eden-emu/opus
19
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Start enabling AVX2 silk_inner_product_FLP()

Browse source 
Not yet with rtcd
This commit is contained in:
Jean-Marc Valin 2023-11-21 02:13:06 -05:00
parent ed90060389
commit b93e4a149c
No known key found for this signature in database
GPG key ID: 531A52533318F00A
6 changed files with 33 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

4
Makefile.am
View file

@ -39,6 +39,9 @@ SILK_SOURCES += $(SILK_SOURCES_FLOAT)
if HAVE_SSE4_1 if HAVE_SSE4_1
SILK_SOURCES += $(SILK_SOURCES_SSE4_1) SILK_SOURCES += $(SILK_SOURCES_SSE4_1)
endif endif
if HAVE_AVX2
SILK_SOURCES += $(SILK_SOURCES_FLOAT_AVX2)
endif
endif endif
if DISABLE_FLOAT_API if DISABLE_FLOAT_API
@ -427,6 +430,7 @@ endif
if HAVE_AVX2 if HAVE_AVX2
AVX2_OBJ = $(CELT_SOURCES_AVX2:.c=.lo) \ AVX2_OBJ = $(CELT_SOURCES_AVX2:.c=.lo) \
$(SILK_SOURCES_AVX2:.c=.lo) \ $(SILK_SOURCES_AVX2:.c=.lo) \
$(SILK_SOURCES_FLOAT_AVX2:.c=.lo) \
$(DNN_SOURCES_AVX2:.c=.lo) $(DNN_SOURCES_AVX2:.c=.lo)
$(AVX2_OBJ): CFLAGS += $(OPUS_X86_AVX2_CFLAGS) $(AVX2_OBJ): CFLAGS += $(OPUS_X86_AVX2_CFLAGS)
endif endif

8
silk/float/SigProc_FLP.h
View file

@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE.
#include "SigProc_FIX.h" #include "SigProc_FIX.h"
#include "float_cast.h" #include "float_cast.h"
#include "main.h"
#include <math.h> #include <math.h>
#ifdef __cplusplus #ifdef __cplusplus
@ -124,12 +125,17 @@ void silk_scale_copy_vector_FLP(
); );
/* inner product of two silk_float arrays, with result as double */ /* inner product of two silk_float arrays, with result as double */
double silk_inner_product_FLP( double silk_inner_product_FLP_c(
const silk_float *data1, const silk_float *data1,
const silk_float *data2, const silk_float *data2,
opus_int dataSize opus_int dataSize
); );
#ifndef OVERRIDE_inner_product_FLP
#define silk_inner_product_FLP(data1, data2, dataSize) silk_inner_product_FLP_c(data1, data2, dataSize)
#endif
/* sum of squares of a silk_float array, with result as double */ /* sum of squares of a silk_float array, with result as double */
double silk_energy_FLP( double silk_energy_FLP(
const silk_float *data, const silk_float *data,

2
silk/float/inner_product_FLP.c
View file

@ -32,7 +32,7 @@ POSSIBILITY OF SUCH DAMAGE.
#include "SigProc_FLP.h" #include "SigProc_FLP.h"
/* inner product of two silk_float arrays, with result as double */ /* inner product of two silk_float arrays, with result as double */
double silk_inner_product_FLP( double silk_inner_product_FLP_c(
const silk_float *data1, const silk_float *data1,
const silk_float *data2, const silk_float *data2,
opus_int dataSize opus_int dataSize

2
silk/float/x86/inner_product_FLP_avx2.c
View file

@ -35,7 +35,7 @@ POSSIBILITY OF SUCH DAMAGE.
/* inner product of two silk_float arrays, with result as double */ /* inner product of two silk_float arrays, with result as double */
double silk_inner_product_FLP( double silk_inner_product_FLP_avx2(
const silk_float *data1, const silk_float *data1,
const silk_float *data2, const silk_float *data2,
opus_int dataSize opus_int dataSize

17
silk/x86/main_sse.h
View file

@ -269,5 +269,22 @@ extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])(
# endif # endif
double silk_inner_product_FLP_avx2(
const silk_float *data1,
const silk_float *data2,
opus_int dataSize
);
#if defined (OPUS_X86_PRESUME_AVX2)
#define OVERRIDE_inner_product_FLP
#define silk_inner_product_FLP(data1, data2, dataSize) silk_inner_product_FLP_avx2(data1, data2, dataSize)
#elif defined(OPUS_HAVE_RTCD) && defined(OPUS_X86_MAY_HAVE_AVX2)
/*#define OVERRIDE_inner_product_FLP*/
#endif
# endif # endif
#endif #endif

3
silk_sources.mk
View file

@ -159,3 +159,6 @@ silk/float/scale_copy_vector_FLP.c \
silk/float/scale_vector_FLP.c \ silk/float/scale_vector_FLP.c \
silk/float/schur_FLP.c \ silk/float/schur_FLP.c \
silk/float/sort_FLP.c silk/float/sort_FLP.c
SILK_SOURCES_FLOAT_AVX2 = \
silk/float/x86/inner_product_FLP_avx2.c