|
sl@0
|
1 |
/*
|
|
sl@0
|
2 |
* Copyright (c) 2005
|
|
sl@0
|
3 |
* Eric Anholt. All rights reserved.
|
|
sl@0
|
4 |
*
|
|
sl@0
|
5 |
* Redistribution and use in source and binary forms, with or without
|
|
sl@0
|
6 |
* modification, are permitted provided that the following conditions
|
|
sl@0
|
7 |
* are met:
|
|
sl@0
|
8 |
* 1. Redistributions of source code must retain the above copyright
|
|
sl@0
|
9 |
* notice, this list of conditions and the following disclaimer.
|
|
sl@0
|
10 |
* 2. Redistributions in binary form must reproduce the above copyright
|
|
sl@0
|
11 |
* notice, this list of conditions and the following disclaimer in the
|
|
sl@0
|
12 |
* documentation and/or other materials provided with the distribution.
|
|
sl@0
|
13 |
*
|
|
sl@0
|
14 |
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
|
|
sl@0
|
15 |
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
sl@0
|
16 |
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
sl@0
|
17 |
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE
|
|
sl@0
|
18 |
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
sl@0
|
19 |
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
sl@0
|
20 |
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
sl@0
|
21 |
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
sl@0
|
22 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
sl@0
|
23 |
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
sl@0
|
24 |
* SUCH DAMAGE.
|
|
sl@0
|
25 |
*/
|
|
sl@0
|
26 |
//Portions Copyright (c) 2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved.
|
|
sl@0
|
27 |
|
|
sl@0
|
28 |
#ifdef HAVE_CONFIG_H
|
|
sl@0
|
29 |
#include "config.h"
|
|
sl@0
|
30 |
#endif
|
|
sl@0
|
31 |
#include <liboilclasses.h>
|
|
sl@0
|
32 |
#include <liboilfunction.h>
|
|
sl@0
|
33 |
#include <emmintrin.h>
|
|
sl@0
|
34 |
#include <xmmintrin.h>
|
|
sl@0
|
35 |
|
|
sl@0
|
36 |
#define SSE_FUNCTION __attribute__((force_align_arg_pointer))
|
|
sl@0
|
37 |
|
|
sl@0
|
38 |
SSE_FUNCTION static void
|
|
sl@0
|
39 |
add_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
40 |
{
|
|
sl@0
|
41 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
42 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
43 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
44 |
}
|
|
sl@0
|
45 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
46 |
__m128 xmm0, xmm1;
|
|
sl@0
|
47 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
48 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
49 |
xmm0 = _mm_add_ps(xmm0, xmm1);
|
|
sl@0
|
50 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
51 |
dest += 4;
|
|
sl@0
|
52 |
src1 += 4;
|
|
sl@0
|
53 |
src2 += 4;
|
|
sl@0
|
54 |
}
|
|
sl@0
|
55 |
for (; n > 0; n--) {
|
|
sl@0
|
56 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
57 |
}
|
|
sl@0
|
58 |
}
|
|
sl@0
|
59 |
OIL_DEFINE_IMPL_FULL (add_f32_sse, add_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
60 |
|
|
sl@0
|
61 |
SSE_FUNCTION static void
|
|
sl@0
|
62 |
add_f64_sse2 (double *dest, double *src1, double *src2, int n)
|
|
sl@0
|
63 |
{
|
|
sl@0
|
64 |
__m128d xmm0, xmm1;
|
|
sl@0
|
65 |
while (((long)dest & 15) && (0 < n)) {
|
|
sl@0
|
66 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
67 |
n--;
|
|
sl@0
|
68 |
}
|
|
sl@0
|
69 |
while (1 < n) {
|
|
sl@0
|
70 |
xmm0 = _mm_loadu_pd(src1);
|
|
sl@0
|
71 |
xmm1 = _mm_loadu_pd(src2);
|
|
sl@0
|
72 |
xmm0 = _mm_add_pd(xmm0, xmm1);
|
|
sl@0
|
73 |
_mm_store_pd(dest, xmm0);
|
|
sl@0
|
74 |
dest += 2;
|
|
sl@0
|
75 |
src1 += 2;
|
|
sl@0
|
76 |
src2 += 2;
|
|
sl@0
|
77 |
n -= 2;
|
|
sl@0
|
78 |
}
|
|
sl@0
|
79 |
while (0 < n) {
|
|
sl@0
|
80 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
