Skip to content

F
ffmpeg.wasm-core
Commit bee330e3 authored by Ronald S. Bultje's avatar Ronald S. Bultje
Browse files
Options

vp8: convert inner loopfilter x86 assembly to use named arguments.

parent ffae713a
Show whitespace changes
Inline Side-by-side
Showing with 164 additions and 187 deletions
libavcodec/x86/vp8dsp.asm
View file @ bee330e3
...@@ -1654,174 +1654,151 @@ SIMPLE_LOOPFILTER h, 5 ...@@ -1654,174 +1654,151 @@ SIMPLE_LOOPFILTER h, 5
; int flimE, int flimI, int hev_thr); ; int flimE, int flimI, int hev_thr);
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
%macro INNER_LOOPFILTER 3 %macro INNER_LOOPFILTER 2
%if %3 == 8 ; chroma %if %2 == 8 ; chroma
cglobal vp8_%1_loop_filter8uv_inner, 6, %2, 13 cglobal vp8_%1_loop_filter8uv_inner, 6, 6, 13, dst, dst8, stride, flimE, flimI, hevthr
%define dst8_reg r1
%define mstride_reg r2
%define E_reg r3
%define I_reg r4
%define hev_thr_reg r5
%else ; luma %else ; luma
cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, dst, stride, flimE, flimI, hevthr
%define mstride_reg r1
%define E_reg r2
%define I_reg r3
%define hev_thr_reg r4
%ifdef m8 ; x86-64, sse2
%define dst8_reg r4
%elif mmsize == 16 ; x86-32, sse2
%define dst8_reg r5
%else ; x86-32, mmx/mmxext
%define cnt_reg r5
%endif
%endif
%define dst_reg r0
%define stride_reg E_reg
%define dst2_reg I_reg
%ifndef m8
%define stack_reg hev_thr_reg
%endif %endif
%if cpuflag(ssse3) %if cpuflag(ssse3)
pxor m7, m7 pxor m7, m7
%endif %endif
%ifndef m8 ; stack layout: [0]=E, [1]=I, [2]=hev_thr
%ifndef m8 ; mmx/mmxext or sse2 on x86-32 %ifidn %1, v ; [3]=hev() result
; splat function arguments %assign pad 16 + mmsize * 4 - gprsize - (stack_offset & 15)
SPLATB_REG m0, E_reg, m7 ; E %else ; h ; extra storage space for transposes
SPLATB_REG m1, I_reg, m7 ; I %assign pad 16 + mmsize * 5 - gprsize - (stack_offset & 15)
SPLATB_REG m2, hev_thr_reg, m7 ; hev_thresh
; align stack
mov stack_reg, rsp ; backup stack pointer
and rsp, ~(mmsize-1) ; align stack
%ifidn %1, v
sub rsp, mmsize * 4 ; stack layout: [0]=E, [1]=I, [2]=hev_thr
; [3]=hev() result
%else ; h
sub rsp, mmsize * 5 ; extra storage space for transposes
%endif %endif
; splat function arguments
%define flim_E [rsp] SPLATB_REG m0, flimEq, m7 ; E
%define flim_I [rsp+mmsize] SPLATB_REG m1, flimIq, m7 ; I
%define hev_thr [rsp+mmsize*2] SPLATB_REG m2, hevthrq, m7 ; hev_thresh
%define mask_res [rsp+mmsize*3]
%define p0backup [rsp+mmsize*3] SUB rsp, pad
%define q0backup [rsp+mmsize*4]
%define m_flimE [rsp]
mova flim_E, m0 %define m_flimI [rsp+mmsize]
mova flim_I, m1 %define m_hevthr [rsp+mmsize*2]
mova hev_thr, m2 %define m_maskres [rsp+mmsize*3]
%define m_p0backup [rsp+mmsize*3]
%else ; sse2 on x86-64 %define m_q0backup [rsp+mmsize*4]
%define flim_E m9 mova m_flimE, m0
%define flim_I m10 mova m_flimI, m1
%define hev_thr m11 mova m_hevthr, m2
%define mask_res m12 %else
