wine/gecko
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/wine-gecko.git synced 2024-09-13 09:24:08 -07:00
Code Issues Packages Projects Releases Wiki Activity

b=1042508 move resampler simd optimizations to separate translation units r=padenot

Browse Source 
This ensures that code in resample.c will run on Intel x86 cpus even when SSE
support has been compiled, and will provide similarly for neon support when
enabled.

--HG--
rename : media/libspeex_resampler/sse-detect-runtime.patch => media/libspeex_resampler/simd-detect-runtime.patch
rename : media/libspeex_resampler/src/resample_neon.h => media/libspeex_resampler/src/resample_neon.c
rename : media/libspeex_resampler/src/resample_sse.h => media/libspeex_resampler/src/resample_sse.c
rename : media/libspeex_resampler/src/sse_detect.cpp => media/libspeex_resampler/src/simd_detect.cpp
rename : media/libspeex_resampler/src/sse_detect.h => media/libspeex_resampler/src/simd_detect.h
extra : rebase_source : bdf1ef791129dedeadce7369354a5992729a99b7
This commit is contained in:
Karl Tomlinson 2014-07-23 21:49:04 +12:00
parent 944001f0dc
commit 6858d0226c
10 changed files with 413 additions and 251 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

331
media/libspeex_resampler/simd-detect-runtime.patch Normal file
View File

@ -0,0 +1,331 @@
diff --git a/media/libspeex_resampler/src/resample.c b/media/libspeex_resampler/src/resample.c
--- a/media/libspeex_resampler/src/resample.c
+++ b/media/libspeex_resampler/src/resample.c
@@ -92,23 +92,17 @@ static void speex_free (void *ptr) {free
#define IMAX(a,b) ((a) > (b) ? (a) : (b))
#define IMIN(a,b) ((a) < (b) ? (a) : (b))
#ifndef NULL
#define NULL 0
#endif
-#ifdef _USE_SSE
-#include "resample_sse.h"
-#endif
-
-#ifdef _USE_NEON
-#include "resample_neon.h"
-#endif
+#include "simd_detect.h"
/* Numer of elements to allocate on the stack */
#ifdef VAR_ARRAYS
#define FIXED_STACK_ALLOC 8192
#else
#define FIXED_STACK_ALLOC 1024
#endif
@@ -344,17 +338,19 @@ static int resampler_basic_direct_single
const spx_uint32_t den_rate = st->den_rate;
spx_word32_t sum;
while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
{
const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
const spx_word16_t *iptr = & in[last_sample];
-#ifndef OVERRIDE_INNER_PRODUCT_SINGLE
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
+ if (!moz_speex_have_single_simd()) {
+#endif
int j;
sum = 0;
for(j=0;j<N;j++) sum += MULT16_16(sinct[j], iptr[j]);
/* This code is slower on most DSPs which have only 2 accumulators.
Plus this this forces truncation to 32 bits and you lose the HW guard bits.
I think we can trust the compiler and let it vectorize and/or unroll itself.
spx_word32_t accum[4] = {0,0,0,0};
@@ -362,18 +358,20 @@ static int resampler_basic_direct_single
accum[0] += MULT16_16(sinct[j], iptr[j]);
accum[1] += MULT16_16(sinct[j+1], iptr[j+1]);
accum[2] += MULT16_16(sinct[j+2], iptr[j+2]);
accum[3] += MULT16_16(sinct[j+3], iptr[j+3]);
}
sum = accum[0] + accum[1] + accum[2] + accum[3];
*/
sum = SATURATE32PSHR(sum, 15, 32767);
-#else
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
+ } else {
sum = inner_product_single(sinct, iptr, N);
+ }
#endif
out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -402,29 +400,33 @@ static int resampler_basic_direct_double
const spx_uint32_t den_rate = st->den_rate;
double sum;
while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
{
const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
const spx_word16_t *iptr = & in[last_sample];
-#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
+ if(moz_speex_have_double_simd()) {
+#endif
int j;
double accum[4] = {0,0,0,0};
for(j=0;j<N;j+=4) {
accum[0] += sinct[j]*iptr[j];
accum[1] += sinct[j+1]*iptr[j+1];
accum[2] += sinct[j+2]*iptr[j+2];
accum[3] += sinct[j+3]*iptr[j+3];
}
sum = accum[0] + accum[1] + accum[2] + accum[3];
-#else
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
+ } else {
sum = inner_product_double(sinct, iptr, N);
+ }
#endif
out[out_stride * out_sample++] = PSHR32(sum, 15);
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -458,34 +460,38 @@ static int resampler_basic_interpolate_s
#ifdef FIXED_POINT
const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
#else
const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
#endif
spx_word16_t interp[4];
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+ if (!moz_speex_have_single_simd()) {
+#endif
int j;
spx_word32_t accum[4] = {0,0,0,0};
for(j=0;j<N;j++) {
const spx_word16_t curr_in=iptr[j];
accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
}
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1));
sum = SATURATE32PSHR(sum, 15, 32767);
-#else
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+ } else {
