1
// Copyright (C) 2003 Dolphin Project.
3
// This program is free software: you can redistribute it and/or modify
4
// it under the terms of the GNU General Public License as published by
5
// the Free Software Foundation, version 2.0.
7
// This program is distributed in the hope that it will be useful,
8
// but WITHOUT ANY WARRANTY; without even the implied warranty of
9
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10
// GNU General Public License 2.0 for more details.
12
// A copy of the GPL 2.0 should have been included with the program.
13
// If not, see http://www.gnu.org/licenses/
15
// Official SVN repository and contact information can be found at
16
// http://code.google.com/p/dolphin-emu/
24
#include "base/logging.h"
25
#include "base/basictypes.h"
28
#include "CPUDetect.h"
29
#include "StringUtils.h"
32
#define WIN32_LEAN_AND_MEAN
34
#define _interlockedbittestandset workaround_ms_header_bug_platform_sdk6_set
35
#define _interlockedbittestandreset workaround_ms_header_bug_platform_sdk6_reset
36
#define _interlockedbittestandset64 workaround_ms_header_bug_platform_sdk6_set64
37
#define _interlockedbittestandreset64 workaround_ms_header_bug_platform_sdk6_reset64
39
#undef _interlockedbittestandset
40
#undef _interlockedbittestandreset
41
#undef _interlockedbittestandset64
42
#undef _interlockedbittestandreset64
44
void do_cpuidex(u32 regs[4], u32 cpuid_leaf, u32 ecxval) {
45
__cpuidex((int *)regs, cpuid_leaf, ecxval);
47
void do_cpuid(u32 regs[4], u32 cpuid_leaf) {
48
__cpuid((int *)regs, cpuid_leaf);
53
#include <xmmintrin.h>
55
#define _XCR_XFEATURE_ENABLED_MASK 0
56
static unsigned long long _xgetbv(unsigned int index)
58
unsigned int eax, edx;
59
__asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index));
60
return ((unsigned long long)edx << 32) | eax;
64
#define _XCR_XFEATURE_ENABLED_MASK 0
69
void do_cpuidex(u32 regs[4], u32 cpuid_leaf, u32 ecxval) {
70
#if defined(__i386__) && defined(__PIC__)
72
"xchgl %%ebx, %1;\n\t"
74
"xchgl %%ebx, %1;\n\t"
75
:"=a" (regs[0]), "=r" (regs[1]), "=c" (regs[2]), "=d" (regs[3])
76
:"a" (cpuid_leaf), "c" (ecxval));
80
:"=a" (regs[0]), "=b" (regs[1]), "=c" (regs[2]), "=d" (regs[3])
81
:"a" (cpuid_leaf), "c" (ecxval));
84
void do_cpuid(u32 regs[4], u32 cpuid_leaf)
86
do_cpuidex(regs, cpuid_leaf, 0);
98
// Detects the various cpu features
99
void CPUInfo::Detect() {
100
memset(this, 0, sizeof(*this));
103
#elif defined (_M_X64)
112
IsWow64Process(GetCurrentProcess(), &f64);
113
OS64bit = (f64 == TRUE) ? true : false;
117
// Set obvious defaults, for extra safety
124
// Assume CPU supports the CPUID instruction. Those that don't can barely
125
// boot modern OS:es anyway.
127
memset(cpu_string, 0, sizeof(cpu_string));
129
// Detect CPU's CPUID capabilities, and grab cpu string
130
do_cpuid(cpu_id, 0x00000000);
131
u32 max_std_fn = cpu_id[0]; // EAX
132
*((int *)cpu_string) = cpu_id[1];
133
*((int *)(cpu_string + 4)) = cpu_id[3];
134
*((int *)(cpu_string + 8)) = cpu_id[2];
135
do_cpuid(cpu_id, 0x80000000);
136
u32 max_ex_fn = cpu_id[0];
137
if (!strcmp(cpu_string, "GenuineIntel"))
138
vendor = VENDOR_INTEL;
139
else if (!strcmp(cpu_string, "AuthenticAMD"))
142
vendor = VENDOR_OTHER;
144
// Set reasonable default brand string even if brand string not available.
145
strcpy(brand_string, cpu_string);
147
// Detect family and other misc stuff.
150
logical_cpu_count = 1;
151
if (max_std_fn >= 1) {
152
do_cpuid(cpu_id, 0x00000001);
153
int family = ((cpu_id[0] >> 8) & 0xf) + ((cpu_id[0] >> 20) & 0xff);
154
int model = ((cpu_id[0] >> 4) & 0xf) + ((cpu_id[0] >> 12) & 0xf0);
155
// Detect people unfortunate enough to be running PPSSPP on an Atom
156
if (family == 6 && (model == 0x1C || model == 0x26 || model == 0x27 || model == 0x35 || model == 0x36 ||
157
model == 0x37 || model == 0x4A || model == 0x4D || model == 0x5A || model == 0x5D))
160
logical_cpu_count = (cpu_id[1] >> 16) & 0xFF;
161
ht = (cpu_id[3] >> 28) & 1;
163
if ((cpu_id[3] >> 25) & 1) bSSE = true;
164
if ((cpu_id[3] >> 26) & 1) bSSE2 = true;
165
if ((cpu_id[2]) & 1) bSSE3 = true;
166
if ((cpu_id[2] >> 9) & 1) bSSSE3 = true;
167
if ((cpu_id[2] >> 19) & 1) bSSE4_1 = true;
168
if ((cpu_id[2] >> 20) & 1) bSSE4_2 = true;
169
if ((cpu_id[2] >> 28) & 1) {
171
if ((cpu_id[2] >> 12) & 1)
174
if ((cpu_id[2] >> 25) & 1) bAES = true;
176
if ((cpu_id[3] >> 24) & 1)
178
// We can use FXSAVE.