%define p0backup m12 %define m_flimE m9
%define q0backup m8 %define m_flimI m10
%define m_hevthr m11
%define m_maskres m12
%define m_p0backup m12
%define m_q0backup m8
; splat function arguments ; splat function arguments
SPLATB_REG flim_E, E_reg, m7 ; E SPLATB_REG m_flimE, flimEq, m7 ; E
SPLATB_REG flim_I, I_reg, m7 ; I SPLATB_REG m_flimI, flimIq, m7 ; I
SPLATB_REG hev_thr, hev_thr_reg, m7 ; hev_thresh SPLATB_REG m_hevthr, hevthrq, m7 ; hev_thresh
%endif %endif
%if mmsize == 8 && %3 == 16 ; mmx/mmxext %if %2 == 8 ; chroma
mov cnt_reg, 2 DEFINE_ARGS dst1, dst8, mstride, stride, dst2
%elif mmsize == 8
DEFINE_ARGS dst1, mstride, stride, dst2, cntr
mov cntrq, 2
%else
DEFINE_ARGS dst1, mstride, stride, dst2, dst8
%endif %endif
mov stride_reg, mstride_reg mov strideq, mstrideq
neg mstride_reg neg mstrideq
%ifidn %1, h %ifidn %1, h
lea dst_reg, [dst_reg + stride_reg*4-4] lea dst1q, [dst1q+strideq*4-4]
%if %3 == 8 %if %2 == 8 ; chroma
lea dst8_reg, [dst8_reg+ stride_reg*4-4] lea dst8q, [dst8q+strideq*4-4]
%endif %endif
%endif %endif
%if mmsize == 8 %if mmsize == 8
.next8px .next8px:
%endif %endif
; read ; read
lea dst2_reg, [dst_reg + stride_reg] lea dst2q, [dst1q+strideq]
%ifidn %1, v %ifidn %1, v
%if %3 == 8 && mmsize == 16 %if %2 == 8 && mmsize == 16
%define movrow movh %define movrow movh
%else %else
%define movrow mova %define movrow mova
%endif %endif
movrow m0, [dst_reg +mstride_reg*4] ; p3 movrow m0, [dst1q+mstrideq*4] ; p3
movrow m1, [dst2_reg+mstride_reg*4] ; p2 movrow m1, [dst2q+mstrideq*4] ; p2
movrow m2, [dst_reg +mstride_reg*2] ; p1 movrow m2, [dst1q+mstrideq*2] ; p1
movrow m5, [dst2_reg] ; q1 movrow m5, [dst2q] ; q1
movrow m6, [dst2_reg+ stride_reg] ; q2 movrow m6, [dst2q+ strideq*1] ; q2
movrow m7, [dst2_reg+ stride_reg*2] ; q3 movrow m7, [dst2q+ strideq*2] ; q3
%if mmsize == 16 && %3 == 8 %if mmsize == 16 && %2 == 8
movhps m0, [dst8_reg+mstride_reg*4] movhps m0, [dst8q+mstrideq*4]
movhps m2, [dst8_reg+mstride_reg*2] movhps m2, [dst8q+mstrideq*2]
add dst8_reg, stride_reg add dst8q, strideq
movhps m1, [dst8_reg+mstride_reg*4] movhps m1, [dst8q+mstrideq*4]
movhps m5, [dst8_reg] movhps m5, [dst8q]
movhps m6, [dst8_reg+ stride_reg] movhps m6, [dst8q+ strideq ]
movhps m7, [dst8_reg+ stride_reg*2] movhps m7, [dst8q+ strideq*2]
add dst8_reg, mstride_reg add dst8q, mstrideq
%endif %endif
%elif mmsize == 8 ; mmx/mmxext (h) %elif mmsize == 8 ; mmx/mmxext (h)
; read 8 rows of 8px each ; read 8 rows of 8px each
movu m0, [dst_reg +mstride_reg*4] movu m0, [dst1q+mstrideq*4]
movu m1, [dst2_reg+mstride_reg*4] movu m1, [dst2q+mstrideq*4]
movu m2, [dst_reg +mstride_reg*2] movu m2, [dst1q+mstrideq*2]
movu m3, [dst_reg +mstride_reg] movu m3, [dst1q+mstrideq ]
movu m4, [dst_reg] movu m4, [dst1q]
movu m5, [dst2_reg] movu m5, [dst2q]
movu m6, [dst2_reg+ stride_reg] movu m6, [dst2q+ strideq ]
; 8x8 transpose ; 8x8 transpose
TRANSPOSE4x4B 0, 1, 2, 3, 7 TRANSPOSE4x4B 0, 1, 2, 3, 7