cubic_coef(frac, interp);
sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+ }
#endif
out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -521,33 +527,37 @@ static int resampler_basic_interpolate_d
#ifdef FIXED_POINT
const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
#else
const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
#endif
spx_word16_t interp[4];
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+ if (!moz_speex_have_double_simd()) {
+#endif
int j;
double accum[4] = {0,0,0,0};
for(j=0;j<N;j++) {
const double curr_in=iptr[j];
accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
}
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]);
-#else
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+ } else {
cubic_coef(frac, interp);
sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+ }
#endif
out[out_stride * out_sample++] = PSHR32(sum,15);
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
diff --git a/media/libspeex_resampler/src/resample_neon.c b/media/libspeex_resampler/src/resample_neon.c
--- a/media/libspeex_resampler/src/resample_neon.c
+++ b/media/libspeex_resampler/src/resample_neon.c
@@ -31,16 +31,18 @@
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
+#include "simd_detect.h"
+
#include <arm_neon.h>
#ifdef FIXED_POINT
#ifdef __thumb2__
static inline int32_t saturate_32bit_to_16bit(int32_t a) {
int32_t ret;
asm ("ssat %[ret], #16, %[a]"
: [ret] "=&r" (ret)
@@ -60,17 +62,17 @@ static inline int32_t saturate_32bit_to_
return ret;
}
#endif
#undef WORD2INT
#define WORD2INT(x) (saturate_32bit_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
-static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
+int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
{
int32_t ret;
uint32_t remainder = len % 16;
len = len - remainder;
asm volatile (" cmp %[len], #0\n"
" bne 1f\n"
" vld1.16 {d16}, [%[b]]!\n"
@@ -134,17 +136,17 @@ static inline int32_t saturate_float_to_
: "q0");
return ret;
}
#undef WORD2INT
#define WORD2INT(x) (saturate_float_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
-static inline float inner_product_single(const float *a, const float *b, unsigned int len)
+float inner_product_single(const float *a, const float *b, unsigned int len)
{
float ret;
uint32_t remainder = len % 16;
len = len - remainder;
asm volatile (" cmp %[len], #0\n"
" bne 1f\n"
" vld1.32 {q4}, [%[b]]!\n"
diff --git a/media/libspeex_resampler/src/resample_sse.c b/media/libspeex_resampler/src/resample_sse.c
--- a/media/libspeex_resampler/src/resample_sse.c
+++ b/media/libspeex_resampler/src/resample_sse.c
@@ -29,37 +29,39 @@
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
+#include "simd_detect.h"
+
#include <xmmintrin.h>
#define OVERRIDE_INNER_PRODUCT_SINGLE
-static inline float inner_product_single(const float *a, const float *b, unsigned int len)
+float inner_product_single(const float *a, const float *b, unsigned int len)
{
int i;
float ret;
__m128 sum = _mm_setzero_ps();
for (i=0;i<len;i+=8)
{
sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i)));
sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4)));
}
sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
_mm_store_ss(&ret, sum);
return ret;
}
#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
-static inline float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
+float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
int i;
float ret;
__m128 sum = _mm_setzero_ps();
__m128 f = _mm_loadu_ps(frac);
for(i=0;i<len;i+=2)
{
sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample)));
sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample)));
@@ -70,17 +72,17 @@ static inline float interpolate_product_
_mm_store_ss(&ret, sum);
return ret;
}
#ifdef _USE_SSE2
#include <emmintrin.h>
#define OVERRIDE_INNER_PRODUCT_DOUBLE
-static inline double inner_product_double(const float *a, const float *b, unsigned int len)
+double inner_product_double(const float *a, const float *b, unsigned int len)
{
int i;
double ret;
__m128d sum = _mm_setzero_pd();
__m128 t;
for (i=0;i<len;i+=8)
{
t = _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i));
@@ -92,17 +94,17 @@ static inline double inner_product_doubl
sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
}
sum = _mm_add_sd(sum, _mm_unpackhi_pd(sum, sum));
_mm_store_sd(&ret, sum);
return ret;
}
#define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
-static inline double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
+double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
int i;
double ret;
__m128d sum;
__m128d sum1 = _mm_setzero_pd();
__m128d sum2 = _mm_setzero_pd();
__m128 f = _mm_loadu_ps(frac);
__m128d f1 = _mm_cvtps_pd(f);
__m128d f2 = _mm_cvtps_pd(_mm_movehl_ps(f,f));

