| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright 2011 The LibYuv Project Authors. All rights reserved. | 2 * Copyright 2011 The LibYuv Project Authors. All rights reserved. |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license | 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source | 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found | 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ | 9 */ |
| 10 | 10 |
| (...skipping 34 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 45 | 45 |
| 46 // Low level cpuid for X86. Returns zeros on other CPUs. | 46 // Low level cpuid for X86. Returns zeros on other CPUs. |
| 47 #if !defined(__pnacl__) && !defined(__CLR_VER) && \ | 47 #if !defined(__pnacl__) && !defined(__CLR_VER) && \ |
| 48 (defined(_M_IX86) || defined(_M_X64) || \ | 48 (defined(_M_IX86) || defined(_M_X64) || \ |
| 49 defined(__i386__) || defined(__x86_64__)) | 49 defined(__i386__) || defined(__x86_64__)) |
| 50 LIBYUV_API | 50 LIBYUV_API |
| 51 void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) { | 51 void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) { |
| 52 #if defined(_MSC_VER) && !defined(__clang__) | 52 #if defined(_MSC_VER) && !defined(__clang__) |
| 53 #if (_MSC_FULL_VER >= 160040219) | 53 #if (_MSC_FULL_VER >= 160040219) |
| 54 __cpuidex((int*)(cpu_info), info_eax, info_ecx); | 54 __cpuidex((int*)(cpu_info), info_eax, info_ecx); |
| 55 #elif defined(_M_IX86) | 55 #endif |
| 56 #if defined(_M_IX86) |
| 56 __asm { | 57 __asm { |
| 57 mov eax, info_eax | 58 mov eax, info_eax |
| 58 mov ecx, info_ecx | 59 mov ecx, info_ecx |
| 59 mov edi, cpu_info | 60 mov edi, cpu_info |
| 60 cpuid | 61 cpuid |
| 61 mov [edi], eax | 62 mov [edi], eax |
| 62 mov [edi + 4], ebx | 63 mov [edi + 4], ebx |
| 63 mov [edi + 8], ecx | 64 mov [edi + 8], ecx |
| 64 mov [edi + 12], edx | 65 mov [edi + 12], edx |
| 65 } | 66 } |
| (...skipping 25 matching lines...) Expand all Loading... |
| 91 #endif // defined(_MSC_VER) | 92 #endif // defined(_MSC_VER) |
| 92 } | 93 } |
| 93 | 94 |
| 94 #if !defined(__native_client__) | 95 #if !defined(__native_client__) |
| 95 #define HAS_XGETBV | 96 #define HAS_XGETBV |
| 96 // X86 CPUs have xgetbv to detect OS saves high parts of ymm registers. | 97 // X86 CPUs have xgetbv to detect OS saves high parts of ymm registers. |
| 97 int TestOsSaveYmm() { | 98 int TestOsSaveYmm() { |
| 98 uint32 xcr0 = 0u; | 99 uint32 xcr0 = 0u; |
| 99 #if defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219) | 100 #if defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219) |
| 100 xcr0 = (uint32)(_xgetbv(0)); // VS2010 SP1 required. | 101 xcr0 = (uint32)(_xgetbv(0)); // VS2010 SP1 required. |
| 101 #elif defined(_M_IX86) && defined(_MSC_VER) | 102 #endif |
| 103 #if defined(_M_IX86) && defined(_MSC_VER) |
| 102 __asm { | 104 __asm { |
| 103 xor ecx, ecx // xcr 0 | 105 xor ecx, ecx // xcr 0 |
| 104 _asm _emit 0x0f _asm _emit 0x01 _asm _emit 0xd0 // For VS2010 and earlier. | 106 _asm _emit 0x0f _asm _emit 0x01 _asm _emit 0xd0 // For VS2010 and earlier. |
| 105 mov xcr0, eax | 107 mov xcr0, eax |
| 106 } | 108 } |
| 107 #elif defined(__i386__) || defined(__x86_64__) | 109 #endif |
| 110 #if defined(__i386__) || defined(__x86_64__) |
