Skip to content
Snippets Groups Projects
Commit f2e3d706 authored by Clément Bœsch's avatar Clément Bœsch Committed by Anton Khirnov
Browse files

vp9lpf/x86: add ff_vp9_loop_filter_h_{48,84}_16_{sse2,ssse3,avx}(). 


Signed-off-by: default avatarAnton Khirnov <anton@khirnov.net>
parent 92d47550
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
libavcodec/x86/vp9dsp_init.c
+ 22
− 20
View file @ f2e3d706
...@@ -226,6 +226,12 @@ void ff_vp9_loop_filter_h_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stri ...@@ -226,6 +226,12 @@ void ff_vp9_loop_filter_h_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stri
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);
lpf_funcs(84, 16, sse2);
lpf_funcs(84, 16, ssse3);
lpf_funcs(84, 16, avx);
lpf_funcs(48, 16, sse2);
lpf_funcs(48, 16, ssse3);
lpf_funcs(48, 16, avx);
lpf_funcs(88, 16, sse2); lpf_funcs(88, 16, sse2);
lpf_funcs(88, 16, ssse3); lpf_funcs(88, 16, ssse3);
lpf_funcs(88, 16, avx); lpf_funcs(88, 16, avx);
...@@ -269,6 +275,19 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -269,6 +275,19 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
init_subpel3_8to64(idx, type, opt); \ init_subpel3_8to64(idx, type, opt); \
init_subpel2(4, idx, 4, type, opt) init_subpel2(4, idx, 4, type, opt)
#define init_lpf(opt) do { \
if (ARCH_X86_64) { \
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_##opt; \
dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_##opt; \
dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_##opt; \
dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_##opt; \
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_##opt; \
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_##opt; \
} \
} while (0)
if (EXTERNAL_MMX(cpu_flags)) { if (EXTERNAL_MMX(cpu_flags)) {
init_fpel(4, 0, 4, put, mmx); init_fpel(4, 0, 4, put, mmx);
init_fpel(3, 0, 8, put, mmx); init_fpel(3, 0, 8, put, mmx);
...@@ -293,36 +312,19 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -293,36 +312,19 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
init_fpel(2, 1, 16, avg, sse2); init_fpel(2, 1, 16, avg, sse2);
init_fpel(1, 1, 32, avg, sse2); init_fpel(1, 1, 32, avg, sse2);
init_fpel(0, 1, 64, avg, sse2); init_fpel(0, 1, 64, avg, sse2);
if (ARCH_X86_64) { init_lpf(sse2);
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_sse2;
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_sse2;
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_sse2;
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_sse2;
}
} }
if (EXTERNAL_SSSE3(cpu_flags)) { if (EXTERNAL_SSSE3(cpu_flags)) {
init_subpel3(0, put, ssse3); init_subpel3(0, put, ssse3);
init_subpel3(1, avg, ssse3); init_subpel3(1, avg, ssse3);
init_lpf(ssse3);
if (ARCH_X86_64) {
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_ssse3;
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_ssse3;
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_ssse3;
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_ssse3;
}
} }
if (EXTERNAL_AVX(cpu_flags)) { if (EXTERNAL_AVX(cpu_flags)) {
init_fpel(1, 0, 32, put, avx); init_fpel(1, 0, 32, put, avx);
init_fpel(0, 0, 64, put, avx); init_fpel(0, 0, 64, put, avx);
init_lpf(avx);
if (ARCH_X86_64) {
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_avx;
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_avx;
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_avx;
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_avx;
}
} }
if (EXTERNAL_AVX2(cpu_flags)) { if (EXTERNAL_AVX2(cpu_flags)) {
......
libavcodec/x86/vp9lpf.asm
+ 31
− 28
View file @ f2e3d706
...@@ -45,6 +45,11 @@ pw_8: times 8 dw 8 ...@@ -45,6 +45,11 @@ pw_8: times 8 dw 8
mask_mix: times 8 db 0 mask_mix: times 8 db 0
times 8 db 1 times 8 db 1
mask_mix84: times 8 db 0xff
times 8 db 0x00
mask_mix48: times 8 db 0x00
times 8 db 0xff
SECTION .text SECTION .text
; %1 = abs(%2-%3) ; %1 = abs(%2-%3)
...@@ -312,7 +317,7 @@ SECTION .text ...@@ -312,7 +317,7 @@ SECTION .text
neg mstride3q neg mstride3q
%ifidn %1, h %ifidn %1, h
%if %2 == 88 %if %2 > 16
%define movx movh %define movx movh
lea dstq, [dstq + 8*strideq - 4] lea dstq, [dstq + 8*strideq - 4]
%else %else
...@@ -360,7 +365,7 @@ SECTION .text ...@@ -360,7 +365,7 @@ SECTION .text
%define Q6 rsp + 224 %define Q6 rsp + 224
%define Q7 rsp + 240 %define Q7 rsp + 240
%if %2 != 88 %if %2 == 16
TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp] TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp]
mova [P7], m0 mova [P7], m0
mova [P6], m1 mova [P6], m1
...@@ -377,7 +382,7 @@ SECTION .text ...@@ -377,7 +382,7 @@ SECTION .text
mova [Q1], m9 mova [Q1], m9
mova [Q2], m10 mova [Q2], m10
mova [Q3], m11 mova [Q3], m11
%if %2 != 88 %if %2 == 16
mova [Q4], m12 mova [Q4], m12
mova [Q5], m13 mova [Q5], m13
mova [Q6], m14 mova [Q6], m14
...@@ -392,7 +397,7 @@ SECTION .text ...@@ -392,7 +397,7 @@ SECTION .text
%endif %endif
SPLATB_REG m2, I, m0 ; I I I I ... SPLATB_REG m2, I, m0 ; I I I I ...
