[FFmpeg-cvslog] fft: port FFT/IMDCT 3dnow functions to yasm, and disable on x86-64.
Ronald S. Bultje
git at videolan.org
Thu Aug 2 23:39:53 CEST 2012
ffmpeg | branch: master | Ronald S. Bultje <rsbultje at gmail.com> | Tue Jul 31 20:23:39 2012 -0700| [9f14cd91b5195e2beba7213e855c0d36b787ab85] | committer: Michael Niedermayer
fft: port FFT/IMDCT 3dnow functions to yasm, and disable on x86-64.
64-bit CPUs always have SSE available, thus there is no need to compile
in the 3dnow functions. This results in smaller binaries.
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=9f14cd91b5195e2beba7213e855c0d36b787ab85
---
libavcodec/x86/Makefile | 2 -
libavcodec/x86/fft_3dn.c | 23 ------
libavcodec/x86/fft_3dn2.c | 173 --------------------------------------
libavcodec/x86/fft_mmx.asm | 196 +++++++++++++++++++++++++++++++++++++-------
4 files changed, 168 insertions(+), 226 deletions(-)
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index e18074a..3e998ef 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -42,8 +42,6 @@ YASM-OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp.o
YASM-OBJS-$(CONFIG_DCT) += x86/dct32_sse.o
YASM-OBJS-$(CONFIG_DIRAC_DECODER) += x86/diracdsp_mmx.o x86/diracdsp_yasm.o
YASM-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_yasm.o
-YASM-OBJS-FFT-$(HAVE_AMD3DNOW) += x86/fft_3dn.o
-YASM-OBJS-FFT-$(HAVE_AMD3DNOWEXT) += x86/fft_3dn2.o
YASM-OBJS-$(CONFIG_FFT) += x86/fft_mmx.o \
$(YASM-OBJS-FFT-yes)
diff --git a/libavcodec/x86/fft_3dn.c b/libavcodec/x86/fft_3dn.c
deleted file mode 100644
index 6f2e2e8..0000000
--- a/libavcodec/x86/fft_3dn.c
+++ /dev/null
@@ -1,23 +0,0 @@
-/*
- * FFT/MDCT transform with 3DNow! optimizations
- * Copyright (c) 2008 Loren Merritt
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * FFmpeg is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#define EMULATE_3DNOWEXT
-#include "fft_3dn2.c"
diff --git a/libavcodec/x86/fft_3dn2.c b/libavcodec/x86/fft_3dn2.c
deleted file mode 100644
index 5e16522..0000000
--- a/libavcodec/x86/fft_3dn2.c
+++ /dev/null
@@ -1,173 +0,0 @@
-/*
- * FFT/MDCT transform with Extended 3DNow! optimizations
- * Copyright (c) 2006-2008 Zuxy MENG Jie, Loren Merritt
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * FFmpeg is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include "libavutil/x86_cpu.h"
-#include "libavcodec/dsputil.h"
-#include "fft.h"
-
-DECLARE_ALIGNED(8, static const unsigned int, m1m1)[2] = { 1U<<31, 1U<<31 };
-
-#ifdef EMULATE_3DNOWEXT
-#define PSWAPD(s,d)\
- "movq "#s","#d"\n"\
- "psrlq $32,"#d"\n"\
- "punpckldq "#s","#d"\n"
-#define ff_fft_calc_3dnow2 ff_fft_calc_3dnow