mova q0backup, m1 mova m_q0backup, m1
movu m7, [dst2_reg+ stride_reg*2] movu m7, [dst2q+ strideq*2]
TRANSPOSE4x4B 4, 5, 6, 7, 1 TRANSPOSE4x4B 4, 5, 6, 7, 1
SBUTTERFLY dq, 0, 4, 1 ; p3/p2 SBUTTERFLY dq, 0, 4, 1 ; p3/p2
SBUTTERFLY dq, 2, 6, 1 ; q0/q1 SBUTTERFLY dq, 2, 6, 1 ; q0/q1
SBUTTERFLY dq, 3, 7, 1 ; q2/q3 SBUTTERFLY dq, 3, 7, 1 ; q2/q3
mova m1, q0backup mova m1, m_q0backup
mova q0backup, m2 ; store q0 mova m_q0backup, m2 ; store q0
SBUTTERFLY dq, 1, 5, 2 ; p1/p0 SBUTTERFLY dq, 1, 5, 2 ; p1/p0
mova p0backup, m5 ; store p0 mova m_p0backup, m5 ; store p0
SWAP 1, 4 SWAP 1, 4
SWAP 2, 4 SWAP 2, 4
SWAP 6, 3 SWAP 6, 3
SWAP 5, 3 SWAP 5, 3
%else ; sse2 (h) %else ; sse2 (h)
%if %3 == 16 %if %2 == 16
lea dst8_reg, [dst_reg + stride_reg*8] lea dst8q, [dst1q+ strideq*8]
%endif %endif
; read 16 rows of 8px each, interleave ; read 16 rows of 8px each, interleave
movh m0, [dst_reg +mstride_reg*4] movh m0, [dst1q+mstrideq*4]
movh m1, [dst8_reg+mstride_reg*4] movh m1, [dst8q+mstrideq*4]
movh m2, [dst_reg +mstride_reg*2] movh m2, [dst1q+mstrideq*2]
movh m5, [dst8_reg+mstride_reg*2] movh m5, [dst8q+mstrideq*2]
movh m3, [dst_reg +mstride_reg] movh m3, [dst1q+mstrideq ]
movh m6, [dst8_reg+mstride_reg] movh m6, [dst8q+mstrideq ]
movh m4, [dst_reg] movh m4, [dst1q]
movh m7, [dst8_reg] movh m7, [dst8q]
punpcklbw m0, m1 ; A/I punpcklbw m0, m1 ; A/I
punpcklbw m2, m5 ; C/K punpcklbw m2, m5 ; C/K
punpcklbw m3, m6 ; D/L punpcklbw m3, m6 ; D/L
punpcklbw m4, m7 ; E/M punpcklbw m4, m7 ; E/M
add dst8_reg, stride_reg add dst8q, strideq
movh m1, [dst2_reg+mstride_reg*4] movh m1, [dst2q+mstrideq*4]
movh m6, [dst8_reg+mstride_reg*4] movh m6, [dst8q+mstrideq*4]
movh m5, [dst2_reg] movh m5, [dst2q]
movh m7, [dst8_reg] movh m7, [dst8q]
punpcklbw m1, m6 ; B/J punpcklbw m1, m6 ; B/J
punpcklbw m5, m7 ; F/N punpcklbw m5, m7 ; F/N
movh m6, [dst2_reg+ stride_reg] movh m6, [dst2q+ strideq ]
movh m7, [dst8_reg+ stride_reg] movh m7, [dst8q+ strideq ]
punpcklbw m6, m7 ; G/O punpcklbw m6, m7 ; G/O
; 8x16 transpose ; 8x16 transpose
...@@ -1829,10 +1806,10 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1829,10 +1806,10 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
%ifdef m8 %ifdef m8
SWAP 1, 8 SWAP 1, 8
%else %else
mova q0backup, m1 mova m_q0backup, m1
%endif %endif
movh m7, [dst2_reg+ stride_reg*2] movh m7, [dst2q+ strideq*2]
movh m1, [dst8_reg+ stride_reg*2] movh m1, [dst8q+ strideq*2]
punpcklbw m7, m1 ; H/P punpcklbw m7, m1 ; H/P
TRANSPOSE4x4B 4, 5, 6, 7, 1 TRANSPOSE4x4B 4, 5, 6, 7, 1
SBUTTERFLY dq, 0, 4, 1 ; p3/p2 SBUTTERFLY dq, 0, 4, 1 ; p3/p2
...@@ -1842,14 +1819,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1842,14 +1819,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
SWAP 1, 8 SWAP 1, 8
SWAP 2, 8 SWAP 2, 8
%else %else
mova m1, q0backup mova m1, m_q0backup
mova q0backup, m2 ; store q0 mova m_q0backup, m2 ; store q0
