Skip to content

GitLab

Commit 27d083c1 authored by Parag Salasakar's avatar Parag Salasakar
Browse files

mips msa vp9 convolve8 vert optimization 

average improvement ~6x-10x

Change-Id: Ie3f3ab3a9005be84935919701e56b404e420affa
parent 3701dcc5
Expand all Hide whitespace changes
Inline Side-by-side
test/convolve_test.cc
View file @ 27d083c1
......@@ -1814,4 +1814,27 @@ INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
make_tuple(32, 64, &convolve8_dspr2),
make_tuple(64, 64, &convolve8_dspr2)));
#endif
#if HAVE_MSA
const ConvolveFunctions convolve8_msa(
vp9_convolve_copy_c, vp9_convolve_avg_c,
vp9_convolve8_horiz_c, vp9_convolve8_avg_horiz_c,
vp9_convolve8_vert_msa, vp9_convolve8_avg_vert_c,
vp9_convolve8_c, vp9_convolve8_avg_c, 0);
INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_msa),
make_tuple(8, 4, &convolve8_msa),
make_tuple(4, 8, &convolve8_msa),
make_tuple(8, 8, &convolve8_msa),
make_tuple(16, 8, &convolve8_msa),
make_tuple(8, 16, &convolve8_msa),
make_tuple(16, 16, &convolve8_msa),
make_tuple(32, 16, &convolve8_msa),
make_tuple(16, 32, &convolve8_msa),
make_tuple(32, 32, &convolve8_msa),
make_tuple(64, 32, &convolve8_msa),
make_tuple(32, 64, &convolve8_msa),
make_tuple(64, 64, &convolve8_msa)));
#endif // HAVE_MSA
} // namespace
vp9/common/mips/msa/vp9_convolve8_vert_msa.c 0 → 100644
View file @ 27d083c1
This diff is collapsed.
vp9/common/mips/msa/vp9_convolve_msa.h 0 → 100644
View file @ 27d083c1
/*
* 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.
*/
#ifndef VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_
#define VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_
#include "vp9/common/vp9_filter.h"
#include "vp9/common/mips/msa/vp9_macros_msa.h"
extern uint8_t mc_filt_mask_arr[16 * 3];
#define HORIZ_8TAP_FILT(src, mask0, mask1, mask2, mask3, \
filt_h0, filt_h1, filt_h2, filt_h3) ({ \
v8i16 vec0, vec1, vec2, vec3, horiz_out; \
\
vec0 = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src), (v16i8)(src)); \
vec0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)(filt_h0)); \
vec1 = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src), (v16i8)(src)); \
vec0 = __msa_dpadd_s_h(vec0, (v16i8)(filt_h1), (v16i8)vec1); \
vec2 = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src), (v16i8)(src)); \
vec2 = __msa_dotp_s_h((v16i8)vec2, (v16i8)(filt_h2)); \
vec3 = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src), (v16i8)(src)); \
vec2 = __msa_dpadd_s_h(vec2, (v16i8)(filt_h3), (v16i8)vec3); \
vec0 = __msa_adds_s_h(vec0, vec2); \
horiz_out = SRARI_SATURATE_SIGNED_H(vec0, FILTER_BITS, 7); \
\
horiz_out; \
})
#define HORIZ_8TAP_FILT_2VECS(src0, src1, mask0, mask1, mask2, mask3, \
filt_h0, filt_h1, filt_h2, filt_h3) ({ \
v8i16 vec0, vec1, vec2, vec3, horiz_out; \
\
vec0 = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src0)); \
vec0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)(filt_h0)); \
vec1 = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src0)); \
vec0 = __msa_dpadd_s_h(vec0, (v16i8)(filt_h1), (v16i8)vec1); \
vec2 = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src0)); \
vec2 = __msa_dotp_s_h((v16i8)vec2, (v16i8)(filt_h2)); \
vec3 = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src0)); \
vec2 = __msa_dpadd_s_h(vec2, ((v16i8)filt_h3), (v16i8)vec3); \
vec0 = __msa_adds_s_h(vec0, vec2); \
horiz_out = (v8i16)SRARI_SATURATE_SIGNED_H(vec0, FILTER_BITS, 7); \
\
horiz_out; \
})
#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \
filt0, filt1, filt2, filt3) ({ \
v8i16 tmp0, tmp1; \
\
tmp0 = __msa_dotp_s_h((v16i8)(vec0), (v16i8)(filt0)); \
tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)(vec1), (v16i8)(filt1)); \
tmp1 = __msa_dotp_s_h((v16i8)(vec2), (v16i8)(filt2)); \
tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)(vec3), ((v16i8)filt3)); \
tmp0 = __msa_adds_s_h(tmp0, tmp1); \
\
tmp0; \
})
#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \
mask0, mask1, mask2, mask3, \
filt0, filt1, filt2, filt3, \
out0, out1) { \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
v8i16 res0_m, res1_m, res2_m, res3_m; \
\
vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src0)); \
vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src3), (v16i8)(src2)); \
\
res0_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt0)); \
res1_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt0)); \
\
vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src0)); \
vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src3), (v16i8)(src2)); \
\
res0_m = __msa_dpadd_s_h(res0_m, (filt1), (v16i8)vec2_m); \