81 |
n--;
|
|
sl@0
|
82 |
}
|
|
sl@0
|
83 |
}
|
|
sl@0
|
84 |
OIL_DEFINE_IMPL_FULL (add_f64_sse2, add_f64, OIL_IMPL_FLAG_SSE2);
|
|
sl@0
|
85 |
|
|
sl@0
|
86 |
SSE_FUNCTION static void
|
|
sl@0
|
87 |
add_f64_sse2_unroll (double *dest, double *src1, double *src2, int n)
|
|
sl@0
|
88 |
{
|
|
sl@0
|
89 |
__m128d xmm0, xmm1;
|
|
sl@0
|
90 |
while (((long)dest & 15) && (0 < n)) {
|
|
sl@0
|
91 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
92 |
n--;
|
|
sl@0
|
93 |
}
|
|
sl@0
|
94 |
while (3 < n) {
|
|
sl@0
|
95 |
xmm0 = _mm_loadu_pd(src1);
|
|
sl@0
|
96 |
xmm1 = _mm_loadu_pd(src2);
|
|
sl@0
|
97 |
xmm0 = _mm_add_pd(xmm0, xmm1);
|
|
sl@0
|
98 |
_mm_store_pd(dest, xmm0);
|
|
sl@0
|
99 |
|
|
sl@0
|
100 |
xmm0 = _mm_loadu_pd(src1+2);
|
|
sl@0
|
101 |
xmm1 = _mm_loadu_pd(src2+2);
|
|
sl@0
|
102 |
xmm0 = _mm_add_pd(xmm0, xmm1);
|
|
sl@0
|
103 |
_mm_store_pd(dest+2, xmm0);
|
|
sl@0
|
104 |
dest += 4;
|
|
sl@0
|
105 |
src1 += 4;
|
|
sl@0
|
106 |
src2 += 4;
|
|
sl@0
|
107 |
n -= 4;
|
|
sl@0
|
108 |
}
|
|
sl@0
|
109 |
while (1 < n) {
|
|
sl@0
|
110 |
xmm0 = _mm_loadu_pd(src1);
|
|
sl@0
|
111 |
xmm1 = _mm_loadu_pd(src2);
|
|
sl@0
|
112 |
xmm0 = _mm_add_pd(xmm0, xmm1);
|
|
sl@0
|
113 |
_mm_store_pd(dest, xmm0);
|
|
sl@0
|
114 |
dest += 2;
|
|
sl@0
|
115 |
src1 += 2;
|
|
sl@0
|
116 |
src2 += 2;
|
|
sl@0
|
117 |
n -= 2;
|
|
sl@0
|
118 |
}
|
|
sl@0
|
119 |
while (0 < n) {
|
|
sl@0
|
120 |
*dest++ = *src1++ + *src2++;
|
|
sl@0
|
121 |
n--;
|
|
sl@0
|
122 |
}
|
|
sl@0
|
123 |
}
|
|
sl@0
|
124 |
OIL_DEFINE_IMPL_FULL (add_f64_sse2_unroll, add_f64, OIL_IMPL_FLAG_SSE2);
|
|
sl@0
|
125 |
|
|
sl@0
|
126 |
SSE_FUNCTION static void
|
|
sl@0
|
127 |
subtract_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
128 |
{
|
|
sl@0
|
129 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
130 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
131 |
*dest++ = *src1++ - *src2++;
|
|
sl@0
|
132 |
}
|
|
sl@0
|
133 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
134 |
__m128 xmm0, xmm1;
|
|
sl@0
|
135 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
136 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
137 |
xmm0 = _mm_sub_ps(xmm0, xmm1);
|
|
sl@0
|
138 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
139 |
dest += 4;
|
|
sl@0
|
140 |
src1 += 4;
|
|
sl@0
|
141 |
src2 += 4;
|
|
sl@0
|
142 |
}
|
|
sl@0
|
143 |
for (; n > 0; n--) {
|
|
sl@0
|
144 |
*dest++ = *src1++ - *src2++;
|
|
sl@0
|
145 |
}
|
|
sl@0
|
146 |
}
|
|
sl@0
|
147 |
OIL_DEFINE_IMPL_FULL (subtract_f32_sse, subtract_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
148 |
|
|
sl@0
|
149 |
SSE_FUNCTION static void
|
|
sl@0
|
150 |
multiply_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
151 |
{
|
|
sl@0
|
152 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
153 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
154 |
*dest++ = *src1++ * *src2++;
|
|
sl@0
|
155 |
}
|
|
sl@0
|
156 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
157 |
__m128 xmm0, xmm1;
|
|
sl@0
|
158 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
159 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
160 |
xmm0 = _mm_mul_ps(xmm0, xmm1);
|
|
sl@0
|
161 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
162 |
dest += 4;
|
|
sl@0
|
163 |
src1 += 4;
|
|
sl@0
|
164 |
src2 += 4;
|
|
sl@0
|
165 |
}
|
|
sl@0
|
166 |
for (; n > 0; n--) {
|
|
sl@0
|
167 |
*dest++ = *src1++ * *src2++;
|
|
sl@0
|
168 |
}
|
|
sl@0
|
169 |
}
|
|
sl@0
|
170 |
OIL_DEFINE_IMPL_FULL (multiply_f32_sse, multiply_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
171 |
|
|
sl@0
|
172 |
SSE_FUNCTION static void
|
|
sl@0
|
173 |
