generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge "ppc: Add vpx_sad16x8/16/32_vsx"

Browse Source
This commit is contained in:
James Zern 2017-05-12 13:33:50 +00:00 committed by Gerrit Code Review
commit a12ea1d5e9
4 changed files with 67 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
test/sad_test.cc
View File

@ -920,4 +920,13 @@ const SadMxNx4Param x4d_msa_tests[] = {
INSTANTIATE_TEST_CASE_P(MSA, SADx4Test, ::testing::ValuesIn(x4d_msa_tests));
#endif // HAVE_MSA
//------------------------------------------------------------------------------
// VSX functions
#if HAVE_VSX
const SadMxNParam vsx_tests[] = {
SadMxNParam(16, 32, &vpx_sad16x32_vsx),
SadMxNParam(16, 16, &vpx_sad16x16_vsx), SadMxNParam(16, 8, &vpx_sad16x8_vsx),
};
INSTANTIATE_TEST_CASE_P(VSX, SADTest, ::testing::ValuesIn(vsx_tests));
#endif // HAVE_VSX
} // namespace

53
vpx_dsp/ppc/sad_vsx.c Normal file
View File

@ -0,0 +1,53 @@
/*
* Copyright (c) 2017 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 <stdlib.h>
#include "vpx_dsp/ppc/types_vsx.h"
#include "vpx/vpx_integer.h"
#define PROCESS16(offset) \
v_a = vec_vsx_ld(offset, a); \
v_b = vec_vsx_ld(offset, b); \
v_ah = unpack_to_s16_h(v_a); \
v_al = unpack_to_s16_l(v_a); \
v_bh = unpack_to_s16_h(v_b); \
v_bl = unpack_to_s16_l(v_b); \
v_subh = vec_sub(v_ah, v_bh); \
v_subl = vec_sub(v_al, v_bl); \
v_absh = vec_abs(v_subh); \
v_absl = vec_abs(v_subl); \
v_sad = vec_sum4s(v_absh, v_sad); \
v_sad = vec_sum4s(v_absl, v_sad);
#define SAD16(height) \
unsigned int vpx_sad16x##height##_vsx(const uint8_t *a, int a_stride, \
const uint8_t *b, int b_stride) { \
int y; \
unsigned int sad[4]; \
uint8x16_t v_a, v_b; \
int16x8_t v_ah, v_al, v_bh, v_bl, v_absh, v_absl, v_subh, v_subl; \
int32x4_t v_sad = vec_splat_s32(0); \
\
for (y = 0; y < height; y++) { \
PROCESS16(0); \
\
a += a_stride; \
b += b_stride; \
} \
vec_vsx_st((uint32x4_t)v_sad, 0, sad); \
\
return sad[3] + sad[2] + sad[1] + sad[0]; \
}
SAD16(8);
SAD16(16);
SAD16(32);

2
vpx_dsp/vpx_dsp.mk
View File

@ -311,6 +311,8 @@ DSP_SRCS-$(HAVE_SSE2) += x86/sad4d_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/sad_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/subtract_sse2.asm
DSP_SRCS-$(HAVE_VSX) += ppc/sad_vsx.c
ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_sad4d_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_sad_sse2.asm

6
vpx_dsp/vpx_dsp_rtcd_defs.pl
View File

@ -711,13 +711,13 @@ add_proto qw/unsigned int vpx_sad32x16/, "const uint8_t *src_ptr, int src_stride
specialize qw/vpx_sad32x16 avx2 msa sse2/;
add_proto qw/unsigned int vpx_sad16x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
specialize qw/vpx_sad16x32 msa sse2/;
specialize qw/vpx_sad16x32 msa sse2 vsx/;
add_proto qw/unsigned int vpx_sad16x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
specialize qw/vpx_sad16x16 neon msa sse2/;
specialize qw/vpx_sad16x16 neon msa sse2 vsx/;
add_proto qw/unsigned int vpx_sad16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
specialize qw/vpx_sad16x8 neon msa sse2/;
specialize qw/vpx_sad16x8 neon msa sse2 vsx/;
add_proto qw/unsigned int vpx_sad8x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
specialize qw/vpx_sad8x16 neon msa sse2/;