-#define ff_fft_dispatch_3dnow2 ff_fft_dispatch_3dnow
-#define ff_fft_dispatch_interleave_3dnow2 ff_fft_dispatch_interleave_3dnow
-#define ff_imdct_calc_3dnow2 ff_imdct_calc_3dnow
-#define ff_imdct_half_3dnow2 ff_imdct_half_3dnow
-#else
-#define PSWAPD(s,d) "pswapd "#s","#d"\n"
-#endif
-
-void ff_fft_dispatch_3dnow2(FFTComplex *z, int nbits);
-void ff_fft_dispatch_interleave_3dnow2(FFTComplex *z, int nbits);
-
-void ff_fft_calc_3dnow2(FFTContext *s, FFTComplex *z)
-{
- int n = 1<<s->nbits;
- int i;
- ff_fft_dispatch_interleave_3dnow2(z, s->nbits);
- __asm__ volatile("femms");
- if(n <= 8)
- for(i=0; i<n; i+=2)
- FFSWAP(FFTSample, z[i].im, z[i+1].re);
-}
-
-void ff_imdct_half_3dnow2(FFTContext *s, FFTSample *output, const FFTSample *input)
-{
- x86_reg j, k;
- long n = s->mdct_size;
- long n2 = n >> 1;
- long n4 = n >> 2;
- long n8 = n >> 3;
- const uint16_t *revtab = s->revtab;
- const FFTSample *tcos = s->tcos;
- const FFTSample *tsin = s->tsin;
- const FFTSample *in1, *in2;
- FFTComplex *z = (FFTComplex *)output;
-
- /* pre rotation */
- in1 = input;
- in2 = input + n2 - 1;
-#ifdef EMULATE_3DNOWEXT
- __asm__ volatile("movd %0, %%mm7" ::"r"(1U<<31));
-#endif
- for(k = 0; k < n4; k++) {
- // FIXME a single block is faster, but gcc 2.95 and 3.4.x on 32bit can't compile it
- __asm__ volatile(
- "movd %0, %%mm0 \n"
- "movd %2, %%mm1 \n"
- "punpckldq %1, %%mm0 \n"
- "punpckldq %3, %%mm1 \n"
- "movq %%mm0, %%mm2 \n"
- PSWAPD( %%mm1, %%mm3 )
- "pfmul %%mm1, %%mm0 \n"
- "pfmul %%mm3, %%mm2 \n"
-#ifdef EMULATE_3DNOWEXT
- "movq %%mm0, %%mm1 \n"
- "punpckhdq %%mm2, %%mm0 \n"
- "punpckldq %%mm2, %%mm1 \n"
- "pxor %%mm7, %%mm0 \n"
- "pfadd %%mm1, %%mm0 \n"
-#else
- "pfpnacc %%mm2, %%mm0 \n"
-#endif
- ::"m"(in2[-2*k]), "m"(in1[2*k]),
- "m"(tcos[k]), "m"(tsin[k])
- );
- __asm__ volatile(
- "movq %%mm0, %0 \n\t"
- :"=m"(z[revtab[k]])
- );
- }
-
- ff_fft_dispatch_3dnow2(z, s->nbits);
-
-#define CMUL(j,mm0,mm1)\
- "movq (%2,"#j",2), %%mm6 \n"\
- "movq 8(%2,"#j",2), "#mm0"\n"\
- "movq %%mm6, "#mm1"\n"\
- "movq "#mm0",%%mm7 \n"\
- "pfmul (%3,"#j"), %%mm6 \n"\
- "pfmul (%4,"#j"), "#mm0"\n"\
- "pfmul (%4,"#j"), "#mm1"\n"\
- "pfmul (%3,"#j"), %%mm7 \n"\
- "pfsub %%mm6, "#mm0"\n"\
- "pfadd %%mm7, "#mm1"\n"
-
- /* post rotation */
- j = -n2;
- k = n2-8;
- __asm__ volatile(
- "1: \n"
- CMUL(%0, %%mm0, %%mm1)
- CMUL(%1, %%mm2, %%mm3)
- "movd %%mm0, (%2,%0,2) \n"
- "movd %%mm1,12(%2,%1,2) \n"
- "movd %%mm2, (%2,%1,2) \n"
- "movd %%mm3,12(%2,%0,2) \n"
- "psrlq $32, %%mm0 \n"
- "psrlq $32, %%mm1 \n"
- "psrlq $32, %%mm2 \n"
- "psrlq $32, %%mm3 \n"
- "movd %%mm0, 8(%2,%0,2) \n"
