2012-05-21 18:58:41 +02:00
|
|
|
/*
|
2012-06-08 23:02:54 +02:00
|
|
|
* This file is part of FFmpeg.
|
2012-05-21 18:58:41 +02:00
|
|
|
*
|
2012-06-08 23:02:54 +02:00
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2012-05-21 18:58:41 +02:00
|
|
|
* 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.
|
|
|
|
*
|
2012-06-08 23:02:54 +02:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2012-05-21 18:58:41 +02:00
|
|
|
* 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
|
2012-06-08 23:02:54 +02:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2012-05-21 18:58:41 +02:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "config.h"
|
|
|
|
|
|
|
|
#include "libavutil/cpu.h"
|
|
|
|
#include "libavutil/float_dsp.h"
|
2012-08-29 19:01:05 +02:00
|
|
|
#include "cpu.h"
|
2013-01-07 05:47:30 +01:00
|
|
|
#include "asm.h"
|
2012-05-21 18:58:41 +02:00
|
|
|
|
2013-03-26 18:41:24 +01:00
|
|
|
void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
|
|
|
|
int len);
|
|
|
|
void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
|
|
|
|
int len);
|
|
|
|
|
|
|
|
void ff_vector_fmac_scalar_sse(float *dst, const float *src, float mul,
|
2012-05-21 18:58:41 +02:00
|
|
|
int len);
|
2013-03-26 18:41:24 +01:00
|
|
|
void ff_vector_fmac_scalar_avx(float *dst, const float *src, float mul,
|
2012-05-21 18:58:41 +02:00
|
|
|
int len);
|
|
|
|
|
2013-03-26 18:41:24 +01:00
|
|
|
void ff_vector_fmul_scalar_sse(float *dst, const float *src, float mul,
|
|
|
|
int len);
|
2012-09-23 00:41:25 +02:00
|
|
|
|
2013-03-26 18:41:24 +01:00
|
|
|
void ff_vector_dmul_scalar_sse2(double *dst, const double *src,
|
|
|
|
double mul, int len);
|
|
|
|
void ff_vector_dmul_scalar_avx(double *dst, const double *src,
|
|
|
|
double mul, int len);
|
2012-09-24 21:00:53 +02:00
|
|
|
|
2013-01-20 07:26:58 +01:00
|
|
|
void ff_vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
|
|
|
|
const float *src2, int len);
|
|
|
|
void ff_vector_fmul_add_avx(float *dst, const float *src0, const float *src1,
|
|
|
|
const float *src2, int len);
|
|
|
|
|
2013-01-20 22:20:30 +01:00
|
|
|
void ff_vector_fmul_reverse_sse(float *dst, const float *src0,
|
|
|
|
const float *src1, int len);
|
|
|
|
void ff_vector_fmul_reverse_avx(float *dst, const float *src0,
|
|
|
|
const float *src1, int len);
|
|
|
|
|
2013-01-21 00:41:52 +01:00
|
|
|
float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
|
|
|
|
|
2013-04-12 21:07:01 +02:00
|
|
|
void ff_butterflies_float_sse(float *src0, float *src1, int len);
|
|
|
|
|
2013-01-17 17:58:25 +01:00
|
|
|
#if HAVE_6REGS && HAVE_INLINE_ASM
|
2013-01-07 05:47:30 +01:00
|
|
|
static void vector_fmul_window_3dnowext(float *dst, const float *src0,
|
|
|
|
const float *src1, const float *win,
|
|
|
|
int len)
|
|
|
|
{
|
|
|
|
x86_reg i = -len * 4;
|
|
|
|
x86_reg j = len * 4 - 8;
|
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n"
|
|
|
|
"pswapd (%5, %1), %%mm1 \n"
|
|
|
|
"movq (%5, %0), %%mm0 \n"
|
|
|
|
"pswapd (%4, %1), %%mm5 \n"
|
|
|
|
"movq (%3, %0), %%mm4 \n"
|
|
|
|
"movq %%mm0, %%mm2 \n"
|
|
|
|
"movq %%mm1, %%mm3 \n"
|
|
|
|
"pfmul %%mm4, %%mm2 \n" // src0[len + i] * win[len + i]
|
|
|
|
"pfmul %%mm5, %%mm3 \n" // src1[j] * win[len + j]
|
|
|
|
"pfmul %%mm4, %%mm1 \n" // src0[len + i] * win[len + j]
|
|
|
|
"pfmul %%mm5, %%mm0 \n" // src1[j] * win[len + i]
|
|
|
|
"pfadd %%mm3, %%mm2 \n"
|
|
|
|
"pfsub %%mm0, %%mm1 \n"
|
|
|
|
