointer instead of cost arrays
ref_mv = &x->e_mbd.mi_8x8[0]->bmi[0].as_mv[0];
- bestsme = vp9_hex_search(x, &best_ref_mv1_full,
+ bestsme = vp9_hex_search(x, &best_ref_mv1_full.as_mv,
step_param, sadpb, 1,
&cpi->fn_ptr[BLOCK_16X16],
- 0, &best_ref_mv1, ref_mv);
+ 0, &best_ref_mv1.as_mv, &ref_mv->as_mv);
#if ALT_REF_SUBPEL_ENABLED
// Try sub-pixel MC?
#ifdef DEBUGFWG
// DEBUG FWG
- printf("max:%d FBCK:%d FFWD:%d ftb:%d ftbbck:%d ftbfwd:%d sei:%d lasei:%d start:%d"
-, max_frames
-, num_frames_backward
-, num_frames_forward
-, frames_to_blur
-, frames_to_blur_backward
-, frames_to_blur_forward
-, cpi->source_encode_index
-, cpi->last_alt_ref_sei
-, start_frame);
+ printf(
+ "max:%d FBCK:%d FFWD:%d ftb:%d ftbbck:%d ftbfwd:%d sei:%d lasei:%d "
+ "start:%d",
+ max_frames, num_frames_backward, num_frames_forward, frames_to_blur,
+ frames_to_blur_backward, frames_to_blur_forward, cpi->source_encode_index,
+ cpi->last_alt_ref_sei, start_frame);
#endif
// Setup scaling factors. Scaling on each of the arnr frames is not supported
cm->width, cm->height);
// Setup frame pointers, NULL indicates frame not included in filter
- vpx_memset(cpi->frames, 0, max_frames * sizeof(YV12_BUFFER_CONFIG *));
+ vp9_zero(cpi->frames);
for (frame = 0; frame < frames_to_blur; frame++) {
int which_buffer = start_frame - frame;
struct lookahead_entry *buf = vp9_lookahead_peek(cpi->lookahead,
const int *vp9_dct_value_cost_ptr;
static void fill_value_tokens() {
-
TOKENVALUE *const t = dct_value_tokens + DCT_MAX_VALUE;
const vp9_extra_bit *const e = vp9_extra_bits;
t[i].token = --j;
eb |= (a - e[j].base_val) << 1;
- } else
+ } else {
t[i].token = a;
-
+ }
t[i].extra = eb;
}
cost += vp9_cost_bit(vp9_prob_half, extra & 1); /* sign */
dct_value_cost[i + DCT_MAX_VALUE] = cost;
}
-
}
-
} while (++i < DCT_MAX_VALUE);
vp9_dct_value_tokens_ptr = dct_value_tokens + DCT_MAX_VALUE;
pt = get_entropy_context(tx_size, pd->above_context + aoff,
pd->left_context + loff);
- get_scan_and_band(xd, tx_size, type, block, &scan, &band_translate);
- nb = vp9_get_coef_neighbors_handle(scan);
+ get_scan_and_band(xd, tx_size, type, block, &scan, &nb, &band_translate);
c = 0;
do {
const int band = get_coef_band(band_translate, c);
unsigned int *sse,
const uint8_t *second_pred);
-typedef void (*vp9_ssimpf_fn_t)(uint8_t *s, int sp, uint8_t *r,
- int rp, unsigned long *sum_s,
- unsigned long *sum_r, unsigned long *sum_sq_s,
- unsigned long *sum_sq_r,
- unsigned long *sum_sxr);
-
typedef unsigned int (*vp9_getmbss_fn_t)(const short *);
typedef unsigned int (*vp9_get16x16prederror_fn_t)(const uint8_t *src_ptr,
* be found in the AUTHORS file in the root of the source tree.
*/
+#include "./vp9_rtcd.h"
-#include "vp9/encoder/vp9_variance.h"
-#include "vp9/common/vp9_filter.h"
-#include "vp9/common/vp9_subpelvar.h"
-#include "vpx/vpx_integer.h"
#include "vpx_ports/mem.h"
-#include "./vp9_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/vp9_filter.h"
+#include "vp9/encoder/vp9_variance.h"
+
+static void variance(const uint8_t *src_ptr,
+ int source_stride,
+ const uint8_t *ref_ptr,
+ int recon_stride,
+ int w,
+ int h,
+ unsigned int *sse,
+ int *sum) {
+ int i, j;
+ int diff;
+
+ *sum = 0;
+ *sse = 0;
+
+ for (i = 0; i < h; i++) {
+ for (j = 0; j < w; j++) {
+ diff = src_ptr[j] - ref_ptr[j];
+ *sum += diff;
+ *sse += diff * diff;
+ }
+
+ src_ptr += source_stride;
+ ref_ptr += recon_stride;
+ }
+}
+
+/****************************************************************************
+ *
+ * ROUTINE : filter_block2d_bil_first_pass
+ *
+ * INPUTS : uint8_t *src_ptr : Pointer to source block.
