ref: 7555e2b8225e9387e30288cd4eb5b26414badf14
parent: 7b71cdb0b4639ffd351717dbc91b52e2109edb99
author: Parag Salasakar <[email protected]>
date: Tue Jun 23 03:32:25 EDT 2015
mips msa vp9 avg optimization average improvement ~2x-3x Change-Id: I76f7fc00c0ffdf2b4ba41bf3819f3b6044bcdeff
--- a/test/vp9_avg_test.cc
+++ b/test/vp9_avg_test.cc
@@ -288,4 +288,16 @@
#endif
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(
+ MSA, AverageTest,
+ ::testing::Values(
+ make_tuple(16, 16, 0, 8, &vp9_avg_8x8_msa),
+ make_tuple(16, 16, 5, 8, &vp9_avg_8x8_msa),
+ make_tuple(32, 32, 15, 8, &vp9_avg_8x8_msa),
+ make_tuple(16, 16, 0, 4, &vp9_avg_4x4_msa),
+ make_tuple(16, 16, 5, 4, &vp9_avg_4x4_msa),
+ make_tuple(32, 32, 15, 4, &vp9_avg_4x4_msa)));
+#endif
+
} // namespace
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -878,10 +878,10 @@
specialize qw/vp9_sub_pixel_avg_variance4x4/, "$sse_x86inc", "$ssse3_x86inc";
add_proto qw/unsigned int vp9_avg_8x8/, "const uint8_t *, int p";
-specialize qw/vp9_avg_8x8 sse2 neon/;
+specialize qw/vp9_avg_8x8 sse2 neon msa/;
add_proto qw/unsigned int vp9_avg_4x4/, "const uint8_t *, int p";
-specialize qw/vp9_avg_4x4 sse2/;
+specialize qw/vp9_avg_4x4 sse2 msa/;
add_proto qw/void vp9_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
specialize qw/vp9_minmax_8x8 sse2/;
--- /dev/null
+++ b/vp9/encoder/mips/msa/vp9_avg_msa.c
@@ -1,0 +1,56 @@
+/*
+ * 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_rtcd.h"
+#include "vp9/common/mips/msa/vp9_macros_msa.h"
+
+uint32_t vp9_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {+ uint32_t sum_out;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v8u16 sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7;
+ v4u32 sum = { 0 };+
+ LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ HADD_UB4_UH(src0, src1, src2, src3, sum0, sum1, sum2, sum3);
+ HADD_UB4_UH(src4, src5, src6, src7, sum4, sum5, sum6, sum7);
+ ADD4(sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7, sum0, sum2, sum4, sum6);
+ ADD2(sum0, sum2, sum4, sum6, sum0, sum4);
+ sum0 += sum4;
+
+ sum = __msa_hadd_u_w(sum0, sum0);
+ sum0 = (v8u16)__msa_pckev_h((v8i16)sum, (v8i16)sum);
+ sum = __msa_hadd_u_w(sum0, sum0);
+ sum = (v4u32)__msa_srari_w((v4i32)sum, 6);
+ sum_out = __msa_copy_u_w((v4i32)sum, 0);
+
+ return sum_out;
+}
+
+uint32_t vp9_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {+ uint32_t sum_out;
+ uint32_t src0, src1, src2, src3;
+ v16u8 vec = { 0 };+ v8u16 sum0;
+ v4u32 sum1;
+ v2u64 sum2;
+
+ LW4(src, src_stride, src0, src1, src2, src3);
+ INSERT_W4_UB(src0, src1, src2, src3, vec);
+
+ sum0 = __msa_hadd_u_h(vec, vec);
+ sum1 = __msa_hadd_u_w(sum0, sum0);
+ sum0 = (v8u16)__msa_pckev_h((v8i16)sum1, (v8i16)sum1);
+ sum1 = __msa_hadd_u_w(sum0, sum0);
+ sum2 = __msa_hadd_u_d(sum1, sum1);
+ sum1 = (v4u32)__msa_srari_w((v4i32)sum2, 4);
+ sum_out = __msa_copy_u_w((v4i32)sum1, 0);
+
+ return sum_out;
+}
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -157,5 +157,6 @@
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct16x16_msa.c
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct32x32_msa.c
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct_msa.h
+VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_avg_msa.c
VP9_CX_SRCS-yes := $(filter-out $(VP9_CX_SRCS_REMOVE-yes),$(VP9_CX_SRCS-yes))