%endif %endif
SBUTTERFLY dq, 1, 5, 2 ; p1/p0 SBUTTERFLY dq, 1, 5, 2 ; p1/p0
%ifdef m12 %ifdef m12
SWAP 5, 12 SWAP 5, 12
%else %else
mova p0backup, m5 ; store p0 mova m_p0backup, m5 ; store p0
%endif %endif
SWAP 1, 4 SWAP 1, 4
SWAP 2, 4 SWAP 2, 4
...@@ -1883,7 +1860,7 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1883,7 +1860,7 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
por m6, m4 ; abs(q2-q1) por m6, m4 ; abs(q2-q1)
%if notcpuflag(mmx2) %if notcpuflag(mmx2)
mova m4, flim_I mova m4, m_flimI
pxor m3, m3 pxor m3, m3
psubusb m0, m4 psubusb m0, m4
psubusb m1, m4 psubusb m1, m4
...@@ -1905,14 +1882,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1905,14 +1882,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
; normal_limit and high_edge_variance for p1-p0, q1-q0 ; normal_limit and high_edge_variance for p1-p0, q1-q0
SWAP 7, 3 ; now m7 is zero SWAP 7, 3 ; now m7 is zero
%ifidn %1, v %ifidn %1, v
movrow m3, [dst_reg +mstride_reg] ; p0 movrow m3, [dst1q+mstrideq ] ; p0
%if mmsize == 16 && %3 == 8 %if mmsize == 16 && %2 == 8
movhps m3, [dst8_reg+mstride_reg] movhps m3, [dst8q+mstrideq ]
%endif %endif
%elifdef m12 %elifdef m12
SWAP 3, 12 SWAP 3, 12
%else %else
mova m3, p0backup mova m3, m_p0backup
%endif %endif
mova m1, m2 mova m1, m2
...@@ -1925,11 +1902,11 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1925,11 +1902,11 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
%if notcpuflag(mmx2) %if notcpuflag(mmx2)
mova m6, m1 mova m6, m1
psubusb m1, m4 psubusb m1, m4
psubusb m6, hev_thr psubusb m6, m_hevthr
pcmpeqb m1, m7 ; abs(p1-p0) <= I pcmpeqb m1, m7 ; abs(p1-p0) <= I
pcmpeqb m6, m7 ; abs(p1-p0) <= hev_thresh pcmpeqb m6, m7 ; abs(p1-p0) <= hev_thresh
pand m0, m1 pand m0, m1
mova mask_res, m6 mova m_maskres, m6
%else ; mmxext/sse2 %else ; mmxext/sse2
pmaxub m0, m1 ; max_I pmaxub m0, m1 ; max_I
SWAP 1, 4 ; max_hev_thresh SWAP 1, 4 ; max_hev_thresh
...@@ -1937,14 +1914,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1937,14 +1914,14 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
SWAP 6, 4 ; now m6 is I SWAP 6, 4 ; now m6 is I
%ifidn %1, v %ifidn %1, v
movrow m4, [dst_reg] ; q0 movrow m4, [dst1q] ; q0
%if mmsize == 16 && %3 == 8 %if mmsize == 16 && %2 == 8
movhps m4, [dst8_reg] movhps m4, [dst8q]
%endif %endif
%elifdef m8 %elifdef m8
SWAP 4, 8 SWAP 4, 8
%else %else
mova m4, q0backup mova m4, m_q0backup
%endif %endif
mova m1, m4 mova m1, m4
SWAP 1, 4 SWAP 1, 4
...@@ -1956,26 +1933,26 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1956,26 +1933,26 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
%if notcpuflag(mmx2) %if notcpuflag(mmx2)
mova m7, m1 mova m7, m1
psubusb m1, m6 psubusb m1, m6
psubusb m7, hev_thr psubusb m7, m_hevthr
pxor m6, m6 pxor m6, m6
pcmpeqb m1, m6 ; abs(q1-q0) <= I pcmpeqb m1, m6 ; abs(q1-q0) <= I
pcmpeqb m7, m6 ; abs(q1-q0) <= hev_thresh pcmpeqb m7, m6 ; abs(q1-q0) <= hev_thresh
