---
Now passes FATE on ARM thanks to a hint from Janne.
libavcodec/aarch64/Makefile | 1 +
libavcodec/aarch64/fft_init_aarch64.c | 12 --
.../aarch64/{fft_init_aarch64.c => mdct_init.c} | 14 +--
libavcodec/arm/Makefile | 2 +
libavcodec/arm/fft_fixed_init_arm.c | 10 --
...{fft_fixed_init_arm.c => mdct_fixed_init_arm.c} | 10 +-
.../fft_init_aarch64.c => arm/mdct_init_arm.c} | 17 ++-
libavcodec/fft.h | 7 ++
libavcodec/fft_template.c | 7 --
libavcodec/mdct_template.c | 20 ++++
libavcodec/ppc/Makefile | 1 +
libavcodec/ppc/fft_init.c | 124 +-------------------
libavcodec/ppc/{fft_init.c => mdct_init.c} | 4 +-
libavcodec/x86/Makefile | 1 +
libavcodec/x86/fft.asm | 128 +++++++++++----------
libavcodec/x86/fft.h | 8 --
libavcodec/x86/fft_init.c | 7 --
libavcodec/x86/{fft.h => mdct.h} | 12 +-
libavcodec/x86/{fft_init.c => mdct_init.c} | 11 +-
19 files changed, 115 insertions(+), 281 deletions(-)
copy libavcodec/aarch64/{fft_init_aarch64.c => mdct_init.c} (80%)
copy libavcodec/arm/{fft_fixed_init_arm.c => mdct_fixed_init_arm.c} (82%)
copy libavcodec/{aarch64/fft_init_aarch64.c => arm/mdct_init_arm.c} (81%)
copy libavcodec/ppc/{fft_init.c => mdct_init.c} (96%)
copy libavcodec/x86/{fft.h => mdct.h} (79%)
copy libavcodec/x86/{fft_init.c => mdct_init.c} (79%)
diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
index 022ed84..311cd2c 100644
--- a/libavcodec/aarch64/Makefile
+++ b/libavcodec/aarch64/Makefile
@@ -7,6 +7,7 @@ OBJS-$(CONFIG_H264PRED) +=
aarch64/h264pred_init.o
OBJS-$(CONFIG_H264QPEL) += aarch64/h264qpel_init_aarch64.o
OBJS-$(CONFIG_HPELDSP) += aarch64/hpeldsp_init_aarch64.o
OBJS-$(CONFIG_IMDCT15) += aarch64/imdct15_init.o
+OBJS-$(CONFIG_MDCT) += aarch64/mdct_init.o
OBJS-$(CONFIG_MPEGAUDIODSP) += aarch64/mpegaudiodsp_init.o
OBJS-$(CONFIG_NEON_CLOBBER_TEST) += aarch64/neontest.o
OBJS-$(CONFIG_VIDEODSP) += aarch64/videodsp_init.o
diff --git a/libavcodec/aarch64/fft_init_aarch64.c
b/libavcodec/aarch64/fft_init_aarch64.c
index f85091e..9cc57d3 100644
--- a/libavcodec/aarch64/fft_init_aarch64.c
+++ b/libavcodec/aarch64/fft_init_aarch64.c
@@ -18,8 +18,6 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "config.h"
-
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/aarch64/cpu.h"
@@ -29,10 +27,6 @@
void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
-void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
-
av_cold void ff_fft_init_aarch64(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
@@ -40,11 +34,5 @@ av_cold void ff_fft_init_aarch64(FFTContext *s)
if (have_neon(cpu_flags)) {
s->fft_permute = ff_fft_permute_neon;
s->fft_calc = ff_fft_calc_neon;
-#if CONFIG_MDCT
- s->imdct_calc = ff_imdct_calc_neon;
- s->imdct_half = ff_imdct_half_neon;
- s->mdct_calc = ff_mdct_calc_neon;
- s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
-#endif
}
}
diff --git a/libavcodec/aarch64/fft_init_aarch64.c
b/libavcodec/aarch64/mdct_init.c
similarity index 80%
copy from libavcodec/aarch64/fft_init_aarch64.c
copy to libavcodec/aarch64/mdct_init.c
index f85091e..81826c2 100644
--- a/libavcodec/aarch64/fft_init_aarch64.c
+++ b/libavcodec/aarch64/mdct_init.c
@@ -1,6 +1,4 @@
/*
- * Copyright (c) 2009 Mans Rullgard <[email protected]>