+ * uint32_t src_pixels_per_line : Stride of input block.
+ * uint32_t pixel_step : Offset between filter input
+ * samples (see notes).
+ * uint32_t output_height : Input block height.
+ * uint32_t output_width : Input block width.
+ * int32_t *vp9_filter : Array of 2 bi-linear filter
+ * taps.
+ *
+ * OUTPUTS : int32_t *output_ptr : Pointer to filtered block.
+ *
+ * RETURNS : void
+ *
+ * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
+ * either horizontal or vertical direction to produce the
+ * filtered output block. Used to implement first-pass
+ * of 2-D separable filter.
+ *
+ * SPECIAL NOTES : Produces int32_t output to retain precision for next pass.
+ * Two filter taps should sum to VP9_FILTER_WEIGHT.
+ * pixel_step defines whether the filter is applied
+ * horizontally (pixel_step=1) or vertically (pixel_step=
+ * stride).
+ * It defines the offset required to move from one input
+ * to the next.
+ *
+ ****************************************************************************/
+static void var_filter_block2d_bil_first_pass(const uint8_t *src_ptr,
+ uint16_t *output_ptr,
+ unsigned int src_pixels_per_line,
+ int pixel_step,
+ unsigned int output_height,
+ unsigned int output_width,
+ const int16_t *vp9_filter) {
+ unsigned int i, j;
+
+ for (i = 0; i < output_height; i++) {
+ for (j = 0; j < output_width; j++) {
+ output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+ (int)src_ptr[pixel_step] * vp9_filter[1],
+ FILTER_BITS);
+
+ src_ptr++;
+ }
+
+ // Next row...
+ src_ptr += src_pixels_per_line - output_width;
+ output_ptr += output_width;
+ }
+}
+
+/****************************************************************************
+ *
+ * ROUTINE : filter_block2d_bil_second_pass
+ *
+ * INPUTS : int32_t *src_ptr : Pointer to source block.
+ * uint32_t src_pixels_per_line : Stride of input block.
+ * uint32_t pixel_step : Offset between filter input
+ * samples (see notes).
+ * uint32_t output_height : Input block height.
+ * uint32_t output_width : Input block width.
+ * int32_t *vp9_filter : Array of 2 bi-linear filter
+ * taps.
+ *
+ * OUTPUTS : uint16_t *output_ptr : Pointer to filtered block.
+ *
+ * RETURNS : void
+ *
+ * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
+ * either horizontal or vertical direction to produce the
+ * filtered output block. Used to implement second-pass
+ * of 2-D separable filter.
+ *
+ * SPECIAL NOTES : Requires 32-bit input as produced by
+ * filter_block2d_bil_first_pass.
+ * Two filter taps should sum to VP9_FILTER_WEIGHT.
+ * pixel_step defines whether the filter is applied
+ * horizontally (pixel_step=1) or vertically (pixel_step=
+ * stride).
+ * It defines the offset required to move from one input
+ * to the next.