"pswapd %%mm2, %%mm2 \n"
|
|
|
|
"movq %%mm1, (%2, %0) \n"
|
|
|
|
"movq %%mm2, (%2, %1) \n"
|
|
|
|
"sub $8, %1 \n"
|
|
|
|
"add $8, %0 \n"
|
|
|
|
"jl 1b \n"
|
|
|
|
"femms \n"
|
|
|
|
: "+r"(i), "+r"(j)
|
|
|
|
: "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void vector_fmul_window_sse(float *dst, const float *src0,
|
|
|
|
const float *src1, const float *win, int len)
|
|
|
|
{
|
|
|
|
x86_reg i = -len * 4;
|
|
|
|
x86_reg j = len * 4 - 16;
|
|
|
|
__asm__ volatile (
|
|
|
|
"1: \n"
|
|
|
|
"movaps (%5, %1), %%xmm1 \n"
|
|
|
|
"movaps (%5, %0), %%xmm0 \n"
|
|
|
|
"movaps (%4, %1), %%xmm5 \n"
|
|
|
|
"movaps (%3, %0), %%xmm4 \n"
|
|
|
|
"shufps $0x1b, %%xmm1, %%xmm1 \n"
|
|
|
|
"shufps $0x1b, %%xmm5, %%xmm5 \n"
|
|
|
|
"movaps %%xmm0, %%xmm2 \n"
|
|
|
|
"movaps %%xmm1, %%xmm3 \n"
|
|
|
|
"mulps %%xmm4, %%xmm2 \n" // src0[len + i] * win[len + i]
|
|
|
|
"mulps %%xmm5, %%xmm3 \n" // src1[j] * win[len + j]
|
|
|
|
"mulps %%xmm4, %%xmm1 \n" // src0[len + i] * win[len + j]
|
|
|
|
"mulps %%xmm5, %%xmm0 \n" // src1[j] * win[len + i]
|
|
|
|
"addps %%xmm3, %%xmm2 \n"
|
|
|
|
"subps %%xmm0, %%xmm1 \n"
|
|
|
|
"shufps $0x1b, %%xmm2, %%xmm2 \n"
|
|
|
|
"movaps %%xmm1, (%2, %0) \n"
|
|
|
|
"movaps %%xmm2, (%2, %1) \n"
|
|
|
|
"sub $16, %1 \n"
|
|
|
|
"add $16, %0 \n"
|
|
|
|
"jl 1b \n"
|
|
|
|
: "+r"(i), "+r"(j)
|
|
|
|
: "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
|
|
|
|
);
|
|
|
|
}
|
2013-01-17 17:58:25 +01:00
|
|
|
#endif /* HAVE_6REGS && HAVE_INLINE_ASM */
|
2013-01-07 05:47:30 +01:00
|
|
|
|
2012-05-21 18:58:41 +02:00
|
|
|
void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
|
|
|
|
{
|
|
|
|
int mm_flags = av_get_cpu_flags();
|
|
|
|
|
2013-01-17 17:58:25 +01:00
|
|
|
#if HAVE_6REGS && HAVE_INLINE_ASM
|
2013-01-07 05:47:30 +01:00
|
|
|
if (INLINE_AMD3DNOWEXT(mm_flags)) {
|
|
|
|
fdsp->vector_fmul_window = vector_fmul_window_3dnowext;
|
|
|
|
}
|
|
|
|
if (INLINE_SSE(mm_flags)) {
|
|
|
|
fdsp->vector_fmul_window = vector_fmul_window_sse;
|
|
|
|
}
|
|
|
|
#endif
|
2012-08-29 19:01:05 +02:00
|
|
|
if (EXTERNAL_SSE(mm_flags)) {
|
2012-05-21 18:58:41 +02:00
|
|
|
fdsp->vector_fmul = ff_vector_fmul_sse;
|
2012-06-09 05:20:59 +02:00
|
|
|
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse;
|
2012-09-23 00:41:25 +02:00
|
|
|
fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse;
|
2013-01-20 07:26:58 +01:00
|
|
|
fdsp->vector_fmul_add = ff_vector_fmul_add_sse;
|
2013-01-20 22:20:30 +01:00
|
|
|
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
|
2013-01-21 00:41:52 +01:00
|
|
|
fdsp->scalarproduct_float = ff_scalarproduct_float_sse;
|
2013-04-12 21:07:01 +02:00
|
|
|
fdsp->butterflies_float = ff_butterflies_float_sse;
|
2012-05-21 18:58:41 +02:00
|
|
|
}
|
2012-09-24 21:00:53 +02:00
|
|
|
if (EXTERNAL_SSE2(mm_flags)) {
|
|
|
|
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2;
|
|
|
|
}
|
2012-08-29 19:01:05 +02:00
|
|
|
if (EXTERNAL_AVX(mm_flags)) {
|
2012-05-21 18:58:41 +02:00
|
|
|
fdsp->vector_fmul = ff_vector_fmul_avx;
|
2012-06-09 05:20:59 +02:00
|
|
|
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx;
|
2012-09-24 21:00:53 +02:00
|
|
|
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx;
|
2013-01-20 07:26:58 +01:00
|
|
|
fdsp->vector_fmul_add = ff_vector_fmul_add_avx;
|
2013-01-20 22:20:30 +01:00
|
|
|
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
|
2012-05-21 18:58:41 +02:00
|
|
|
}
|
|
|
|
}
|