- *
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
@@ -18,33 +16,23 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "config.h"
-
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/aarch64/cpu.h"
-
#include "libavcodec/fft.h"
-void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
-
void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
-av_cold void ff_fft_init_aarch64(FFTContext *s)
+av_cold void ff_mdct_init_aarch64(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
if (have_neon(cpu_flags)) {
- s->fft_permute = ff_fft_permute_neon;
- s->fft_calc = ff_fft_calc_neon;
-#if CONFIG_MDCT
s->imdct_calc = ff_imdct_calc_neon;
s->imdct_half = ff_imdct_half_neon;
s->mdct_calc = ff_mdct_calc_neon;
s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
-#endif
}
}
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 89ec237..a684c8b 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -21,6 +21,8 @@ OBJS-$(CONFIG_IDCTDSP) +=
arm/idctdsp_init_arm.o \
OBJS-$(CONFIG_FLACDSP) += arm/flacdsp_init_arm.o \
arm/flacdsp_arm.o
OBJS-$(CONFIG_G722DSP) += arm/g722dsp_init_arm.o
+OBJS-$(CONFIG_MDCT) += arm/mdct_init_arm.o \
+ arm/mdct_fixed_init_arm.o
OBJS-$(CONFIG_ME_CMP) += arm/me_cmp_init_arm.o
OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_init_arm.o
OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o
diff --git a/libavcodec/arm/fft_fixed_init_arm.c
b/libavcodec/arm/fft_fixed_init_arm.c
index 1f941a1..5132b09 100644
--- a/libavcodec/arm/fft_fixed_init_arm.c
+++ b/libavcodec/arm/fft_fixed_init_arm.c
@@ -26,8 +26,6 @@
#include "libavcodec/fft.h"
void ff_fft_fixed_calc_neon(FFTContext *s, FFTComplex *z);
-void ff_mdct_fixed_calc_neon(FFTContext *s, FFTSample *o, const FFTSample *i);
-void ff_mdct_fixed_calcw_neon(FFTContext *s, FFTDouble *o, const FFTSample *i);
av_cold void ff_fft_fixed_init_arm(FFTContext *s)
{
@@ -36,13 +34,5 @@ av_cold void ff_fft_fixed_init_arm(FFTContext *s)
if (have_neon(cpu_flags)) {
s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
s->fft_calc = ff_fft_fixed_calc_neon;
-
-#if CONFIG_MDCT
- if (!s->inverse && s->nbits >= 3) {
- s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
- s->mdct_calc = ff_mdct_fixed_calc_neon;
- s->mdct_calcw = ff_mdct_fixed_calcw_neon;
- }
-#endif
}
}
diff --git a/libavcodec/arm/fft_fixed_init_arm.c
b/libavcodec/arm/mdct_fixed_init_arm.c
similarity index 82%
copy from libavcodec/arm/fft_fixed_init_arm.c
copy to libavcodec/arm/mdct_fixed_init_arm.c
index 1f941a1..606c80c 100644
--- a/libavcodec/arm/fft_fixed_init_arm.c
+++ b/libavcodec/arm/mdct_fixed_init_arm.c
@@ -1,6 +1,4 @@
/*
- * Copyright (c) 2009 Mans Rullgard <[email protected]>
- *
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
@@ -25,24 +23,18 @@
#define FFT_FLOAT 0
#include "libavcodec/fft.h"
-void ff_fft_fixed_calc_neon(FFTContext *s, FFTComplex *z);
void ff_mdct_fixed_calc_neon(FFTContext *s, FFTSample *o, const FFTSample *i);
void ff_mdct_fixed_calcw_neon(FFTContext *s, FFTDouble *o, const FFTSample *i);
-av_cold void ff_fft_fixed_init_arm(FFTContext *s)
+av_cold void ff_mdct_fixed_init_arm(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