10
media/libspeex_resampler/src/moz.build
View File

@ -10,12 +10,9 @@ EXPORTS.speex += [
'speex_resampler.h',
]
SOURCES += [
'sse_detect.cpp',
]
SOURCES += [
'resample.c',
'simd_detect.cpp',
]
MSVC_ENABLE_PGO = True
@ -40,7 +37,10 @@ else:
if CONFIG['INTEL_ARCHITECTURE'] and not CONFIG['MOZ_SAMPLE_TYPE_S16']:
DEFINES['_USE_SSE'] = True
DEFINES['_USE_SSE2'] = True
SOURCES['resample.c'].flags += CONFIG['SSE2_FLAGS']
SOURCES += [
'resample_sse.c'
]
SOURCES['resample_sse.c'].flags += CONFIG['SSE2_FLAGS']
# Suppress warnings in third-party code.
if CONFIG['GNU_CC']:

26
media/libspeex_resampler/src/resample.c
View File

@ -97,23 +97,7 @@ static void speex_free (void *ptr) {free(ptr);}
#define NULL 0
#endif
#include "sse_detect.h"
/* We compile SSE code on x86 all the time, but we only use it if we find at
* runtime that the CPU supports it. */
#ifdef _USE_SSE
#ifdef _MSC_VER
#define inline __inline
#endif
#include "resample_sse.h"
#ifdef _MSC_VER
#undef inline
#endif
#endif
#ifdef _USE_NEON
#include "resample_neon.h"
#endif
#include "simd_detect.h"
/* Numer of elements to allocate on the stack */
#ifdef VAR_ARRAYS
@ -360,7 +344,7 @@ static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t c
const spx_word16_t *iptr = & in[last_sample];
#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
if (!moz_has_sse()) {
if (!moz_speex_have_single_simd()) {
#endif
int j;
sum = 0;
@ -422,7 +406,7 @@ static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t c
const spx_word16_t *iptr = & in[last_sample];
#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
if(moz_has_sse2()) {
if(moz_speex_have_double_simd()) {
#endif
int j;
double accum[4] = {0,0,0,0};
@ -482,7 +466,7 @@ static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint3
#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
if (!moz_has_sse()) {
if (!moz_speex_have_single_simd()) {
#endif
int j;
spx_word32_t accum[4] = {0,0,0,0};
@ -549,7 +533,7 @@ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint3
#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
if (!moz_has_sse2()) {
if (!moz_speex_have_double_simd()) {
#endif
int j;
double accum[4] = {0,0,0,0};

6
media/libspeex_resampler/src/resample_neon.h → media/libspeex_resampler/src/resample_neon.c
View File

@ -36,6 +36,8 @@
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "simd_detect.h"
#include <arm_neon.h>
#ifdef FIXED_POINT
@ -65,7 +67,7 @@ static inline int32_t saturate_32bit_to_16bit(int32_t a) {
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
{
int32_t ret;
uint32_t remainder = len % 16;
@ -139,7 +141,7 @@ static inline int32_t saturate_float_to_16bit(float a) {
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
static inline float inner_product_single(const float *a, const float *b, unsigned int len)
float inner_product_single(const float *a, const float *b, unsigned int len)
{
float ret;
uint32_t remainder = len % 16;

10
media/libspeex_resampler/src/resample_sse.h → media/libspeex_resampler/src/resample_sse.c
View File

@ -34,10 +34,12 @@
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "simd_detect.h"
#include <xmmintrin.h>
#define OVERRIDE_INNER_PRODUCT_SINGLE
static inline float inner_product_single(const float *a, const float *b, unsigned int len)
float inner_product_single(const float *a, const float *b, unsigned int len)
{
int i;
float ret;
@ -54,7 +56,7 @@ static inline float inner_product_single(const float *a, const float *b, unsigne
}
#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
static inline float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
int i;
float ret;
__m128 sum = _mm_setzero_ps();
@ -75,7 +77,7 @@ static inline float interpolate_product_single(const float *a, const float *b, u
#include <emmintrin.h>
#define OVERRIDE_INNER_PRODUCT_DOUBLE
static inline double inner_product_double(const float *a, const float *b, unsigned int len)
double inner_product_double(const float *a, const float *b, unsigned int len)
{
int i;
double ret;
@ -97,7 +99,7 @@ static inline double inner_product_double(const float *a, const float *b, unsign
}
#define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
static inline double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
int i;
double ret;
__m128d sum;