- "movd %%mm1, 4(%2,%1,2) \n"
- "movd %%mm2, 8(%2,%1,2) \n"
- "movd %%mm3, 4(%2,%0,2) \n"
- "sub $8, %1 \n"
- "add $8, %0 \n"
- "jl 1b \n"
- :"+r"(j), "+r"(k)
- :"r"(z+n8), "r"(tcos+n8), "r"(tsin+n8)
- :"memory"
- );
- __asm__ volatile("femms");
-}
-
-void ff_imdct_calc_3dnow2(FFTContext *s, FFTSample *output, const FFTSample *input)
-{
- x86_reg j, k;
- long n = s->mdct_size;
- long n4 = n >> 2;
-
- ff_imdct_half_3dnow2(s, output+n4, input);
-
- j = -n;
- k = n-8;
- __asm__ volatile(
- "movq %4, %%mm7 \n"
- "1: \n"
- PSWAPD((%2,%1), %%mm0)
- PSWAPD((%3,%0), %%mm1)
- "pxor %%mm7, %%mm0 \n"
- "movq %%mm1, (%3,%1) \n"
- "movq %%mm0, (%2,%0) \n"
- "sub $8, %1 \n"
- "add $8, %0 \n"
- "jl 1b \n"
- :"+r"(j), "+r"(k)
- :"r"(output+n4), "r"(output+n4*3),
- "m"(*m1m1)
- );
- __asm__ volatile("femms");
-}
diff --git a/libavcodec/x86/fft_mmx.asm b/libavcodec/x86/fft_mmx.asm
index a709287..8392817 100644
--- a/libavcodec/x86/fft_mmx.asm
+++ b/libavcodec/x86/fft_mmx.asm
@@ -29,6 +29,7 @@
; i.e. {4x real, 4x imaginary, 4x real, ...} (or 2x respectively)
%include "libavutil/x86/x86inc.asm"
+%include "libavutil/x86/x86util.asm"
%if ARCH_X86_64
%define pointer resq
@@ -105,7 +106,7 @@ SECTION_TEXT
pfadd %5, %4 ; {t6,t5}
pxor %3, [ps_m1p1] ; {t8,t7}
mova %6, %1
- pswapd %3, %3
+ PSWAPD %3, %3
pfadd %1, %5 ; {r0,i0}
pfsub %6, %5 ; {r2,i2}
mova %4, %2
@@ -397,7 +398,6 @@ fft32_interleave_avx:
%endif
INIT_XMM sse
-%define movdqa movaps
align 16
fft4_avx:
@@ -470,8 +470,8 @@ fft8 %+ SUFFIX:
mova Z(2), m2
T2_3DN m4, m5, Z(4), Z(5)
T2_3DN m6, m7, Z2(6), Z2(7)
- pswapd m0, m5
- pswapd m2, m7
+ PSWAPD m0, m5
+ PSWAPD m2, m7
pxor m0, [ps_m1p1]
pxor m2, [ps_m1p1]
pfsub m5, m0
@@ -499,11 +499,11 @@ fft8 %+ SUFFIX:
ret
%endmacro
-INIT_MMX 3dnow2
-FFT48_3DN
-
-%macro pswapd 2
-%ifidn %1, %2
+%if ARCH_X86_32
+%macro PSWAPD 2
+%if cpuflag(3dnow2)
+ pswapd %1, %2
+%elifidn %1, %2
movd [r0+12], %1
punpckhdq %1, [r0+8]
%else
@@ -513,9 +513,12 @@ FFT48_3DN
%endif
%endmacro
-INIT_MMX 3dnow
+INIT_MMX 3dnow2
FFT48_3DN
+INIT_MMX 3dnow
+FFT48_3DN
+%endif
%define Z(x) [zcq + o1q*(x&6) + mmsize*(x&1)]
%define Z2(x) [zcq + o3q + mmsize*(x&1)]
@@ -589,6 +592,7 @@ INIT_XMM sse
DECL_PASS pass_sse, PASS_BIG 1
DECL_PASS pass_interleave_sse, PASS_BIG 0
+%macro FFT_CALC_FUNC 0
cglobal fft_calc, 2,5,8
mov r3d, [r0 + FFTContext.nbits]
PUSH r1
@@ -598,23 +602,43 @@ cglobal fft_calc, 2,5,8
FFT_DISPATCH _interleave %+ SUFFIX, r1
POP rcx
POP r4
- cmp rcx, 4
+ cmp rcx, 3+(mmsize/16)
jg .end
mov r2, -1
add rcx, 3
shl r2, cl
sub r4, r2
.loop
+%if mmsize == 8
+ PSWAPD m0, [r4 + r2 + 4]
+ mova [r4 + r2 + 4], m0
