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

Merge "BUG FIX: sse2 subpel variance is not PIC compliant"

Browse Source
This commit is contained in:
Scott LaVarnway 2018-01-24 22:54:42 +00:00 committed by Gerrit Code Review
commit 15b261d854
2 changed files with 78 additions and 90 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

80
vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
View File

@ -91,7 +91,7 @@ SECTION .text
%define filter_idx_shift 5
%ifdef PIC ; 64bit PIC
%if ARCH_X86_64
%if %2 == 1 ; avg
cglobal highbd_sub_pixel_avg_variance%1xh, 9, 10, 13, src, src_stride, \
x_offset, y_offset, \
@ -99,19 +99,20 @@ SECTION .text
sec, sec_stride, height, sse
%define sec_str sec_strideq
%else
cglobal highbd_sub_pixel_variance%1xh, 7, 8, 13, src, src_stride, x_offset, \
y_offset, dst, dst_stride, height, sse
cglobal highbd_sub_pixel_variance%1xh, 7, 8, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, height, sse
%endif
%define block_height heightd
%define bilin_filter sseq
%else
%if ARCH_X86=1 && CONFIG_PIC=1
%if CONFIG_PIC=1
%if %2 == 1 ; avg
cglobal highbd_sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, \
height, sse, g_bilin_filter, g_pw_8
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, height, sse, \
g_bilin_filter, g_pw_8
%define block_height dword heightm
%define sec_str sec_stridemp
@ -130,8 +131,9 @@ SECTION .text
LOAD_IF_USED 0, 1 ; load eax, ecx back
%else
cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, dst, dst_stride, height, \
sse, g_bilin_filter, g_pw_8
x_offset, y_offset, \
dst, dst_stride, height, sse, \
g_bilin_filter, g_pw_8
%define block_height heightd
; Store bilin_filter and pw_8 location in stack
@ -150,22 +152,16 @@ SECTION .text
%endif
%else
%if %2 == 1 ; avg
cglobal highbd_sub_pixel_avg_variance%1xh, 7 + 2 * ARCH_X86_64, \
7 + 2 * ARCH_X86_64, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, \
height, sse
%if ARCH_X86_64
%define block_height heightd
%define sec_str sec_strideq
%else
cglobal highbd_sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, height, sse
%define block_height dword heightm
%define sec_str sec_stridemp
%endif
%else
cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, dst, dst_stride, height, sse
x_offset, y_offset, \
dst, dst_stride, height, sse
%define block_height heightd
%endif
@ -284,14 +280,14 @@ SECTION .text
.x_zero_y_nonhalf:
; x_offset == 0 && y_offset == bilin interpolation
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl y_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
mova m8, [bilin_filter+y_offsetq]
mova m9, [bilin_filter+y_offsetq+16]
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
@ -308,7 +304,7 @@ SECTION .text
add y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -511,14 +507,14 @@ SECTION .text
.x_half_y_nonhalf:
; x_offset == 0.5 && y_offset == bilin interpolation
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl y_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
mova m8, [bilin_filter+y_offsetq]
mova m9, [bilin_filter+y_offsetq+16]
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
@ -535,7 +531,7 @@ SECTION .text
add y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -633,14 +629,14 @@ SECTION .text
jnz .x_nonhalf_y_nonzero
; x_offset == bilin interpolation && y_offset == 0
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
mova m8, [bilin_filter+x_offsetq]
mova m9, [bilin_filter+x_offsetq+16]
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
@ -657,7 +653,7 @@ SECTION .text
add x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -732,14 +728,14 @@ SECTION .text
jne .x_nonhalf_y_nonhalf
; x_offset == bilin interpolation && y_offset == 0.5
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
mova m8, [bilin_filter+x_offsetq]
mova m9, [bilin_filter+x_offsetq+16]
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
@ -756,7 +752,7 @@ SECTION .text
add x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -859,8 +855,8 @@ SECTION .text
.x_nonhalf_y_nonhalf:
; loading filter - this is same as in 8-bit depth
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift ; filter_idx_shift = 5
shl y_offsetd, filter_idx_shift
@ -869,7 +865,7 @@ SECTION .text
mova m9, [bilin_filter+x_offsetq+16]
mova m10, [bilin_filter+y_offsetq]
mova m11, [bilin_filter+y_offsetq+16]
mova m12, [pw_8]
mova m12, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_y_a m10
@ -897,7 +893,7 @@ SECTION .text
%define filter_x_b [x_offsetq+16]
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
; end of load filter

