sigrid/ffmpeg

libavcodec/x86/Makefile+1−0

@@ -43,6 +43,7 @@ YASM-OBJS-$(CONFIG_DIRAC_DECODER) += x86/diracdsp_mmx.o x86/diracdsp_yasm.o · 42 unmodified lines
43	YASM-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_yasm.o	43	YASM-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_yasm.o
44	YASM-OBJS-FFT-$(HAVE_AMD3DNOW) += x86/fft_3dn.o	44	YASM-OBJS-FFT-$(HAVE_AMD3DNOW) += x86/fft_3dn.o
45	YASM-OBJS-FFT-$(HAVE_AMD3DNOWEXT) += x86/fft_3dn2.o	45	YASM-OBJS-FFT-$(HAVE_AMD3DNOWEXT) += x86/fft_3dn2.o
		46	YASM-OBJS-FFT-$(HAVE_SSE) += x86/fft_sse.o
46	YASM-OBJS-$(CONFIG_FFT) += x86/fft_mmx.o \	47	YASM-OBJS-$(CONFIG_FFT) += x86/fft_mmx.o \
47	$(YASM-OBJS-FFT-yes)	48	$(YASM-OBJS-FFT-yes)
48		49

libavcodec/x86/fft_mmx.asm+10−129

@@ -47,10 +47,6 @@ struc FFTContext · 46 unmodified lines
47	.mdctbits: resd 1	47	.mdctbits: resd 1
48	.tcos: pointer 1	48	.tcos: pointer 1
49	.tsin: pointer 1	49	.tsin: pointer 1
50	.fftperm: pointer 1
51	.fftcalc: pointer 1
52	.imdctcalc:pointer 1
53	.imdcthalf:pointer 1
54	endstruc	50	endstruc
55		51
56	%define M_SQRT1_2 0.70710678118654752440	52	%define M_SQRT1_2 0.70710678118654752440
@@ -69,7 +65,6 @@ perm1: dd 0x00, 0x02, 0x03, 0x01, 0x03, 0x00, 0x02, 0x01 · 12 unmodified lines
69	perm2: dd 0x00, 0x01, 0x02, 0x03, 0x01, 0x00, 0x02, 0x03	65	perm2: dd 0x00, 0x01, 0x02, 0x03, 0x01, 0x00, 0x02, 0x03
70	ps_p1p1m1p1root2: dd 1.0, 1.0, -1.0, 1.0, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2	66	ps_p1p1m1p1root2: dd 1.0, 1.0, -1.0, 1.0, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2
71	ps_m1m1p1m1p1m1m1m1: dd 1<<31, 1<<31, 0, 1<<31, 0, 1<<31, 1<<31, 1<<31	67	ps_m1m1p1m1p1m1m1m1: dd 1<<31, 1<<31, 0, 1<<31, 0, 1<<31, 1<<31, 1<<31
72	ps_m1m1m1m1: times 4 dd 1<<31
73	ps_m1p1: dd 1<<31, 0	68	ps_m1p1: dd 1<<31, 0
74		69
75	%assign i 16	70	%assign i 16
@@ -538,16 +533,6 @@ DEFINE_ARGS z, w, n, o1, o3 · 462 unmodified lines
538	rep ret	533	rep ret
539	%endmacro	534	%endmacro
540		535
541	%macro FFT_DISPATCH 2; clobbers 5 GPRs, 8 XMMs
542	lea r2, [dispatch_tab%1]
543	mov r2, [r2 + (%2q-2)*gprsize]
544	%ifdef PIC
545	lea r3, [$$]
546	add r2, r3
547	%endif
548	call r2
549	%endmacro ; FFT_DISPATCH
550
551	INIT_YMM avx	536	INIT_YMM avx
552		537
553	%if HAVE_AVX	538	%if HAVE_AVX
@@ -564,14 +549,6 @@ INIT_YMM avx · 10 unmodified lines
564		549
565	DECL_PASS pass_avx, PASS_BIG 1	550	DECL_PASS pass_avx, PASS_BIG 1
566	DECL_PASS pass_interleave_avx, PASS_BIG 0	551	DECL_PASS pass_interleave_avx, PASS_BIG 0
567
568	cglobal fft_calc, 2,5,8
569	mov r3d, [r0 + FFTContext.nbits]
570	mov r0, r1
571	mov r1, r3
572	FFT_DISPATCH _interleave %+ SUFFIX, r1
573	REP_RET
574
575	%endif	552	%endif
576		553
577	INIT_XMM sse	554	INIT_XMM sse
@@ -589,112 +566,6 @@ INIT_XMM sse · 11 unmodified lines
589	DECL_PASS pass_sse, PASS_BIG 1	566	DECL_PASS pass_sse, PASS_BIG 1