+%else
movaps xmm0, [r4 + r2]
movaps xmm1, xmm0
unpcklps xmm0, [r4 + r2 + 16]
unpckhps xmm1, [r4 + r2 + 16]
movaps [r4 + r2], xmm0
movaps [r4 + r2 + 16], xmm1
- add r2, 32
+%endif
+ add r2, mmsize*2
jl .loop
.end:
+%if cpuflag(3dnow)
+ femms
+ RET
+%else
REP_RET
+%endif
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX 3dnow
+FFT_CALC_FUNC
+INIT_MMX 3dnow2
+FFT_CALC_FUNC
+%endif
+INIT_XMM sse
+FFT_CALC_FUNC
cglobal fft_permute, 2,7,1
mov r4, [r0 + FFTContext.revtab]
@@ -649,6 +673,7 @@ cglobal fft_permute, 2,7,1
jl .loopcopy
REP_RET
+%macro IMDCT_CALC_FUNC 0
cglobal imdct_calc, 3,5,3
mov r3d, [r0 + FFTContext.mdctsize]
mov r4, [r0 + FFTContext.imdcthalf]
@@ -672,22 +697,45 @@ cglobal imdct_calc, 3,5,3
POP r3
lea r0, [r1 + 2*r3]
mov r2, r3
- sub r3, 16
+ sub r3, mmsize
neg r2
- movaps xmm2, [ps_m1m1m1m1]
+ mova m2, [ps_m1m1m1m1]
.loop:
- movaps xmm0, [r1 + r3]
- movaps xmm1, [r0 + r2]
- shufps xmm0, xmm0, 0x1b
- shufps xmm1, xmm1, 0x1b
- xorps xmm0, xmm2
- movaps [r0 + r3], xmm1
- movaps [r1 + r2], xmm0
- sub r3, 16
- add r2, 16
+%if mmsize == 8
+ PSWAPD m0, [r1 + r3]
+ PSWAPD m1, [r0 + r2]
+ pxor m0, m2
+%else
+ mova m0, [r1 + r3]
+ mova m1, [r0 + r2]
+ shufps m0, m0, 0x1b
+ shufps m1, m1, 0x1b
+ xorps m0, m2
+%endif
+ mova [r0 + r3], m1
+ mova [r1 + r2], m0
+ sub r3, mmsize
+ add r2, mmsize
jl .loop
+%if cpuflag(3dnow)
+ femms
+ RET
+%else
REP_RET
+%endif
+%endmacro
+%if ARCH_X86_32
+INIT_MMX 3dnow
+IMDCT_CALC_FUNC
+INIT_MMX 3dnow2
+IMDCT_CALC_FUNC
+%endif
+
+INIT_XMM sse
+IMDCT_CALC_FUNC
+
+%if ARCH_X86_32
INIT_MMX 3dnow
%define mulps pfmul
%define addps pfadd
@@ -698,6 +746,7 @@ DECL_PASS pass_3dnow, PASS_SMALL 1, [wq], [wq+o1q]
DECL_PASS pass_interleave_3dnow, PASS_BIG 0
%define pass_3dnow2 pass_3dnow
%define pass_interleave_3dnow2 pass_interleave_3dnow
+%endif
%ifdef PIC
%define SECTION_REL - $$
@@ -761,12 +810,14 @@ DECL_FFT 6, _interleave
INIT_XMM sse
DECL_FFT 5
DECL_FFT 5, _interleave
+%if ARCH_X86_32
INIT_MMX 3dnow
DECL_FFT 4
DECL_FFT 4, _interleave
INIT_MMX 3dnow2
DECL_FFT 4
DECL_FFT 4, _interleave
+%endif
INIT_XMM sse
%undef mulps
@@ -776,6 +827,37 @@ INIT_XMM sse
%undef unpckhps
%macro PREROTATER 5 ;-2*k, 2*k, input+n4, tcos+n8, tsin+n8
+%if mmsize == 8 ; j*2+2-n4, n4-2-j*2, input+n4, tcos+n8, tsin+n8
+ PSWAPD m0, [%3+%2*4]
+ movq m2, [%3+%1*4-8]
+ movq m3, m0
+ punpckldq m0, m2
+ punpckhdq m2, m3
+ movd m1, [%4+%1*2-4] ; tcos[j]
+ movd m3, [%4+%2*2] ; tcos[n4-j-1]
+ punpckldq m1, [%5+%1*2-4] ; tsin[j]
+ punpckldq m3, [%5+%2*2] ; tsin[n4-j-1]
+
+ mova m4, m0
+ PSWAPD m5, m1
+ pfmul m0, m1
+ pfmul m4, m5
+ mova m6, m2
+ PSWAPD m5, m3