+ *
+ ****************************************************************************/
+static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
+ uint8_t *output_ptr,
+ unsigned int src_pixels_per_line,
+ unsigned int pixel_step,
+ unsigned int output_height,
+ unsigned int output_width,
+ const int16_t *vp9_filter) {
+ unsigned int i, j;
+
+ for (i = 0; i < output_height; i++) {
+ for (j = 0; j < output_width; j++) {
+ output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+ (int)src_ptr[pixel_step] * vp9_filter[1],
+ FILTER_BITS);
+ src_ptr++;
+ }
+
+ src_ptr += src_pixels_per_line - output_width;
+ output_ptr += output_width;
+ }
+}
unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
unsigned int i, sum = 0;
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "vpx_config.h"
+#include "./vpx_config.h"
#include "vp9/encoder/vp9_variance.h"
#include "vp9/common/vp9_pragmas.h"
#include "vpx_ports/mem.h"
-extern unsigned int vp9_get_mb_ss_mmx(const short *src_ptr);
+extern unsigned int vp9_get_mb_ss_mmx(const int16_t *src_ptr);
extern unsigned int vp9_get8x8var_mmx
(
const unsigned char *src_ptr,
vp9_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 4));
-
}
unsigned int vp9_variance8x8_mmx(
*sse = var;
return (var - (((unsigned int)avg * avg) >> 6));
-
}
unsigned int vp9_mse16x16_mmx(
int sum0, sum1, sum2, sum3;
- vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
- vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
- vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
+ vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
+ &sum0);
+ vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride,
+ &sse1, &sum1);
+ vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride,
+ ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
+ vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride,
+ ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
var = sse0 + sse1 + sse2 + sse3;
*sse = var;
unsigned int sse0, sse1, sse2, sse3, var;
int sum0, sum1, sum2, sum3, avg;
-
- vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
- vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
- vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
+ vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
+ &sum0);
+ vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride,
+ &sse1, &sum1);
+ vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride,
+ ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
+ vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride,
+ ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
var = sse0 + sse1 + sse2 + sse3;
avg = sum0 + sum1 + sum2 + sum3;
unsigned int sse0, sse1, var;
int sum0, sum1, avg;
- vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
+ vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
+ &sum0);
+ vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride,
+ &sse1, &sum1);
var = sse0 + sse1;
avg = sum0 + sum1;
*sse = var;
return (var - (((unsigned int)avg * avg) >> 7));
-
}
unsigned int sse0, sse1, var;
int sum0, sum1, avg;
- vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1);
+ vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
+ &sum0);
+ vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride,
+ ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1);
var = sse0 + sse1;
avg = sum0 + sum1;
*sse = var;
return (var - (((unsigned int)avg * avg) >> 7));
-
}
VP9_COMMON_SRCS-yes += common/vp9_rtcd.c
VP9_COMMON_SRCS-yes += common/vp9_rtcd_defs.sh
VP9_COMMON_SRCS-yes += common/vp9_sadmxn.h
-VP9_COMMON_SRCS-yes += common/vp9_subpelvar.h
VP9_COMMON_SRCS-yes += common/vp9_scale.h
VP9_COMMON_SRCS-yes += common/vp9_scale.c
VP9_COMMON_SRCS-yes += common/vp9_seg_common.h
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
endif
+# common (c)
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_common_dspr2.h
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_avg_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_vert_dspr2.c
+
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_convolve_neon.c
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_mb_lpf_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_copy_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_avg_neon$(ASM)
+VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_save_reg_neon$(ASM)