20
media/libspeex_resampler/src/sse_detect.cpp → media/libspeex_resampler/src/simd_detect.cpp
View File

@ -3,13 +3,25 @@
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "mozilla/SSE.h"
#include "sse_detect.h"
#include "simd_detect.h"
int moz_has_sse2() {
#include "mozilla/SSE.h"
#include "mozilla/arm.h"
#ifdef _USE_SSE2
int moz_speex_have_double_simd() {
return mozilla::supports_sse2() ? 1 : 0;
}
#endif
int moz_has_sse() {
#ifdef _USE_SSE
int moz_speex_have_single_simd() {
return mozilla::supports_sse() ? 1 : 0;
}
#endif
#ifdef _USE_NEON
int moz_speex_have_single_simd() {
return mozilla::supports_neon() ? 1 : 0;
}
#endif

43
media/libspeex_resampler/src/simd_detect.h Normal file
View File

@ -0,0 +1,43 @@
/* vim: set shiftwidth=2 tabstop=8 autoindent cindent expandtab: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef simd_detect_h
#define simd_detect_h
#include "speex_resampler.h"
#include "arch.h"
#ifdef __cplusplus
extern "C" {
#endif
int moz_speex_have_single_simd();
int moz_speex_have_double_simd();
#if defined(_USE_SSE) || defined(_USE_NEON)
#define OVERRIDE_INNER_PRODUCT_SINGLE
#define inner_product_single CAT_PREFIX(RANDOM_PREFIX,_inner_product_single)
spx_word32_t inner_product_single(const spx_word16_t *a, const spx_word16_t *b, unsigned int len);
#endif
#if defined(_USE_SSE)
#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
#define interpolate_product_single CAT_PREFIX(RANDOM_PREFIX,_interpolate_product_single)
spx_word32_t interpolate_product_single(const spx_word16_t *a, const spx_word16_t *b, unsigned int len, const spx_uint32_t oversample, float *frac);
#endif
#if defined(_USE_SSE2)
#define OVERRIDE_INNER_PRODUCT_DOUBLE
#define inner_product_double CAT_PREFIX(RANDOM_PREFIX,_inner_product_double)
double inner_product_double(const float *a, const float *b, unsigned int len);
#define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
#define interpolate_product_double CAT_PREFIX(RANDOM_PREFIX,_interpolate_product_double)
double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac);
#endif
#ifdef __cplusplus
}
#endif
#endif // simd_detect_h

20
media/libspeex_resampler/src/sse_detect.h
View File

@ -1,20 +0,0 @@
/* vim: set shiftwidth=2 tabstop=8 autoindent cindent expandtab: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef SSE_DETECT
#define SSE_DETECT
#ifdef __cplusplus
extern "C" {
#endif
int moz_has_sse2();
int moz_has_sse();
#ifdef __cplusplus
}
#endif
#endif // SSE_DETECT