divide_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
174 |
{
|
|
sl@0
|
175 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
176 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
177 |
*dest++ = *src1++ / *src2++;
|
|
sl@0
|
178 |
}
|
|
sl@0
|
179 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
180 |
__m128 xmm0, xmm1;
|
|
sl@0
|
181 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
182 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
183 |
xmm0 = _mm_div_ps(xmm0, xmm1);
|
|
sl@0
|
184 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
185 |
dest += 4;
|
|
sl@0
|
186 |
src1 += 4;
|
|
sl@0
|
187 |
src2 += 4;
|
|
sl@0
|
188 |
}
|
|
sl@0
|
189 |
for (; n > 0; n--) {
|
|
sl@0
|
190 |
*dest++ = *src1++ / *src2++;
|
|
sl@0
|
191 |
}
|
|
sl@0
|
192 |
}
|
|
sl@0
|
193 |
OIL_DEFINE_IMPL_FULL (divide_f32_sse, divide_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
194 |
|
|
sl@0
|
195 |
SSE_FUNCTION static void
|
|
sl@0
|
196 |
minimum_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
197 |
{
|
|
sl@0
|
198 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
199 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
200 |
*dest++ = *src1 < *src2 ? *src1 : *src2;
|
|
sl@0
|
201 |
src1++;
|
|
sl@0
|
202 |
src2++;
|
|
sl@0
|
203 |
}
|
|
sl@0
|
204 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
205 |
__m128 xmm0, xmm1;
|
|
sl@0
|
206 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
207 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
208 |
xmm0 = _mm_min_ps(xmm0, xmm1);
|
|
sl@0
|
209 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
210 |
dest += 4;
|
|
sl@0
|
211 |
src1 += 4;
|
|
sl@0
|
212 |
src2 += 4;
|
|
sl@0
|
213 |
}
|
|
sl@0
|
214 |
for (; n > 0; n--) {
|
|
sl@0
|
215 |
*dest++ = *src1 < *src2 ? *src1 : *src2;
|
|
sl@0
|
216 |
src1++;
|
|
sl@0
|
217 |
src2++;
|
|
sl@0
|
218 |
}
|
|
sl@0
|
219 |
}
|
|
sl@0
|
220 |
OIL_DEFINE_IMPL_FULL (minimum_f32_sse, minimum_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
221 |
|
|
sl@0
|
222 |
SSE_FUNCTION static void
|
|
sl@0
|
223 |
maximum_f32_sse (float *dest, float *src1, float *src2, int n)
|
|
sl@0
|
224 |
{
|
|
sl@0
|
225 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
226 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
227 |
*dest++ = *src1 > *src2 ? *src1 : *src2;
|
|
sl@0
|
228 |
src1++;
|
|
sl@0
|
229 |
src2++;
|
|
sl@0
|
230 |
}
|
|
sl@0
|
231 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
232 |
__m128 xmm0, xmm1;
|
|
sl@0
|
233 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
234 |
xmm1 = _mm_loadu_ps(src2);
|
|
sl@0
|
235 |
xmm0 = _mm_max_ps(xmm0, xmm1);
|
|
sl@0
|
236 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
237 |
dest += 4;
|
|
sl@0
|
238 |
src1 += 4;
|
|
sl@0
|
239 |
src2 += 4;
|
|
sl@0
|
240 |
}
|
|
sl@0
|
241 |
for (; n > 0; n--) {
|
|
sl@0
|
242 |
*dest++ = *src1 > *src2 ? *src1 : *src2;
|
|
sl@0
|
243 |
src1++;
|
|
sl@0
|
244 |
src2++;
|
|
sl@0
|
245 |
}
|
|
sl@0
|
246 |
}
|
|
sl@0
|
247 |
OIL_DEFINE_IMPL_FULL (maximum_f32_sse, maximum_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
248 |
|
|
sl@0
|
249 |
SSE_FUNCTION static void
|
|
sl@0
|
250 |
inverse_f32_sse (float *dest, float *src1, int n)
|
|
sl@0
|
251 |
{
|
|
sl@0
|
252 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
253 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
254 |
*dest++ = 1.0 / *src1++;
|
|
sl@0
|
255 |
}
|
|
sl@0
|
256 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
257 |
__m128 xmm0, xmm1;
|
|
sl@0
|
258 |
/* While _mm_rcp_ps sounds promising, the results it gives are rather
|
|
sl@0
|
259 |
* different from the 1.0 / src1 reference implementation, so do that.