mova m6, mask_res mova m6, m_maskres
pand m0, m1 ; abs([pq][321]-[pq][210]) <= I pand m0, m1 ; abs([pq][321]-[pq][210]) <= I
pand m6, m7 pand m6, m7
%else ; mmxext/sse2 %else ; mmxext/sse2
pxor m7, m7 pxor m7, m7
pmaxub m0, m1 pmaxub m0, m1
pmaxub m6, m1 pmaxub m6, m1
psubusb m0, flim_I psubusb m0, m_flimI
psubusb m6, hev_thr psubusb m6, m_hevthr
pcmpeqb m0, m7 ; max(abs(..)) <= I pcmpeqb m0, m7 ; max(abs(..)) <= I
pcmpeqb m6, m7 ; !(max(abs..) > thresh) pcmpeqb m6, m7 ; !(max(abs..) > thresh)
%endif %endif
%ifdef m12 %ifdef m12
SWAP 6, 12 SWAP 6, 12
%else %else
mova mask_res, m6 ; !(abs(p1-p0) > hev_t || abs(q1-q0) > hev_t) mova m_maskres, m6 ; !(abs(p1-p0) > hev_t || abs(q1-q0) > hev_t)
%endif %endif
; simple_limit ; simple_limit
...@@ -1999,28 +1976,28 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -1999,28 +1976,28 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
pand m7, [pb_FE] pand m7, [pb_FE]
psrlq m7, 1 ; abs(q1-p1)/2 psrlq m7, 1 ; abs(q1-p1)/2
paddusb m7, m1 ; abs(q0-p0)*2+abs(q1-p1)/2 paddusb m7, m1 ; abs(q0-p0)*2+abs(q1-p1)/2
psubusb m7, flim_E psubusb m7, m_flimE
pcmpeqb m7, m6 ; abs(q0-p0)*2+abs(q1-p1)/2 <= E pcmpeqb m7, m6 ; abs(q0-p0)*2+abs(q1-p1)/2 <= E
pand m0, m7 ; normal_limit result pand m0, m7 ; normal_limit result
; filter_common; at this point, m2-m5=p1-q1 and m0 is filter_mask ; filter_common; at this point, m2-m5=p1-q1 and m0 is filter_mask
%ifdef m8 ; x86-64 && sse2 %ifdef m8 ; x86-64 && sse2
mova m8, [pb_80] mova m8, [pb_80]
%define pb_80_var m8 %define m_pb_80 m8
%else ; x86-32 or mmx/mmxext %else ; x86-32 or mmx/mmxext
%define pb_80_var [pb_80] %define m_pb_80 [pb_80]
%endif %endif
mova m1, m4 mova m1, m4
mova m7, m3 mova m7, m3
pxor m1, pb_80_var pxor m1, m_pb_80
pxor m7, pb_80_var pxor m7, m_pb_80
psubsb m1, m7 ; (signed) q0-p0 psubsb m1, m7 ; (signed) q0-p0
mova m6, m2 mova m6, m2
mova m7, m5 mova m7, m5
pxor m6, pb_80_var pxor m6, m_pb_80
pxor m7, pb_80_var pxor m7, m_pb_80
psubsb m6, m7 ; (signed) p1-q1 psubsb m6, m7 ; (signed) p1-q1
mova m7, mask_res mova m7, m_maskres
pandn m7, m6 pandn m7, m6
paddsb m7, m1 paddsb m7, m1
paddsb m7, m1 paddsb m7, m1
...@@ -2059,7 +2036,7 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -2059,7 +2036,7 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
%ifdef m12 %ifdef m12
SWAP 6, 12 SWAP 6, 12
%else %else
mova m6, mask_res mova m6, m_maskres
%endif %endif
%if notcpuflag(mmx2) %if notcpuflag(mmx2)
mova m7, [pb_1] mova m7, [pb_1]
...@@ -2087,81 +2064,81 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13 ...@@ -2087,81 +2064,81 @@ cglobal vp8_%1_loop_filter16y_inner, 5, %2, 13
; store ; store
%ifidn %1, v %ifidn %1, v
movrow [dst_reg +mstride_reg*2], m2 movrow [dst1q+mstrideq*2], m2
movrow [dst_reg +mstride_reg ], m3 movrow [dst1q+mstrideq ], m3
movrow [dst_reg], m4 movrow [dst1q], m4
movrow [dst_reg + stride_reg ], m5 movrow [dst1q+ strideq ], m5
