Skip to content

F
ffmpeg.wasm-core
Commit 7ca422bb authored by Ronald S. Bultje's avatar Ronald S. Bultje
Browse files
Options

vp9: add mxext versions of the single-block (w=4,npx=8) h/v loopfilters. 

Each takes about 0.5% of runtime in my profiles, and they didn't have
any SIMD yet so far (we only had simd for npx=16 double-block versions).
parent 726501a3
Hide whitespace changes
Inline Side-by-side
Showing with 103 additions and 49 deletions
libavcodec/x86/vp9dsp_init.c
View file @ 7ca422bb
...@@ -126,6 +126,7 @@ void ff_vp9_loop_filter_v_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stri ...@@ -126,6 +126,7 @@ void ff_vp9_loop_filter_v_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stri
void ff_vp9_loop_filter_h_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stride, \ void ff_vp9_loop_filter_h_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stride, \
int E, int I, int H) int E, int I, int H)
lpf_funcs(4, 8, mmxext);
lpf_funcs(16, 16, sse2); lpf_funcs(16, 16, sse2);
lpf_funcs(16, 16, ssse3); lpf_funcs(16, 16, ssse3);
lpf_funcs(16, 16, avx); lpf_funcs(16, 16, avx);
...@@ -281,6 +282,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact) ...@@ -281,6 +282,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
} }
if (EXTERNAL_MMXEXT(cpu_flags)) { if (EXTERNAL_MMXEXT(cpu_flags)) {
dsp->loop_filter_8[0][0] = ff_vp9_loop_filter_h_4_8_mmxext;
dsp->loop_filter_8[0][1] = ff_vp9_loop_filter_v_4_8_mmxext;
init_subpel2(4, 0, 4, put, 8, mmxext); init_subpel2(4, 0, 4, put, 8, mmxext);
init_subpel2(4, 1, 4, avg, 8, mmxext); init_subpel2(4, 1, 4, avg, 8, mmxext);
init_fpel_func(4, 1, 4, avg, _8, mmxext); init_fpel_func(4, 1, 4, avg, _8, mmxext);
......
libavcodec/x86/vp9lpf.asm
View file @ 7ca422bb
...@@ -52,7 +52,7 @@ mask_mix48: times 8 db 0x00 ...@@ -52,7 +52,7 @@ mask_mix48: times 8 db 0x00
SECTION .text SECTION .text
%macro SCRATCH 3 %macro SCRATCH 3
%if ARCH_X86_64 %ifdef m8
SWAP %1, %2 SWAP %1, %2
%else %else
mova [%3], m%1 mova [%3], m%1
...@@ -60,7 +60,7 @@ SECTION .text ...@@ -60,7 +60,7 @@ SECTION .text
%endmacro %endmacro
%macro UNSCRATCH 3 %macro UNSCRATCH 3
%if ARCH_X86_64 %ifdef m8
SWAP %1, %2 SWAP %1, %2
%else %else
mova m%1, [%3] mova m%1, [%3]
...@@ -69,7 +69,7 @@ SECTION .text ...@@ -69,7 +69,7 @@ SECTION .text
; %1 = abs(%2-%3) ; %1 = abs(%2-%3)
%macro ABSSUB 4 ; dst, src1 (RO), src2 (RO), tmp %macro ABSSUB 4 ; dst, src1 (RO), src2 (RO), tmp
%if ARCH_X86_64 %ifdef m8
psubusb %1, %3, %2 psubusb %1, %3, %2
psubusb %4, %2, %3 psubusb %4, %2, %3
%else %else
...@@ -102,7 +102,7 @@ SECTION .text ...@@ -102,7 +102,7 @@ SECTION .text
%endmacro %endmacro
%macro UNPACK 4 %macro UNPACK 4
%if ARCH_X86_64 %ifdef m8
punpck%1bw %2, %3, %4 punpck%1bw %2, %3, %4
%else %else
mova %2, %3 mova %2, %3
...@@ -334,22 +334,24 @@ SECTION .text ...@@ -334,22 +334,24 @@ SECTION .text
%endmacro %endmacro
%macro DEFINE_TRANSPOSED_P7_TO_Q7 0-1 0 %macro DEFINE_TRANSPOSED_P7_TO_Q7 0-1 0
%define P3 rsp + 0 + %1 %define P3 rsp + 0*mmsize + %1