|
|
sl@0
|
260 |
*/
|
|
sl@0
|
261 |
xmm0 = _mm_set_ps1(1.0);
|
|
sl@0
|
262 |
xmm1 = _mm_loadu_ps(src1);
|
|
sl@0
|
263 |
xmm0 = _mm_div_ps(xmm0, xmm1);
|
|
sl@0
|
264 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
265 |
dest += 4;
|
|
sl@0
|
266 |
src1 += 4;
|
|
sl@0
|
267 |
}
|
|
sl@0
|
268 |
for (; n > 0; n--) {
|
|
sl@0
|
269 |
*dest++ = 1.0 / *src1++;
|
|
sl@0
|
270 |
}
|
|
sl@0
|
271 |
}
|
|
sl@0
|
272 |
OIL_DEFINE_IMPL_FULL (inverse_f32_sse, inverse_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
273 |
|
|
sl@0
|
274 |
SSE_FUNCTION static void
|
|
sl@0
|
275 |
negative_f32_sse (float *dest, float *src1, int n)
|
|
sl@0
|
276 |
{
|
|
sl@0
|
277 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
278 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
279 |
*dest++ = -(*src1++);
|
|
sl@0
|
280 |
}
|
|
sl@0
|
281 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
282 |
__m128 xmm0, xmm1;
|
|
sl@0
|
283 |
xmm0 = _mm_setzero_ps();
|
|
sl@0
|
284 |
xmm1 = _mm_loadu_ps(src1);
|
|
sl@0
|
285 |
xmm0 = _mm_sub_ps(xmm0, xmm1);
|
|
sl@0
|
286 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
287 |
dest += 4;
|
|
sl@0
|
288 |
src1 += 4;
|
|
sl@0
|
289 |
}
|
|
sl@0
|
290 |
for (; n > 0; n--) {
|
|
sl@0
|
291 |
*dest++ = -(*src1++);
|
|
sl@0
|
292 |
}
|
|
sl@0
|
293 |
}
|
|
sl@0
|
294 |
OIL_DEFINE_IMPL_FULL (negative_f32_sse, negative_f32, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
295 |
|
|
sl@0
|
296 |
SSE_FUNCTION static void
|
|
sl@0
|
297 |
scalaradd_f32_ns_sse (float *dest, float *src1, float *val, int n)
|
|
sl@0
|
298 |
{
|
|
sl@0
|
299 |
__m128 xmm1;
|
|
sl@0
|
300 |
|
|
sl@0
|
301 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
302 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
303 |
*dest++ = *src1++ + *val;
|
|
sl@0
|
304 |
}
|
|
sl@0
|
305 |
xmm1 = _mm_load_ps1(val);
|
|
sl@0
|
306 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
307 |
__m128 xmm0;
|
|
sl@0
|
308 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
309 |
xmm0 = _mm_add_ps(xmm0, xmm1);
|
|
sl@0
|
310 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
311 |
dest += 4;
|
|
sl@0
|
312 |
src1 += 4;
|
|
sl@0
|
313 |
}
|
|
sl@0
|
314 |
for (; n > 0; n--) {
|
|
sl@0
|
315 |
*dest++ = *src1++ + *val;
|
|
sl@0
|
316 |
}
|
|
sl@0
|
317 |
}
|
|
sl@0
|
318 |
OIL_DEFINE_IMPL_FULL (scalaradd_f32_ns_sse, scalaradd_f32_ns, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
319 |
|
|
sl@0
|
320 |
SSE_FUNCTION static void
|
|
sl@0
|
321 |
scalarmultiply_f32_ns_sse (float *dest, float *src1, float *val, int n)
|
|
sl@0
|
322 |
{
|
|
sl@0
|
323 |
__m128 xmm1;
|
|
sl@0
|
324 |
|
|
sl@0
|
325 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
326 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
327 |
*dest++ = *src1++ * *val;
