vorbisdsp: convert x86 simd functions from inline asm to yasm.

Ronald S. Bultje committed 13 years ago

Commit 2e4bb991 parent 0ee82935 files · +92 −74Browse files

Unified · Split

5 files

libavcodec/x86/Makefile +1 −0
libavcodec/x86/dsputil_mmx.c +0 −3
libavcodec/x86/dsputil_mmx.h +0 −2
libavcodec/x86/vorbisdsp.asm +83 −0
libavcodec/x86/vorbisdsp_init.c +8 −69

@@ -62,6 +62,7 @@ YASM-OBJS-$(CONFIG_RV40_DECODER) += x86/rv34dsp.o \ · 61 unmodified lines
62	62	x86/rv40dsp.o
63	63	YASM-OBJS-$(CONFIG_VC1_DECODER) += x86/vc1dsp.o
64	64	YASM-OBJS-$(CONFIG_VIDEODSP) += x86/videodsp.o
	65	+YASM-OBJS-$(CONFIG_VORBIS_DECODER) += x86/vorbisdsp.o
65	66	YASM-OBJS-$(CONFIG_VP3DSP) += x86/vp3dsp.o
66	67	YASM-OBJS-$(CONFIG_VP6_DECODER) += x86/vp56dsp.o
67	68	YASM-OBJS-$(CONFIG_VP8_DECODER) += x86/vp8dsp.o

libavcodec/x86/dsputil_mmx.c+0−3

@@ -38,9 +38,6 @@ · 37 unmodified lines
38	38	DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL;
39	39	DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
40	40
41		−DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
42		− { 0x8000000080000000ULL, 0x8000000080000000ULL };
43		−
44	41	DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1) = { 0x0001000100010001ULL, 0x0001000100010001ULL };
45	42	DECLARE_ALIGNED(16, const xmm_reg, ff_pw_2) = { 0x0002000200020002ULL, 0x0002000200020002ULL };
46	43	DECLARE_ALIGNED(16, const xmm_reg, ff_pw_3) = { 0x0003000300030003ULL, 0x0003000300030003ULL };

libavcodec/x86/dsputil_mmx.h+0−2

@@ -31,8 +31,6 @@ typedef struct xmm_reg { uint64_t a, b; } xmm_reg; · 30 unmodified lines
31	31	extern const uint64_t ff_bone;
32	32	extern const uint64_t ff_wtwo;
33	33
34		−extern const uint64_t ff_pdw_80000000[2];
35		−
36	34	extern const xmm_reg ff_pw_3;
37	35	extern const xmm_reg ff_pw_4;
38	36	extern const xmm_reg ff_pw_5;

libavcodec/x86/vorbisdsp.asm+83−0

@@ -0,0 +1,83 @@
	1	+;******************************************************************************
	2	+;* Vorbis x86 optimizations
	3	+;* Copyright (C) 2006 Loren Merritt <lorenm@u.washington.edu>
	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/x86util.asm"
	23	+
	24	+SECTION_RODATA
	25	+
	26	+pdw_80000000: times 4 dd 0x80000000
	27	+
	28	+SECTION .text
	29	+
	30	+%if ARCH_X86_32
	31	+INIT_MMX 3dnow
	32	+cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
	33	+ pxor m7, m7
	34	+ lea magq, [magq+block_sizeq*4]
	35	+ lea angq, [angq+block_sizeq*4]
	36	+ neg block_sizeq
	37	+.loop:
	38	+ mova m0, [magq+block_sizeq*4]
	39	+ mova m1, [angq+block_sizeq*4]
	40	+ mova m2, m0
	41	+ mova m3, m1
	42	+ pfcmpge m2, m7 ; m <= 0.0
	43	+ pfcmpge m3, m7 ; a <= 0.0
	44	+ pslld m2, 31 ; keep only the sign bit
	45	+ pxor m1, m2
	46	+ mova m4, m3
	47	+ pand m3, m1
	48	+ pandn m4, m1
	49	+ pfadd m3, m0 ; a = m + ((a < 0) & (a ^ sign(m)))
	50	+ pfsub m0, m4 ; m = m + ((a > 0) & (a ^ sign(m)))
	51	+ mova [angq+block_sizeq*4], m3
	52	+ mova [magq+block_sizeq*4], m0
	53	+ add block_sizeq, 2
	54	+ jl .loop
	55	+ femms
	56	+ RET
	57	+%endif
	58	+
	59	+INIT_XMM sse
	60	+cglobal vorbis_inverse_coupling, 3, 4, 6, mag, ang, block_size, cntr
	61	+ mova m5, [pdw_80000000]
	62	+ xor cntrq, cntrq
	63	+align 16
	64	+.loop:
	65	+ mova m0, [magq+cntrq*4]
	66	+ mova m1, [angq+cntrq*4]
	67	+ xorps m2, m2
	68	+ xorps m3, m3
	69	+ cmpleps m2, m0 ; m <= 0.0
	70	+ cmpleps m3, m1 ; a <= 0.0
	71	+ andps m2, m5 ; keep only the sign bit
	72	+ xorps m1, m2
	73	+ mova m4, m3
	74	+ andps m3, m1
	75	+ andnps m4, m1
	76	+ addps m3, m0 ; a = m + ((a < 0) & (a ^ sign(m)))
	77	+ subps m0, m4 ; m = m + ((a > 0) & (a ^ sign(m)))
	78	+ mova [angq+cntrq*4], m3
	79	+ mova [magq+cntrq*4], m0
	80	+ add cntrq, 4
	81	+ cmp cntrq, block_sizeq
	82	+ jl .loop
	83	+ RET