%define P2 rsp + 16 + %1 %define P2 rsp + 1*mmsize + %1
%define P1 rsp + 32 + %1 %define P1 rsp + 2*mmsize + %1
%define P0 rsp + 48 + %1 %define P0 rsp + 3*mmsize + %1
%define Q0 rsp + 64 + %1 %define Q0 rsp + 4*mmsize + %1
%define Q1 rsp + 80 + %1 %define Q1 rsp + 5*mmsize + %1
%define Q2 rsp + 96 + %1 %define Q2 rsp + 6*mmsize + %1
%define Q3 rsp + 112 + %1 %define Q3 rsp + 7*mmsize + %1
%define P7 rsp + 128 + %1 %if mmsize == 16
%define P6 rsp + 144 + %1 %define P7 rsp + 8*mmsize + %1
%define P5 rsp + 160 + %1 %define P6 rsp + 9*mmsize + %1
%define P4 rsp + 176 + %1 %define P5 rsp + 10*mmsize + %1
%define Q4 rsp + 192 + %1 %define P4 rsp + 11*mmsize + %1
%define Q5 rsp + 208 + %1 %define Q4 rsp + 12*mmsize + %1
%define Q6 rsp + 224 + %1 %define Q5 rsp + 13*mmsize + %1
%define Q7 rsp + 240 + %1 %define Q6 rsp + 14*mmsize + %1
%define Q7 rsp + 15*mmsize + %1
%endif
%endmacro %endmacro
; ..............AB -> AAAAAAAABBBBBBBB ; ..............AB -> AAAAAAAABBBBBBBB
...@@ -365,12 +367,12 @@ SECTION .text ...@@ -365,12 +367,12 @@ SECTION .text
%macro LOOPFILTER 5 ; %1=v/h %2=size1 %3+%4=stack, %5=32bit stack only %macro LOOPFILTER 5 ; %1=v/h %2=size1 %3+%4=stack, %5=32bit stack only
%if UNIX64 %if UNIX64
cglobal vp9_loop_filter_%1_%2_16, 5, 9, 16, %3 + %4, dst, stride, E, I, H, mstride, dst2, stride3, mstride3 cglobal vp9_loop_filter_%1_%2_ %+ mmsize, 5, 9, 16, %3 + %4, dst, stride, E, I, H, mstride, dst2, stride3, mstride3
%else %else
%if WIN64 %if WIN64
cglobal vp9_loop_filter_%1_%2_16, 4, 8, 16, %3 + %4, dst, stride, E, I, mstride, dst2, stride3, mstride3 cglobal vp9_loop_filter_%1_%2_ %+ mmsize, 4, 8, 16, %3 + %4, dst, stride, E, I, mstride, dst2, stride3, mstride3
%else %else
cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, dst2, stride3, mstride3 cglobal vp9_loop_filter_%1_%2_ %+ mmsize, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, dst2, stride3, mstride3
%define Ed dword r2m %define Ed dword r2m
%define Id dword r3m %define Id dword r3m
%endif %endif
...@@ -384,18 +386,22 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -384,18 +386,22 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
lea mstride3q, [mstrideq*3] lea mstride3q, [mstrideq*3]
%ifidn %1, h %ifidn %1, h
%if %2 > 16 %if %2 != 16
%if mmsize == 16
%define movx movh %define movx movh
%else
%define movx mova
%endif
lea dstq, [dstq + 4*strideq - 4] lea dstq, [dstq + 4*strideq - 4]
%else %else
%define movx movu %define movx movu
lea dstq, [dstq + 4*strideq - 8] ; go from top center (h pos) to center left (v pos) lea dstq, [dstq + 4*strideq - 8] ; go from top center (h pos) to center left (v pos)
%endif %endif
lea dst2q, [dstq + 8*strideq]
%else %else
lea dstq, [dstq + 4*mstrideq] lea dstq, [dstq + 4*mstrideq]
lea dst2q, [dstq + 8*strideq]
%endif %endif
; FIXME we shouldn't need two dts registers if mmsize == 8
lea dst2q, [dstq + 8*strideq]
DEFINE_REAL_P7_TO_Q7 DEFINE_REAL_P7_TO_Q7
...@@ -406,11 +412,11 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -406,11 +412,11 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