|
|
sl@0
|
328 |
}
|
|
sl@0
|
329 |
xmm1 = _mm_load_ps1(val);
|
|
sl@0
|
330 |
for (; n >= 4; n -= 4) {
|
|
sl@0
|
331 |
__m128 xmm0;
|
|
sl@0
|
332 |
xmm0 = _mm_loadu_ps(src1);
|
|
sl@0
|
333 |
xmm0 = _mm_mul_ps(xmm0, xmm1);
|
|
sl@0
|
334 |
_mm_store_ps(dest, xmm0);
|
|
sl@0
|
335 |
dest += 4;
|
|
sl@0
|
336 |
src1 += 4;
|
|
sl@0
|
337 |
}
|
|
sl@0
|
338 |
for (; n > 0; n--) {
|
|
sl@0
|
339 |
*dest++ = *src1++ * *val;
|
|
sl@0
|
340 |
}
|
|
sl@0
|
341 |
}
|
|
sl@0
|
342 |
OIL_DEFINE_IMPL_FULL (scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns, OIL_IMPL_FLAG_SSE);
|
|
sl@0
|
343 |
|
|
sl@0
|
344 |
SSE_FUNCTION static void
|
|
sl@0
|
345 |
scalarmultiply_f64_ns_sse2 (double *dest, double *src1, double *val, int n)
|
|
sl@0
|
346 |
{
|
|
sl@0
|
347 |
__m128d xmm1;
|
|
sl@0
|
348 |
|
|
sl@0
|
349 |
/* Initial operations to align the destination pointer */
|
|
sl@0
|
350 |
for (; ((long)dest & 15) && (n > 0); n--) {
|
|
sl@0
|
351 |
*dest++ = *src1++ * *val;
|
|
sl@0
|
352 |
}
|
|
sl@0
|
353 |
xmm1 = _mm_load_pd1(val);
|
|
sl@0
|
354 |
for (; n >= 2; n -= 2) {
|
|
sl@0
|
355 |
__m128d xmm0;
|
|
sl@0
|
356 |
xmm0 = _mm_loadu_pd(src1);
|
|
sl@0
|
357 |
xmm0 = _mm_mul_pd(xmm0, xmm1);
|
|
sl@0
|
358 |
_mm_store_pd(dest, xmm0);
|
|
sl@0
|
359 |
dest += 2;
|
|
sl@0
|
360 |
src1 += 2;
|
|
sl@0
|
361 |
}
|
|
sl@0
|
362 |
for (; n > 0; n--) {
|
|
sl@0
|
363 |
*dest++ = *src1++ * *val;
|
|
sl@0
|
364 |
}
|
|
sl@0
|
365 |
}
|
|
sl@0
|
366 |
OIL_DEFINE_IMPL_FULL (scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns, OIL_IMPL_FLAG_SSE2);
|
|
sl@0
|
367 |
|
|
sl@0
|
368 |
|
|
sl@0
|
369 |
|
|
sl@0
|
370 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
371 |
|
|
sl@0
|
372 |
OilFunctionImpl* __oil_function_impl_add_f32_sse, add_f32() {
|
|
sl@0
|
373 |
return &_oil_function_impl_add_f32_sse, add_f32;
|
|
sl@0
|
374 |
}
|
|
sl@0
|
375 |
#endif
|
|
sl@0
|
376 |
|
|
sl@0
|
377 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
378 |
|
|
sl@0
|
379 |
OilFunctionImpl* __oil_function_impl_add_f64_sse2, add_f64() {
|
|
sl@0
|
380 |
return &_oil_function_impl_add_f64_sse2, add_f64;
|
|
sl@0
|
381 |
}
|
|
sl@0
|
382 |
#endif
|
|
sl@0
|
383 |
|
|
sl@0
|
384 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
385 |
|
|
sl@0
|
386 |
OilFunctionImpl* __oil_function_impl_add_f64_sse2_unroll, add_f64() {
|
|
sl@0
|
387 |
return &_oil_function_impl_add_f64_sse2_unroll, add_f64;
|
|
sl@0
|
388 |
}
|
|
sl@0
|
389 |
#endif
|
|
sl@0
|
390 |
|
|
sl@0
|
391 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
392 |
|
|
sl@0
|
393 |
OilFunctionImpl* __oil_function_impl_subtract_f32_sse, subtract_f32() {
|
|
sl@0
|
394 |
return &_oil_function_impl_subtract_f32_sse, subtract_f32;
|
|
sl@0
|
395 |
}
|
|
sl@0
|
396 |
#endif
|
|
sl@0
|
397 |
|