88
vpx_dsp/x86/subpel_variance_sse2.asm
View File

@ -114,27 +114,26 @@ SECTION .text
; 11, not 13, if the registers are ordered correctly. May make a minor speed
; difference on Win64
%ifdef PIC ; 64bit PIC
%if ARCH_X86_64
%if %2 == 1 ; avg
cglobal sub_pixel_avg_variance%1xh, 9, 10, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, height, sse
x_offset, y_offset, dst, dst_stride, \
sec, sec_stride, height, sse
%define sec_str sec_strideq
%else
cglobal sub_pixel_variance%1xh, 7, 8, 13, src, src_stride, x_offset, \
y_offset, dst, dst_stride, height, sse
cglobal sub_pixel_variance%1xh, 7, 8, 13, src, src_stride, \
x_offset, y_offset, dst, dst_stride, \
height, sse
%endif
%define block_height heightd
%define bilin_filter sseq
%else
%if ARCH_X86=1 && CONFIG_PIC=1
%if CONFIG_PIC=1
%if %2 == 1 ; avg
cglobal sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, \
height, sse, g_bilin_filter, g_pw_8
x_offset, y_offset, dst, dst_stride, \
sec, sec_stride, height, sse, \
g_bilin_filter, g_pw_8
%define block_height dword heightm
%define sec_str sec_stridemp
@ -152,9 +151,9 @@ SECTION .text
LOAD_IF_USED 0, 1 ; load eax, ecx back
%else
cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, x_offset, \
y_offset, dst, dst_stride, height, sse, \
g_bilin_filter, g_pw_8
cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, dst, dst_stride, \
height, sse, g_bilin_filter, g_pw_8
%define block_height heightd
;Store bilin_filter and pw_8 location in stack
@ -173,25 +172,18 @@ SECTION .text
%endif
%else
%if %2 == 1 ; avg
cglobal sub_pixel_avg_variance%1xh, 7 + 2 * ARCH_X86_64, \
7 + 2 * ARCH_X86_64, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
sec, sec_stride, \
height, sse
%if ARCH_X86_64
%define block_height heightd
%define sec_str sec_strideq
%else
cglobal sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, sec, sec_stride, \
height, sse
%define block_height dword heightm
%define sec_str sec_stridemp
%endif
%else
cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, x_offset, \
y_offset, dst, dst_stride, height, sse
cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
x_offset, y_offset, dst, dst_stride, \
height, sse
%define block_height heightd
%endif
%define bilin_filter bilin_filter_m
%endif
%endif
@ -371,8 +363,8 @@ SECTION .text
.x_zero_y_nonhalf:
; x_offset == 0 && y_offset == bilin interpolation
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl y_offsetd, filter_idx_shift
%if ARCH_X86_64 && %1 > 4
@ -380,7 +372,7 @@ SECTION .text
%if notcpuflag(ssse3) ; FIXME(rbultje) don't scatter registers on x86-64
mova m9, [bilin_filter+y_offsetq+16]
%endif
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
@ -397,7 +389,7 @@ SECTION .text
add y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -694,8 +686,8 @@ SECTION .text
.x_half_y_nonhalf:
; x_offset == 0.5 && y_offset == bilin interpolation
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl y_offsetd, filter_idx_shift
%if ARCH_X86_64 && %1 > 4
@ -703,7 +695,7 @@ SECTION .text
%if notcpuflag(ssse3) ; FIXME(rbultje) don't scatter registers on x86-64
mova m9, [bilin_filter+y_offsetq+16]
%endif
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
@ -720,7 +712,7 @@ SECTION .text
add y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -852,8 +844,8 @@ SECTION .text
jnz .x_nonhalf_y_nonzero
; x_offset == bilin interpolation && y_offset == 0
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift
%if ARCH_X86_64 && %1 > 4
@ -861,7 +853,7 @@ SECTION .text
%if notcpuflag(ssse3) ; FIXME(rbultje) don't scatter registers on x86-64
mova m9, [bilin_filter+x_offsetq+16]
%endif
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
@ -878,7 +870,7 @@ SECTION .text
add x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -994,8 +986,8 @@ SECTION .text
jne .x_nonhalf_y_nonhalf
; x_offset == bilin interpolation && y_offset == 0.5
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift
%if ARCH_X86_64 && %1 > 4
@ -1003,7 +995,7 @@ SECTION .text
%if notcpuflag(ssse3) ; FIXME(rbultje) don't scatter registers on x86-64
mova m9, [bilin_filter+x_offsetq+16]
%endif
mova m10, [pw_8]
mova m10, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
@ -1020,7 +1012,7 @@ SECTION .text
add x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif
@ -1192,8 +1184,8 @@ SECTION .text
STORE_AND_RET %1
.x_nonhalf_y_nonhalf:
%ifdef PIC
lea bilin_filter, [bilin_filter_m]
%if ARCH_X86_64
lea bilin_filter, [GLOBAL(bilin_filter_m)]
%endif
shl x_offsetd, filter_idx_shift
shl y_offsetd, filter_idx_shift
@ -1206,7 +1198,7 @@ SECTION .text
%if notcpuflag(ssse3) ; FIXME(rbultje) don't scatter registers on x86-64
mova m11, [bilin_filter+y_offsetq+16]
%endif
mova m12, [pw_8]
mova m12, [GLOBAL(pw_8)]
%define filter_x_a m8
%define filter_x_b m9
%define filter_y_a m10
@ -1234,7 +1226,7 @@ SECTION .text
%define filter_x_b [x_offsetq+16]
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%define filter_rnd [GLOBAL(pw_8)]
%endif
%endif