VP9 common for ARMv8 by using NEON intrinsics 10

Add vp9_idct32x32_1_add_neon.c
- vp9_idct32x32_1_add_neon

Change-Id: If9ffe9a857228f5c67f61dc2b428b40965816eda
Signed-off-by: James Yu <james.yu@linaro.org>

diff --git a/vp9/common/arm/neon/vp9_idct32x32_1_add_neon.c b/vp9/common/arm/neon/vp9_idct32x32_1_add_neon.c
new file mode 100644 (file)
index 0000000..1bfee22
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_idct32x32_1_add_neon.c
@@ -0,0 +1,161 @@
+/*
+ *  Copyright (c) 2014 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 <arm_neon.h>
+#include "vp9/common/vp9_idct.h"
+
+static inline void LD_16x8(
+        uint8_t *d,
+        int d_stride,
+        uint8x16_t *q8u8,
+        uint8x16_t *q9u8,
+        uint8x16_t *q10u8,
+        uint8x16_t *q11u8,
+        uint8x16_t *q12u8,
+        uint8x16_t *q13u8,
+        uint8x16_t *q14u8,
+        uint8x16_t *q15u8) {
+    *q8u8 = vld1q_u8(d);
+    d += d_stride;
+    *q9u8 = vld1q_u8(d);
+    d += d_stride;
+    *q10u8 = vld1q_u8(d);
+    d += d_stride;
+    *q11u8 = vld1q_u8(d);
+    d += d_stride;
+    *q12u8 = vld1q_u8(d);
+    d += d_stride;
+    *q13u8 = vld1q_u8(d);
+    d += d_stride;
+    *q14u8 = vld1q_u8(d);
+    d += d_stride;
+    *q15u8 = vld1q_u8(d);
+    return;
+}
+
+static inline void ADD_DIFF_16x8(
+        uint8x16_t qdiffu8,
+        uint8x16_t *q8u8,
+        uint8x16_t *q9u8,
+        uint8x16_t *q10u8,
+        uint8x16_t *q11u8,
+        uint8x16_t *q12u8,
+        uint8x16_t *q13u8,
+        uint8x16_t *q14u8,
+        uint8x16_t *q15u8) {
+    *q8u8 = vqaddq_u8(*q8u8, qdiffu8);
+    *q9u8 = vqaddq_u8(*q9u8, qdiffu8);
+    *q10u8 = vqaddq_u8(*q10u8, qdiffu8);
+    *q11u8 = vqaddq_u8(*q11u8, qdiffu8);
+    *q12u8 = vqaddq_u8(*q12u8, qdiffu8);
+    *q13u8 = vqaddq_u8(*q13u8, qdiffu8);
+    *q14u8 = vqaddq_u8(*q14u8, qdiffu8);
+    *q15u8 = vqaddq_u8(*q15u8, qdiffu8);
+    return;
+}
+
+static inline void SUB_DIFF_16x8(
+        uint8x16_t qdiffu8,
+        uint8x16_t *q8u8,
+        uint8x16_t *q9u8,
+        uint8x16_t *q10u8,
+        uint8x16_t *q11u8,
+        uint8x16_t *q12u8,
+        uint8x16_t *q13u8,
+        uint8x16_t *q14u8,
+        uint8x16_t *q15u8) {
+    *q8u8 = vqsubq_u8(*q8u8, qdiffu8);
+    *q9u8 = vqsubq_u8(*q9u8, qdiffu8);
+    *q10u8 = vqsubq_u8(*q10u8, qdiffu8);
+    *q11u8 = vqsubq_u8(*q11u8, qdiffu8);
+    *q12u8 = vqsubq_u8(*q12u8, qdiffu8);
+    *q13u8 = vqsubq_u8(*q13u8, qdiffu8);
+    *q14u8 = vqsubq_u8(*q14u8, qdiffu8);
+    *q15u8 = vqsubq_u8(*q15u8, qdiffu8);
+    return;
+}
+
+static inline void ST_16x8(
+        uint8_t *d,
+        int d_stride,
+        uint8x16_t *q8u8,
+        uint8x16_t *q9u8,
+        uint8x16_t *q10u8,
+        uint8x16_t *q11u8,
+        uint8x16_t *q12u8,
+        uint8x16_t *q13u8,
+        uint8x16_t *q14u8,
+        uint8x16_t *q15u8) {
+    vst1q_u8(d, *q8u8);
+    d += d_stride;
+    vst1q_u8(d, *q9u8);
+    d += d_stride;
+    vst1q_u8(d, *q10u8);
+    d += d_stride;
+    vst1q_u8(d, *q11u8);
+    d += d_stride;
+    vst1q_u8(d, *q12u8);
+    d += d_stride;
+    vst1q_u8(d, *q13u8);
+    d += d_stride;
+    vst1q_u8(d, *q14u8);
+    d += d_stride;
+    vst1q_u8(d, *q15u8);
+    return;
+}
+
+void vp9_idct32x32_1_add_neon(
+        int16_t *input,
+        uint8_t *dest,
+        int dest_stride) {
+    uint8x16_t q0u8, q8u8, q9u8, q10u8, q11u8, q12u8, q13u8, q14u8, q15u8;
+    int i, j, dest_stride8;
+    uint8_t *d;
+    int16_t a1, cospi_16_64 = 11585;
+    int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+
+    out = dct_const_round_shift(out * cospi_16_64);
+    a1 = ROUND_POWER_OF_TWO(out, 6);
+
+    dest_stride8 = dest_stride * 8;
+    if (a1 >= 0) {  // diff_positive_32_32
+        a1 = a1 < 0 ? 0 : a1 > 255 ? 255 : a1;
+        q0u8 = vdupq_n_u8(a1);
+        for (i = 0; i < 2; i++, dest += 16) {  // diff_positive_32_32_loop
+            d = dest;
+            for (j = 0; j < 4; j++) {
+                LD_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8,
+                                        &q12u8, &q13u8, &q14u8, &q15u8);
+                ADD_DIFF_16x8(q0u8, &q8u8, &q9u8, &q10u8, &q11u8,
+                                    &q12u8, &q13u8, &q14u8, &q15u8);
+                ST_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8,
+                                        &q12u8, &q13u8, &q14u8, &q15u8);
+                d += dest_stride8;
+            }
+        }
+    } else {  // diff_negative_32_32
+        a1 = -a1;
+        a1 = a1 < 0 ? 0 : a1 > 255 ? 255 : a1;
+        q0u8 = vdupq_n_u8(a1);
+        for (i = 0; i < 2; i++, dest += 16) {  // diff_negative_32_32_loop
+            d = dest;
+            for (j = 0; j < 4; j++) {
+                LD_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8,
+                                        &q12u8, &q13u8, &q14u8, &q15u8);
+                SUB_DIFF_16x8(q0u8, &q8u8, &q9u8, &q10u8, &q11u8,
+                                    &q12u8, &q13u8, &q14u8, &q15u8);
+                ST_16x8(d, dest_stride, &q8u8, &q9u8, &q10u8, &q11u8,
+                                        &q12u8, &q13u8, &q14u8, &q15u8);
+                d += dest_stride8;
+            }
+        }
+    }
+    return;
+}

