VP9 common for ARMv8 by using NEON intrinsics 02

Add vp9_avg_neon.c
- vp9_convolve_avg_neon

Change-Id: Id2c9d5bcfa37cff1a16417aba1656ff07bdf10fd
Signed-off-by: James Yu <james.yu@linaro.org>

test/convolve_test.cc

@@ -1758,12 +1758,20 @@ INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest, ::testing::Values(
     make_tuple(64, 64, &convolve8_avx2)));
 #endif  // HAVE_AVX2 && HAVE_SSSE3
 
+#if HAVE_NEON
 #if HAVE_NEON_ASM
 const ConvolveFunctions convolve8_neon(
     vp9_convolve_copy_neon, vp9_convolve_avg_neon,
     vp9_convolve8_horiz_neon, vp9_convolve8_avg_horiz_neon,
     vp9_convolve8_vert_neon, vp9_convolve8_avg_vert_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_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);
+#endif  // HAVE_NEON_ASM
 
 INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_neon),
@@ -1779,7 +1787,7 @@ INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values(
     make_tuple(64, 32, &convolve8_neon),
     make_tuple(32, 64, &convolve8_neon),
     make_tuple(64, 64, &convolve8_neon)));
-#endif
+#endif  // HAVE_NEON
 
 #if HAVE_DSPR2
 const ConvolveFunctions convolve8_dspr2(

vp9/common/arm/neon/vp9_avg_neon.c

+/*
+ *  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_avg_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) {
+    uint8_t *d;
+    uint8x8_t d0u8, d1u8, d2u8, d3u8;
+    uint32x2_t d0u32, d2u32;
+    uint8x16_t q0u8, q1u8, q2u8, q3u8, q8u8, q9u8, q10u8, q11u8;
+    (void)filter_x;  (void)filter_x_stride;
+    (void)filter_y;  (void)filter_y_stride;
+
+    d = dst;
+    if (w > 32) {  // avg64
+        for (; h > 0; h -= 1) {
+            q0u8  = vld1q_u8(src);
+            q1u8  = vld1q_u8(src + 16);
+            q2u8  = vld1q_u8(src + 32);
+            q3u8  = vld1q_u8(src + 48);
+            src += src_stride;
+            q8u8  = vld1q_u8(d);
+            q9u8  = vld1q_u8(d + 16);
+            q10u8 = vld1q_u8(d + 32);
+            q11u8 = vld1q_u8(d + 48);
+            d += dst_stride;
+
+            q0u8 = vrhaddq_u8(q0u8, q8u8);
+            q1u8 = vrhaddq_u8(q1u8, q9u8);
+            q2u8 = vrhaddq_u8(q2u8, q10u8);
+            q3u8 = vrhaddq_u8(q3u8, q11u8);
+
+            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) {  // avg32
+        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;
+            q8u8 = vld1q_u8(d);
+            q9u8 = vld1q_u8(d + 16);
+            d += dst_stride;
+            q10u8 = vld1q_u8(d);
+            q11u8 = vld1q_u8(d + 16);
+            d += dst_stride;
+
+            q0u8 = vrhaddq_u8(q0u8, q8u8);
+            q1u8 = vrhaddq_u8(q1u8, q9u8);
+            q2u8 = vrhaddq_u8(q2u8, q10u8);
+            q3u8 = vrhaddq_u8(q3u8, q11u8);
+
+            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) {  // avg16
+        for (; h > 0; h -= 2) {
+            q0u8 = vld1q_u8(src);
+            src += src_stride;
+            q1u8 = vld1q_u8(src);
+            src += src_stride;
+            q2u8 = vld1q_u8(d);
+            d += dst_stride;
+            q3u8 = vld1q_u8(d);
+            d += dst_stride;
+
+            q0u8 = vrhaddq_u8(q0u8, q2u8);
+            q1u8 = vrhaddq_u8(q1u8, q3u8);
+
+            vst1q_u8(dst, q0u8);
+            dst += dst_stride;
+            vst1q_u8(dst, q1u8);
+            dst += dst_stride;
+        }
+    } else if (w == 8) {  // avg8
+        for (; h > 0; h -= 2) {
+            d0u8 = vld1_u8(src);
+            src += src_stride;
+            d1u8 = vld1_u8(src);
+            src += src_stride;
+            d2u8 = vld1_u8(d);
+            d += dst_stride;
+            d3u8 = vld1_u8(d);
+            d += dst_stride;
+
+            q0u8 = vcombine_u8(d0u8, d1u8);
+            q1u8 = vcombine_u8(d2u8, d3u8);
+            q0u8 = vrhaddq_u8(q0u8, q1u8);
+
+            vst1_u8(dst, vget_low_u8(q0u8));
+            dst += dst_stride;
+            vst1_u8(dst, vget_high_u8(q0u8));
+            dst += dst_stride;
+        }
+    } else {  // avg4
+        for (; h > 0; h -= 2) {
+            d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 0);
+            src += src_stride;
+            d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 1);
+            src += src_stride;
+            d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 0);
+            d += dst_stride;
+            d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 1);
+            d += dst_stride;
+
+            d0u8 = vrhadd_u8(vreinterpret_u8_u32(d0u32),
+                             vreinterpret_u8_u32(d2u32));
+
+            d0u32 = vreinterpret_u32_u8(d0u8);
+            vst1_lane_u32((uint32_t *)dst, d0u32, 0);
+            dst += dst_stride;
+            vst1_lane_u32((uint32_t *)dst, d0u32, 1);
+            dst += dst_stride;
+        }
+    }
+    return;
+}

vp9/common/vp9_rtcd_defs.pl

@@ -293,8 +293,7 @@ specialize qw/vp9_convolve_copy neon_asm 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_asm dspr2/, "$sse2_x86inc";
-$vp9_convolve_avg_neon_asm=vp9_convolve_avg_neon;
+specialize qw/vp9_convolve_avg neon dspr2/, "$sse2_x86inc";
 
 add_proto qw/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 qw/vp9_convolve8 sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";

vp9/vp9_common.mk

@@ -149,17 +149,18 @@ 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-$(HAVE_NEON_ASM) += common/arm/neon/vp9_avg_neon$(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)
 
 # neon with assembly and intrinsics implementations. If both are available
 # 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_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_loopfilter_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
 endif  # HAVE_NEON