if (have_neon(cpu_flags)) {
- s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
- s->fft_calc = ff_fft_fixed_calc_neon;
-
-#if CONFIG_MDCT
if (!s->inverse && s->nbits >= 3) {
s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
s->mdct_calc = ff_mdct_fixed_calc_neon;
s->mdct_calcw = ff_mdct_fixed_calcw_neon;
}
-#endif
}
}
diff --git a/libavcodec/aarch64/fft_init_aarch64.c
b/libavcodec/arm/mdct_init_arm.c
similarity index 81%
copy from libavcodec/aarch64/fft_init_aarch64.c
copy to libavcodec/arm/mdct_init_arm.c
index f85091e..24678dd 100644
--- a/libavcodec/aarch64/fft_init_aarch64.c
+++ b/libavcodec/arm/mdct_init_arm.c
@@ -18,33 +18,30 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "config.h"
-
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
-#include "libavutil/aarch64/cpu.h"
+#include "libavutil/arm/cpu.h"
#include "libavcodec/fft.h"
-void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
+void ff_imdct_half_vfp(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample
*input);
-av_cold void ff_fft_init_aarch64(FFTContext *s)
+av_cold void ff_mdct_init_arm(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
+ if (have_vfp_vm(cpu_flags)) {
+ s->imdct_half = ff_imdct_half_vfp;
+ }
+
if (have_neon(cpu_flags)) {
- s->fft_permute = ff_fft_permute_neon;
- s->fft_calc = ff_fft_calc_neon;
-#if CONFIG_MDCT
s->imdct_calc = ff_imdct_calc_neon;
s->imdct_half = ff_imdct_half_neon;
s->mdct_calc = ff_mdct_calc_neon;
s->mdct_permutation = FF_MDCT_PERM_INTERLEAVE;
-#endif
}
}
diff --git a/libavcodec/fft.h b/libavcodec/fft.h
index 7daae24..57dc17f 100644
--- a/libavcodec/fft.h
+++ b/libavcodec/fft.h
@@ -154,4 +154,11 @@ void ff_fft_end(FFTContext *s);
int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale);
void ff_mdct_end(FFTContext *s);
+void ff_mdct_init_aarch64(FFTContext *s);
+void ff_mdct_init_arm(FFTContext *s);
+void ff_mdct_init_ppc(FFTContext *s);
+void ff_mdct_init_x86(FFTContext *s);
+
+void ff_mdct_fixed_init_arm(FFTContext *s);
+
#endif /* AVCODEC_FFT_H */
diff --git a/libavcodec/fft_template.c b/libavcodec/fft_template.c
index 808f317..3642b43 100644
--- a/libavcodec/fft_template.c
+++ b/libavcodec/fft_template.c
@@ -151,20 +151,13 @@ av_cold int ff_fft_init(FFTContext *s, int nbits, int
inverse)
s->fft_permute = fft_permute_c;
s->fft_calc = fft_calc_c;
-#if CONFIG_MDCT
- s->imdct_calc = ff_imdct_calc_c;
- s->imdct_half = ff_imdct_half_c;
- s->mdct_calc = ff_mdct_calc_c;
-#endif
#if FFT_FLOAT
if (ARCH_AARCH64) ff_fft_init_aarch64(s);
if (ARCH_ARM) ff_fft_init_arm(s);
if (ARCH_PPC) ff_fft_init_ppc(s);
if (ARCH_X86) ff_fft_init_x86(s);
- if (CONFIG_MDCT) s->mdct_calcw = s->mdct_calc;
#else
- if (CONFIG_MDCT) s->mdct_calcw = ff_mdct_calcw_c;
if (ARCH_ARM) ff_fft_fixed_init_arm(s);
#endif
diff --git a/libavcodec/mdct_template.c b/libavcodec/mdct_template.c
index bad890e..5b3a6ff 100644
--- a/libavcodec/mdct_template.c
+++ b/libavcodec/mdct_template.c
@@ -56,6 +56,26 @@ av_cold int ff_mdct_init(FFTContext *s, int nbits, int
inverse, double scale)
if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
goto fail;
+ s->imdct_calc = ff_imdct_calc_c;
+ s->imdct_half = ff_imdct_half_c;
