VP9 common for ARMv8 by using NEON intrinsics 03

Add vp9_copy_neon.c
- vp9_convolve_copy_neon

Change-Id: I291fc5423d06240876411bbceab03eae5ef585be
Signed-off-by: James Yu <james.yu@linaro.org>

test/convolve_test.cc

@@ -1767,7 +1767,7 @@ const ConvolveFunctions convolve8_neon(
     vp9_convolve8_neon, vp9_convolve8_avg_neon, 0);
 #else  // HAVE_NEON
 const ConvolveFunctions convolve8_neon(
-    vp9_convolve_copy_c, vp9_convolve_avg_neon,
+    vp9_convolve_copy_neon, vp9_convolve_avg_neon,
     vp9_convolve8_horiz_c, vp9_convolve8_avg_horiz_c,
     vp9_convolve8_vert_c, vp9_convolve8_avg_vert_c,
     vp9_convolve8_c, vp9_convolve8_avg_c, 0);

vp9/common/arm/neon/vp9_copy_neon.c

@@ -0,0 +1,92 @@
+/*
+ *  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 <stddef.h>
+#include <arm_neon.h>
+
+void vp9_convolve_copy_neon(
+        const uint8_t *src,    // r0
+        ptrdiff_t src_stride,  // r1
+        uint8_t *dst,          // r2
+        ptrdiff_t dst_stride,  // r3
+        const int16_t *filter_x,
+        int filter_x_stride,
+        const int16_t *filter_y,
+        int filter_y_stride,
+        int w,
+        int h) {
+    uint8x8_t d0u8, d2u8;
+    uint8x16_t q0u8, q1u8, q2u8, q3u8;
+    (void)filter_x;  (void)filter_x_stride;
+    (void)filter_y;  (void)filter_y_stride;
+
+    if (w > 32) {  // copy64
+        for (; h > 0; h--) {
+            q0u8 = vld1q_u8(src);
+            q1u8 = vld1q_u8(src + 16);
+            q2u8 = vld1q_u8(src + 32);
+            q3u8 = vld1q_u8(src + 48);
+            src += src_stride;
+
+            vst1q_u8(dst, q0u8);
+            vst1q_u8(dst + 16, q1u8);
+            vst1q_u8(dst + 32, q2u8);
+            vst1q_u8(dst + 48, q3u8);
+            dst += dst_stride;
+        }
+    } else if (w == 32) {  // copy32
+        for (; h > 0; h -= 2) {
+            q0u8 = vld1q_u8(src);
+            q1u8 = vld1q_u8(src + 16);
+            src += src_stride;
+            q2u8 = vld1q_u8(src);
+            q3u8 = vld1q_u8(src + 16);
+            src += src_stride;
+
+            vst1q_u8(dst, q0u8);
+            vst1q_u8(dst + 16, q1u8);
+            dst += dst_stride;
+            vst1q_u8(dst, q2u8);
+            vst1q_u8(dst + 16, q3u8);
+            dst += dst_stride;
+        }
+    } else if (w > 8) {  // copy16
+        for (; h > 0; h -= 2) {
+            q0u8 = vld1q_u8(src);
+            src += src_stride;
+            q1u8 = vld1q_u8(src);
+            src += src_stride;
+
+            vst1q_u8(dst, q0u8);
+            dst += dst_stride;
+            vst1q_u8(dst, q1u8);
+            dst += dst_stride;
+        }
+    } else if (w == 8) {  // copy8
+        for (; h > 0; h -= 2) {
+            d0u8 = vld1_u8(src);
+            src += src_stride;
+            d2u8 = vld1_u8(src);
+            src += src_stride;
+
+            vst1_u8(dst, d0u8);
+            dst += dst_stride;
+            vst1_u8(dst, d2u8);
+            dst += dst_stride;
+        }
+    } else {  // copy4
+        for (; h > 0; h--) {
+            *(uint32_t *)dst = *(const uint32_t *)src;
+            src += src_stride;
+            dst += dst_stride;
+        }
+    }
+    return;
+}

vp9/common/vp9_rtcd_defs.pl

@@ -289,8 +289,7 @@ $vp9_plane_add_noise_sse2=vp9_plane_add_noise_wmt;
 # Sub Pixel Filters
 #
 add_proto qw/void vp9_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve_copy neon_asm dspr2/, "$sse2_x86inc";
+specialize qw/vp9_convolve_copy neon dspr2/, "$sse2_x86inc";
-$vp9_convolve_copy_neon_asm=vp9_convolve_copy_neon;
 
 add_proto qw/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 qw/vp9_convolve_avg neon dspr2/, "$sse2_x86inc";

vp9/vp9_common.mk

@@ -148,7 +148,6 @@ 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)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_mb_lpf_neon$(ASM)
-VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_copy_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_save_reg_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_reconintra_neon$(ASM)
 
@@ -156,11 +155,13 @@ VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_reconintra_neon$(ASM)
 # prefer assembly.
 ifeq ($(HAVE_NEON_ASM), yes)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_avg_neon_asm$(ASM)
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
 else
 ifeq ($(HAVE_NEON), yes)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_avg_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
 endif  # HAVE_NEON