590	DECL_PASS pass_interleave_sse, PASS_BIG 0	567	DECL_PASS pass_interleave_sse, PASS_BIG 0
591		568
592	cglobal fft_calc, 2,5,8
593	mov r3d, [r0 + FFTContext.nbits]
594	PUSH r1
595	PUSH r3
596	mov r0, r1
597	mov r1, r3
598	FFT_DISPATCH _interleave %+ SUFFIX, r1
599	POP rcx
600	POP r4
601	cmp rcx, 4
602	jg .end
603	mov r2, -1
604	add rcx, 3
605	shl r2, cl
606	sub r4, r2
607	.loop
608	movaps xmm0, [r4 + r2]
609	movaps xmm1, xmm0
610	unpcklps xmm0, [r4 + r2 + 16]
611	unpckhps xmm1, [r4 + r2 + 16]
612	movaps [r4 + r2], xmm0
613	movaps [r4 + r2 + 16], xmm1
614	add r2, 32
615	jl .loop
616	.end:
617	REP_RET
618
619	cextern_naked memcpy
620
621	cglobal fft_permute, 2,7,1
622	mov r4, [r0 + FFTContext.revtab]
623	mov r5, [r0 + FFTContext.tmpbuf]
624	mov ecx, [r0 + FFTContext.nbits]
625	mov r2, 1
626	shl r2, cl
627	xor r0, r0
628	%if ARCH_X86_32
629	mov r1, r1m
630	%endif
631	.loop:
632	movaps xmm0, [r1 + 8*r0]
633	movzx r6, word [r4 + 2*r0]
634	movzx r3, word [r4 + 2*r0 + 2]
635	movlps [r5 + 8*r6], xmm0
636	movhps [r5 + 8*r3], xmm0
637	add r0, 2
638	cmp r0, r2
639	jl .loop
640	shl r2, 3
641	%if ARCH_X86_64
642	mov r0, r1
643	mov r1, r5
644	%else
645	push r2
646	push r5
647	push r1
648	%endif
649	%if ARCH_X86_64 && WIN64 == 0
650	jmp memcpy
651	%else
652	call memcpy
653	%if ARCH_X86_32
654	add esp, 12
655	%endif
656	REP_RET
657	%endif
658
659	cglobal imdct_calc, 3,5,3
660	mov r3d, [r0 + FFTContext.mdctsize]
661	mov r4, [r0 + FFTContext.imdcthalf]
662	add r1, r3
663	PUSH r3
664	PUSH r1
665	%if ARCH_X86_32
666	push r2
667	push r1
668	push r0
669	%else
670	sub rsp, 8
671	%endif
672	call r4
673	%if ARCH_X86_32
674	add esp, 12
675	%else
676	add rsp, 8
677	%endif
678	POP r1
679	POP r3
680	lea r0, [r1 + 2*r3]
681	mov r2, r3
682	sub r3, 16
683	neg r2
684	movaps xmm2, [ps_m1m1m1m1]
685	.loop:
686	movaps xmm0, [r1 + r3]
687	movaps xmm1, [r0 + r2]
688	shufps xmm0, xmm0, 0x1b
689	shufps xmm1, xmm1, 0x1b
690	xorps xmm0, xmm2
691	movaps [r0 + r3], xmm1
692	movaps [r1 + r2], xmm0
693	sub r3, 16
694	add r2, 16
695	jl .loop
696	REP_RET
697
698	INIT_MMX 3dnow	569	INIT_MMX 3dnow
699	%define mulps pfmul	570	%define mulps pfmul
700	%define addps pfadd	571	%define addps pfadd
@@ -712,6 +583,16 @@ DECL_PASS pass_interleave_3dnow, PASS_BIG 0 · 11 unmodified lines
712	%define SECTION_REL	583	%define SECTION_REL
713	%endif	584	%endif
714		585
		586	%macro FFT_DISPATCH 2; clobbers 5 GPRs, 8 XMMs
		587	lea r2, [dispatch_tab%1]
		588	mov r2, [r2 + (%2q-2)*gprsize]
		589	%ifdef PIC
		590	lea r3, [$$]
		591	add r2, r3
		592	%endif
		593	call r2
		594	%endmacro ; FFT_DISPATCH
		595
715	%macro DECL_FFT 1-2 ; nbits, suffix	596	%macro DECL_FFT 1-2 ; nbits, suffix
716	%ifidn %0, 1	597	%ifidn %0, 1
717	%xdefine fullsuffix SUFFIX	598	%xdefine fullsuffix SUFFIX