+ s->mdct_calc = ff_mdct_calc_c;
+
+#if FFT_FLOAT
+ if (ARCH_AARCH64)
+ ff_mdct_init_aarch64(s);
+ if (ARCH_ARM)
+ ff_mdct_init_arm(s);
+ if (ARCH_PPC)
+ ff_mdct_init_ppc(s);
+ if (ARCH_X86)
+ ff_mdct_init_x86(s);
+ s->mdct_calcw = s->mdct_calc;
+#else
+ s->mdct_calcw = ff_mdct_calcw_c;
+ if (ARCH_ARM)
+ ff_mdct_fixed_init_arm(s);
+#endif
+
s->tcos = av_malloc(n/2 * sizeof(FFTSample));
if (!s->tcos)
goto fail;
diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile
index 2b6c81b..759888b 100644
--- a/libavcodec/ppc/Makefile
+++ b/libavcodec/ppc/Makefile
@@ -11,6 +11,7 @@ OBJS-$(CONFIG_HPELDSP) +=
ppc/hpeldsp_altivec.o
OBJS-$(CONFIG_HUFFYUVDSP) += ppc/huffyuvdsp_altivec.o
OBJS-$(CONFIG_FDCTDSP) += ppc/fdctdsp.o
OBJS-$(CONFIG_IDCTDSP) += ppc/idctdsp.o
+OBJS-$(CONFIG_MDCT) += ppc/mdct_init.o
OBJS-$(CONFIG_ME_CMP) += ppc/me_cmp.o
OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o
OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o \
diff --git a/libavcodec/ppc/fft_init.c b/libavcodec/ppc/fft_init.c
index a749900..56eafb9 100644
--- a/libavcodec/ppc/fft_init.c
+++ b/libavcodec/ppc/fft_init.c
@@ -1,8 +1,4 @@
/*
- * FFT/IFFT transforms
- * AltiVec-enabled
- * Copyright (c) 2009 Loren Merritt
- *
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
@@ -21,126 +17,14 @@
*/
#include "config.h"
+
#include "libavutil/cpu.h"
#include "libavutil/ppc/cpu.h"
-#include "libavutil/ppc/types_altivec.h"
-#include "libavutil/ppc/util_altivec.h"
-#include "libavcodec/fft.h"
-/**
- * Do a complex FFT with the parameters defined in ff_fft_init().
- * The input data must be permuted before with s->revtab table.
- * No 1.0 / sqrt(n) normalization is done.
- * AltiVec-enabled:
- * This code assumes that the 'z' pointer is 16 bytes-aligned.
- * It also assumes all FFTComplex are 8 bytes-aligned pairs of floats.
- */
+#include "libavcodec/fft.h"
-void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
-#if HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN
-static void imdct_half_altivec(FFTContext *s, FFTSample *output, const
FFTSample *input)
-{
- int j, k;
- int n = 1 << s->mdct_bits;
- int n4 = n >> 2;
- int n8 = n >> 3;
- int n32 = n >> 5;
- const uint16_t *revtabj = s->revtab;
- const uint16_t *revtabk = s->revtab+n4;
- const vec_f *tcos = (const vec_f*)(s->tcos+n8);
- const vec_f *tsin = (const vec_f*)(s->tsin+n8);
- const vec_f *pin = (const vec_f*)(input+n4);
- vec_f *pout = (vec_f*)(output+n4);
-
- /* pre rotation */
- k = n32-1;
- do {
- vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d;
-#define CMULA(p,o0,o1,o2,o3)\
- a = pin[ k*2+p]; /* { z[k].re, z[k].im,
z[k+1].re, z[k+1].im } */\
- b = pin[-k*2-p-1]; /* { z[-k-2].re, z[-k-2].im,
z[-k-1].re, z[-k-1].im } */\
- re = vec_perm(a, b, vcprm(0,2,s0,s2)); /* { z[k].re, z[k+1].re,
z[-k-2].re, z[-k-1].re } */\
- im = vec_perm(a, b, vcprm(s3,s1,3,1)); /* { z[-k-1].im, z[-k-2].im,
z[k+1].im, z[k].im } */\
- cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k],
cos[k+1], cos[-k-2], cos[-k-1] } */\
- sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\
- r##p = im*cos - re*sin;\
- i##p = re*cos + im*sin;
-#define STORE2(v,dst)\
- j = dst;\
- vec_ste(v, 0, output+j*2);\