$(eval $(call rtcd_h_template,vp9_rtcd,vp9/common/vp9_rtcd_defs.sh))
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <stdlib.h>
+#include <string.h>
#include "vpx/vpx_codec.h"
#include "vpx/internal/vpx_codec_internal.h"
-#include "vpx_version.h"
+#include "./vpx_version.h"
#include "vp9/encoder/vp9_onyx_int.h"
#include "vpx/vp8cx.h"
#include "vp9/encoder/vp9_firstpass.h"
#include "vp9/common/vp9_onyx.h"
#include "vp9/vp9_iface_common.h"
-#include <stdlib.h>
-#include <string.h>
struct vp9_extracfg {
struct vpx_codec_pkt_list *pkt_list;
- int cpu_used; /** available cpu percentage in 1/16*/
- unsigned int enable_auto_alt_ref; /** if encoder decides to uses alternate reference frame */
+ int cpu_used; /* available cpu percentage in 1/16 */
+ unsigned int enable_auto_alt_ref;
unsigned int noise_sensitivity;
unsigned int Sharpness;
unsigned int static_thresh;
unsigned int tile_columns;
unsigned int tile_rows;
- unsigned int arnr_max_frames; /* alt_ref Noise Reduction Max Frame Count */
- unsigned int arnr_strength; /* alt_ref Noise Reduction Strength */
- unsigned int arnr_type; /* alt_ref filter type */
+ unsigned int arnr_max_frames;
+ unsigned int arnr_strength;
+ unsigned int arnr_type;
unsigned int experimental;
vp8e_tuning tuning;
unsigned int cq_level; /* constrained quality level */
static const struct extraconfig_map extracfg_map[] = {
{
0,
- {
+ { // NOLINT
NULL,
0, /* cpu_used */
1, /* enable_auto_alt_ref */
uint32_t pending_frame_magnitude;
vpx_image_t preview_img;
vp8_postproc_cfg_t preview_ppcfg;
- vpx_codec_pkt_list_decl(64) pkt_list; // changed to accomendate the maximum number of lagged frames allowed
+ vpx_codec_pkt_list_decl(64) pkt_list;
unsigned int fixed_kf_cntr;
};
-static const VP9_REFFRAME ref_frame_to_vp9_reframe(vpx_ref_frame_type_t frame) {
+static VP9_REFFRAME ref_frame_to_vp9_reframe(vpx_ref_frame_type_t frame) {
switch (frame) {
case VP8_LAST_FRAME:
return VP9_LAST_FLAG;
#define ERROR(str) do {\
ctx->base.err_detail = str;\
return VPX_CODEC_INVALID_PARAM;\
- } while(0)
+ } while (0)
-#define RANGE_CHECK(p,memb,lo,hi) do {\
- if(!(((p)->memb == lo || (p)->memb > (lo)) && (p)->memb <= hi)) \
+#define RANGE_CHECK(p, memb, lo, hi) do {\
+ if (!(((p)->memb == lo || (p)->memb > (lo)) && (p)->memb <= hi)) \
ERROR(#memb " out of range ["#lo".."#hi"]");\
- } while(0)
+ } while (0)
-#define RANGE_CHECK_HI(p,memb,hi) do {\
- if(!((p)->memb <= (hi))) \
+#define RANGE_CHECK_HI(p, memb, hi) do {\
+ if (!((p)->memb <= (hi))) \
ERROR(#memb " out of range [.."#hi"]");\
- } while(0)
+ } while (0)
-#define RANGE_CHECK_LO(p,memb,lo) do {\
- if(!((p)->memb >= (lo))) \
+#define RANGE_CHECK_LO(p, memb, lo) do {\
+ if (!((p)->memb >= (lo))) \
ERROR(#memb " out of range ["#lo"..]");\
- } while(0)
+ } while (0)
-#define RANGE_CHECK_BOOL(p,memb) do {\
- if(!!((p)->memb) != (p)->memb) ERROR(#memb " expected boolean");\
- } while(0)
+#define RANGE_CHECK_BOOL(p, memb) do {\
+ if (!!((p)->memb) != (p)->memb) ERROR(#memb " expected boolean");\
+ } while (0)
static vpx_codec_err_t validate_config(vpx_codec_alg_priv_t *ctx,
const vpx_codec_enc_cfg_t *cfg,
oxcf->width = cfg.g_w;
oxcf->height = cfg.g_h;
/* guess a frame rate if out of whack, use 30 */
- oxcf->framerate = (double)(cfg.g_timebase.den) / (double)(cfg.g_timebase.num);
+ oxcf->framerate = (double)(cfg.g_timebase.den)
+ / (double)(cfg.g_timebase.num);
if (oxcf->framerate > 180) {
oxcf->framerate = 30;
*/
for (i = 0;
extracfg_map[i].usage && extracfg_map[i].usage != cfg->g_usage;
- i++);
+ i++) {}
priv->vp8_cfg = extracfg_map[i].cfg;
priv->vp8_cfg.pkt_list = &priv->pkt_list.head;
static vpx_codec_err_t vp9e_destroy(vpx_codec_alg_priv_t *ctx) {
-
free(ctx->cx_data);
vp9_remove_compressor(&ctx->cpi);
free(ctx);
lib_flags = (flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
/* vp8 use 10,000,000 ticks/second as time stamp */
- dst_time_stamp = pts * 10000000 * ctx->cfg.g_timebase.num / ctx->cfg.g_timebase.den;
- dst_end_time_stamp = (pts + duration) * 10000000 * ctx->cfg.g_timebase.num / ctx->cfg.g_timebase.den;
+ dst_time_stamp = pts * 10000000 * ctx->cfg.g_timebase.num
+ / ctx->cfg.g_timebase.den;
+ dst_end_time_stamp = (pts + duration) * 10000000 * ctx->cfg.g_timebase.num /
+ ctx->cfg.g_timebase.den;
if (img != NULL) {
res = image2yuvconfig(img, &sd);
cx_data += size;
cx_data_sz -= size;
}
-