movx m3, [P4] movx m3, [P4]
movx m4, [P3] movx m4, [P3]
movx m5, [P2] movx m5, [P2]
%if ARCH_X86_64 || %2 != 16 %if (ARCH_X86_64 && mmsize == 16) || %2 > 16
movx m6, [P1] movx m6, [P1]
%endif %endif
movx m7, [P0] movx m7, [P0]
%if ARCH_X86_64 %ifdef m8
movx m8, [Q0] movx m8, [Q0]
movx m9, [Q1] movx m9, [Q1]
movx m10, [Q2] movx m10, [Q2]
...@@ -502,7 +508,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -502,7 +508,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
movhps [Q5], m6 movhps [Q5], m6
movhps [Q7], m7 movhps [Q7], m7
DEFINE_TRANSPOSED_P7_TO_Q7 DEFINE_TRANSPOSED_P7_TO_Q7
%else ; %2 == 44/48/84/88 %elif %2 > 16 ; %2 == 44/48/84/88
punpcklbw m0, m1 punpcklbw m0, m1
punpcklbw m2, m3 punpcklbw m2, m3
punpcklbw m4, m5 punpcklbw m4, m5
...@@ -529,12 +535,31 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -529,12 +535,31 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
mova [Q1], m5 mova [Q1], m5
mova [Q2], m7 mova [Q2], m7
mova [Q3], m3 mova [Q3], m3
%else ; %2 == 4 || %2 == 8
SBUTTERFLY bw, 0, 1, 6
SBUTTERFLY bw, 2, 3, 6
SBUTTERFLY bw, 4, 5, 6
mova [rsp+4*mmsize], m5
mova m6, [P1]
SBUTTERFLY bw, 6, 7, 5
DEFINE_TRANSPOSED_P7_TO_Q7
TRANSPOSE4x4W 0, 2, 4, 6, 5
mova [P3], m0
mova [P2], m2
mova [P1], m4
mova [P0], m6
mova m5, [rsp+4*mmsize]
TRANSPOSE4x4W 1, 3, 5, 7, 0
mova [Q0], m1
mova [Q1], m3
mova [Q2], m5
mova [Q3], m7
%endif ; %2 %endif ; %2
%endif ; x86-32/64 %endif ; x86-32/64
%endif ; %1 == h %endif ; %1 == h
; calc fm mask ; calc fm mask
%if %2 == 16 %if %2 == 16 || mmsize == 8
%if cpuflag(ssse3) %if cpuflag(ssse3)
pxor m0, m0 pxor m0, m0
%endif %endif
...@@ -552,7 +577,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -552,7 +577,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
mova m0, [pb_80] mova m0, [pb_80]
pxor m2, m0 pxor m2, m0
pxor m3, m0 pxor m3, m0
%if ARCH_X86_64 %ifdef m8
%ifidn %1, v %ifidn %1, v
mova m8, [P3] mova m8, [P3]
mova m9, [P2] mova m9, [P2]
...@@ -613,10 +638,10 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -613,10 +638,10 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
; (m3: fm, m8..15: p3 p2 p1 p0 q0 q1 q2 q3) ; (m3: fm, m8..15: p3 p2 p1 p0 q0 q1 q2 q3)
; calc flat8in (if not 44_16) and hev masks ; calc flat8in (if not 44_16) and hev masks
%if %2 != 44 %if %2 != 44 && %2 != 4
mova m6, [pb_81] ; [1 1 1 1 ...] ^ 0x80 mova m6, [pb_81] ; [1 1 1 1 ...] ^ 0x80
ABSSUB_GT m2, rp3, rp0, m6, m5 ; abs(p3 - p0) <= 1 ABSSUB_GT m2, rp3, rp0, m6, m5 ; abs(p3 - p0) <= 1
%if ARCH_X86_64 %ifdef m8
mova m8, [pb_80] mova m8, [pb_80]
%define rb80 m8 %define rb80 m8
%else %else
...@@ -655,8 +680,15 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -655,8 +680,15 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
%endif %endif
%else %else
mova m6, [pb_80] mova m6, [pb_80]
%if %2 == 44
movd m7, Hd movd m7, Hd
SPLATB_MIX m7 SPLATB_MIX m7
%else
%if cpuflag(ssse3)
pxor m0, m0
%endif
SPLATB_REG m7, H, m0 ; H H H H ...
