From 40edab5e391efc11b5c07fddd920f6dc96c1c719 Mon Sep 17 00:00:00 2001 From: Parag Salasakar Date: Fri, 13 Sep 2013 15:18:32 +0530 Subject: [PATCH] mips dsp-ase r2 vp9 decoder convolve module optimizations Change-Id: I401536778e3c68ba2b3ae3955c689d005e1f1d59 --- build/make/rtcd.sh | 4 + test/convolve_test.cc | 22 + vp9/common/mips/dspr2/vp9_common_dspr2.h | 85 ++ .../mips/dspr2/vp9_convolve8_avg_dspr2.c | 689 +++++++++ .../dspr2/vp9_convolve8_avg_horiz_dspr2.c | 1032 +++++++++++++ vp9/common/mips/dspr2/vp9_convolve8_dspr2.c | 1281 +++++++++++++++++ .../mips/dspr2/vp9_convolve8_horiz_dspr2.c | 917 ++++++++++++ .../mips/dspr2/vp9_convolve8_vert_dspr2.c | 390 +++++ vp9/common/vp9_rtcd_defs.sh | 16 +- vp9/vp9_common.mk | 8 + 10 files changed, 4436 insertions(+), 8 deletions(-) create mode 100644 vp9/common/mips/dspr2/vp9_common_dspr2.h create mode 100644 vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c create mode 100644 vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c create mode 100644 vp9/common/mips/dspr2/vp9_convolve8_dspr2.c create mode 100644 vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c create mode 100644 vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c diff --git a/build/make/rtcd.sh b/build/make/rtcd.sh index 9a8d97e32..c531e95a1 100755 --- a/build/make/rtcd.sh +++ b/build/make/rtcd.sh @@ -294,6 +294,10 @@ $(set_function_pointers c $ALL_ARCHS) void dsputil_static_init(); dsputil_static_init(); #endif +#if CONFIG_VP9 +void vp9_dsputil_static_init(); +vp9_dsputil_static_init(); +#endif #endif } #endif diff --git a/test/convolve_test.cc b/test/convolve_test.cc index 3100571da..f0b412dea 100644 --- a/test/convolve_test.cc +++ b/test/convolve_test.cc @@ -642,4 +642,26 @@ INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values( 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 diff --git a/vp9/common/mips/dspr2/vp9_common_dspr2.h b/vp9/common/mips/dspr2/vp9_common_dspr2.h new file mode 100644 index 000000000..d2fa4c1dc --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_common_dspr2.h @@ -0,0 +1,85 @@ +/* + * 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 + +#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_ diff --git a/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c new file mode 100644 index 000000000..0930ad123 --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c @@ -0,0 +1,689 @@ +/* + * 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 +#include + +#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 diff --git a/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c new file mode 100644 index 000000000..37c665be9 --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c @@ -0,0 +1,1032 @@ +/* + * 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 +#include + +#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 diff --git a/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c new file mode 100644 index 000000000..2c48bd038 --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c @@ -0,0 +1,1281 @@ +/* + * 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 +#include + +#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 diff --git a/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c new file mode 100644 index 000000000..743d64116 --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c @@ -0,0 +1,917 @@ +/* + * 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 +#include + +#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 diff --git a/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c new file mode 100644 index 000000000..bdc7930b7 --- /dev/null +++ b/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c @@ -0,0 +1,390 @@ +/* + * 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 +#include + +#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 diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh index 225305b19..e5439ec8c 100644 --- a/vp9/common/vp9_rtcd_defs.sh +++ b/vp9/common/vp9_rtcd_defs.sh @@ -257,28 +257,28 @@ specialize vp9_blend_b # 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 $sse2_x86inc neon dspr2 prototype void vp9_convolve8 "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8 ssse3 neon +specialize vp9_convolve8 ssse3 neon dspr2 prototype void vp9_convolve8_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8_horiz ssse3 neon +specialize vp9_convolve8_horiz ssse3 neon dspr2 prototype void vp9_convolve8_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8_vert ssse3 neon +specialize vp9_convolve8_vert ssse3 neon dspr2 prototype void vp9_convolve8_avg "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8_avg ssse3 neon +specialize vp9_convolve8_avg ssse3 neon dspr2 prototype void vp9_convolve8_avg_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8_avg_horiz ssse3 neon +specialize vp9_convolve8_avg_horiz ssse3 neon dspr2 prototype void vp9_convolve8_avg_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h" -specialize vp9_convolve8_avg_vert ssse3 neon +specialize vp9_convolve8_avg_vert ssse3 neon dspr2 # # dct diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk index 7d040f7db..10fa461d8 100644 --- a/vp9/vp9_common.mk +++ b/vp9/vp9_common.mk @@ -87,6 +87,14 @@ VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_intrapred_sse2.asm 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 -- 2.40.0