mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-29 07:42:04 +00:00
b=1042504 update speex resampler to speexdsp 305e54ea r=padenot
Upstream SATURATE32PSHR changes conflict with Gecko's SSE runtime detection patch. That patch is updated to reduce deviation from upstream. --HG-- extra : rebase_source : fdaf355f802b944fcbe286e076ca19dc6bc33f80
This commit is contained in:
Karl Tomlinson
parent
3f4977b36d
commit
a3e81ff91a
@ -1,5 +1,5 @@
|
||||
This source is from the Speex DSP library
|
||||
(http://git.xiph.org/?p=speexdsp.git), from commit bbe7e099.
|
||||
(http://git.xiph.org/?p=speexdsp.git), from commit 305e54ea.
|
||||
|
||||
It consists in the audio resampling code (resampler.c) and its header files
|
||||
dependancies, imported into the tree using the update.sh script.
|
||||
|
||||
@ -163,6 +163,7 @@ typedef float spx_word32_t;
|
||||
#define VSHR32(a,shift) (a)
|
||||
#define SATURATE16(x,a) (x)
|
||||
#define SATURATE32(x,a) (x)
|
||||
#define SATURATE32PSHR(x,shift,a) (x)
|
||||
|
||||
#define PSHR(a,shift) (a)
|
||||
#define SHR(a,shift) (a)
|
||||
|
||||
@ -52,6 +52,10 @@
|
||||
#define SATURATE16(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x)))
|
||||
#define SATURATE32(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x)))
|
||||
|
||||
#define SATURATE32PSHR(x,shift,a) (((x)>=(SHL32(a,shift))) ? (a) : \
|
||||
(x)<=-(SHL32(a,shift)) ? -(a) : \
|
||||
(PSHR32(x, shift)))
|
||||
|
||||
#define SHR(a,shift) ((a) >> (shift))
|
||||
#define SHL(a,shift) ((spx_word32_t)(a) << (shift))
|
||||
#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift))
|
||||
|
||||
@ -79,7 +79,7 @@ static void speex_free (void *ptr) {free(ptr);}
|
||||
#include <math.h>
|
||||
|
||||
#ifndef M_PI
|
||||
#define M_PI 3.14159263
|
||||
#define M_PI 3.14159265358979323846
|
||||
#endif
|
||||
|
||||
#ifdef FIXED_POINT
|
||||
@ -109,6 +109,10 @@ static void speex_free (void *ptr) {free(ptr);}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifdef _USE_NEON
|
||||
#include "resample_neon.h"
|
||||
#endif
|
||||
|
||||
/* Numer of elements to allocate on the stack */
|
||||
#ifdef VAR_ARRAYS
|
||||
#define FIXED_STACK_ALLOC 8192
|
||||
@ -354,9 +358,7 @@ 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()) {
|
||||
sum = inner_product_single(sinct, iptr, N);
|
||||
} else {
|
||||
if (!moz_has_sse()) {
|
||||
#endif
|
||||
int j;
|
||||
sum = 0;
|
||||
@ -374,11 +376,14 @@ static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t c
|
||||
}
|
||||
sum = accum[0] + accum[1] + accum[2] + accum[3];
|
||||
*/
|
||||
sum = SATURATE32PSHR(sum, 15, 32767);
|
||||
#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
|
||||
}
|
||||
} else {
|
||||
sum = inner_product_single(sinct, iptr, N);
|
||||
}
|
||||
#endif
|
||||
|
||||
out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 15), 32767);
|
||||
out[out_stride * out_sample++] = sum;
|
||||
last_sample += int_advance;
|
||||
samp_frac_num += frac_advance;
|
||||
if (samp_frac_num >= den_rate)
|
||||
@ -416,20 +421,20 @@ static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t c
|
||||
|
||||
#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
if(moz_has_sse2()) {
|
||||
sum = inner_product_double(sinct, iptr, N);
|
||||
} else {
|
||||
#endif
|
||||
int j;
|
||||
double accum[4] = {0,0,0,0};
|
||||
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];
|
||||
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];
|
||||
#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
} else {
|
||||
sum = inner_product_double(sinct, iptr, N);
|
||||
}
|
||||
#endif
|
||||
|
||||
@ -475,28 +480,30 @@ static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint3