- vec_ste(v, 4, output+j*2);
-#define STORE8(p)\
- a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\
- b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\
- c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\
- d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\
- STORE2(a, revtabk[ p*2-4]);\
- STORE2(b, revtabk[ p*2-3]);\
- STORE2(c, revtabj[-p*2+2]);\
- STORE2(d, revtabj[-p*2+3]);
-
- cos0 = tcos[k];
- sin0 = tsin[k];
- cos1 = tcos[-k-1];
- sin1 = tsin[-k-1];
- CMULA(0, 0,1,2,3);
- CMULA(1, 2,3,0,1);
- STORE8(0);
- STORE8(1);
- revtabj += 4;
- revtabk -= 4;
- k--;
- } while(k >= 0);
-
- ff_fft_calc_altivec(s, (FFTComplex*)output);
-
- /* post rotation + reordering */
- j = -n32;
- k = n32-1;
- do {
- vec_f cos,sin,re,im,a,b,c,d;
-#define CMULB(d0,d1,o)\
- re = pout[o*2];\
- im = pout[o*2+1];\
- cos = tcos[o];\
- sin = tsin[o];\
- d0 = im*sin - re*cos;\
- d1 = re*sin + im*cos;
-
- CMULB(a,b,j);
- CMULB(c,d,k);
- pout[2*j] = vec_perm(a, d, vcprm(0,s3,1,s2));
- pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0));
- pout[2*k] = vec_perm(c, b, vcprm(0,s3,1,s2));
- pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0));
- j++;
- k--;
- } while(k >= 0);
-}
-
-static void imdct_calc_altivec(FFTContext *s, FFTSample *output, const
FFTSample *input)
-{
- int k;
- int n = 1 << s->mdct_bits;
- int n4 = n >> 2;
- int n16 = n >> 4;
- vec_u32 sign = {1U<<31,1U<<31,1U<<31,1U<<31};
- vec_u32 *p0 = (vec_u32*)(output+n4);
- vec_u32 *p1 = (vec_u32*)(output+n4*3);
-
- imdct_half_altivec(s, output + n4, input);
-
- for (k = 0; k < n16; k++) {
- vec_u32 a = p0[k] ^ sign;
- vec_u32 b = p1[-k-1];
- p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0));
- p1[k] = vec_perm(b, b, vcprm(3,2,1,0));
- }
-}
-#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
-
av_cold void ff_fft_init_ppc(FFTContext *s)
{
#if HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN
@@ -148,9 +32,5 @@ av_cold void ff_fft_init_ppc(FFTContext *s)
return;
s->fft_calc = ff_fft_calc_interleave_altivec;
- if (s->mdct_bits >= 5) {
- s->imdct_calc = imdct_calc_altivec;
- s->imdct_half = imdct_half_altivec;
- }
#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
}
diff --git a/libavcodec/ppc/fft_init.c b/libavcodec/ppc/mdct_init.c
similarity index 96%
copy from libavcodec/ppc/fft_init.c
copy to libavcodec/ppc/mdct_init.c
index a749900..d3582bc 100644
--- a/libavcodec/ppc/fft_init.c
+++ b/libavcodec/ppc/mdct_init.c
@@ -37,7 +37,6 @@
*/
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
#if HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN
static void imdct_half_altivec(FFTContext *s, FFTSample *output, const
FFTSample *input)
@@ -141,13 +140,12 @@ static void imdct_calc_altivec(FFTContext *s, FFTSample
*output, const FFTSample
}
#endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
-av_cold void ff_fft_init_ppc(FFTContext *s)
+av_cold void ff_mdct_init_ppc(FFTContext *s)
{
#if HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN
if (!PPC_ALTIVEC(av_get_cpu_flags()))
return;
- s->fft_calc = ff_fft_calc_interleave_altivec;
if (s->mdct_bits >= 5) {
s->imdct_calc = imdct_calc_altivec;