libavcodec/x86/vorbisdsp_init.c+8−69

@@ -21,83 +21,22 @@ · 20 unmodified lines
21	21	#include "config.h"
22	22	#include "libavutil/cpu.h"
23	23	#include "libavcodec/vorbisdsp.h"
24		−#include "dsputil_mmx.h" // for ff_pdw_80000000
25	24
26		−#if HAVE_INLINE_ASM
27		−#if ARCH_X86_32
28		−static void vorbis_inverse_coupling_3dnow(float mag, float ang,
29		− intptr_t blocksize)
30		−{
31		− int i;
32		− __asm__ volatile ("pxor %%mm7, %%mm7":);
33		− for (i = 0; i < blocksize; i += 2) {
34		− __asm__ volatile (
35		− "movq %0, %%mm0 \n\t"
36		− "movq %1, %%mm1 \n\t"
37		− "movq %%mm0, %%mm2 \n\t"
38		− "movq %%mm1, %%mm3 \n\t"
39		− "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
40		− "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
41		− "pslld $31, %%mm2 \n\t" // keep only the sign bit
42		− "pxor %%mm2, %%mm1 \n\t"
43		− "movq %%mm3, %%mm4 \n\t"
44		− "pand %%mm1, %%mm3 \n\t"
45		− "pandn %%mm1, %%mm4 \n\t"
46		− "pfadd %%mm0, %%mm3 \n\t" // a = m + ((a < 0) & (a ^ sign(m)))
47		− "pfsub %%mm4, %%mm0 \n\t" // m = m + ((a > 0) & (a ^ sign(m)))
48		− "movq %%mm3, %1 \n\t"
49		− "movq %%mm0, %0 \n\t"
50		− : "+m"(mag[i]), "+m"(ang[i])
51		− :: "memory"
52		− );
53		− }
54		− __asm__ volatile ("femms");
55		−}
56		−#endif
57		−
58		−static void vorbis_inverse_coupling_sse(float mag, float ang,
59		− intptr_t blocksize)
60		−{
61		− int i;
62		−
63		− __asm__ volatile (
64		− "movaps %0, %%xmm5 \n\t"
65		− :: "m"(ff_pdw_80000000[0])
66		− );
67		− for (i = 0; i < blocksize; i += 4) {
68		− __asm__ volatile (
69		− "movaps %0, %%xmm0 \n\t"
70		− "movaps %1, %%xmm1 \n\t"
71		− "xorps %%xmm2, %%xmm2 \n\t"
72		− "xorps %%xmm3, %%xmm3 \n\t"
73		− "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
74		− "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
75		− "andps %%xmm5, %%xmm2 \n\t" // keep only the sign bit
76		− "xorps %%xmm2, %%xmm1 \n\t"
77		− "movaps %%xmm3, %%xmm4 \n\t"
78		− "andps %%xmm1, %%xmm3 \n\t"
79		− "andnps %%xmm1, %%xmm4 \n\t"
80		− "addps %%xmm0, %%xmm3 \n\t" // a = m + ((a < 0) & (a ^ sign(m)))
81		− "subps %%xmm4, %%xmm0 \n\t" // m = m + ((a > 0) & (a ^ sign(m)))
82		− "movaps %%xmm3, %1 \n\t"
83		− "movaps %%xmm0, %0 \n\t"
84		− : "+m"(mag[i]), "+m"(ang[i])
85		− :: "memory"
86		− );
87		− }
88		−}
89		−#endif
	25	+void ff_vorbis_inverse_coupling_3dnow(float mag, float ang,
	26	+ intptr_t blocksize);
	27	+void ff_vorbis_inverse_coupling_sse(float mag, float ang,
	28	+ intptr_t blocksize);
90	29
91	30	void ff_vorbisdsp_init_x86(VorbisDSPContext *dsp)
92	31	{
93		−#if HAVE_INLINE_ASM
	32	+#if HAVE_YASM
94	33	int mm_flags = av_get_cpu_flags();
95	34
96	35	#if ARCH_X86_32
97	36	if (mm_flags & AV_CPU_FLAG_3DNOW)
98		− dsp->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
	37	+ dsp->vorbis_inverse_coupling = ff_vorbis_inverse_coupling_3dnow;
99	38	#endif /* ARCH_X86_32 */
100	39	if (mm_flags & AV_CPU_FLAG_SSE)
101		− dsp->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
102		−#endif /* HAVE_INLINE_ASM */
	40	+ dsp->vorbis_inverse_coupling = ff_vorbis_inverse_coupling_sse;
	41	+#endif /* HAVE_YASM */
103	42	}

vorbisdsp: convert x86 simd functions from inline asm to yasm.

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code