ref: 95cb130f327e233443a750816bc2af72fa972baf
parent: d45870be8d39fe8545008af4e432ba9a79f0b0e8
parent: 2301d10f739a13f9f04eb016036e044484cf9d5d
author: Parag Salasakar <img.mips1@gmail.com>
date: Thu Apr 30 00:39:11 EDT 2015
Merge "mips msa vp9 copy and avg convolve optimization"
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -1817,7 +1817,7 @@
#if HAVE_MSA
const ConvolveFunctions convolve8_msa(
- vp9_convolve_copy_c, vp9_convolve_avg_c,
+ vp9_convolve_copy_msa, vp9_convolve_avg_msa,
vp9_convolve8_horiz_msa, vp9_convolve8_avg_horiz_c,
vp9_convolve8_vert_msa, vp9_convolve8_avg_vert_c,
vp9_convolve8_msa, vp9_convolve8_avg_c, 0);
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_convolve_avg_msa.c
@@ -1,0 +1,335 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/mips/msa/vp9_macros_msa.h"
+
+static void avg_width4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint32_t out0, out1, out2, out3;
+ v16u8 src0, src1, src2, src3;
+ v16u8 dst0, dst1, dst2, dst3;
+
+ if (0 == (height % 4)) {+ for (cnt = (height / 4); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ LOAD_4VECS_UB(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+ dst2 = __msa_aver_u_b(src2, dst2);
+ dst3 = __msa_aver_u_b(src3, dst3);
+
+ out0 = __msa_copy_u_w((v4i32)dst0, 0);
+ out1 = __msa_copy_u_w((v4i32)dst1, 0);
+ out2 = __msa_copy_u_w((v4i32)dst2, 0);
+ out3 = __msa_copy_u_w((v4i32)dst3, 0);
+
+ STORE_WORD(dst, out0);
+ dst += dst_stride;
+ STORE_WORD(dst, out1);
+ dst += dst_stride;
+ STORE_WORD(dst, out2);
+ dst += dst_stride;
+ STORE_WORD(dst, out3);
+ dst += dst_stride;
+ }
+ } else if (0 == (height % 2)) {+ for (cnt = (height / 2); cnt--;) {+ LOAD_2VECS_UB(src, src_stride, src0, src1);
+ src += (2 * src_stride);
+
+ LOAD_2VECS_UB(dst, dst_stride, dst0, dst1);
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+
+ out0 = __msa_copy_u_w((v4i32)dst0, 0);
+ out1 = __msa_copy_u_w((v4i32)dst1, 0);
+
+ STORE_WORD(dst, out0);
+ dst += dst_stride;
+ STORE_WORD(dst, out1);
+ dst += dst_stride;
+ }
+ }
+}
+
+static void avg_width8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint64_t out0, out1, out2, out3;
+ v16u8 src0, src1, src2, src3;
+ v16u8 dst0, dst1, dst2, dst3;
+
+ for (cnt = (height / 4); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ LOAD_4VECS_UB(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+ dst2 = __msa_aver_u_b(src2, dst2);
+ dst3 = __msa_aver_u_b(src3, dst3);
+
+ out0 = __msa_copy_u_d((v2i64)dst0, 0);
+ out1 = __msa_copy_u_d((v2i64)dst1, 0);
+ out2 = __msa_copy_u_d((v2i64)dst2, 0);
+ out3 = __msa_copy_u_d((v2i64)dst3, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ STORE_DWORD(dst, out2);
+ dst += dst_stride;
+ STORE_DWORD(dst, out3);
+ dst += dst_stride;
+ }
+}
+
+static void avg_width16_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+
+ for (cnt = (height / 8); cnt--;) {+ LOAD_8VECS_UB(src, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ LOAD_8VECS_UB(dst, dst_stride,
+ dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+ dst2 = __msa_aver_u_b(src2, dst2);
+ dst3 = __msa_aver_u_b(src3, dst3);
+ dst4 = __msa_aver_u_b(src4, dst4);
+ dst5 = __msa_aver_u_b(src5, dst5);
+ dst6 = __msa_aver_u_b(src6, dst6);
+ dst7 = __msa_aver_u_b(src7, dst7);
+
+ STORE_8VECS_UB(dst, dst_stride,
+ dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ dst += (8 * dst_stride);
+ }
+}
+
+static void avg_width32_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint8_t *dst_dup = dst;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
+
+ for (cnt = (height / 8); cnt--;) {+ src0 = LOAD_UB(src);