s->imdct_half = imdct_half_altivec;
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 8ae1283..1a2cfb1 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -19,6 +19,7 @@ OBJS-$(CONFIG_HUFFYUVDSP) +=
x86/huffyuvdsp_init.o
OBJS-$(CONFIG_HUFFYUVENCDSP) += x86/huffyuvencdsp_mmx.o
OBJS-$(CONFIG_IDCTDSP) += x86/idctdsp_init.o
OBJS-$(CONFIG_LPC) += x86/lpc.o
+OBJS-$(CONFIG_MDCT) += x86/mdct_init.o
OBJS-$(CONFIG_ME_CMP) += x86/me_cmp_init.o
OBJS-$(CONFIG_MPEGAUDIODSP) += x86/mpegaudiodsp.o
OBJS-$(CONFIG_MPEGVIDEO) += x86/mpegvideo.o \
diff --git a/libavcodec/x86/fft.asm b/libavcodec/x86/fft.asm
index d3be72e..ef007f4 100644
--- a/libavcodec/x86/fft.asm
+++ b/libavcodec/x86/fft.asm
@@ -655,68 +655,6 @@ 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]
- add r1, r3
- PUSH r3
- PUSH r1
-%if ARCH_X86_32
- push r2
- push r1
- push r0
-%else
- sub rsp, 8+32*WIN64 ; allocate win64 shadow space
-%endif
- call r4
-%if ARCH_X86_32
- add esp, 12
-%else
- add rsp, 8+32*WIN64
-%endif
- POP r1
- POP r3
- lea r0, [r1 + 2*r3]
- mov r2, r3
- sub r3, mmsize
- neg r2
- mova m2, [ps_m1m1m1m1]
-.loop:
-%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 3dnowext
-IMDCT_CALC_FUNC
-%endif
-
-INIT_XMM sse
-IMDCT_CALC_FUNC
-
%if ARCH_X86_32
INIT_MMX 3dnow
%define mulps pfmul
@@ -791,6 +729,70 @@ DECL_FFT 4
DECL_FFT 4, _interleave
%endif
+%if CONFIG_MDCT
+
+%macro IMDCT_CALC_FUNC 0
+cglobal imdct_calc, 3,5,3
+ mov r3d, [r0 + FFTContext.mdctsize]
+ mov r4, [r0 + FFTContext.imdcthalf]
+ add r1, r3
+ PUSH r3
+ PUSH r1
+%if ARCH_X86_32
+ push r2
+ push r1
+ push r0
+%else
+ sub rsp, 8+32*WIN64 ; allocate win64 shadow space
+%endif
+ call r4
+%if ARCH_X86_32
+ add esp, 12
+%else
+ add rsp, 8+32*WIN64
+%endif
+ POP r1
+ POP r3
+ lea r0, [r1 + 2*r3]
+ mov r2, r3
+ sub r3, mmsize
+ neg r2
+ mova m2, [ps_m1m1m1m1]
+.loop:
+%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 3dnowext
+IMDCT_CALC_FUNC
+%endif
+
+INIT_XMM sse
+IMDCT_CALC_FUNC
+
INIT_XMM sse
%undef mulps
%undef addps
@@ -1081,3 +1083,5 @@ DECL_IMDCT POSROTATESHUF_3DNOW
INIT_YMM avx
DECL_IMDCT POSROTATESHUF_AVX
+
+%endif ; CONFIG_MDCT
diff --git a/libavcodec/x86/fft.h b/libavcodec/x86/fft.h
index a604956..94405d0 100644
--- a/libavcodec/x86/fft.h
+++ b/libavcodec/x86/fft.h
@@ -27,12 +27,4 @@ void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
void ff_fft_calc_3dnow(FFTContext *s, FFTComplex *z);
void ff_fft_calc_3dnowext(FFTContext *s, FFTComplex *z);
-void ff_imdct_calc_3dnow(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_half_3dnow(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_calc_3dnowext(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_half_3dnowext(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample
*input);
-void ff_imdct_half_avx(FFTContext *s, FFTSample *output, const FFTSample
*input);
-
#endif /* AVCODEC_X86_FFT_H */
diff --git a/libavcodec/x86/fft_init.c b/libavcodec/x86/fft_init.c
index 2234d76..ed12909 100644
--- a/libavcodec/x86/fft_init.c
+++ b/libavcodec/x86/fft_init.c
@@ -30,28 +30,21 @@ av_cold void ff_fft_init_x86(FFTContext *s)
#if ARCH_X86_32
if (EXTERNAL_AMD3DNOW(cpu_flags)) {
- s->imdct_calc = ff_imdct_calc_3dnow;
- s->imdct_half = ff_imdct_half_3dnow;