%endif
pxor m7, m6 pxor m7, m6
ABSSUB m4, rp1, rp0, m1 ; abs(p1 - p0) ABSSUB m4, rp1, rp0, m1 ; abs(p1 - p0)
pxor m4, m6 pxor m4, m6
...@@ -670,7 +702,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -670,7 +702,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
%if %2 == 16 %if %2 == 16
; (m0: hev, m2: flat8in, m3: fm, m6: pb_81, m9..15: p2 p1 p0 q0 q1 q2 q3) ; (m0: hev, m2: flat8in, m3: fm, m6: pb_81, m9..15: p2 p1 p0 q0 q1 q2 q3)
; calc flat8out mask ; calc flat8out mask
%if ARCH_X86_64 %ifdef m8
mova m8, [P7] mova m8, [P7]
mova m9, [P6] mova m9, [P6]
%define rp7 m8 %define rp7 m8
...@@ -682,7 +714,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -682,7 +714,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
ABSSUB_GT m1, rp7, rp0, m6, m5 ; abs(p7 - p0) <= 1 ABSSUB_GT m1, rp7, rp0, m6, m5 ; abs(p7 - p0) <= 1
ABSSUB_GT m7, rp6, rp0, m6, m5 ; abs(p6 - p0) <= 1 ABSSUB_GT m7, rp6, rp0, m6, m5 ; abs(p6 - p0) <= 1
por m1, m7 por m1, m7
%if ARCH_X86_64 %ifdef m8
mova m8, [P5] mova m8, [P5]
mova m9, [P4] mova m9, [P4]
%define rp5 m8 %define rp5 m8
...@@ -695,7 +727,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -695,7 +727,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
por m1, m7 por m1, m7
ABSSUB_GT m7, rp4, rp0, m6, m5 ; abs(p4 - p0) <= 1 ABSSUB_GT m7, rp4, rp0, m6, m5 ; abs(p4 - p0) <= 1
por m1, m7 por m1, m7
%if ARCH_X86_64 %ifdef m8
mova m14, [Q4] mova m14, [Q4]
mova m15, [Q5] mova m15, [Q5]
%define rq4 m14 %define rq4 m14
...@@ -708,7 +740,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -708,7 +740,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
por m1, m7 por m1, m7
ABSSUB_GT m7, rq5, rq0, m6, m5 ; abs(q5 - q0) <= 1 ABSSUB_GT m7, rq5, rq0, m6, m5 ; abs(q5 - q0) <= 1
por m1, m7 por m1, m7
%if ARCH_X86_64 %ifdef m8
mova m14, [Q6] mova m14, [Q6]
mova m15, [Q7] mova m15, [Q7]
%define rq6 m14 %define rq6 m14
...@@ -738,7 +770,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -738,7 +770,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
; (m0: hev, [m1: flat8out], [m2: flat8in], m3: fm, m8..15: p5 p4 p1 p0 q0 q1 q6 q7) ; (m0: hev, [m1: flat8out], [m2: flat8in], m3: fm, m8..15: p5 p4 p1 p0 q0 q1 q6 q7)
; filter2() ; filter2()
%if %2 != 44 %if %2 != 44 && %2 != 4
mova m6, [pb_80] ; already in m6 if 44_16 mova m6, [pb_80] ; already in m6 if 44_16
SCRATCH 2, 15, rsp+%3+%4 SCRATCH 2, 15, rsp+%3+%4
%if %2 == 16 %if %2 == 16
...@@ -756,7 +788,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -756,7 +788,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