%if mmsize == 16 && %3 == 8 %if mmsize == 16 && %2 == 8
movhps [dst8_reg+mstride_reg*2], m2 movhps [dst8q+mstrideq*2], m2
movhps [dst8_reg+mstride_reg ], m3 movhps [dst8q+mstrideq ], m3
movhps [dst8_reg], m4 movhps [dst8q], m4
movhps [dst8_reg+ stride_reg ], m5 movhps [dst8q+ strideq ], m5
%endif %endif
%else ; h %else ; h
add dst_reg, 2 add dst1q, 2
add dst2_reg, 2 add dst2q, 2
; 4x8/16 transpose ; 4x8/16 transpose
TRANSPOSE4x4B 2, 3, 4, 5, 6 TRANSPOSE4x4B 2, 3, 4, 5, 6
%if mmsize == 8 ; mmx/mmxext (h) %if mmsize == 8 ; mmx/mmxext (h)
WRITE_4x2D 2, 3, 4, 5, dst_reg, dst2_reg, mstride_reg, stride_reg WRITE_4x2D 2, 3, 4, 5, dst1q, dst2q, mstrideq, strideq
%else ; sse2 (h) %else ; sse2 (h)
lea dst8_reg, [dst8_reg+mstride_reg+2] lea dst8q, [dst8q+mstrideq +2]
WRITE_4x4D 2, 3, 4, 5, dst_reg, dst2_reg, dst8_reg, mstride_reg, stride_reg, %3 WRITE_4x4D 2, 3, 4, 5, dst1q, dst2q, dst8q, mstrideq, strideq, %2
%endif %endif
%endif %endif
%if mmsize == 8 %if mmsize == 8
%if %3 == 8 ; chroma %if %2 == 8 ; chroma
%ifidn %1, h %ifidn %1, h
sub dst_reg, 2 sub dst1q, 2
%endif %endif
cmp dst_reg, dst8_reg cmp dst1q, dst8q
mov dst_reg, dst8_reg mov dst1q, dst8q
jnz .next8px jnz .next8px
%else %else
%ifidn %1, h %ifidn %1, h
lea dst_reg, [dst_reg + stride_reg*8-2] lea dst1q, [dst1q+ strideq*8-2]
%else ; v %else ; v
add dst_reg, 8 add dst1q, 8
%endif %endif
dec cnt_reg dec cntrq
jg .next8px jg .next8px
%endif %endif
%endif %endif
%ifndef m8 ; sse2 on x86-32 or mmx/mmxext %ifndef m8 ; sse2 on x86-32 or mmx/mmxext
mov rsp, stack_reg ; restore stack pointer ADD rsp, pad
%endif %endif
RET RET
%endmacro %endmacro
%if ARCH_X86_32 %if ARCH_X86_32
INIT_MMX mmx INIT_MMX mmx
INNER_LOOPFILTER v, 6, 16 INNER_LOOPFILTER v, 16
INNER_LOOPFILTER h, 6, 16 INNER_LOOPFILTER h, 16
INNER_LOOPFILTER v, 6, 8 INNER_LOOPFILTER v, 8
INNER_LOOPFILTER h, 6, 8 INNER_LOOPFILTER h, 8
INIT_MMX mmx2 INIT_MMX mmx2
INNER_LOOPFILTER v, 6, 16 INNER_LOOPFILTER v, 16
INNER_LOOPFILTER h, 6, 16 INNER_LOOPFILTER h, 16
INNER_LOOPFILTER v, 6, 8 INNER_LOOPFILTER v, 8
INNER_LOOPFILTER h, 6, 8 INNER_LOOPFILTER h, 8
%endif %endif
INIT_XMM sse2 INIT_XMM sse2
INNER_LOOPFILTER v, 5, 16 INNER_LOOPFILTER v, 16
INNER_LOOPFILTER h, 5 + ARCH_X86_32, 16 INNER_LOOPFILTER h, 16
INNER_LOOPFILTER v, 6, 8 INNER_LOOPFILTER v, 8
INNER_LOOPFILTER h, 6, 8 INNER_LOOPFILTER h, 8
INIT_XMM ssse3 INIT_XMM ssse3
INNER_LOOPFILTER v, 5, 16 INNER_LOOPFILTER v, 16
INNER_LOOPFILTER h, 5 + ARCH_X86_32, 16 INNER_LOOPFILTER h, 16
INNER_LOOPFILTER v, 6, 8 INNER_LOOPFILTER v, 8
INNER_LOOPFILTER h, 6, 8 INNER_LOOPFILTER h, 8
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
; void vp8_h/v_loop_filter<size>_mbedge_<opt>(uint8_t *dst, [uint8_t *v,] int stride, ; void vp8_h/v_loop_filter<size>_mbedge_<opt>(uint8_t *dst, [uint8_t *v,] int stride,
......
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