libavcodec/x86/fft_sse.c+110−0

@@ -0,0 +1,110 @@
	1	/*
	2	* FFT/MDCT transform with SSE optimizations
	3	* Copyright (c) 2008 Loren Merritt
	4	*
	5	* This file is part of Libav.
	6	*
	7	* Libav is free software; you can redistribute it and/or
	8	* modify it under the terms of the GNU Lesser General Public
	9	* License as published by the Free Software Foundation; either
	10	* version 2.1 of the License, or (at your option) any later version.
	11	*
	12	* Libav is distributed in the hope that it will be useful,
	13	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	15	* Lesser General Public License for more details.
	16	*
	17	* You should have received a copy of the GNU Lesser General Public
	18	* License along with Libav; if not, write to the Free Software
	19	* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	20	*/
	21
	22	#include "libavutil/x86_cpu.h"
	23	#include "libavcodec/dsputil.h"
	24	#include "fft.h"
	25	#include "config.h"
	26
	27	DECLARE_ASM_CONST(16, unsigned int, ff_m1m1m1m1)[4] =
	28	{ 1U << 31, 1U << 31, 1U << 31, 1U << 31 };
	29
	30	void ff_fft_dispatch_sse(FFTComplex *z, int nbits);
	31	void ff_fft_dispatch_interleave_sse(FFTComplex *z, int nbits);
	32	void ff_fft_dispatch_interleave_avx(FFTComplex *z, int nbits);
	33
	34	#if HAVE_AVX
	35	void ff_fft_calc_avx(FFTContext s, FFTComplex z)
	36	{
	37	ff_fft_dispatch_interleave_avx(z, s->nbits);
	38	}
	39	#endif
	40
	41	void ff_fft_calc_sse(FFTContext s, FFTComplex z)
	42	{
	43	int n = 1 << s->nbits;
	44
	45	ff_fft_dispatch_interleave_sse(z, s->nbits);
	46
	47	if(n <= 16) {
	48	x86_reg i = -8*n;
	49	__asm__ volatile(
	50	"1: \n"
	51	"movaps (%0,%1), %%xmm0 \n"
	52	"movaps %%xmm0, %%xmm1 \n"
	53	"unpcklps 16(%0,%1), %%xmm0 \n"
	54	"unpckhps 16(%0,%1), %%xmm1 \n"
	55	"movaps %%xmm0, (%0,%1) \n"
	56	"movaps %%xmm1, 16(%0,%1) \n"
	57	"add $32, %0 \n"
	58	"jl 1b \n"
	59	:"+r"(i)
	60	:"r"(z+n)
	61	:"memory"
	62	);
	63	}
	64	}
	65
	66	void ff_fft_permute_sse(FFTContext s, FFTComplex z)
	67	{
	68	int n = 1 << s->nbits;
	69	int i;
	70	for(i=0; i<n; i+=2) {
	71	__asm__ volatile(
	72	"movaps %2, %%xmm0 \n"
	73	"movlps %%xmm0, %0 \n"
	74	"movhps %%xmm0, %1 \n"
	75	:"=m"(s->tmp_buf[s->revtab[i]]),
	76	"=m"(s->tmp_buf[s->revtab[i+1]])
	77	:"m"(z[i])
	78	);
	79	}
	80	memcpy(z, s->tmp_buf, n*sizeof(FFTComplex));
	81	}
	82
	83	void ff_imdct_calc_sse(FFTContext s, FFTSample output, const FFTSample *input)
	84	{
	85	x86_reg j, k;
	86	long n = s->mdct_size;
	87	long n4 = n >> 2;
	88
	89	s->imdct_half(s, output + n4, input);
	90
	91	j = -n;
	92	k = n-16;
	93	__asm__ volatile(
	94	"movaps "MANGLE(ff_m1m1m1m1)", %%xmm7 \n"
	95	"1: \n"
	96	"movaps (%2,%1), %%xmm0 \n"
	97	"movaps (%3,%0), %%xmm1 \n"
	98	"shufps $0x1b, %%xmm0, %%xmm0 \n"
	99	"shufps $0x1b, %%xmm1, %%xmm1 \n"
	100	"xorps %%xmm7, %%xmm0 \n"
	101	"movaps %%xmm1, (%3,%1) \n"
	102	"movaps %%xmm0, (%2,%0) \n"
	103	"sub $16, %1 \n"
	104	"add $16, %0 \n"
	105	"jl 1b \n"
	106	:"+r"(j), "+r"(k)
	107	:"r"(output+n4), "r"(output+n4*3)
	108	XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm7")
	109	);
	110	}

Revert "x86: fft: convert sse inline asm to yasm"

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code