|
sl@0
|
398 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
399 |
|
|
sl@0
|
400 |
OilFunctionImpl* __oil_function_impl_multiply_f32_sse, multiply_f32() {
|
|
sl@0
|
401 |
return &_oil_function_impl_multiply_f32_sse, multiply_f32;
|
|
sl@0
|
402 |
}
|
|
sl@0
|
403 |
#endif
|
|
sl@0
|
404 |
|
|
sl@0
|
405 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
406 |
|
|
sl@0
|
407 |
OilFunctionImpl* __oil_function_impl_divide_f32_sse, divide_f32() {
|
|
sl@0
|
408 |
return &_oil_function_impl_divide_f32_sse, divide_f32;
|
|
sl@0
|
409 |
}
|
|
sl@0
|
410 |
#endif
|
|
sl@0
|
411 |
|
|
sl@0
|
412 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
413 |
|
|
sl@0
|
414 |
OilFunctionImpl* __oil_function_impl_minimum_f32_sse, minimum_f32() {
|
|
sl@0
|
415 |
return &_oil_function_impl_minimum_f32_sse, minimum_f32;
|
|
sl@0
|
416 |
}
|
|
sl@0
|
417 |
#endif
|
|
sl@0
|
418 |
|
|
sl@0
|
419 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
420 |
|
|
sl@0
|
421 |
OilFunctionImpl* __oil_function_impl_maximum_f32_sse, maximum_f32() {
|
|
sl@0
|
422 |
return &_oil_function_impl_maximum_f32_sse, maximum_f32;
|
|
sl@0
|
423 |
}
|
|
sl@0
|
424 |
#endif
|
|
sl@0
|
425 |
|
|
sl@0
|
426 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
427 |
|
|
sl@0
|
428 |
OilFunctionImpl* __oil_function_impl_inverse_f32_sse, inverse_f32() {
|
|
sl@0
|
429 |
return &_oil_function_impl_inverse_f32_sse, inverse_f32;
|
|
sl@0
|
430 |
}
|
|
sl@0
|
431 |
#endif
|
|
sl@0
|
432 |
|
|
sl@0
|
433 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
434 |
|
|
sl@0
|
435 |
OilFunctionImpl* __oil_function_impl_negative_f32_sse, negative_f32() {
|
|
sl@0
|
436 |
return &_oil_function_impl_negative_f32_sse, negative_f32;
|
|
sl@0
|
437 |
}
|
|
sl@0
|
438 |
#endif
|
|
sl@0
|
439 |
|
|
sl@0
|
440 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
441 |
|
|
sl@0
|
442 |
OilFunctionImpl* __oil_function_impl_scalaradd_f32_ns_sse, scalaradd_f32_ns() {
|
|
sl@0
|
443 |
return &_oil_function_impl_scalaradd_f32_ns_sse, scalaradd_f32_ns;
|
|
sl@0
|
444 |
}
|
|
sl@0
|
445 |
#endif
|
|
sl@0
|
446 |
|
|
sl@0
|
447 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
448 |
|
|
sl@0
|
449 |
OilFunctionImpl* __oil_function_impl_scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns() {
|
|
sl@0
|
450 |
return &_oil_function_impl_scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns;
|
|
sl@0
|
451 |
}
|
|
sl@0
|
452 |
#endif
|
|
sl@0
|
453 |
|
|
sl@0
|
454 |
#ifdef __SYMBIAN32__
|
|
sl@0
|
455 |
|
|
sl@0
|
456 |
OilFunctionImpl* __oil_function_impl_scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns() {
|
|
sl@0
|
457 |
return &_oil_function_impl_scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns;
|
|
sl@0
|
458 |
}
|
|
sl@0
|
459 |
#endif
|
|
sl@0
|
460 |
|