s->fft_calc = ff_fft_calc_3dnow;
}
if (EXTERNAL_AMD3DNOWEXT(cpu_flags)) {
- s->imdct_calc = ff_imdct_calc_3dnowext;
- s->imdct_half = ff_imdct_half_3dnowext;
s->fft_calc = ff_fft_calc_3dnowext;
}
#endif /* ARCH_X86_32 */
if (EXTERNAL_SSE(cpu_flags)) {
- s->imdct_calc = ff_imdct_calc_sse;
- s->imdct_half = ff_imdct_half_sse;
s->fft_permute = ff_fft_permute_sse;
s->fft_calc = ff_fft_calc_sse;
s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
}
if (EXTERNAL_AVX_FAST(cpu_flags) && s->nbits >= 5) {
- s->imdct_half = ff_imdct_half_avx;
s->fft_calc = ff_fft_calc_avx;
s->fft_permutation = FF_FFT_PERM_AVX;
}
diff --git a/libavcodec/x86/fft.h b/libavcodec/x86/mdct.h
similarity index 79%
copy from libavcodec/x86/fft.h
copy to libavcodec/x86/mdct.h
index a604956..cc107cb 100644
--- a/libavcodec/x86/fft.h
+++ b/libavcodec/x86/mdct.h
@@ -16,17 +16,11 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#ifndef AVCODEC_X86_FFT_H
-#define AVCODEC_X86_FFT_H
+#ifndef AVCODEC_X86_MDCT_H
+#define AVCODEC_X86_MDCT_H
#include "libavcodec/fft.h"
-void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_avx(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_3dnow(FFTContext *s, FFTComplex *z);
-void ff_fft_calc_3dnowext(FFTContext *s, FFTComplex *z);
-
void ff_imdct_calc_3dnow(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_half_3dnow(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_calc_3dnowext(FFTContext *s, FFTSample *output, const FFTSample
*input);
@@ -35,4 +29,4 @@ void ff_imdct_calc_sse(FFTContext *s, FFTSample *output,
const FFTSample *input)
void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample
*input);
void ff_imdct_half_avx(FFTContext *s, FFTSample *output, const FFTSample
*input);
-#endif /* AVCODEC_X86_FFT_H */
+#endif /* AVCODEC_X86_MDCT_H */
diff --git a/libavcodec/x86/fft_init.c b/libavcodec/x86/mdct_init.c
similarity index 79%
copy from libavcodec/x86/fft_init.c
copy to libavcodec/x86/mdct_init.c
index 2234d76..db642d8 100644
--- a/libavcodec/x86/fft_init.c
+++ b/libavcodec/x86/mdct_init.c
@@ -22,9 +22,9 @@
#include "libavutil/cpu.h"
#include "libavutil/x86/cpu.h"
-#include "fft.h"
+#include "mdct.h"
-av_cold void ff_fft_init_x86(FFTContext *s)
+av_cold void ff_mdct_init_x86(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
@@ -32,27 +32,20 @@ av_cold void ff_fft_init_x86(FFTContext *s)
if (EXTERNAL_AMD3DNOW(cpu_flags)) {
s->imdct_calc = ff_imdct_calc_3dnow;
s->imdct_half = ff_imdct_half_3dnow;
- s->fft_calc = ff_fft_calc_3dnow;
}
if (EXTERNAL_AMD3DNOWEXT(cpu_flags)) {
s->imdct_calc = ff_imdct_calc_3dnowext;
s->imdct_half = ff_imdct_half_3dnowext;
- s->fft_calc = ff_fft_calc_3dnowext;
}
#endif /* ARCH_X86_32 */
if (EXTERNAL_SSE(cpu_flags)) {
s->imdct_calc = ff_imdct_calc_sse;
s->imdct_half = ff_imdct_half_sse;
- s->fft_permute = ff_fft_permute_sse;
- s->fft_calc = ff_fft_calc_sse;
- s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
}
if (EXTERNAL_AVX_FAST(cpu_flags) && s->nbits >= 5) {
s->imdct_half = ff_imdct_half_avx;
- s->fft_calc = ff_fft_calc_avx;
- s->fft_permutation = FF_FFT_PERM_AVX;
}
}
--
2.5.0
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