+ src1 = LOAD_UB(src + 16);
+ src += src_stride;
+ src2 = LOAD_UB(src);
+ src3 = LOAD_UB(src + 16);
+ src += src_stride;
+ src4 = LOAD_UB(src);
+ src5 = LOAD_UB(src + 16);
+ src += src_stride;
+ src6 = LOAD_UB(src);
+ src7 = LOAD_UB(src + 16);
+ src += src_stride;
+
+ dst0 = LOAD_UB(dst_dup);
+ dst1 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst2 = LOAD_UB(dst_dup);
+ dst3 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst4 = LOAD_UB(dst_dup);
+ dst5 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst6 = LOAD_UB(dst_dup);
+ dst7 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+
+ src8 = LOAD_UB(src);
+ src9 = LOAD_UB(src + 16);
+ src += src_stride;
+ src10 = LOAD_UB(src);
+ src11 = LOAD_UB(src + 16);
+ src += src_stride;
+ src12 = LOAD_UB(src);
+ src13 = LOAD_UB(src + 16);
+ src += src_stride;
+ src14 = LOAD_UB(src);
+ src15 = LOAD_UB(src + 16);
+ src += src_stride;
+
+ dst8 = LOAD_UB(dst_dup);
+ dst9 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst10 = LOAD_UB(dst_dup);
+ dst11 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst12 = LOAD_UB(dst_dup);
+ dst13 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+ dst14 = LOAD_UB(dst_dup);
+ dst15 = LOAD_UB(dst_dup + 16);
+ dst_dup += dst_stride;
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+ dst2 = __msa_aver_u_b(src2, dst2);
+ dst3 = __msa_aver_u_b(src3, dst3);
+ dst4 = __msa_aver_u_b(src4, dst4);
+ dst5 = __msa_aver_u_b(src5, dst5);
+ dst6 = __msa_aver_u_b(src6, dst6);
+ dst7 = __msa_aver_u_b(src7, dst7);
+ dst8 = __msa_aver_u_b(src8, dst8);
+ dst9 = __msa_aver_u_b(src9, dst9);
+ dst10 = __msa_aver_u_b(src10, dst10);
+ dst11 = __msa_aver_u_b(src11, dst11);
+ dst12 = __msa_aver_u_b(src12, dst12);
+ dst13 = __msa_aver_u_b(src13, dst13);
+ dst14 = __msa_aver_u_b(src14, dst14);
+ dst15 = __msa_aver_u_b(src15, dst15);
+
+ STORE_UB(dst0, dst);
+ STORE_UB(dst1, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst2, dst);
+ STORE_UB(dst3, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst4, dst);
+ STORE_UB(dst5, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst6, dst);
+ STORE_UB(dst7, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst8, dst);
+ STORE_UB(dst9, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst10, dst);
+ STORE_UB(dst11, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst12, dst);
+ STORE_UB(dst13, dst + 16);
+ dst += dst_stride;
+ STORE_UB(dst14, dst);
+ STORE_UB(dst15, dst + 16);
+ dst += dst_stride;
+ }
+}
+
+static void avg_width64_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint8_t *dst_dup = dst;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
+
+ for (cnt = (height / 4); cnt--;) {+ LOAD_4VECS_UB(src, 16, src0, src1, src2, src3);
+ src += src_stride;
+ LOAD_4VECS_UB(src, 16, src4, src5, src6, src7);
+ src += src_stride;
+ LOAD_4VECS_UB(src, 16, src8, src9, src10, src11);
+ src += src_stride;
+ LOAD_4VECS_UB(src, 16, src12, src13, src14, src15);
+ src += src_stride;
+
+ LOAD_4VECS_UB(dst_dup, 16, dst0, dst1, dst2, dst3);
+ dst_dup += dst_stride;
+ LOAD_4VECS_UB(dst_dup, 16, dst4, dst5, dst6, dst7);
+ dst_dup += dst_stride;
+ LOAD_4VECS_UB(dst_dup, 16, dst8, dst9, dst10, dst11);
+ dst_dup += dst_stride;
+ LOAD_4VECS_UB(dst_dup, 16, dst12, dst13, dst14, dst15);
+ dst_dup += dst_stride;
+
+ dst0 = __msa_aver_u_b(src0, dst0);
+ dst1 = __msa_aver_u_b(src1, dst1);
+ dst2 = __msa_aver_u_b(src2, dst2);
+ dst3 = __msa_aver_u_b(src3, dst3);
+ dst4 = __msa_aver_u_b(src4, dst4);
+ dst5 = __msa_aver_u_b(src5, dst5);
