Commit eea8ea88 authored by James Zern's avatar James Zern
Browse files

vpx_dsp: remove mmx variance implementations

there are sse2 equivalents for all remaining variance implementations

Change-Id: I10b947e73fc0067688181f819b59e47966bec3d2
parent 7aef9790
......@@ -976,16 +976,6 @@ INSTANTIATE_TEST_CASE_P(
make_tuple(2, 2, &vpx_highbd_12_sub_pixel_avg_variance4x4_c, 12)));
#endif // CONFIG_VP9_HIGHBITDEPTH
#if HAVE_MMX
INSTANTIATE_TEST_CASE_P(
MMX, VpxSubpelVarianceTest,
::testing::Values(make_tuple(4, 4, &vpx_sub_pixel_variance16x16_mmx, 0),
make_tuple(4, 3, &vpx_sub_pixel_variance16x8_mmx, 0),
make_tuple(3, 4, &vpx_sub_pixel_variance8x16_mmx, 0),
make_tuple(3, 3, &vpx_sub_pixel_variance8x8_mmx, 0),
make_tuple(2, 2, &vpx_sub_pixel_variance4x4_mmx, 0)));
#endif // HAVE_MMX
#if HAVE_SSE2
INSTANTIATE_TEST_CASE_P(SSE2, SumOfSquaresTest,
::testing::Values(vpx_get_mb_ss_sse2));
......
......@@ -322,8 +322,6 @@ DSP_SRCS-$(HAVE_NEON) += arm/variance_neon.c
DSP_SRCS-$(HAVE_MSA) += mips/variance_msa.c
DSP_SRCS-$(HAVE_MSA) += mips/sub_pixel_variance_msa.c
DSP_SRCS-$(HAVE_MMX) += x86/variance_mmx.c
DSP_SRCS-$(HAVE_MMX) += x86/variance_impl_mmx.asm
DSP_SRCS-$(HAVE_SSE) += x86/variance_sse2.c
DSP_SRCS-$(HAVE_SSE2) += x86/variance_sse2.c # Contains SSE2 and SSSE3
DSP_SRCS-$(HAVE_SSE2) += x86/halfpix_variance_sse2.c
......
......@@ -1478,16 +1478,16 @@ add_proto qw/uint32_t vpx_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int
specialize qw/vpx_sub_pixel_variance16x32 msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance16x16 mmx media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";
specialize qw/vpx_sub_pixel_variance16x16 media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance16x8 mmx msa/, "$sse2_x86inc", "$ssse3_x86inc";
specialize qw/vpx_sub_pixel_variance16x8 msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance8x16 mmx msa/, "$sse2_x86inc", "$ssse3_x86inc";
specialize qw/vpx_sub_pixel_variance8x16 msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance8x8 mmx media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";
specialize qw/vpx_sub_pixel_variance8x8 media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance8x4 msa/, "$sse2_x86inc", "$ssse3_x86inc";
......@@ -1496,7 +1496,7 @@ add_proto qw/uint32_t vpx_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int s
specialize qw/vpx_sub_pixel_variance4x8 msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_sub_pixel_variance4x4 mmx msa/, "$sse2_x86inc", "$ssse3_x86inc";
specialize qw/vpx_sub_pixel_variance4x4 msa/, "$sse2_x86inc", "$ssse3_x86inc";
add_proto qw/uint32_t vpx_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
specialize qw/vpx_sub_pixel_avg_variance64x64 avx2 msa/, "$sse2_x86inc", "$ssse3_x86inc";
......@@ -1541,13 +1541,13 @@ add_proto qw/uint32_t vpx_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, i
# Specialty Subpixel
#
add_proto qw/uint32_t vpx_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_variance_halfpixvar16x16_h mmx sse2 media/;
specialize qw/vpx_variance_halfpixvar16x16_h sse2 media/;
add_proto qw/uint32_t vpx_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_variance_halfpixvar16x16_v mmx sse2 media/;
specialize qw/vpx_variance_halfpixvar16x16_v sse2 media/;
add_proto qw/uint32_t vpx_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, uint32_t *sse";
specialize qw/vpx_variance_halfpixvar16x16_hv mmx sse2 media/;
specialize qw/vpx_variance_halfpixvar16x16_hv sse2 media/;
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/unsigned int vpx_highbd_12_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
......
