ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Add AVX2/BMI1/BMI2 detection support 

Also clean up the formatting in a bit of the CPU detection code.
This commit is contained in:
Fiora 2014-08-24 09:14:54 -07:00
parent 48f52b9662
commit ce6d09ca5d
2 changed files with 38 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
Source/Core/Common/CPUDetect.h
View File

@ -40,6 +40,9 @@ struct CPUInfo
bool bLZCNT; bool bLZCNT;
bool bSSE4A; bool bSSE4A;
bool bAVX; bool bAVX;
bool bAVX2;
bool bBMI1;
bool bBMI2;
bool bFMA; bool bFMA;
bool bAES; bool bAES;
// FXSAVE/FXRSTOR // FXSAVE/FXRSTOR

45
Source/Core/Common/x64CPUDetect.cpp
View File

@ -82,7 +82,8 @@ static unsigned long long _xgetbv(unsigned int index)
CPUInfo cpu_info; CPUInfo cpu_info;
CPUInfo::CPUInfo() { CPUInfo::CPUInfo()
{
Detect(); Detect();
} }
@ -97,7 +98,8 @@ void CPUInfo::Detect()
num_cores = 1; num_cores = 1;
// Set obvious defaults, for extra safety // Set obvious defaults, for extra safety
if (Mode64bit) { if (Mode64bit)
{
bSSE = true; bSSE = true;
bSSE2 = true; bSSE2 = true;
bLongMode = true; bLongMode = true;
@ -130,7 +132,8 @@ void CPUInfo::Detect()
bool ht = false; bool ht = false;
HTT = ht; HTT = ht;
logical_cpu_count = 1; logical_cpu_count = 1;
if (max_std_fn >= 1) { if (max_std_fn >= 1)
{
__cpuid(cpu_id, 0x00000001); __cpuid(cpu_id, 0x00000001);
logical_cpu_count = (cpu_id[1] >> 16) & 0xFF; logical_cpu_count = (cpu_id[1] >> 16) & 0xFF;
ht = (cpu_id[3] >> 28) & 1; ht = (cpu_id[3] >> 28) & 1;
@ -163,11 +166,24 @@ void CPUInfo::Detect()
bFMA = true; bFMA = true;
} }
} }
if (max_std_fn >= 7)
{
__cpuid(cpu_id, 0x00000007);
// careful; we can't enable AVX2 unless the XSAVE/XGETBV checks above passed
if ((cpu_id[1] >> 5) & 1)
bAVX2 = bAVX;
if ((cpu_id[1] >> 3) & 1)
bBMI1 = true;
if ((cpu_id[1] >> 8) & 1)
bBMI2 = true;
}
} }
bFlushToZero = bSSE; bFlushToZero = bSSE;
if (max_ex_fn >= 0x80000004) { if (max_ex_fn >= 0x80000004)
{
// Extract brand string // Extract brand string
__cpuid(cpu_id, 0x80000002); __cpuid(cpu_id, 0x80000002);
memcpy(brand_string, cpu_id, sizeof(cpu_id)); memcpy(brand_string, cpu_id, sizeof(cpu_id));
@ -176,7 +192,8 @@ void CPUInfo::Detect()
__cpuid(cpu_id, 0x80000004); __cpuid(cpu_id, 0x80000004);
memcpy(brand_string + 32, cpu_id, sizeof(cpu_id)); memcpy(brand_string + 32, cpu_id, sizeof(cpu_id));
} }
if (max_ex_fn >= 0x80000001) { if (max_ex_fn >= 0x80000001)
{
// Check for more features. // Check for more features.
__cpuid(cpu_id, 0x80000001); __cpuid(cpu_id, 0x80000001);
if (cpu_id[2] & 1) bLAHFSAHF64 = true; if (cpu_id[2] & 1) bLAHFSAHF64 = true;
@ -185,14 +202,18 @@ void CPUInfo::Detect()
num_cores = (logical_cpu_count == 0) ? 1 : logical_cpu_count; num_cores = (logical_cpu_count == 0) ? 1 : logical_cpu_count;
if (max_ex_fn >= 0x80000008) { if (max_ex_fn >= 0x80000008)
{
// Get number of cores. This is a bit complicated. Following AMD manual here. // Get number of cores. This is a bit complicated. Following AMD manual here.
__cpuid(cpu_id, 0x80000008); __cpuid(cpu_id, 0x80000008);
int apic_id_core_id_size = (cpu_id[2] >> 12) & 0xF; int apic_id_core_id_size = (cpu_id[2] >> 12) & 0xF;
if (apic_id_core_id_size == 0) { if (apic_id_core_id_size == 0)
if (ht) { {
if (ht)
{
// New mechanism for modern Intel CPUs. // New mechanism for modern Intel CPUs.
if (vendor == VENDOR_INTEL) { if (vendor == VENDOR_INTEL)
{
__cpuid(cpu_id, 0x00000004); __cpuid(cpu_id, 0x00000004);
int cores_x_package = ((cpu_id[0] >> 26) & 0x3F) + 1; int cores_x_package = ((cpu_id[0] >> 26) & 0x3F) + 1;
HTT = (cores_x_package < logical_cpu_count); HTT = (cores_x_package < logical_cpu_count);
@ -201,7 +222,8 @@ void CPUInfo::Detect()
logical_cpu_count /= cores_x_package; logical_cpu_count /= cores_x_package;
} }
} }
} else { } else
{
// Use AMD's new method. // Use AMD's new method.
num_cores = (cpu_id[2] & 0xFF) + 1; num_cores = (cpu_id[2] & 0xFF) + 1;
} }
@ -225,6 +247,9 @@ std::string CPUInfo::Summarize()
if (bSSE4_2) sum += ", SSE4.2"; if (bSSE4_2) sum += ", SSE4.2";
if (HTT) sum += ", HTT"; if (HTT) sum += ", HTT";
if (bAVX) sum += ", AVX"; if (bAVX) sum += ", AVX";
if (bAVX2) sum += ", AVX2";
if (bBMI1) sum += ", BMI1";
if (bBMI2) sum += ", BMI2";
if (bFMA) sum += ", FMA"; if (bFMA) sum += ", FMA";
if (bAES) sum += ", AES"; if (bAES) sum += ", AES";
if (bMOVBE) sum += ", MOVBE"; if (bMOVBE) sum += ", MOVBE";