- // printf("timestamp: %lld, duration: %d\n", pkt->data.frame.pts, pkt->data.frame.duration);
}
}
}
vp9_set_reference_enc(ctx->cpi, ref_frame_to_vp9_reframe(frame->frame_type),
&sd);
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
-
+ }
}
static vpx_codec_err_t vp9e_copy_reference(vpx_codec_alg_priv_t *ctx,
int ctr_id,
va_list args) {
-
vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
if (data) {
vp9_copy_reference_enc(ctx->cpi,
ref_frame_to_vp9_reframe(frame->frame_type), &sd);
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
}
static vpx_codec_err_t get_reference(vpx_codec_alg_priv_t *ctx,
if (data) {
ctx->preview_ppcfg = *((vp8_postproc_cfg_t *)data);
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
#else
(void)ctx;
(void)ctr_id;
static vpx_image_t *vp9e_get_preview(vpx_codec_alg_priv_t *ctx) {
-
YV12_BUFFER_CONFIG sd;
vp9_ppflags_t flags = {0};
if (0 == vp9_get_preview_raw_frame(ctx->cpi, &sd, &flags)) {
yuvconfig2image(&ctx->preview_img, &sd, NULL);
return &ctx->preview_img;
- } else
+ } else {
return NULL;
+ }
}
static vpx_codec_err_t vp9e_update_entropy(vpx_codec_alg_priv_t *ctx,
int update = va_arg(args, int);
vp9_update_entropy(ctx->cpi, update);
return VPX_CODEC_OK;
-
}
static vpx_codec_err_t vp9e_update_reference(vpx_codec_alg_priv_t *ctx,
return VPX_CODEC_OK;
else
return VPX_CODEC_INVALID_PARAM;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
}
vpx_active_map_t *data = va_arg(args, vpx_active_map_t *);
if (data) {
-
vpx_active_map_t *map = (vpx_active_map_t *)data;
if (!vp9_set_active_map(ctx->cpi, map->active_map, map->rows, map->cols))
return VPX_CODEC_OK;
else
return VPX_CODEC_INVALID_PARAM;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
}
static vpx_codec_err_t vp9e_set_scalemode(vpx_codec_alg_priv_t *ctx,
int ctr_id,
va_list args) {
-
vpx_scaling_mode_t *data = va_arg(args, vpx_scaling_mode_t *);
if (data) {
if (!res) {
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
- } else
+ }
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
}
static vpx_codec_err_t vp9e_set_width(vpx_codec_alg_priv_t *ctx, int ctr_id,
static vpx_codec_enc_cfg_map_t vp9e_usage_cfg_map[] = {
{
0,
- {
+ { // NOLINT
0, /* g_usage */
0, /* g_threads */
0, /* g_profile */
vp9e_ctf_maps, /* vpx_codec_ctrl_fn_map_t *ctrl_maps; */
NOT_IMPLEMENTED, /* vpx_codec_get_mmap_fn_t get_mmap; */
NOT_IMPLEMENTED, /* vpx_codec_set_mmap_fn_t set_mmap; */
- {
+ { // NOLINT
NOT_IMPLEMENTED, /* vpx_codec_peek_si_fn_t peek_si; */
NOT_IMPLEMENTED, /* vpx_codec_get_si_fn_t get_si; */
NOT_IMPLEMENTED, /* vpx_codec_decode_fn_t decode; */
NOT_IMPLEMENTED, /* vpx_codec_frame_get_fn_t frame_get; */
},
- {
+ { // NOLINT
vp9e_usage_cfg_map, /* vpx_codec_enc_cfg_map_t peek_si; */
vp9e_encode, /* vpx_codec_encode_fn_t encode; */
vp9e_get_cxdata, /* vpx_codec_get_cx_data_fn_t frame_get; */
vp9e_ctf_maps, /* vpx_codec_ctrl_fn_map_t *ctrl_maps; */
NOT_IMPLEMENTED, /* vpx_codec_get_mmap_fn_t get_mmap; */
NOT_IMPLEMENTED, /* vpx_codec_set_mmap_fn_t set_mmap; */
- {
+ { // NOLINT
NOT_IMPLEMENTED, /* vpx_codec_peek_si_fn_t peek_si; */
NOT_IMPLEMENTED, /* vpx_codec_get_si_fn_t get_si; */
NOT_IMPLEMENTED, /* vpx_codec_decode_fn_t decode; */
NOT_IMPLEMENTED, /* vpx_codec_frame_get_fn_t frame_get; */
},
- {
+ { // NOLINT
vp9e_usage_cfg_map, /* vpx_codec_enc_cfg_map_t peek_si; */
vp9e_encode, /* vpx_codec_encode_fn_t encode; */
vp9e_get_cxdata, /* vpx_codec_get_cx_data_fn_t frame_get; */
#include "vpx/vpx_decoder.h"
#include "vpx/vp8dx.h"
#include "vpx/internal/vpx_codec_internal.h"
-#include "vpx_version.h"
+#include "./vpx_version.h"
#include "vp9/decoder/vp9_onyxd.h"
#include "vp9/decoder/vp9_onyxd_int.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
static vpx_codec_err_t vp9_get_si(vpx_codec_alg_priv_t *ctx,
vpx_codec_stream_info_t *si) {
-
unsigned int sz;
if (si->sz >= sizeof(vp9_stream_info_t))
vp9_ppflags_t flags = {0};
if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC) {
- flags.post_proc_flag = ctx->postproc_cfg.post_proc_flag
+ flags.post_proc_flag =
#if CONFIG_POSTPROC_VISUALIZER
-
- | ((ctx->dbg_color_ref_frame_flag != 0) ? VP9D_DEBUG_CLR_FRM_REF_BLKS : 0)
- | ((ctx->dbg_color_mb_modes_flag != 0) ? VP9D_DEBUG_CLR_BLK_MODES : 0)
- | ((ctx->dbg_color_b_modes_flag != 0) ? VP9D_DEBUG_CLR_BLK_MODES : 0)
- | ((ctx->dbg_display_mv_flag != 0) ? VP9D_DEBUG_DRAW_MV : 0)
+ ((ctx->dbg_color_ref_frame_flag != 0) ?
