media/libyuv/source/cpu_id.cc@b8a032363ba2
media/libyuv/source/cpu_id.cc
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
1 /*
2 * Copyright 2011 The LibYuv Project Authors. All rights reserved.
3 *
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
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
11 #include "libyuv/cpu_id.h"
13 #ifdef _MSC_VER
14 #include <intrin.h> // For __cpuidex()
15 #endif
16 #if !defined(__pnacl__) && !defined(__CLR_VER) && \
17 !defined(__native_client__) && defined(_M_X64) && \
18 defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219)
19 #include <immintrin.h> // For _xgetbv()
20 #endif
22 #if !defined(__native_client__)
23 #include <stdlib.h> // For getenv()
24 #endif
26 // For ArmCpuCaps() but unittested on all platforms
27 #include <stdio.h>
28 #include <string.h>
30 #include "libyuv/basic_types.h" // For CPU_X86
32 #ifdef __cplusplus
33 namespace libyuv {
34 extern "C" {
35 #endif
37 // For functions that use the stack and have runtime checks for overflow,
38 // use SAFEBUFFERS to avoid additional check.
39 #if defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219)
40 #define SAFEBUFFERS __declspec(safebuffers)
41 #else
42 #define SAFEBUFFERS
43 #endif
45 // Low level cpuid for X86. Returns zeros on other CPUs.
46 #if !defined(__pnacl__) && !defined(__CLR_VER) && \
47 (defined(_M_IX86) || defined(_M_X64) || \
48 defined(__i386__) || defined(__x86_64__))
49 LIBYUV_API
50 void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
51 #if defined(_MSC_VER)
52 #if (_MSC_FULL_VER >= 160040219)
53 __cpuidex((int*)(cpu_info), info_eax, info_ecx);
54 #elif defined(_M_IX86)
55 __asm {
56 mov eax, info_eax
57 mov ecx, info_ecx
58 mov edi, cpu_info
59 cpuid
60 mov [edi], eax
61 mov [edi + 4], ebx
62 mov [edi + 8], ecx
63 mov [edi + 12], edx
64 }
65 #else
66 if (info_ecx == 0) {
67 __cpuid((int*)(cpu_info), info_eax);
68 } else {
69 cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0;
70 }
71 #endif
72 #else // defined(_MSC_VER)
73 uint32 info_ebx, info_edx;
74 asm volatile ( // NOLINT
75 #if defined( __i386__) && defined(__PIC__)
76 // Preserve ebx for fpic 32 bit.
77 "mov %%ebx, %%edi \n"
78 "cpuid \n"
79 "xchg %%edi, %%ebx \n"
80 : "=D" (info_ebx),
81 #else
82 "cpuid \n"
83 : "=b" (info_ebx),
84 #endif // defined( __i386__) && defined(__PIC__)
85 "+a" (info_eax), "+c" (info_ecx), "=d" (info_edx));
86 cpu_info[0] = info_eax;
87 cpu_info[1] = info_ebx;
88 cpu_info[2] = info_ecx;
89 cpu_info[3] = info_edx;
90 #endif // defined(_MSC_VER)
91 }
93 #if !defined(__native_client__)
94 #define HAS_XGETBV
95 // X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
96 int TestOsSaveYmm() {
97 uint32 xcr0 = 0u;
98 #if defined(_MSC_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
99 xcr0 = (uint32)(_xgetbv(_XCR_XFEATURE_ENABLED_MASK));
100 #elif defined(_MSC_VER) && defined(_M_IX86)
101 __asm {
102 xor ecx, ecx // xcr 0
103 _asm _emit 0x0f _asm _emit 0x01 _asm _emit 0xd0 // For VS2010 and earlier.
104 mov xcr0, eax
105 }
106 #elif defined(__i386__) || defined(__x86_64__)
107 asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx");
108 #endif // defined(_MSC_VER)
109 return((xcr0 & 6) == 6); // Is ymm saved?
110 }
111 #endif // !defined(__native_client__)
112 #else
113 LIBYUV_API
114 void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
115 cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
116 }
117 #endif
119 // based on libvpx arm_cpudetect.c
120 // For Arm, but public to allow testing on any CPU
121 LIBYUV_API SAFEBUFFERS
122 int ArmCpuCaps(const char* cpuinfo_name) {
123 FILE* f = fopen(cpuinfo_name, "r");
124 if (f) {
125 char cpuinfo_line[512];
126 while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) {
127 if (memcmp(cpuinfo_line, "Features", 8) == 0) {
128 char* p = strstr(cpuinfo_line, " neon");
129 if (p && (p[5] == ' ' || p[5] == '\n')) {
130 fclose(f);
131 return kCpuHasNEON;
132 }
133 }
134 }
135 fclose(f);
136 }
137 return 0;
138 }
140 #if defined(__mips__) && defined(__linux__)
141 static int MipsCpuCaps(const char* search_string) {
142 const char* file_name = "/proc/cpuinfo";
143 char cpuinfo_line[256];
144 FILE* f = NULL;
145 if ((f = fopen(file_name, "r")) != NULL) {
146 while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f) != NULL) {
147 if (strstr(cpuinfo_line, search_string) != NULL) {
148 fclose(f);
149 return kCpuHasMIPS_DSP;
150 }
151 }
152 fclose(f);
153 }
154 /* Did not find string in the proc file, or not Linux ELF. */
155 return 0;
156 }
157 #endif
159 // CPU detect function for SIMD instruction sets.