+ pfmul m2, m3
+ pfmul m6, m5
+%if cpuflag(3dnow2)
+ pfpnacc m0, m4
+ pfpnacc m2, m6
+%else
+ SBUTTERFLY dq, 0, 4, 1
+ SBUTTERFLY dq, 2, 6, 3
+ pxor m4, m7
+ pxor m6, m7
+ pfadd m0, m4
+ pfadd m2, m6
+%endif
+%else
movaps xmm0, [%3+%2*4]
movaps xmm1, [%3+%1*4-0x10]
movaps xmm2, xmm0
@@ -796,6 +878,7 @@ INIT_XMM sse
movaps xmm0, xmm1
unpcklps xmm1, xmm2
unpckhps xmm0, xmm2
+%endif
%endmacro
%macro CMUL 6 ;j, xmm0, xmm1, 3, 4, 5
@@ -864,6 +947,40 @@ INIT_XMM sse
jl .post
%endmacro
+%macro CMUL_3DNOW 6
+ mova m6, [%1+%2*2]
+ mova %3, [%1+%2*2+8]
+ mova %4, m6
+ mova m7, %3
+ pfmul m6, [%5+%2]
+ pfmul %3, [%6+%2]
+ pfmul %4, [%6+%2]
+ pfmul m7, [%5+%2]
+ pfsub %3, m6
+ pfadd %4, m7
+%endmacro
+
+%macro POSROTATESHUF_3DNOW 5 ;j, k, z+n8, tcos+n8, tsin+n8
+.post:
+ CMUL_3DNOW %3, %1, m0, m1, %4, %5
+ CMUL_3DNOW %3, %2, m2, m3, %4, %5
+ movd [%3+%1*2+ 0], m0
+ movd [%3+%2*2+12], m1
+ movd [%3+%2*2+ 0], m2
+ movd [%3+%1*2+12], m3
+ psrlq m0, 32
+ psrlq m1, 32
+ psrlq m2, 32
+ psrlq m3, 32
+ movd [%3+%1*2+ 8], m0
+ movd [%3+%2*2+ 4], m1
+ movd [%3+%2*2+ 8], m2
+ movd [%3+%1*2+ 4], m3
+ sub %2, 8
+ add %1, 8
+ jl .post
+%endmacro
+
%macro DECL_IMDCT 1
cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *input
%if ARCH_X86_64
@@ -893,22 +1010,34 @@ cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *i
push rrevtab
%endif
- sub r3, 4
-%if ARCH_X86_64
+ sub r3, mmsize/4
+%if ARCH_X86_64 || mmsize == 8
xor r4, r4
sub r4, r3
%endif
+%if notcpuflag(3dnow2) && mmsize == 8
+ movd m7, [ps_m1m1m1m1]
+%endif
.pre:
%if ARCH_X86_64 == 0
;unspill
+%if mmsize != 8
xor r4, r4
sub r4, r3
- mov rtsin, [esp+4]
+%endif
mov rtcos, [esp+8]
+ mov rtsin, [esp+4]
%endif
PREROTATER r4, r3, r2, rtcos, rtsin
-%if ARCH_X86_64
+%if mmsize == 8
+ mov r6, [esp] ; rrevtab = ptr+n8
+ movzx r5, word [rrevtab+r4-2] ; rrevtab[j]
+ movzx r6, word [rrevtab+r3] ; rrevtab[n4-j-1]
+ mova [r1+r5*8], m0
+ mova [r1+r6*8], m2
+ add r4, 2
+%elif ARCH_X86_64
movzx r5, word [rrevtab+r4-4]
movzx r6, word [rrevtab+r4-2]
movzx r10, word [rrevtab+r3]
@@ -929,7 +1058,7 @@ cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *i
movlps [r1+r5*8], xmm1
movhps [r1+r4*8], xmm1
%endif
- sub r3, 4
+ sub r3, mmsize/4
jns .pre
mov r5, r0
@@ -955,11 +1084,22 @@ cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *i
%if ARCH_X86_64 == 0
add esp, 12
%endif
+%if mmsize == 8
+ femms
+%endif
RET
%endmacro
DECL_IMDCT POSROTATESHUF
+%if ARCH_X86_32
+INIT_MMX 3dnow
+DECL_IMDCT POSROTATESHUF_3DNOW
+
+INIT_MMX 3dnow2
+DECL_IMDCT POSROTATESHUF_3DNOW
+%endif
+
INIT_YMM avx
%if HAVE_AVX
More information about the ffmpeg-cvslog
mailing list