+ VP9D_DEBUG_CLR_FRM_REF_BLKS : 0)
+ | ((ctx->dbg_color_mb_modes_flag != 0) ?
+ VP9D_DEBUG_CLR_BLK_MODES : 0)
+ | ((ctx->dbg_color_b_modes_flag != 0) ?
+ VP9D_DEBUG_CLR_BLK_MODES : 0)
+ | ((ctx->dbg_display_mv_flag != 0) ?
+ VP9D_DEBUG_DRAW_MV : 0)
+ |
#endif
-;
+ ctx->postproc_cfg.post_proc_flag;
+
flags.deblocking_level = ctx->postproc_cfg.deblocking_level;
flags.noise_level = ctx->postproc_cfg.noise_level;
#if CONFIG_POSTPROC_VISUALIZER
mmap->sz = seg_iter->calc_sz(ctx->config.dec, ctx->init_flags);
res = VPX_CODEC_OK;
- } else
+ } else {
res = VPX_CODEC_LIST_END;
+ }
} while (!mmap->sz && res != VPX_CODEC_LIST_END);
return res;
static vpx_codec_err_t set_reference(vpx_codec_alg_priv_t *ctx,
int ctr_id,
va_list args) {
-
vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
if (data) {
return vp9_set_reference_dec(ctx->pbi,
(VP9_REFFRAME)frame->frame_type, &sd);
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
-
+ }
}
static vpx_codec_err_t copy_reference(vpx_codec_alg_priv_t *ctx,
int ctr_id,
va_list args) {
-
vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
if (data) {
return vp9_copy_reference_dec(ctx->pbi,
(VP9_REFFRAME)frame->frame_type, &sd);
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
-
+ }
}
static vpx_codec_err_t get_reference(vpx_codec_alg_priv_t *ctx,
ctx->postproc_cfg_set = 1;
ctx->postproc_cfg = *((vp8_postproc_cfg_t *)data);
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
-
+ }
#else
return VPX_CODEC_INCAPABLE;
#endif
*update_info = pbi->refresh_frame_flags;
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
+ }
}
static vpx_codec_err_t get_frame_corrupted(vpx_codec_alg_priv_t *ctx,
int ctrl_id,
va_list args) {
-
int *corrupted = va_arg(args, int *);
if (corrupted) {
*corrupted = pbi->common.frame_to_show->corrupted;
return VPX_CODEC_OK;
- } else
+ } else {
return VPX_CODEC_INVALID_PARAM;
-
+ }
}
static vpx_codec_err_t set_invert_tile_order(vpx_codec_alg_priv_t *ctx,
ctf_maps, /* vpx_codec_ctrl_fn_map_t *ctrl_maps; */
vp9_xma_get_mmap, /* vpx_codec_get_mmap_fn_t get_mmap; */
vp9_xma_set_mmap, /* vpx_codec_set_mmap_fn_t set_mmap; */
- {
+ { // NOLINT
vp9_peek_si, /* vpx_codec_peek_si_fn_t peek_si; */
vp9_get_si, /* vpx_codec_get_si_fn_t get_si; */
vp9_decode, /* vpx_codec_decode_fn_t decode; */
vp9_get_frame, /* vpx_codec_frame_get_fn_t frame_get; */
},
- {
+ { // NOLINT
/* encoder functions */
NOT_IMPLEMENTED,
NOT_IMPLEMENTED,
VP9_DX_SRCS-yes += decoder/vp9_thread.h
VP9_DX_SRCS-yes += decoder/vp9_treereader.h
VP9_DX_SRCS-yes += decoder/vp9_onyxd_if.c
-VP9_DX_SRCS-yes += decoder/vp9_idct_blk.c
-VP9_DX_SRCS-yes += decoder/vp9_idct_blk.h
VP9_DX_SRCS-yes += decoder/vp9_dsubexp.c
VP9_DX_SRCS-yes += decoder/vp9_dsubexp.h
VP9_DX_SRCS-yes := $(filter-out $(VP9_DX_SRCS_REMOVE-yes),$(VP9_DX_SRCS-yes))
-
-VP9_DX_SRCS-$(HAVE_SSE2) += decoder/x86/vp9_dequantize_sse2.c
-VP9_DX_SRCS-$(HAVE_NEON) += decoder/arm/neon/vp9_add_constant_residual_neon$(ASM)
*/
#ifndef VP8_H
#define VP8_H
-#include "vpx_codec_impl_top.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
/*!\brief Control functions
*
/*! @} - end defgroup vp8 */
-#include "vpx_codec_impl_bottom.h"
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