160 LIBYUV_API
161 int cpu_info_ = kCpuInit; // cpu_info is not initialized yet.
163 // Test environment variable for disabling CPU features. Any non-zero value
164 // to disable. Zero ignored to make it easy to set the variable on/off.
165 #if !defined(__native_client__) && !defined(_M_ARM)
167 static LIBYUV_BOOL TestEnv(const char* name) {
168 const char* var = getenv(name);
169 if (var) {
170 if (var[0] != '0') {
171 return LIBYUV_TRUE;
172 }
173 }
174 return LIBYUV_FALSE;
175 }
176 #else // nacl does not support getenv().
177 static LIBYUV_BOOL TestEnv(const char*) {
178 return LIBYUV_FALSE;
179 }
180 #endif
182 LIBYUV_API SAFEBUFFERS
183 int InitCpuFlags(void) {
184 #if !defined(__pnacl__) && !defined(__CLR_VER) && defined(CPU_X86)
186 uint32 cpu_info1[4] = { 0, 0, 0, 0 };
187 uint32 cpu_info7[4] = { 0, 0, 0, 0 };
188 CpuId(1, 0, cpu_info1);
189 CpuId(7, 0, cpu_info7);
190 cpu_info_ = ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
191 ((cpu_info1[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
192 ((cpu_info1[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
193 ((cpu_info1[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
194 ((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
195 ((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
196 kCpuHasX86;
197 #ifdef HAS_XGETBV
198 if ((cpu_info1[2] & 0x18000000) == 0x18000000 && // AVX and OSSave
199 TestOsSaveYmm()) { // Saves YMM.
200 cpu_info_ |= ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) |
201 kCpuHasAVX;
202 }
203 #endif
204 // Environment variable overrides for testing.
205 if (TestEnv("LIBYUV_DISABLE_X86")) {
206 cpu_info_ &= ~kCpuHasX86;
207 }
208 if (TestEnv("LIBYUV_DISABLE_SSE2")) {
209 cpu_info_ &= ~kCpuHasSSE2;
210 }
211 if (TestEnv("LIBYUV_DISABLE_SSSE3")) {
212 cpu_info_ &= ~kCpuHasSSSE3;
213 }
214 if (TestEnv("LIBYUV_DISABLE_SSE41")) {
215 cpu_info_ &= ~kCpuHasSSE41;
216 }
217 if (TestEnv("LIBYUV_DISABLE_SSE42")) {
218 cpu_info_ &= ~kCpuHasSSE42;
219 }
220 if (TestEnv("LIBYUV_DISABLE_AVX")) {
221 cpu_info_ &= ~kCpuHasAVX;
222 }
223 if (TestEnv("LIBYUV_DISABLE_AVX2")) {
224 cpu_info_ &= ~kCpuHasAVX2;
225 }
226 if (TestEnv("LIBYUV_DISABLE_ERMS")) {
227 cpu_info_ &= ~kCpuHasERMS;
228 }
229 if (TestEnv("LIBYUV_DISABLE_FMA3")) {
230 cpu_info_ &= ~kCpuHasFMA3;
231 }
232 #elif defined(__mips__) && defined(__linux__)
233 // Linux mips parse text file for dsp detect.
234 cpu_info_ = MipsCpuCaps("dsp"); // set kCpuHasMIPS_DSP.
235 #if defined(__mips_dspr2)
236 cpu_info_ |= kCpuHasMIPS_DSPR2;
237 #endif
238 cpu_info_ |= kCpuHasMIPS;
240 if (getenv("LIBYUV_DISABLE_MIPS")) {
241 cpu_info_ &= ~kCpuHasMIPS;
242 }
243 if (getenv("LIBYUV_DISABLE_MIPS_DSP")) {
244 cpu_info_ &= ~kCpuHasMIPS_DSP;
245 }
246 if (getenv("LIBYUV_DISABLE_MIPS_DSPR2")) {
247 cpu_info_ &= ~kCpuHasMIPS_DSPR2;
248 }
249 #elif defined(__arm__)
250 #if defined(__linux__) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON)) && \
251 !defined(__native_client__)
252 // Linux arm parse text file for neon detect.
253 cpu_info_ = ArmCpuCaps("/proc/cpuinfo");
254 #elif defined(__ARM_NEON__) || defined(__native_client__)
255 // gcc -mfpu=neon defines __ARM_NEON__
256 // Enable Neon if you want support for Neon and Arm, and use MaskCpuFlags
257 // to disable Neon on devices that do not have it.
258 cpu_info_ = kCpuHasNEON;
259 #endif
260 cpu_info_ |= kCpuHasARM;
261 if (TestEnv("LIBYUV_DISABLE_NEON")) {
262 cpu_info_ &= ~kCpuHasNEON;
263 }
264 #endif // __arm__
265 if (TestEnv("LIBYUV_DISABLE_ASM")) {
266 cpu_info_ = 0;
267 }
268 return cpu_info_;
269 }
271 LIBYUV_API
272 void MaskCpuFlags(int enable_flags) {
273 cpu_info_ = InitCpuFlags() & enable_flags;
274 }
276 #ifdef __cplusplus
277 } // extern "C"
278 } // namespace libyuv
279 #endif