192
media/libspeex_resampler/sse-detect-runtime.patch
View File

@ -1,192 +0,0 @@
diff --git a/media/libspeex_resampler/src/resample.c b/media/libspeex_resampler/src/resample.c
--- a/media/libspeex_resampler/src/resample.c
+++ b/media/libspeex_resampler/src/resample.c
@@ -90,18 +90,28 @@ static void speex_free (void *ptr) {free
#define IMAX(a,b) ((a) > (b) ? (a) : (b))
#define IMIN(a,b) ((a) < (b) ? (a) : (b))
#ifndef NULL
#define NULL 0
#endif
+#include "sse_detect.h"
+
+/* We compile SSE code on x86 all the time, but we only use it if we find at
+ * runtime that the CPU supports it. */
#ifdef _USE_SSE
+#ifdef _MSC_VER
+#define inline __inline
+#endif
#include "resample_sse.h"
+#ifdef _MSC_VER
+#undef inline
+#endif
#endif
#ifdef _USE_NEON
#include "resample_neon.h"
#endif
/* Numer of elements to allocate on the stack */
#ifdef VAR_ARRAYS
@@ -342,17 +352,19 @@ static int resampler_basic_direct_single
const spx_uint32_t den_rate = st->den_rate;
spx_word32_t sum;
while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
{
const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
const spx_word16_t *iptr = & in[last_sample];
-#ifndef OVERRIDE_INNER_PRODUCT_SINGLE
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
+ if (!moz_has_sse()) {
+#endif
int j;
sum = 0;
for(j=0;j<N;j++) sum += MULT16_16(sinct[j], iptr[j]);
/* This code is slower on most DSPs which have only 2 accumulators.
Plus this this forces truncation to 32 bits and you lose the HW guard bits.
I think we can trust the compiler and let it vectorize and/or unroll itself.
spx_word32_t accum[4] = {0,0,0,0};
@@ -360,18 +372,20 @@ static int resampler_basic_direct_single
accum[0] += MULT16_16(sinct[j], iptr[j]);
accum[1] += MULT16_16(sinct[j+1], iptr[j+1]);
accum[2] += MULT16_16(sinct[j+2], iptr[j+2]);
accum[3] += MULT16_16(sinct[j+3], iptr[j+3]);
}
sum = accum[0] + accum[1] + accum[2] + accum[3];
*/
sum = SATURATE32PSHR(sum, 15, 32767);
-#else
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
+ } else {
sum = inner_product_single(sinct, iptr, N);
+ }
#endif
out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -400,29 +414,33 @@ static int resampler_basic_direct_double
const spx_uint32_t den_rate = st->den_rate;
double sum;
while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
{
const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
const spx_word16_t *iptr = & in[last_sample];
-#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
+ if(moz_has_sse2()) {
+#endif
int j;
double accum[4] = {0,0,0,0};
for(j=0;j<N;j+=4) {
accum[0] += sinct[j]*iptr[j];
accum[1] += sinct[j+1]*iptr[j+1];
accum[2] += sinct[j+2]*iptr[j+2];
accum[3] += sinct[j+3]*iptr[j+3];
}
sum = accum[0] + accum[1] + accum[2] + accum[3];
-#else
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
+ } else {
sum = inner_product_double(sinct, iptr, N);
+ }
#endif
out[out_stride * out_sample++] = PSHR32(sum, 15);
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -456,34 +474,38 @@ static int resampler_basic_interpolate_s
#ifdef FIXED_POINT
const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
#else
const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
#endif
spx_word16_t interp[4];
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+ if (!moz_has_sse()) {
+#endif
int j;
spx_word32_t accum[4] = {0,0,0,0};
for(j=0;j<N;j++) {
const spx_word16_t curr_in=iptr[j];
accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
}
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1));
sum = SATURATE32PSHR(sum, 15, 32767);
-#else
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+ } else {
cubic_coef(frac, interp);
sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+ }
#endif
out[out_stride * out_sample++] = sum;
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;
@@ -519,33 +541,37 @@ static int resampler_basic_interpolate_d
#ifdef FIXED_POINT
const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
#else
const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
#endif
spx_word16_t interp[4];
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+ if (!moz_has_sse2()) {
+#endif
int j;
double accum[4] = {0,0,0,0};
for(j=0;j<N;j++) {
const double curr_in=iptr[j];
accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
}
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]);
-#else
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+ } else {
cubic_coef(frac, interp);
sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+ }
#endif
out[out_stride * out_sample++] = PSHR32(sum,15);
last_sample += int_advance;
samp_frac_num += frac_advance;
if (samp_frac_num >= den_rate)
{
samp_frac_num -= den_rate;

6
media/libspeex_resampler/update.sh
View File

@ -10,8 +10,8 @@
set -e -x
cp $1/libspeexdsp/resample.c src
cp $1/libspeexdsp/resample_sse.h src
cp $1/libspeexdsp/resample_neon.h src
cp $1/libspeexdsp/resample_sse.h src/resample_sse.c
cp $1/libspeexdsp/resample_neon.h src/resample_neon.c
cp $1/libspeexdsp/arch.h src
cp $1/libspeexdsp/stack_alloc.h src
cp $1/libspeexdsp/fixed_generic.h src
@ -21,6 +21,6 @@ cp $1/COPYING .
# apply outstanding local patches
patch -p3 < outside-speex.patch
patch -p3 < sse-detect-runtime.patch
patch -p3 < simd-detect-runtime.patch
patch -p3 < set-skip-frac.patch
patch -p3 < hugemem.patch