|
||||
|
||||
|
||||
#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
|
||||
if (moz_has_sse()) {
|
||||
cubic_coef(frac, interp);
|
||||
sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
} else {
|
||||
if (!moz_has_sse()) {
|
||||
#endif
|
||||
int j;
|
||||
spx_word32_t accum[4] = {0,0,0,0};
|
||||
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));
|
||||
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);
|
||||
#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++] = SATURATE32(PSHR32(sum, 14), 32767);
|
||||
|
||||
out[out_stride * out_sample++] = sum;
|
||||
last_sample += int_advance;
|
||||
samp_frac_num += frac_advance;
|
||||
if (samp_frac_num >= den_rate)
|
||||
@ -540,10 +547,7 @@ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint3
|
||||
|
||||
|
||||
#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
|
||||
if (moz_has_sse2()) {
|
||||
cubic_coef(frac, interp);
|
||||
sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
} else {
|
||||
if (!moz_has_sse2()) {
|
||||
#endif
|
||||
int j;
|
||||
double accum[4] = {0,0,0,0};
|
||||
@ -558,9 +562,13 @@ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint3
|
||||
|
||||
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]);
|
||||
#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
#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;
|
||||
@ -579,9 +587,10 @@ static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint3
|
||||
|
||||
static void update_filter(SpeexResamplerState *st)
|
||||
{
|
||||
spx_uint32_t old_length;
|
||||
|
||||
old_length = st->filt_len;
|
||||
spx_uint32_t old_length = st->filt_len;
|
||||
spx_uint32_t old_alloc_size = st->mem_alloc_size;
|
||||
spx_uint32_t min_alloc_size;
|
||||
|
||||
st->oversample = quality_map[st->quality].oversample;
|
||||
st->filt_len = quality_map[st->quality].base_length;
|
||||
|
||||
@ -609,12 +618,14 @@ static void update_filter(SpeexResamplerState *st)
|
||||
}
|
||||
|
||||
/* Choose the resampling type that requires the least amount of memory */
|
||||
#ifdef RESAMPLE_FULL_SINC_TABLE
|
||||
if (1)
|
||||
#else
|
||||
if (st->filt_len*st->den_rate <= st->filt_len*st->oversample+8)
|
||||
#endif
|
||||
{
|
||||
spx_uint32_t i;
|
||||
if (!st->sinc_table)
|
||||
st->sinc_table = (spx_word16_t *)speex_alloc(st->filt_len*st->den_rate*sizeof(spx_word16_t));
|
||||
else if (st->sinc_table_length < st->filt_len*st->den_rate)
|
||||
if (st->sinc_table_length < st->filt_len*st->den_rate)
|
||||
{
|
||||
st->sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,st->filt_len*st->den_rate*sizeof(spx_word16_t));
|
||||
st->sinc_table_length = st->filt_len*st->den_rate;
|
||||
@ -638,9 +649,7 @@ static void update_filter(SpeexResamplerState *st)
|
||||
/*fprintf (stderr, "resampler uses direct sinc table and normalised cutoff %f\n", cutoff);*/
|
||||
} else {
|
||||
spx_int32_t i;
|
||||
if (!st->sinc_table)
|
||||
st->sinc_table = (spx_word16_t *)speex_alloc((st->filt_len*st->oversample+8)*sizeof(spx_word16_t));
|
||||
else if (st->sinc_table_length < st->filt_len*st->oversample+8)
|
||||
if (st->sinc_table_length < st->filt_len*st->oversample+8)
|
||||
{
|
||||
st->sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,(st->filt_len*st->oversample+8)*sizeof(spx_word16_t));
|
||||
st->sinc_table_length = st->filt_len*st->oversample+8;
|
||||
@ -664,36 +673,26 @@ static void update_filter(SpeexResamplerState *st)
|
||||
/* Here's the place where we update the filter memory to take into account
|
||||
the change in filter length. It's probably the messiest part of the code
|
||||
due to handling of lots of corner cases. */
|
||||
if (!st->mem)
|
||||
min_alloc_size = st->filt_len-1 + st->buffer_size;
|
||||
if (min_alloc_size > st->mem_alloc_size)
|
||||
{
|
||||
st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*min_alloc_size * sizeof(spx_word16_t));
|
||||
st->mem_alloc_size = min_alloc_size;
|
||||
}
|
||||
if (!st->started)
|
||||
{
|
||||
spx_uint32_t i;
|
||||
st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
|
||||
st->mem = (spx_word16_t*)speex_alloc(st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
|
||||
for (i=0;i<st->nb_channels*st->mem_alloc_size;i++)
|
||||
st->mem[i] = 0;
|
||||
/*speex_warning("init filter");*/
|
||||
} else if (!st->started)
|
||||
{
|
||||
spx_uint32_t i;
|
||||
st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
|
||||
st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
|
||||
for (i=0;i<st->nb_channels*st->mem_alloc_size;i++)
|
||||
st->mem[i] = 0;
|
||||
/*speex_warning("reinit filter");*/
|
||||
} else if (st->filt_len > old_length)
|
||||
{
|
||||
spx_int32_t i;
|
||||
spx_uint32_t i;
|
||||
/* Increase the filter length */
|
||||
/*speex_warning("increase filter size");*/
|
||||
int old_alloc_size = st->mem_alloc_size;
|
||||
if ((st->filt_len-1 + st->buffer_size) > st->mem_alloc_size)
|
||||
for (i=st->nb_channels;i--;)
|
||||
{
|
||||
st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
|
||||
st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
|
||||
}
|
||||
for (i=st->nb_channels-1;i>=0;i--)
|
||||
{
|
||||
spx_int32_t j;
|
||||
spx_uint32_t j;
|
||||
spx_uint32_t olen = old_length;
|
||||
/*if (st->magic_samples[i])*/
|
||||
{
|
||||
@ -701,7 +700,7 @@ static void update_filter(SpeexResamplerState *st)
|
||||
|
||||
/* FIXME: This is wrong but for now we need it to avoid going over the array bounds */
|
||||
olen = old_length + 2*st->magic_samples[i];
|
||||
for (j=old_length-2+st->magic_samples[i];j>=0;j--)
|
||||
for (j=old_length-1+st->magic_samples[i];j--;)
|
||||
st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]] = st->mem[i*old_alloc_size+j];
|
||||
for (j=0;j<st->magic_samples[i];j++)
|
||||
st->mem[i*st->mem_alloc_size+j] = 0;
|
||||
|
||||
@ -1,5 +1,6 @@
|
||||
--- a/src/resample.c 2014-07-01 17:25:53.999320032 +1200
|
||||
+++ b/src/resample.c 2014-07-01 17:42:18.822611775 +1200
|
||||
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))
|
||||
@ -23,13 +24,13 @@
|
||||
+#endif
|
||||
#endif
|
||||
|
||||
#ifdef _USE_NEON
|
||||
#include "resample_neon.h"
|
||||
#endif
|
||||
|
||||
/* Numer of elements to allocate on the stack */
|
||||
#ifdef VAR_ARRAYS
|
||||
#define FIXED_STACK_ALLOC 8192
|
||||
#else
|
||||
#define FIXED_STACK_ALLOC 1024
|
||||
#endif
|
||||
@@ -338,35 +348,39 @@ static int resampler_basic_direct_single
|
||||
@@ -342,17 +352,19 @@ static int resampler_basic_direct_single
|
||||
const spx_uint32_t den_rate = st->den_rate;
|
||||
spx_word32_t sum;
|
||||
|
||||
@ -40,9 +41,7 @@
|
||||
|
||||
-#ifndef OVERRIDE_INNER_PRODUCT_SINGLE
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
|
||||
+ if (moz_has_sse()) {
|
||||
+ sum = inner_product_single(sinct, iptr, N);
|
||||
+ } else {
|
||||
+ if (!moz_has_sse()) {
|
||||
+#endif
|
||||
int j;
|
||||
sum = 0;
|
||||
@ -52,7 +51,7 @@
|
||||
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};
|
||||
for(j=0;j<N;j+=4) {
|
||||
@@ -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]);
|
||||
@ -60,19 +59,21 @@
|
||||
}
|
||||
sum = accum[0] + accum[1] + accum[2] + accum[3];
|
||||
*/
|
||||
sum = SATURATE32PSHR(sum, 15, 32767);
|
||||
-#else
|
||||
- sum = inner_product_single(sinct, iptr, N);
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_SINGLE
|
||||