| 108 asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx"); | 111 asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx"); |
| 109 #endif // defined(_MSC_VER) | 112 #endif // defined(_MSC_VER) |
| 110 return((xcr0 & 6) == 6); // Is ymm saved? | 113 return((xcr0 & 6) == 6); // Is ymm saved? |
| 111 } | 114 } |
| 112 #endif // !defined(__native_client__) | 115 #endif // !defined(__native_client__) |
| 113 #else | 116 #else |
| 114 LIBYUV_API | 117 LIBYUV_API |
| 115 void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) { | 118 void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) { |
| 116 cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0; | 119 cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0; |
| 117 } | 120 } |
| (...skipping 10 matching lines...) Expand all Loading... |
| 128 // This will occur for Chrome sandbox for Pepper or Render process. | 131 // This will occur for Chrome sandbox for Pepper or Render process. |
| 129 return kCpuHasNEON; | 132 return kCpuHasNEON; |
| 130 } | 133 } |
| 131 while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) { | 134 while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) { |
| 132 if (memcmp(cpuinfo_line, "Features", 8) == 0) { | 135 if (memcmp(cpuinfo_line, "Features", 8) == 0) { |
| 133 char* p = strstr(cpuinfo_line, " neon"); | 136 char* p = strstr(cpuinfo_line, " neon"); |
| 134 if (p && (p[5] == ' ' || p[5] == '\n')) { | 137 if (p && (p[5] == ' ' || p[5] == '\n')) { |
| 135 fclose(f); | 138 fclose(f); |
| 136 return kCpuHasNEON; | 139 return kCpuHasNEON; |
| 137 } | 140 } |
| 141 // aarch64 uses asimd for Neon. |
| 142 p = strstr(cpuinfo_line, " asimd"); |
| 143 if (p && (p[6] == ' ' || p[6] == '\n')) { |
| 144 fclose(f); |
| 145 return kCpuHasNEON; |
| 146 } |
| 138 } | 147 } |
| 139 } | 148 } |
| 140 fclose(f); | 149 fclose(f); |
| 141 return 0; | 150 return 0; |
| 142 } | 151 } |
| 143 | 152 |
| 144 #if defined(__mips__) && defined(__linux__) | 153 #if defined(__mips__) && defined(__linux__) |
| 145 static int MipsCpuCaps(const char* search_string) { | 154 static int MipsCpuCaps(const char* search_string) { |
| 146 char cpuinfo_line[512]; | 155 char cpuinfo_line[512]; |
| 147 const char* file_name = "/proc/cpuinfo"; | 156 const char* file_name = "/proc/cpuinfo"; |
| (...skipping 85 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 233 } | 242 } |
| 234 if (TestEnv("LIBYUV_DISABLE_AVX2")) { | 243 if (TestEnv("LIBYUV_DISABLE_AVX2")) { |
| 235 cpu_info_ &= ~kCpuHasAVX2; | 244 cpu_info_ &= ~kCpuHasAVX2; |
| 236 } | 245 } |
| 237 if (TestEnv("LIBYUV_DISABLE_ERMS")) { | 246 if (TestEnv("LIBYUV_DISABLE_ERMS")) { |
| 238 cpu_info_ &= ~kCpuHasERMS; | 247 cpu_info_ &= ~kCpuHasERMS; |
| 239 } | 248 } |
| 240 if (TestEnv("LIBYUV_DISABLE_FMA3")) { | 249 if (TestEnv("LIBYUV_DISABLE_FMA3")) { |
| 241 cpu_info_ &= ~kCpuHasFMA3; | 250 cpu_info_ &= ~kCpuHasFMA3; |
| 242 } | 251 } |
| 243 #elif defined(__mips__) && defined(__linux__) | 252 #endif |
| 253 #if defined(__mips__) && defined(__linux__) |
| 244 // Linux mips parse text file for dsp detect. | 254 // Linux mips parse text file for dsp detect. |
| 245 cpu_info_ = MipsCpuCaps("dsp"); // set kCpuHasMIPS_DSP. | 255 cpu_info_ = MipsCpuCaps("dsp"); // set kCpuHasMIPS_DSP. |