SPLATB_REG m3, E, m0 ; E E E E ... SPLATB_REG m3, E, m0 ; E E E E ...
%elif %2 == 88 %else
%if cpuflag(ssse3) %if cpuflag(ssse3)
mova m0, [mask_mix] mova m0, [mask_mix]
%endif %endif
...@@ -452,7 +457,7 @@ SECTION .text ...@@ -452,7 +457,7 @@ SECTION .text
ABSSUB_CMP m1, m9, m11, m6, m4, m5, m8 ; abs(p2 - p0) <= 1 ABSSUB_CMP m1, m9, m11, m6, m4, m5, m8 ; abs(p2 - p0) <= 1
pand m2, m1 pand m2, m1
ABSSUB m4, m10, m11, m5 ; abs(p1 - p0) ABSSUB m4, m10, m11, m5 ; abs(p1 - p0)
%if %2 != 88 %if %2 == 16
%if cpuflag(ssse3) %if cpuflag(ssse3)
pxor m0, m0 pxor m0, m0
%endif %endif
...@@ -476,8 +481,11 @@ SECTION .text ...@@ -476,8 +481,11 @@ SECTION .text
pand m2, m1 pand m2, m1
ABSSUB_CMP m1, m15, m12, m6, m4, m5, m8 ; abs(q3 - q0) <= 1 ABSSUB_CMP m1, m15, m12, m6, m4, m5, m8 ; abs(q3 - q0) <= 1
pand m2, m1 ; flat8in final value pand m2, m1 ; flat8in final value
%if %2 == 84 || %2 == 48
pand m2, [mask_mix%2]
%endif
%if %2 != 88 %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
mova m8, [P7] mova m8, [P7]
...@@ -570,7 +578,7 @@ SECTION .text ...@@ -570,7 +578,7 @@ SECTION .text
; ([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()
pxor m0, m0 pxor m0, m0
%if %2 == 88 %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)
...@@ -608,7 +616,7 @@ SECTION .text ...@@ -608,7 +616,7 @@ SECTION .text
; q5 +5 -p2 -q4 +q5 +q7 . q5 . . ; q5 +5 -p2 -q4 +q5 +q7 . q5 . .
; q6 +6 -p1 -q5 +q6 +q7 . q6 . . ; q6 +6 -p1 -q5 +q6 +q7 . q6 . .
%if %2 != 88 %if %2 == 16
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]
...@@ -631,7 +639,7 @@ SECTION .text ...@@ -631,7 +639,7 @@ SECTION .text
%endif %endif
%ifidn %1, h %ifidn %1, h
%if %2 != 88 %if %2 == 16
mova m0, [P7] mova m0, [P7]
mova m1, [P6] mova m1, [P6]
mova m2, [P5] mova m2, [P5]
...@@ -720,28 +728,23 @@ SECTION .text ...@@ -720,28 +728,23 @@ SECTION .text
RET RET
%endmacro %endmacro
%macro LPF_16_16_VH 1 %macro LPF_16_VH 2
INIT_XMM %1 INIT_XMM %2
cglobal vp9_loop_filter_v_16_16, 5,10,16, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3 cglobal vp9_loop_filter_v_%1_16, 5,10,16, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3
LOOPFILTER v, 16 LOOPFILTER v, %1
cglobal vp9_loop_filter_h_16_16, 5,10,16, 256, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3 cglobal vp9_loop_filter_h_%1_16, 5,10,16, 256, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3
LOOPFILTER h, 16 LOOPFILTER h, %1
%endmacro %endmacro
%macro LPF_88_16_VH 1 %macro LPF_16_VH_ALL_OPTS 1
INIT_XMM %1 LPF_16_VH %1, sse2
cglobal vp9_loop_filter_v_88_16, 5,10,16, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3 LPF_16_VH %1, ssse3
LOOPFILTER v, 88 LPF_16_VH %1, avx
cglobal vp9_loop_filter_h_88_16, 5,10,16, 256, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3
LOOPFILTER h, 88
%endmacro %endmacro
LPF_16_16_VH sse2 LPF_16_VH_ALL_OPTS 16
LPF_16_16_VH ssse3 LPF_16_VH_ALL_OPTS 48
LPF_16_16_VH avx LPF_16_VH_ALL_OPTS 84
LPF_16_VH_ALL_OPTS 88
LPF_88_16_VH sse2
LPF_88_16_VH ssse3
LPF_88_16_VH avx
%endif ; x86-64 %endif ; x86-64
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment