1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
|
diff --git a/src/resample.c b/src/resample.c
--- a/src/resample.c
+++ b/src/resample.c
@@ -91,23 +91,17 @@ static void speex_free(void *ptr) {free(
#ifndef NULL
#define NULL 0
#endif
#ifndef UINT32_MAX
#define UINT32_MAX 4294967295U
#endif
-#ifdef USE_SSE
-#include "resample_sse.h"
-#endif
-
-#ifdef USE_NEON
-#include "resample_neon.h"
-#endif
+#include "simd_detect.h"
/* Number of elements to allocate on the stack */
#ifdef VAR_ARRAYS
#define FIXED_STACK_ALLOC 8192
#else
#define FIXED_STACK_ALLOC 1024
#endif
@@ -341,17 +335,19 @@ 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};
@@ -359,18 +355,20 @@ 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;
@@ -399,29 +397,33 @@ 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;
@@ -455,34 +457,38 @@ 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],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]);
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;
@@ -518,33 +524,37 @@ 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/src/resample_neon.c b/src/resample_neon.c
--- a/src/resample_neon.c
+++ b/src/resample_neon.c
@@ -32,16 +32,17 @@
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 <stdint.h>
+#include "simd_detect.h"
#ifdef FIXED_POINT
#if defined(__aarch64__)
static inline int32_t saturate_32bit_to_16bit(int32_t a) {
int32_t ret;
asm ("fmov s0, %w[a]\n"
"sqxtn h0, s0\n"
"sxtl v0.4s, v0.4h\n"
@@ -73,17 +74,17 @@
}
#endif
#undef WORD2INT
#define WORD2INT(x) (saturate_32bit_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 and len >= 4 */
#if defined(__aarch64__)
-static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
+inline 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 %w[len], #0\n"
" b.ne 1f\n"
" ld1 {v16.4h}, [%[b]], #8\n"
@@ -128,17 +129,17 @@
: [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+r" (len), [remainder] "+r" (remainder)
:
: "cc", "v0",
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
return ret;
}
#else
-static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
+inline 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"
@@ -218,17 +219,17 @@
#endif
#undef WORD2INT
#define WORD2INT(x) (saturate_float_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 and len >= 4 */
#if defined(__aarch64__)
-static inline float inner_product_single(const float *a, const float *b, unsigned int len)
+inline 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 %w[len], #0\n"
" b.ne 1f\n"
" ld1 {v16.4s}, [%[b]], #16\n"
@@ -273,17 +274,17 @@
: [ret] "=w" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+r" (len), [remainder] "+r" (remainder)
:
: "cc", "v1", "v2", "v3", "v4",
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
return ret;
}
#else
-static inline float inner_product_single(const float *a, const float *b, unsigned int len)
+inline 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/src/resample_sse.c b/src/resample_sse.c
--- a/src/resample_sse.c
+++ b/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));
|