+ dst6 = __msa_aver_u_b(src6, dst6);
+ dst7 = __msa_aver_u_b(src7, dst7);
+ dst8 = __msa_aver_u_b(src8, dst8);
+ dst9 = __msa_aver_u_b(src9, dst9);
+ dst10 = __msa_aver_u_b(src10, dst10);
+ dst11 = __msa_aver_u_b(src11, dst11);
+ dst12 = __msa_aver_u_b(src12, dst12);
+ dst13 = __msa_aver_u_b(src13, dst13);
+ dst14 = __msa_aver_u_b(src14, dst14);
+ dst15 = __msa_aver_u_b(src15, dst15);
+
+ STORE_4VECS_UB(dst, 16, dst0, dst1, dst2, dst3);
+ dst += dst_stride;
+ STORE_4VECS_UB(dst, 16, dst4, dst5, dst6, dst7);
+ dst += dst_stride;
+ STORE_4VECS_UB(dst, 16, dst8, dst9, dst10, dst11);
+ dst += dst_stride;
+ STORE_4VECS_UB(dst, 16, dst12, dst13, dst14, dst15);
+ dst += dst_stride;
+ }
+}
+
+void vp9_convolve_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t filter_x_stride,
+ const int16_t *filter_y, int32_t filter_y_stride,
+ int32_t w, int32_t h) {+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+
+ switch (w) {+ case 4: {+ avg_width4_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 8: {+ avg_width8_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 16: {+ avg_width16_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 32: {+ avg_width32_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 64: {+ avg_width64_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ default: {+ int32_t lp, cnt;
+ for (cnt = h; cnt--;) {+ for (lp = 0; lp < w; ++lp) {+ dst[lp] = (((dst[lp] + src[lp]) + 1) >> 1);
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_convolve_copy_msa.c
@@ -1,0 +1,300 @@
+/*
+ * Copyright (c) 2015 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 <string.h>
+#include "vp9/common/mips/msa/vp9_macros_msa.h"
+
+static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height) {+ int32_t cnt;
+ uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LOAD_8VECS_UB(src, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+ out4 = __msa_copy_u_d((v2i64)src4, 0);
+ out5 = __msa_copy_u_d((v2i64)src5, 0);
+ out6 = __msa_copy_u_d((v2i64)src6, 0);
+ out7 = __msa_copy_u_d((v2i64)src7, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ STORE_DWORD(dst, out2);
+ dst += dst_stride;
+ STORE_DWORD(dst, out3);
+ dst += dst_stride;
+ STORE_DWORD(dst, out4);
+ dst += dst_stride;
+ STORE_DWORD(dst, out5);
+ dst += dst_stride;
+ STORE_DWORD(dst, out6);
+ dst += dst_stride;
+ STORE_DWORD(dst, out7);
+ dst += dst_stride;
+
+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ STORE_DWORD(dst, out2);
+ dst += dst_stride;
+ STORE_DWORD(dst, out3);
+ dst += dst_stride;
+ }
+ } else if (0 == height % 8) {+ for (cnt = height >> 3; cnt--;) {+ LOAD_8VECS_UB(src, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+ out4 = __msa_copy_u_d((v2i64)src4, 0);
+ out5 = __msa_copy_u_d((v2i64)src5, 0);
+ out6 = __msa_copy_u_d((v2i64)src6, 0);
+ out7 = __msa_copy_u_d((v2i64)src7, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ STORE_DWORD(dst, out2);
+ dst += dst_stride;
+ STORE_DWORD(dst, out3);
+ dst += dst_stride;
+ STORE_DWORD(dst, out4);
+ dst += dst_stride;
+ STORE_DWORD(dst, out5);
+ dst += dst_stride;
+ STORE_DWORD(dst, out6);
+ dst += dst_stride;
+ STORE_DWORD(dst, out7);
+ dst += dst_stride;
+ }
+ } else if (0 == height % 4) {+ for (cnt = (height / 4); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ STORE_DWORD(dst, out2);
+ dst += dst_stride;
+ STORE_DWORD(dst, out3);
+ dst += dst_stride;
+ }
+ } else if (0 == height % 2) {+ for (cnt = (height / 2); cnt--;) {+ LOAD_2VECS_UB(src, src_stride, src0, src1);
+ src += (2 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+
+ STORE_DWORD(dst, out0);