;
; Copyright (c) 2010 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 "vpx_ports/x86_abi_support.asm"
%define mmx_filter_shift 7
;void vpx_filter_block2d_bil4x4_var_mmx
;(
; unsigned char *ref_ptr,
; int ref_pixels_per_line,
; unsigned char *src_ptr,
; int src_pixels_per_line,
; unsigned short *HFilter,
; unsigned short *VFilter,
; int *sum,
; unsigned int *sumsquared
;)
global sym(vpx_filter_block2d_bil4x4_var_mmx) PRIVATE
sym(vpx_filter_block2d_bil4x4_var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 8
GET_GOT rbx
push rsi
push rdi
sub rsp, 16
; end prolog
pxor mm6, mm6 ;
pxor mm7, mm7 ;
mov rax, arg(4) ;HFilter ;
mov rdx, arg(5) ;VFilter ;
mov rsi, arg(0) ;ref_ptr ;
mov rdi, arg(2) ;src_ptr ;
mov rcx, 4 ;
pxor mm0, mm0 ;
movd mm1, [rsi] ;
movd mm3, [rsi+1] ;
punpcklbw mm1, mm0 ;
pmullw mm1, [rax] ;
punpcklbw mm3, mm0 ;
pmullw mm3, [rax+8] ;
paddw mm1, mm3 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
movq mm5, mm1
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
add rsi, r8
%endif
.filter_block2d_bil4x4_var_mmx_loop:
movd mm1, [rsi] ;
movd mm3, [rsi+1] ;
punpcklbw mm1, mm0 ;
pmullw mm1, [rax] ;
punpcklbw mm3, mm0 ;
pmullw mm3, [rax+8] ;
paddw mm1, mm3 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
movq mm3, mm5 ;
movq mm5, mm1 ;
pmullw mm3, [rdx] ;
pmullw mm1, [rdx+8] ;
paddw mm1, mm3 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
movd mm3, [rdi] ;
punpcklbw mm3, mm0 ;
psubw mm1, mm3 ;
paddw mm6, mm1 ;
pmaddwd mm1, mm1 ;
paddd mm7, mm1 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
add rdi, dword ptr arg(3) ;src_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
movsxd r9, dword ptr arg(3) ;src_pixels_per_line
add rsi, r8
add rdi, r9
%endif
sub rcx, 1 ;
jnz .filter_block2d_bil4x4_var_mmx_loop ;
pxor mm3, mm3 ;
pxor mm2, mm2 ;
punpcklwd mm2, mm6 ;
punpckhwd mm3, mm6 ;
paddd mm2, mm3 ;
movq mm6, mm2 ;
psrlq mm6, 32 ;
paddd mm2, mm6 ;
psrad mm2, 16 ;
movq mm4, mm7 ;
psrlq mm4, 32 ;
paddd mm4, mm7 ;
mov rdi, arg(6) ;sum
mov rsi, arg(7) ;sumsquared
movd dword ptr [rdi], mm2 ;
movd dword ptr [rsi], mm4 ;
; begin epilog
add rsp, 16
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;void vpx_filter_block2d_bil_var_mmx
;(
; unsigned char *ref_ptr,
; int ref_pixels_per_line,
; unsigned char *src_ptr,
; int src_pixels_per_line,
; unsigned int Height,
; unsigned short *HFilter,
; unsigned short *VFilter,
; int *sum,
; unsigned int *sumsquared
;)
global sym(vpx_filter_block2d_bil_var_mmx) PRIVATE
sym(vpx_filter_block2d_bil_var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 9
GET_GOT rbx
push rsi
push rdi
sub rsp, 16
; end prolog
pxor mm6, mm6 ;
pxor mm7, mm7 ;
mov rax, arg(5) ;HFilter ;
mov rdx, arg(6) ;VFilter ;
mov rsi, arg(0) ;ref_ptr ;
mov rdi, arg(2) ;src_ptr ;
movsxd rcx, dword ptr arg(4) ;Height ;
pxor mm0, mm0 ;
movq mm1, [rsi] ;
movq mm3, [rsi+1] ;
movq mm2, mm1 ;
movq mm4, mm3 ;
punpcklbw mm1, mm0 ;