diff --git a/vp9/common/arm/neon/vp9_idct32x32_1_add_neon.asm b/vp9/common/arm/neon/vp9_idct32x32_1_add_neon_asm.asm
similarity index 100%
rename from vp9/common/arm/neon/vp9_idct32x32_1_add_neon.asm
rename to vp9/common/arm/neon/vp9_idct32x32_1_add_neon_asm.asm

diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 7f74af293bea7d12a27c22219984202076374565..05f7ddf99818fdd4a5fa599b63a48e9d39934d9e 100644 (file)
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -460,8 +460,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
     $vp9_idct32x32_34_add_neon_asm=vp9_idct32x32_1024_add_neon;
 
     add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vp9_idct32x32_1_add sse2 neon_asm dspr2/;
-    $vp9_idct32x32_1_add_neon_asm=vp9_idct32x32_1_add_neon;
+    specialize qw/vp9_idct32x32_1_add sse2 neon dspr2/;
 
     add_proto qw/void vp9_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
     specialize qw/vp9_iht4x4_16_add sse2 neon_asm dspr2/;

diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 2f7a0e7d5106fc074cb447bdb88bee71a4765de2..370be3351d96a53bcb46ad98b1e13047be372e44 100644 (file)
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -138,7 +138,6 @@ VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct4x4_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct8x8_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct16x16_1_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct16x16_add_neon$(ASM)
-VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct32x32_1_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct32x32_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_iht4x4_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_iht8x8_add_neon$(ASM)
@@ -154,6 +153,7 @@ VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon_asm$(ASM)
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon_asm$(ASM)
@@ -165,6 +165,7 @@ VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon.c