paddsb m4, m2 ; 3*(q0 - p0) + (p1 - q1) paddsb m4, m2 ; 3*(q0 - p0) + (p1 - q1)
paddsb m6, m4, [pb_4] ; m6: f1 = clip(f + 4, 127) paddsb m6, m4, [pb_4] ; m6: f1 = clip(f + 4, 127)
paddsb m4, [pb_3] ; m4: f2 = clip(f + 3, 127) paddsb m4, [pb_3] ; m4: f2 = clip(f + 3, 127)
%if ARCH_X86_64 %ifdef m8
mova m14, [pb_10] ; will be reused in filter4() mova m14, [pb_10] ; will be reused in filter4()
%define rb10 m14 %define rb10 m14
%else %else
...@@ -765,8 +797,8 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -765,8 +797,8 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
SRSHIFT3B_2X m6, m4, rb10, m7 ; f1 and f2 sign byte shift by 3 SRSHIFT3B_2X m6, m4, rb10, m7 ; f1 and f2 sign byte shift by 3
SIGN_SUB m7, rq0, m6, m5 ; m7 = q0 - f1 SIGN_SUB m7, rq0, m6, m5 ; m7 = q0 - f1
SIGN_ADD m1, rp0, m4, m5 ; m1 = p0 + f2 SIGN_ADD m1, rp0, m4, m5 ; m1 = p0 + f2
%if %2 != 44 %if %2 != 44 && %2 != 4
%if ARCH_X86_64 %ifdef m8
pandn m6, m15, m3 ; ~mask(in) & mask(fm) pandn m6, m15, m3 ; ~mask(in) & mask(fm)
%else %else
mova m6, [rsp+%3+%4] mova m6, [rsp+%3+%4]
...@@ -787,8 +819,8 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -787,8 +819,8 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
paddsb m6, m2, [pb_4] ; m6: f1 = clip(f + 4, 127) paddsb m6, m2, [pb_4] ; m6: f1 = clip(f + 4, 127)
paddsb m2, [pb_3] ; m2: f2 = clip(f + 3, 127) paddsb m2, [pb_3] ; m2: f2 = clip(f + 3, 127)
SRSHIFT3B_2X m6, m2, rb10, m4 ; f1 and f2 sign byte shift by 3 SRSHIFT3B_2X m6, m2, rb10, m4 ; f1 and f2 sign byte shift by 3
%if %2 != 44 %if %2 != 44 && %2 != 4
%if ARCH_X86_64 %ifdef m8
pandn m5, m15, m3 ; ~mask(in) & mask(fm) pandn m5, m15, m3 ; ~mask(in) & mask(fm)
%else %else
mova m5, [rsp+%3+%4] mova m5, [rsp+%3+%4]
...@@ -815,26 +847,26 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -815,26 +847,26 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
mova [P1], m1 mova [P1], m1
mova [Q1], m4 mova [Q1], m4
%if %2 != 44 %if %2 != 44 && %2 != 4
UNSCRATCH 2, 15, rsp+%3+%4 UNSCRATCH 2, 15, rsp+%3+%4
%endif %endif
; ([m1: flat8out], m2: flat8in, m3: fm, m10..13: p1 p0 q0 q1) ; ([m1: flat8out], m2: flat8in, m3: fm, m10..13: p1 p0 q0 q1)
; filter6() ; filter6()
%if %2 != 44 %if %2 != 44 && %2 != 4
pxor m0, m0 pxor m0, m0
%if %2 > 16 %if %2 > 16
pand m3, m2 pand m3, m2
%else %else
pand m2, m3 ; mask(fm) & mask(in) pand m2, m3 ; mask(fm) & mask(in)
%if ARCH_X86_64 %ifdef m8
pandn m3, m8, m2 ; ~mask(out) & (mask(fm) & mask(in)) pandn m3, m8, m2 ; ~mask(out) & (mask(fm) & mask(in))
%else %else