#endif
*/
#ifndef VP8CX_H
#define VP8CX_H
-#include "vpx_codec_impl_top.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
/*!\name Algorithm interface for VP8
*
VPX_CTRL_USE_TYPE(VP9E_SET_MAX_Q, unsigned int)
VPX_CTRL_USE_TYPE(VP9E_SET_MIN_Q, unsigned int)
/*! @} - end defgroup vp8_encoder */
-#include "vpx_codec_impl_bottom.h"
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
#endif
*/
#ifndef VP8DX_H
#define VP8DX_H
-#include "vpx_codec_impl_top.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
/*!\name Algorithm interface for VP8
*
/*! @} - end defgroup vp8_decoder */
+#ifdef __cplusplus
+} // extern "C"
+#endif
-#include "vpx_codec_impl_bottom.h"
#endif
API_SRCS-yes += src/vpx_image.c
API_SRCS-yes += vpx_codec.h
API_SRCS-yes += vpx_codec.mk
-API_SRCS-yes += vpx_codec_impl_bottom.h
-API_SRCS-yes += vpx_codec_impl_top.h
API_SRCS-yes += vpx_image.h
API_SRCS-$(BUILD_LIBVPX) += vpx_integer.h
+++ /dev/null
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/*
- * This file is to be included at the bottom of the header files defining the
- * interface to individual codecs and contains matching blocks to those defined
- * in vpx_codec_impl_top.h
- */
-#ifdef __cplusplus
-}
-#endif
+++ /dev/null
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/*
- * This file is to be included at the top of the header files defining the
- * interface to individual codecs and contains various workarounds common
- * to all codec implementations.
- */
-#ifdef __cplusplus
-extern "C" {
-#endif
--- /dev/null
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx_scale/yv12config.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/vpx_scale.h"
+
+#if HAVE_DSPR2
+static void extend_plane(uint8_t *const src, int src_stride,
+ int width, int height,
+ int extend_top, int extend_left,
+ int extend_bottom, int extend_right) {
+ int i, j;
+ uint8_t *left_src, *right_src;
+ uint8_t *left_dst_start, *right_dst_start;
+ uint8_t *left_dst, *right_dst;
+ uint8_t *top_src, *bot_src;
+ uint8_t *top_dst, *bot_dst;
+ uint32_t left_pix;
+ uint32_t right_pix;
+ uint32_t linesize;
+
+ /* copy the left and right most columns out */
+ left_src = src;
+ right_src = src + width - 1;
+ left_dst_start = src - extend_left;
+ right_dst_start = src + width;
+
+ for (i = height; i--; ) {
+ left_dst = left_dst_start;
+ right_dst = right_dst_start;
+
+ __asm__ __volatile__ (
+ "lb %[left_pix], 0(%[left_src]) \n\t"
+ "lb %[right_pix], 0(%[right_src]) \n\t"
+ "replv.qb %[left_pix], %[left_pix] \n\t"
+ "replv.qb %[right_pix], %[right_pix] \n\t"
+
+ : [left_pix] "=&r" (left_pix), [right_pix] "=&r" (right_pix)
+ : [left_src] "r" (left_src), [right_src] "r" (right_src)
+ );
+
+ for (j = extend_left/4; j--; ) {
+ __asm__ __volatile__ (
+ "sw %[left_pix], 0(%[left_dst]) \n\t"
+ "sw %[right_pix], 0(%[right_dst]) \n\t"
+
+ :
+ : [left_dst] "r" (left_dst), [left_pix] "r" (left_pix),
+ [right_dst] "r" (right_dst), [right_pix] "r" (right_pix)
+ );
+
+ left_dst += 4;
+ right_dst += 4;
+ }
+
+ for (j = extend_left%4; j--; ) {
+ __asm__ __volatile__ (
+ "sb %[left_pix], 0(%[left_dst]) \n\t"
+ "sb %[right_pix], 0(%[right_dst]) \n\t"
+
+ :