182
// AVX support requires 3 separate checks:
183
// - Is the AVX bit set in CPUID? (>>28)
184
// - Is the XSAVE bit set in CPUID? ( >>26)
185
// - Is the OSXSAVE bit set in CPUID? ( >>27)
186
// - XGETBV result has the XCR bit set.
187
if (((cpu_id[2] >> 28) & 1) && ((cpu_id[2] >> 27) & 1) && ((cpu_id[2] >> 26) & 1))
189
if ((_xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6)
192
if ((cpu_id[2] >> 12) & 1)
199
do_cpuid(cpu_id, 0x00000007);
200
// careful; we can't enable AVX2 unless the XSAVE/XGETBV checks above passed
201
if ((cpu_id[1] >> 5) & 1)
203
if ((cpu_id[1] >> 3) & 1)
205
if ((cpu_id[1] >> 8) & 1)
209
if (max_ex_fn >= 0x80000004) {
210
// Extract brand string
211
do_cpuid(cpu_id, 0x80000002);
212
memcpy(brand_string, cpu_id, sizeof(cpu_id));
213
do_cpuid(cpu_id, 0x80000003);
214
memcpy(brand_string + 16, cpu_id, sizeof(cpu_id));
215
do_cpuid(cpu_id, 0x80000004);
216
memcpy(brand_string + 32, cpu_id, sizeof(cpu_id));
218
if (max_ex_fn >= 0x80000001) {
219
// Check for more features.
220
do_cpuid(cpu_id, 0x80000001);
221
if (cpu_id[2] & 1) bLAHFSAHF64 = true;
222
// CmpLegacy (bit 2) is deprecated.
223
if ((cpu_id[3] >> 29) & 1) bLongMode = true;
226
num_cores = (logical_cpu_count == 0) ? 1 : logical_cpu_count;
228
if (max_ex_fn >= 0x80000008) {
229
// Get number of cores. This is a bit complicated. Following AMD manual here.
230
do_cpuid(cpu_id, 0x80000008);
231
int apic_id_core_id_size = (cpu_id[2] >> 12) & 0xF;
232
if (apic_id_core_id_size == 0) {
234
// 0x0B is the preferred method on Core i series processors.
235
// Inspired by https://github.com/D-Programming-Language/druntime/blob/23b0d1f41e27638bda2813af55823b502195a58d/src/core/cpuid.d#L562.
236
bool hasLeafB = false;
237
if (vendor == VENDOR_INTEL && max_std_fn >= 0x0B) {
238
do_cpuidex(cpu_id, 0x0B, 0);
239
if (cpu_id[1] != 0) {
240
logical_cpu_count = cpu_id[1] & 0xFFFF;
241
do_cpuidex(cpu_id, 0x0B, 1);
242
int totalThreads = cpu_id[1] & 0xFFFF;
243
num_cores = totalThreads / logical_cpu_count;
247
// Old new mechanism for modern Intel CPUs.
248
if (!hasLeafB && vendor == VENDOR_INTEL) {
249
do_cpuid(cpu_id, 0x00000004);
250
int cores_x_package = ((cpu_id[0] >> 26) & 0x3F) + 1;
251
HTT = (cores_x_package < logical_cpu_count);
252
cores_x_package = ((logical_cpu_count % cores_x_package) == 0) ? cores_x_package : 1;
253
num_cores = (cores_x_package > 1) ? cores_x_package : num_cores;
254
logical_cpu_count /= cores_x_package;
258
// Use AMD's new method.
259
num_cores = (cpu_id[2] & 0xFF) + 1;
264
// Turn the cpu info into a string we can show
265
std::string CPUInfo::Summarize()
269
sum = StringFromFormat("%s, %i core", cpu_string, num_cores);
272
sum = StringFromFormat("%s, %i cores", cpu_string, num_cores);
273
if (HTT) sum += StringFromFormat(" (%i logical threads per physical core)", logical_cpu_count);
275
if (bSSE) sum += ", SSE";
276
if (bSSE2) sum += ", SSE2";
277
if (bSSE3) sum += ", SSE3";
278
if (bSSSE3) sum += ", SSSE3";
279
if (bSSE4_1) sum += ", SSE4.1";
280
if (bSSE4_2) sum += ", SSE4.2";
281
if (HTT) sum += ", HTT";
282
if (bAVX) sum += ", AVX";
283
if (bFMA) sum += ", FMA";
284
if (bAES) sum += ", AES";
285
if (bLongMode) sum += ", 64-bit support";
added
removed
Common/CPUDetect.cpp
