uppsrc_Core_r14486_v2.diff
| uppsrc/Core/Cpu.cpp | ||
|---|---|---|
| 149 | 149 |
void EndianSwap(int64 *v, size_t count) ENDIAN_SWAP |
| 150 | 150 |
void EndianSwap(uint64 *v, size_t count) ENDIAN_SWAP |
| 151 | 151 | |
| 152 |
#ifdef CPU_X86 |
|
| 152 |
#if defined(CPU_X86) && defined(__SSE2__) |
|
| 153 | ||
| 154 |
#include <emmintrin.h> |
|
| 155 | ||
| 153 | 156 |
void huge_memsetd(void *p, dword c, int len) |
| 154 | 157 |
{ // bypasses the cache, good for >4MB
|
| 155 | 158 |
dword *t = (dword *)p; |
| ... | ... | |
| 173 | 176 |
while(len--) |
| 174 | 177 |
*t++ = c; |
| 175 | 178 |
} |
| 179 | ||
| 180 |
void memsetd(void *p, dword data, int len) |
|
| 181 |
{
|
|
| 182 |
dword *t = (dword *)p; |
|
| 183 |
if(len < 4) {
|
|
| 184 |
if(len & 2) {
|
|
| 185 |
t[0] = t[1] = data; |
|
| 186 |
t += 2; |
|
| 187 |
} |
|
| 188 |
if(len & 1) |
|
| 189 |
t[0] = data; |
|
| 190 |
return; |
|
| 191 |
} |
|
| 192 | ||
| 193 |
__m128i val4 = _mm_set1_epi32(data); |
|
| 194 |
auto Set4 = [&](int at) { _mm_storeu_si128((__m128i *)(t + at), val4); };
|
|
| 195 | ||
| 196 |
Set4(len - 4); // fill tail |
|
| 197 |
if(len >= 32) {
|
|
| 198 |
if(len >= 1024*1024) { // for really huge data, bypass the cache
|
|
| 199 |
huge_memsetd(t, data, len); |
|
| 200 |
return; |
|
| 201 |
} |
|
| 202 |
const dword *e = t + len - 32; |
|
| 203 |
do {
|
|
| 204 |
Set4(0); Set4(4); Set4(8); Set4(12); |
|
| 205 |
Set4(16); Set4(20); Set4(24); Set4(28); |
|
| 206 |
t += 32; |
|
| 207 |
} |
|
| 208 |
while(t <= e); |
|
| 209 |
} |
|
| 210 |
if(len & 16) {
|
|
| 211 |
Set4(0); Set4(4); Set4(8); Set4(12); |
|
| 212 |
t += 16; |
|
| 213 |
} |
|
| 214 |
if(len & 8) {
|
|
| 215 |
Set4(0); Set4(4); |
|
| 216 |
t += 8; |
|
| 217 |
} |
|
| 218 |
if(len & 4) |
|
| 219 |
Set4(0); |
|
| 220 |
} |
|
| 176 | 221 |
#endif |
| 177 | 222 | |
| 178 | 223 |
} |
| uppsrc/Core/Ops.h | ||
|---|---|---|
| 342 | 342 |
return x >= -9223372036854775808.0 && x < 9223372036854775808.0; |
| 343 | 343 |
} |
| 344 | 344 |
|
| 345 |
#ifdef CPU_X86 |
|
| 346 |
|
|
| 347 |
#include <smmintrin.h> |
|
| 348 |
|
|
| 345 |
#if defined(CPU_X86) && defined(__SSE2__) |
|
| 349 | 346 |
void huge_memsetd(void *p, dword data, int len); |
| 350 |
|
|
| 351 |
inline |
|
| 352 |
void memsetd(void *p, dword data, int len) |
|
| 353 |
{
|
|
| 354 |
dword *t = (dword *)p; |
|
| 355 |
if(len < 4) {
|
|
| 356 |
if(len & 2) {
|
|
| 357 |
t[0] = t[1] = data; |
|
| 358 |
t += 2; |
|
| 359 |
} |
|
| 360 |
if(len & 1) |
|
| 361 |
t[0] = data; |
|
| 362 |
return; |
|
| 363 |
} |
|
| 364 |
|
|
| 365 |
__m128i val4 = _mm_set1_epi32(data); |
|
| 366 |
auto Set4 = [&](int at) { _mm_storeu_si128((__m128i *)(t + at), val4); };
|
|
| 367 |
|
|
| 368 |
Set4(len - 4); // fill tail |
|
| 369 |
if(len >= 32) {
|
|
| 370 |
if(len >= 1024*1024) { // for really huge data, bypass the cache
|
|
| 371 |
huge_memsetd(t, data, len); |
|
| 372 |
return; |
|
| 373 |
} |
|
| 374 |
const dword *e = t + len - 32; |
|
| 375 |
do {
|
|
| 376 |
Set4(0); Set4(4); Set4(8); Set4(12); |
|
| 377 |
Set4(16); Set4(20); Set4(24); Set4(28); |
|
| 378 |
t += 32; |
|
| 379 |
} |
|
| 380 |
while(t <= e); |
|
| 381 |
} |
|
| 382 |
if(len & 16) {
|
|
| 383 |
Set4(0); Set4(4); Set4(8); Set4(12); |
|
| 384 |
t += 16; |
|
| 385 |
} |
|
| 386 |
if(len & 8) {
|
|
| 387 |
Set4(0); Set4(4); |
|
| 388 |
t += 8; |
|
| 389 |
} |
|
| 390 |
if(len & 4) |
|
| 391 |
Set4(0); |
|
| 392 |
} |
|
| 347 |
void memsetd(void *p, dword data, int len); |
|
| 393 | 348 |
#else |
| 394 | 349 |
inline |
| 395 | 350 |
void memsetd(void *p, RGBA c, int len) |