punpckhbw mm2, mm0 ;
pmullw mm1, [rax] ;
pmullw mm2, [rax] ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
pmullw mm3, [rax+8] ;
pmullw mm4, [rax+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
psraw mm2, mmx_filter_shift ;
movq mm5, mm1
packuswb mm5, mm2 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
add rsi, r8
%endif
.filter_block2d_bil_var_mmx_loop:
movq mm1, [rsi] ;
movq mm3, [rsi+1] ;
movq mm2, mm1 ;
movq mm4, mm3 ;
punpcklbw mm1, mm0 ;
punpckhbw mm2, mm0 ;
pmullw mm1, [rax] ;
pmullw mm2, [rax] ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
pmullw mm3, [rax+8] ;
pmullw mm4, [rax+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
psraw mm2, mmx_filter_shift ;
movq mm3, mm5 ;
movq mm4, mm5 ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
movq mm5, mm1 ;
packuswb mm5, mm2 ;
pmullw mm3, [rdx] ;
pmullw mm4, [rdx] ;
pmullw mm1, [rdx+8] ;
pmullw mm2, [rdx+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
psraw mm1, mmx_filter_shift ;
psraw mm2, mmx_filter_shift ;
movq mm3, [rdi] ;
movq mm4, mm3 ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
psubw mm1, mm3 ;
psubw mm2, mm4 ;
paddw mm6, mm1 ;
pmaddwd mm1, mm1 ;
paddw mm6, mm2 ;
pmaddwd mm2, mm2 ;
paddd mm7, mm1 ;
paddd mm7, mm2 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
add rdi, dword ptr arg(3) ;src_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
movsxd r9, dword ptr arg(3) ;src_pixels_per_line ;
add rsi, r8
add rdi, r9
%endif
sub rcx, 1 ;
jnz .filter_block2d_bil_var_mmx_loop ;
pxor mm3, mm3 ;
pxor mm2, mm2 ;
punpcklwd mm2, mm6 ;
punpckhwd mm3, mm6 ;
paddd mm2, mm3 ;
movq mm6, mm2 ;
psrlq mm6, 32 ;
paddd mm2, mm6 ;
psrad mm2, 16 ;
movq mm4, mm7 ;
psrlq mm4, 32 ;
paddd mm4, mm7 ;
mov rdi, arg(7) ;sum
mov rsi, arg(8) ;sumsquared
movd dword ptr [rdi], mm2 ;
movd dword ptr [rsi], mm4 ;
; begin epilog
add rsp, 16
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
;short mmx_bi_rd[4] = { 64, 64, 64, 64};
align 16
mmx_bi_rd:
times 4 dw 64
/*
* Copyright (c) 2010 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 "./vpx_dsp_rtcd.h"
#include "vpx_ports/mem.h"
DECLARE_ALIGNED(16, static const int16_t, bilinear_filters_mmx[8][8]) = {
{ 128, 128, 128, 128, 0, 0, 0, 0 },
{ 112, 112, 112, 112, 16, 16, 16, 16 },
{ 96, 96, 96, 96, 32, 32, 32, 32 },
{ 80, 80, 80, 80, 48, 48, 48, 48 },
{ 64, 64, 64, 64, 64, 64, 64, 64 },
{ 48, 48, 48, 48, 80, 80, 80, 80 },
{ 32, 32, 32, 32, 96, 96, 96, 96 },
{ 16, 16, 16, 16, 112, 112, 112, 112 }
};
extern void vpx_filter_block2d_bil4x4_var_mmx(const unsigned char *ref_ptr,
int ref_pixels_per_line,
const unsigned char *src_ptr,
int src_pixels_per_line,
const int16_t *HFilter,
const int16_t *VFilter,
int *sum,
unsigned int *sumsquared);
extern void vpx_filter_block2d_bil_var_mmx(const unsigned char *ref_ptr,
int ref_pixels_per_line,
const unsigned char *src_ptr,
int src_pixels_per_line,
unsigned int Height,
const int16_t *HFilter,
const int16_t *VFilter,