+ : [left_dst] "r" (left_dst), [left_pix] "r" (left_pix),
+ [right_dst] "r" (right_dst), [right_pix] "r" (right_pix)
+ );
+
+ left_dst += 1;
+ right_dst += 1;
+ }
+
+ left_src += src_stride;
+ right_src += src_stride;
+ left_dst_start += src_stride;
+ right_dst_start += src_stride;
+ }
+
+ /* Now copy the top and bottom lines into each line of the respective
+ * borders
+ */
+ top_src = src - extend_left;
+ bot_src = src + src_stride * (height - 1) - extend_left;
+ top_dst = src + src_stride * (-extend_top) - extend_left;
+ bot_dst = src + src_stride * (height) - extend_left;
+ linesize = extend_left + extend_right + width;
+
+ for (i = 0; i < extend_top; i++) {
+ vpx_memcpy(top_dst, top_src, linesize);
+ top_dst += src_stride;
+ }
+
+ for (i = 0; i < extend_bottom; i++) {
+ vpx_memcpy(bot_dst, bot_src, linesize);
+ bot_dst += src_stride;
+ }
+}
+
+static void extend_frame(YV12_BUFFER_CONFIG *const ybf,
+ int subsampling_x, int subsampling_y,
+ int ext_size) {
+ const int c_w = (ybf->y_crop_width + subsampling_x) >> subsampling_x;
+ const int c_h = (ybf->y_crop_height + subsampling_y) >> subsampling_y;
+ const int c_et = ext_size >> subsampling_y;
+ const int c_el = ext_size >> subsampling_x;
+ const int c_eb = (ext_size + ybf->y_height - ybf->y_crop_height +
+ subsampling_y) >> subsampling_y;
+ const int c_er = (ext_size + ybf->y_width - ybf->y_crop_width +
+ subsampling_x) >> subsampling_x;
+
+ assert(ybf->y_height - ybf->y_crop_height < 16);
+ assert(ybf->y_width - ybf->y_crop_width < 16);
+ assert(ybf->y_height - ybf->y_crop_height >= 0);
+ assert(ybf->y_width - ybf->y_crop_width >= 0);
+
+ extend_plane(ybf->y_buffer, ybf->y_stride,
+ ybf->y_crop_width, ybf->y_crop_height,
+ ext_size, ext_size,
+ ext_size + ybf->y_height - ybf->y_crop_height,
+ ext_size + ybf->y_width - ybf->y_crop_width);
+
+ extend_plane(ybf->u_buffer, ybf->uv_stride,
+ c_w, c_h, c_et, c_el, c_eb, c_er);
+
+ extend_plane(ybf->v_buffer, ybf->uv_stride,
+ c_w, c_h, c_et, c_el, c_eb, c_er);
+}
+
+void vp9_extend_frame_borders_dspr2(YV12_BUFFER_CONFIG *ybf,
+ int subsampling_x, int subsampling_y) {
+ extend_frame(ybf, subsampling_x, subsampling_y, ybf->border);
+}
+
+void vp9_extend_frame_inner_borders_dspr2(YV12_BUFFER_CONFIG *ybf,
+ int subsampling_x,
+ int subsampling_y) {
+ const int inner_bw = (ybf->border > VP9INNERBORDERINPIXELS) ?
+ VP9INNERBORDERINPIXELS : ybf->border;
+ extend_frame(ybf, subsampling_x, subsampling_y, inner_bw);
+}
+#endif
SCALE_SRCS-$(HAVE_NEON) += arm/neon/vp8_vpxyv12_extendframeborders_neon$(ASM)
SCALE_SRCS-$(HAVE_NEON) += arm/neon/yv12extend_arm.c
+#mips(dspr2)
+SCALE_SRCS-$(HAVE_DSPR2) += mips/dspr2/yv12extend_dspr2.c
+
SCALE_SRCS-no += $(SCALE_SRCS_REMOVE-yes)
$(eval $(call asm_offsets_template,\
if [ "$CONFIG_VP9" = "yes" ]; then
prototype void vp9_extend_frame_borders "struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y"
- specialize vp9_extend_frame_borders
+ specialize vp9_extend_frame_borders dspr2
prototype void vp9_extend_frame_inner_borders "struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y"
- specialize vp9_extend_frame_inner_borders_c
+ specialize vp9_extend_frame_inner_borders dspr2
fi
projects / libvpx / commitdiff