+ }
|
||||
+ } else {
|
||||
sum = inner_product_single(sinct, iptr, N);
|
||||
+ }
|
||||
#endif
|
||||
|
||||
out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 15), 32767);
|
||||
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;
|
||||
@@ -395,29 +409,33 @@ static int resampler_basic_direct_double
|
||||
@@ -400,29 +414,33 @@ static int resampler_basic_direct_double
|
||||
const spx_uint32_t den_rate = st->den_rate;
|
||||
double sum;
|
||||
|
||||
@ -82,34 +83,24 @@
|
||||
const spx_word16_t *iptr = & in[last_sample];
|
||||
|
||||
-#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
- 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];
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
+ if(moz_has_sse2()) {
|
||||
+ sum = inner_product_double(sinct, iptr, N);
|
||||
+ } else {
|
||||
+#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];
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
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];
|
||||
sum = accum[0] + accum[1] + accum[2] + accum[3];
|
||||
-#else
|
||||
- sum = inner_product_double(sinct, iptr, N);
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
+ } else {
|
||||
sum = inner_product_double(sinct, iptr, N);
|
||||
+ }
|
||||
#endif
|
||||
|
||||
out[out_stride * out_sample++] = PSHR32(sum, 15);
|
||||
@ -118,7 +109,7 @@
|
||||
if (samp_frac_num >= den_rate)
|
||||
{
|
||||
samp_frac_num -= den_rate;
|
||||
@@ -451,35 +469,38 @@ static int resampler_basic_interpolate_s
|
||||
@@ -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
|
||||
@ -128,53 +119,38 @@
|
||||
|
||||
|
||||
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
|
||||
- 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]);
|
||||
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
|
||||
+ if (moz_has_sse()) {
|
||||
+ cubic_coef(frac, interp);
|
||||
+ sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
+ } else {
|
||||
+ 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));
|
||||
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
|
||||
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));
|
||||
|
||||
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
|
||||
- cubic_coef(frac, interp);
|
||||
- sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
+#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++] = SATURATE32(PSHR32(sum, 14), 32767);
|
||||
|
||||
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;
|
||||
last_sample++;
|
||||
}
|
||||
@@ -513,35 +534,38 @@ static int resampler_basic_interpolate_d
|
||||
@@ -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
|
||||
@ -185,10 +161,7 @@
|
||||
|
||||
-#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
|
||||
+#ifdef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
|
||||
+ if (moz_has_sse2()) {
|
||||
+ cubic_coef(frac, interp);
|
||||
+ sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
+ } else {
|
||||
+ if (!moz_has_sse2()) {
|
||||
+#endif
|
||||
int j;
|
||||
double accum[4] = {0,0,0,0};
|
||||
@ -204,17 +177,16 @@
|
||||
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
|
||||
- cubic_coef(frac, interp);
|
||||
- sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
|
||||
+#ifdef OVERRIDE_INNER_PRODUCT_DOUBLE
|
||||
+#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;
|
||||
last_sample++;
|
||||
}
|
||||
|
||||
@ -20,5 +20,5 @@ cp $1/COPYING .
|
||||
|
||||
# apply outstanding local patches
|
||||
patch -p3 < outside-speex.patch
|
||||
patch -p1 < sse-detect-runtime.patch
|
||||
patch -p3 < sse-detect-runtime.patch
|
||||
patch -p3 < set-skip-frac.patch
|
||||
|
||||
Loading…
Reference in New Issue
Block a user