+ dst += dst_stride;
+ STORE_DWORD(dst, out1);
+ dst += dst_stride;
+ }
+ }
+}
+
+static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height, int32_t width) {+ int32_t cnt, loop_cnt;
+ const uint8_t *src_tmp;
+ uint8_t *dst_tmp;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ for (cnt = (width >> 4); cnt--;) {+ src_tmp = src;
+ dst_tmp = dst;
+
+ for (loop_cnt = (height >> 3); loop_cnt--;) {+ LOAD_8VECS_UB(src_tmp, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src_tmp += (8 * src_stride);
+
+ STORE_8VECS_UB(dst_tmp, dst_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ dst_tmp += (8 * dst_stride);
+ }
+
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LOAD_8VECS_UB(src, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ STORE_8VECS_UB(dst, dst_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ dst += (8 * dst_stride);
+
+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 8) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
+ } else if (0 == height % 4) {+ for (cnt = (height >> 2); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ dst += (4 * dst_stride);
+ }
+ }
+}
+
+static void copy_width32_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7);
+ dst += (4 * dst_stride);
+
+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7);
+ dst += (4 * dst_stride);
+
+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 8) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32);
+ } else if (0 == height % 4) {+ for (cnt = (height >> 2); cnt--;) {+ LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3);
+ LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+
+ STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3);
+ STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7);
+ dst += (4 * dst_stride);
+ }
+ }
+}
+
+static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64);
+}
+
+void vp9_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t filter_x_stride,
+ const int16_t *filter_y, int32_t filter_y_stride,
+ int32_t w, int32_t h) {+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+
+ switch (w) {+ case 4: {+ uint32_t cnt, tmp;
+ /* 1 word storage */
+ for (cnt = h; cnt--;) {+ tmp = LOAD_WORD(src);
+ STORE_WORD(dst, tmp);
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ case 8: {+ copy_width8_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 16: {+ copy_width16_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 32: {+ copy_width32_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 64: {+ copy_width64_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ default: {+ uint32_t cnt;
+ for (cnt = h; cnt--;) {+ memcpy(dst, src, w);
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ }
+}
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -286,10 +286,10 @@
# 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 dspr2/, "$sse2_x86inc";
+specialize qw/vp9_convolve_copy neon dspr2 msa/, "$sse2_x86inc";
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/, "$sse2_x86inc";
+specialize qw/vp9_convolve_avg neon dspr2 msa/, "$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 dspr2 msa/, "$avx2_ssse3";
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -134,6 +134,8 @@
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_horiz_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_msa.c
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_avg_msa.c
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_copy_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_msa.h
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c