mova m3, [rsp+%3+%4+16] mova m3, [rsp+%3+%4+16]
pandn m3, m2 pandn m3, m2
%endif %endif
%endif %endif
%if ARCH_X86_64 %ifdef m8
mova m14, [P3] mova m14, [P3]
mova m9, [Q3] mova m9, [Q3]
%define rp3 m14 %define rp3 m14
...@@ -882,7 +914,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -882,7 +914,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
pand m1, m2 ; mask(out) & (mask(fm) & mask(in)) pand m1, m2 ; mask(out) & (mask(fm) & mask(in))
mova m2, [P7] mova m2, [P7]
mova m3, [P6] mova m3, [P6]
%if ARCH_X86_64 %ifdef m8
mova m8, [P5] mova m8, [P5]
mova m9, [P4] mova m9, [P4]
%define rp5 m8 %define rp5 m8
...@@ -1008,7 +1040,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -1008,7 +1040,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
movhps [Q5], m6 movhps [Q5], m6
movhps [Q7], m7 movhps [Q7], m7
%endif %endif
%elif %2 == 44 %elif %2 == 44 || %2 == 4
SWAP 0, 1 ; m0 = p1 SWAP 0, 1 ; m0 = p1
SWAP 1, 7 ; m1 = p0 SWAP 1, 7 ; m1 = p0
SWAP 2, 5 ; m2 = q0 SWAP 2, 5 ; m2 = q0
...@@ -1018,6 +1050,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -1018,6 +1050,7 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
SBUTTERFLY bw, 2, 3, 4 SBUTTERFLY bw, 2, 3, 4
SBUTTERFLY wd, 0, 2, 4 SBUTTERFLY wd, 0, 2, 4
SBUTTERFLY wd, 1, 3, 4 SBUTTERFLY wd, 1, 3, 4
%if mmsize == 16
movd [P7], m0 movd [P7], m0
movd [P3], m2 movd [P3], m2
movd [Q0], m1 movd [Q0], m1
...@@ -1046,6 +1079,20 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, ...@@ -1046,6 +1079,20 @@ cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride,
movd [P0], m2 movd [P0], m2
movd [Q3], m1 movd [Q3], m1
movd [Q7], m3 movd [Q7], m3
%else
movd [P7], m0
movd [P5], m2
movd [P3], m1
movd [P1], m3
psrlq m0, 32
psrlq m2, 32
psrlq m1, 32
psrlq m3, 32
movd [P6], m0
movd [P4], m2
movd [P2], m1
movd [P0], m3
%endif
%else %else
; the following code do a transpose of 8 full lines to 16 half ; the following code do a transpose of 8 full lines to 16 half
; lines (high part). It is inlined to avoid the need of a staging area ; lines (high part). It is inlined to avoid the need of a staging area
...@@ -1137,3 +1184,7 @@ LPF_16_VH_ALL_OPTS 44, 0, 128, 0 ...@@ -1137,3 +1184,7 @@ LPF_16_VH_ALL_OPTS 44, 0, 128, 0
LPF_16_VH_ALL_OPTS 48, 256, 128, 16 LPF_16_VH_ALL_OPTS 48, 256, 128, 16
LPF_16_VH_ALL_OPTS 84, 256, 128, 16 LPF_16_VH_ALL_OPTS 84, 256, 128, 16
LPF_16_VH_ALL_OPTS 88, 256, 128, 16 LPF_16_VH_ALL_OPTS 88, 256, 128, 16
INIT_MMX mmxext
LOOPFILTER v, 4, 0, 0, 0
LOOPFILTER h, 4, 0, 64, 0
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