res1_m = __msa_dpadd_s_h(res1_m, (filt1), (v16i8)vec3_m); \
\
vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src0)); \
vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src3), (v16i8)(src2)); \
\
res2_m = __msa_dotp_s_h((v16i8)(filt2), (v16i8)vec4_m); \
res3_m = __msa_dotp_s_h((v16i8)(filt2), (v16i8)vec5_m); \
\
vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src0)); \
vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src3), (v16i8)(src2)); \
\
res2_m = __msa_dpadd_s_h(res2_m, (v16i8)(filt3), (v16i8)vec6_m); \
res3_m = __msa_dpadd_s_h(res3_m, (v16i8)(filt3), (v16i8)vec7_m); \
\
out0 = __msa_adds_s_h(res0_m, res2_m); \
out1 = __msa_adds_s_h(res1_m, res3_m); \
}
#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \
mask0, mask1, mask2, mask3, \
filt0, filt1, filt2, filt3, \
out0, out1, out2, out3) { \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m; \
v8i16 vec4_m, vec5_m, vec6_m, vec7_m; \
v8i16 res0_m, res1_m, res2_m, res3_m; \
v8i16 res4_m, res5_m, res6_m, res7_m; \
\
vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src0), (v16i8)(src0)); \
vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src1)); \
vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src2), (v16i8)(src2)); \
vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src3), (v16i8)(src3)); \
\
res0_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt0)); \
res1_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt0)); \
res2_m = __msa_dotp_s_h((v16i8)vec2_m, (v16i8)(filt0)); \
res3_m = __msa_dotp_s_h((v16i8)vec3_m, (v16i8)(filt0)); \
\
vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src0), (v16i8)(src0)); \
vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src1)); \
vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src2), (v16i8)(src2)); \
vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src3), (v16i8)(src3)); \
\
res4_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt2)); \
res5_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt2)); \
res6_m = __msa_dotp_s_h((v16i8)vec2_m, (v16i8)(filt2)); \
res7_m = __msa_dotp_s_h((v16i8)vec3_m, (v16i8)(filt2)); \
\
vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src0), (v16i8)(src0)); \
vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src1)); \
vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src2), (v16i8)(src2)); \
vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src3), (v16i8)(src3)); \
\
res0_m = __msa_dpadd_s_h(res0_m, (v16i8)(filt1), (v16i8)vec4_m); \
res1_m = __msa_dpadd_s_h(res1_m, (v16i8)(filt1), (v16i8)vec5_m); \
res2_m = __msa_dpadd_s_h(res2_m, (v16i8)(filt1), (v16i8)vec6_m); \
res3_m = __msa_dpadd_s_h(res3_m, (v16i8)(filt1), (v16i8)vec7_m); \
\
vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src0), (v16i8)(src0)); \
vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src1)); \
vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src2), (v16i8)(src2)); \
vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src3), (v16i8)(src3)); \
\
res4_m = __msa_dpadd_s_h(res4_m, (v16i8)(filt3), (v16i8)vec4_m); \
res5_m = __msa_dpadd_s_h(res5_m, (v16i8)(filt3), (v16i8)vec5_m); \
res6_m = __msa_dpadd_s_h(res6_m, (v16i8)(filt3), (v16i8)vec6_m); \
res7_m = __msa_dpadd_s_h(res7_m, (v16i8)(filt3), (v16i8)vec7_m); \
\
out0 = __msa_adds_s_h(res0_m, res4_m); \
out1 = __msa_adds_s_h(res1_m, res5_m); \
out2 = __msa_adds_s_h(res2_m, res6_m); \
out3 = __msa_adds_s_h(res3_m, res7_m); \
}
#endif /* VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_ */
vp9/common/vp9_rtcd_defs.pl
View file @ 27d083c1
......@@ -298,7 +298,7 @@ add_proto qw/void vp9_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_strid
specialize qw/vp9_convolve8_horiz sse2 ssse3 neon dspr2/, "$avx2_ssse3";
add_proto qw/void vp9_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
specialize qw/vp9_convolve8_vert sse2 ssse3 neon dspr2/, "$avx2_ssse3";
specialize qw/vp9_convolve8_vert sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3";
add_proto qw/void vp9_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
specialize qw/vp9_convolve8_avg sse2 ssse3 neon dspr2/;
......
vp9/vp9_common.mk
View file @ 27d083c1
......@@ -131,6 +131,8 @@ VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_mblpf_vert_loopfilter_ds
# common (msa)
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_macros_msa.h
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_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
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.h
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment