avfilter/bwdif: add avx2 filter_line function

James Darnley committed 3 years ago

Commit 073ec3b1 parent a9377232 files · +36 −5Browse files

Unified · Split

2 files

libavfilter/x86/vf_bwdif.asm +24 −5
libavfilter/x86/vf_bwdif_init.c +12 −0

@@ -26,18 +26,22 @@ · 25 unmodified lines
26	26
27	27	%include "libavutil/x86/x86util.asm"
28	28
29		−SECTION_RODATA
	29	+SECTION_RODATA 32
30	30
31		−pw_coefhf: times 4 dw 1016, 5570
32		−pw_coefhf1: times 8 dw -3801
33		−pw_coefsp: times 4 dw 5077, -981
34		−pw_splfdif: times 4 dw -768, 768
	31	+pw_coefhf: times 8 dw 1016, 5570
	32	+pw_coefhf1: times 16 dw -3801
	33	+pw_coefsp: times 8 dw 5077, -981
	34	+pw_splfdif: times 8 dw -768, 768
35	35
36	36	SECTION .text
37	37
38	38	%macro LOAD8 2
	39	+ %if mmsize == 32
	40	+ pmovzxbw %1, %2
	41	+ %else
39	42	movh %1, %2
40	43	punpcklbw %1, m7
	44	+ %endif
41	45	%endmacro
42	46
43	47	%macro LOAD12 2
@@ -45,8 +49,14 @@ SECTION .text · 1 unmodified line
45	49	%endmacro
46	50
47	51	%macro DISP8 0
	52	+ %if mmsize == 32
	53	+ vextracti128 xm1, m2, 1
	54	+ packuswb xm2, xm1
	55	+ movu [dstq], xm2
	56	+ %else
48	57	packuswb m2, m2
49	58	movh [dstq], m2
	59	+ %endif
50	60	%endmacro
51	61
52	62	%macro DISP12 0
@@ -244,8 +254,12 @@ cglobal bwdif_filter_line_12bit, 4, 9, 13, 0, dst, prev, cur, next, w, \ · 191 unmodified lines
244	254	prefs, mrefs, prefs2, mrefs2, \
245	255	prefs3, mrefs3, prefs4, \
246	256	mrefs4, parity, clip_max
	257	+ %if mmsize == 32
	258	+ vpbroadcastw m12, WORD clip_maxm
	259	+ %else
247	260	movd m12, DWORD clip_maxm
248	261	SPLATW m12, m12, 0
	262	+ %endif
249	263	%else
250	264	cglobal bwdif_filter_line_12bit, 4, 6, 8, 80, dst, prev, cur, next, w, \
251	265	prefs, mrefs, prefs2, mrefs2, \
@@ -264,3 +278,8 @@ INIT_XMM ssse3 · 12 unmodified lines
264	278	BWDIF
265	279	INIT_XMM sse2
266	280	BWDIF
	281	+
	282	+%if HAVE_AVX2_EXTERNAL && ARCH_X86_64
	283	+INIT_YMM avx2
	284	+BWDIF
	285	+%endif

libavfilter/x86/vf_bwdif_init.c+12−0

@@ -32,6 +32,10 @@ void ff_bwdif_filter_line_ssse3(void dst, void prev, void cur, void next, · 31 unmodified lines
32	32	int w, int prefs, int mrefs, int prefs2,
33	33	int mrefs2, int prefs3, int mrefs3, int prefs4,
34	34	int mrefs4, int parity, int clip_max);
	35	+void ff_bwdif_filter_line_avx2(void dst, void prev, void cur, void next,
	36	+ int w, int prefs, int mrefs, int prefs2,
	37	+ int mrefs2, int prefs3, int mrefs3, int prefs4,
	38	+ int mrefs4, int parity, int clip_max);
35	39
36	40	void ff_bwdif_filter_line_12bit_sse2(void dst, void prev, void cur, void next,
37	41	int w, int prefs, int mrefs, int prefs2,
@@ -41,6 +45,10 @@ void ff_bwdif_filter_line_12bit_ssse3(void dst, void prev, void cur, void ne · 3 unmodified lines
41	45	int w, int prefs, int mrefs, int prefs2,
42	46	int mrefs2, int prefs3, int mrefs3, int prefs4,
43	47	int mrefs4, int parity, int clip_max);
	48	+void ff_bwdif_filter_line_12bit_avx2(void dst, void prev, void cur, void next,
	49	+ int w, int prefs, int mrefs, int prefs2,
	50	+ int mrefs2, int prefs3, int mrefs3, int prefs4,
	51	+ int mrefs4, int parity, int clip_max);
44	52
45	53	av_cold void ff_bwdif_init_x86(BWDIFContext *bwdif, int bit_depth)
46	54	{
@@ -51,10 +59,14 @@ av_cold void ff_bwdif_init_x86(BWDIFContext *bwdif, int bit_depth) · 4 unmodified lines
51	59	bwdif->filter_line = ff_bwdif_filter_line_sse2;
52	60	if (EXTERNAL_SSSE3(cpu_flags))
53	61	bwdif->filter_line = ff_bwdif_filter_line_ssse3;
	62	+ if (ARCH_X86_64 && EXTERNAL_AVX2(cpu_flags))
	63	+ bwdif->filter_line = ff_bwdif_filter_line_avx2;
54	64	} else if (bit_depth <= 12) {
55	65	if (EXTERNAL_SSE2(cpu_flags))
56	66	bwdif->filter_line = ff_bwdif_filter_line_12bit_sse2;
57	67	if (EXTERNAL_SSSE3(cpu_flags))
58	68	bwdif->filter_line = ff_bwdif_filter_line_12bit_ssse3;
	69	+ if (ARCH_X86_64 && EXTERNAL_AVX2(cpu_flags))
	70	+ bwdif->filter_line = ff_bwdif_filter_line_12bit_avx2;
59	71	}
60	72	}

avfilter/bwdif: add avx2 filter_line function

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code