| 246 #if defined(__mips_dspr2) | 256 #if defined(__mips_dspr2) |
| 247 cpu_info_ |= kCpuHasMIPS_DSPR2; | 257 cpu_info_ |= kCpuHasMIPS_DSPR2; |
| 248 #endif | 258 #endif |
| 249 cpu_info_ |= kCpuHasMIPS; | 259 cpu_info_ |= kCpuHasMIPS; |
| 250 | 260 |
| 251 if (getenv("LIBYUV_DISABLE_MIPS")) { | 261 if (getenv("LIBYUV_DISABLE_MIPS")) { |
| 252 cpu_info_ &= ~kCpuHasMIPS; | 262 cpu_info_ &= ~kCpuHasMIPS; |
| 253 } | 263 } |
| 254 if (getenv("LIBYUV_DISABLE_MIPS_DSP")) { | 264 if (getenv("LIBYUV_DISABLE_MIPS_DSP")) { |
| 255 cpu_info_ &= ~kCpuHasMIPS_DSP; | 265 cpu_info_ &= ~kCpuHasMIPS_DSP; |
| 256 } | 266 } |
| 257 if (getenv("LIBYUV_DISABLE_MIPS_DSPR2")) { | 267 if (getenv("LIBYUV_DISABLE_MIPS_DSPR2")) { |
| 258 cpu_info_ &= ~kCpuHasMIPS_DSPR2; | 268 cpu_info_ &= ~kCpuHasMIPS_DSPR2; |
| 259 } | 269 } |
| 260 #elif defined(__arm__) || defined(__aarch64__) | 270 #endif |
| 271 #if defined(__arm__) || defined(__aarch64__) |
| 261 // gcc -mfpu=neon defines __ARM_NEON__ | 272 // gcc -mfpu=neon defines __ARM_NEON__ |
| 262 // __ARM_NEON__ generates code that requires Neon. NaCL also requires Neon. | 273 // __ARM_NEON__ generates code that requires Neon. NaCL also requires Neon. |
| 263 // For Linux, /proc/cpuinfo can be tested but without that assume Neon. | 274 // For Linux, /proc/cpuinfo can be tested but without that assume Neon. |
| 264 #if defined(__ARM_NEON__) || defined(__native_client__) || !defined(__linux__) | 275 #if defined(__ARM_NEON__) || defined(__native_client__) || !defined(__linux__) |
| 265 cpu_info_ = kCpuHasNEON; | 276 cpu_info_ = kCpuHasNEON; |
| 266 // For aarch64(arm64), /proc/cpuinfo's feature is not complete, e.g. no neon | 277 // For aarch64(arm64), /proc/cpuinfo's feature is not complete, e.g. no neon |
| 267 // flag in it. | 278 // flag in it. |
| 268 // So for aarch64, neon enabling is hard coded here. | 279 // So for aarch64, neon enabling is hard coded here. |
| 269 #elif defined(__aarch64__) | 280 #endif |
| 281 #if defined(__aarch64__) |
| 270 cpu_info_ = kCpuHasNEON; | 282 cpu_info_ = kCpuHasNEON; |
| 271 #else | 283 #else |
| 272 // Linux arm parse text file for neon detect. | 284 // Linux arm parse text file for neon detect. |
| 273 cpu_info_ = ArmCpuCaps("/proc/cpuinfo"); | 285 cpu_info_ = ArmCpuCaps("/proc/cpuinfo"); |
| 274 #endif | 286 #endif |
| 275 cpu_info_ |= kCpuHasARM; | 287 cpu_info_ |= kCpuHasARM; |
| 276 if (TestEnv("LIBYUV_DISABLE_NEON")) { | 288 if (TestEnv("LIBYUV_DISABLE_NEON")) { |
| 277 cpu_info_ &= ~kCpuHasNEON; | 289 cpu_info_ &= ~kCpuHasNEON; |
| 278 } | 290 } |
| 279 #endif // __arm__ | 291 #endif // __arm__ |
| 280 if (TestEnv("LIBYUV_DISABLE_ASM")) { | 292 if (TestEnv("LIBYUV_DISABLE_ASM")) { |
| 281 cpu_info_ = 0; | 293 cpu_info_ = 0; |
| 282 } | 294 } |
| 283 return cpu_info_; | 295 return cpu_info_; |
| 284 } | 296 } |
| 285 | 297 |
| 286 LIBYUV_API | 298 LIBYUV_API |
| 287 void MaskCpuFlags(int enable_flags) { | 299 void MaskCpuFlags(int enable_flags) { |
| 288 cpu_info_ = InitCpuFlags() & enable_flags; | 300 cpu_info_ = InitCpuFlags() & enable_flags; |
| 289 } | 301 } |
| 290 | 302 |
| 291 #ifdef __cplusplus | 303 #ifdef __cplusplus |
| 292 } // extern "C" | 304 } // extern "C" |
| 293 } // namespace libyuv | 305 } // namespace libyuv |
| 294 #endif | 306 #endif |
| OLD | NEW |