int *sum,
unsigned int *sumsquared);
uint32_t vpx_sub_pixel_variance4x4_mmx(const uint8_t *a, int a_stride,
int xoffset, int yoffset,
const uint8_t *b, int b_stride,
uint32_t *sse) {
int xsum;
unsigned int xxsum;
vpx_filter_block2d_bil4x4_var_mmx(a, a_stride, b, b_stride,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum, &xxsum);
*sse = xxsum;
return (xxsum - (((unsigned int)xsum * xsum) >> 4));
}
uint32_t vpx_sub_pixel_variance8x8_mmx(const uint8_t *a, int a_stride,
int xoffset, int yoffset,
const uint8_t *b, int b_stride,
uint32_t *sse) {
int xsum;
uint32_t xxsum;
vpx_filter_block2d_bil_var_mmx(a, a_stride, b, b_stride, 8,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum, &xxsum);
*sse = xxsum;
return (xxsum - (((uint32_t)xsum * xsum) >> 6));
}
uint32_t vpx_sub_pixel_variance16x16_mmx(const uint8_t *a, int a_stride,
int xoffset, int yoffset,
const uint8_t *b, int b_stride,
uint32_t *sse) {
int xsum0, xsum1;
unsigned int xxsum0, xxsum1;
vpx_filter_block2d_bil_var_mmx(a, a_stride, b, b_stride, 16,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum0, &xxsum0);
vpx_filter_block2d_bil_var_mmx(a + 8, a_stride, b + 8, b_stride, 16,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum1, &xxsum1);
xsum0 += xsum1;
xxsum0 += xxsum1;
*sse = xxsum0;
return (xxsum0 - (((uint32_t)xsum0 * xsum0) >> 8));
}
uint32_t vpx_sub_pixel_variance16x8_mmx(const uint8_t *a, int a_stride,
int xoffset, int yoffset,
const uint8_t *b, int b_stride,
uint32_t *sse) {
int xsum0, xsum1;
unsigned int xxsum0, xxsum1;
vpx_filter_block2d_bil_var_mmx(a, a_stride, b, b_stride, 8,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum0, &xxsum0);
vpx_filter_block2d_bil_var_mmx(a + 8, a_stride, b + 8, b_stride, 8,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum1, &xxsum1);
xsum0 += xsum1;
xxsum0 += xxsum1;
*sse = xxsum0;
return (xxsum0 - (((uint32_t)xsum0 * xsum0) >> 7));
}
uint32_t vpx_sub_pixel_variance8x16_mmx(const uint8_t *a, int a_stride,
int xoffset, int yoffset,
const uint8_t *b, int b_stride,
uint32_t *sse) {
int xsum;
unsigned int xxsum;
vpx_filter_block2d_bil_var_mmx(a, a_stride, b, b_stride, 16,
bilinear_filters_mmx[xoffset],
bilinear_filters_mmx[yoffset],
&xsum, &xxsum);
*sse = xxsum;
return (xxsum - (((uint32_t)xsum * xsum) >> 7));
}
uint32_t vpx_variance_halfpixvar16x16_h_mmx(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
uint32_t *sse) {
return vpx_sub_pixel_variance16x16_mmx(a, a_stride, 4, 0, b, b_stride, sse);
}
uint32_t vpx_variance_halfpixvar16x16_v_mmx(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
uint32_t *sse) {
return vpx_sub_pixel_variance16x16_mmx(a, a_stride, 0, 4, b, b_stride, sse);
}
uint32_t vpx_variance_halfpixvar16x16_hv_mmx(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
uint32_t *sse) {
return vpx_sub_pixel_variance16x16_mmx(a, a_stride, 4, 4, b, b_stride, sse);
}
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