ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

x86emitter: Purge x86caps 

We can use cpuinfo for querying AVX/AVX2.
This commit is contained in:
Stenzek 2023-12-26 18:13:37 +10:00 committed by Connor McLaughlin
parent b121e5af25
commit f461bc9176
13 changed files with 33 additions and 464 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
common/CMakeLists.txt
View File

@ -37,7 +37,6 @@ target_sources(common PRIVATE
WindowInfo.cpp WindowInfo.cpp
emitter/avx.cpp emitter/avx.cpp
emitter/bmi.cpp emitter/bmi.cpp
emitter/cpudetect.cpp
emitter/fpu.cpp emitter/fpu.cpp
emitter/groups.cpp emitter/groups.cpp
emitter/jmp.cpp emitter/jmp.cpp
@ -45,8 +44,6 @@ target_sources(common PRIVATE
emitter/legacy_sse.cpp emitter/legacy_sse.cpp
emitter/movs.cpp emitter/movs.cpp
emitter/simd.cpp emitter/simd.cpp
emitter/LnxCpuDetect.cpp
emitter/WinCpuDetect.cpp
emitter/x86emitter.cpp emitter/x86emitter.cpp
Darwin/DarwinThreads.cpp Darwin/DarwinThreads.cpp
Darwin/DarwinMisc.cpp Darwin/DarwinMisc.cpp

3
common/common.vcxproj
View File

@ -88,7 +88,6 @@
<ClCompile Include="Semaphore.cpp" /> <ClCompile Include="Semaphore.cpp" />
<ClCompile Include="emitter\avx.cpp" /> <ClCompile Include="emitter\avx.cpp" />
<ClCompile Include="emitter\bmi.cpp" /> <ClCompile Include="emitter\bmi.cpp" />
<ClCompile Include="emitter\cpudetect.cpp" />
<ClCompile Include="emitter\fpu.cpp" /> <ClCompile Include="emitter\fpu.cpp" />
<ClCompile Include="emitter\groups.cpp" /> <ClCompile Include="emitter\groups.cpp" />
<ClCompile Include="emitter\jmp.cpp" /> <ClCompile Include="emitter\jmp.cpp" />
@ -97,8 +96,6 @@
<ClCompile Include="emitter\movs.cpp" /> <ClCompile Include="emitter\movs.cpp" />
<ClCompile Include="emitter\simd.cpp" /> <ClCompile Include="emitter\simd.cpp" />
<ClCompile Include="emitter\x86emitter.cpp" /> <ClCompile Include="emitter\x86emitter.cpp" />
<ClCompile Include="emitter\LnxCpuDetect.cpp" />
<ClCompile Include="emitter\WinCpuDetect.cpp" />
</ItemGroup> </ItemGroup>
<ItemGroup> <ItemGroup>
<MASM Include="FastJmp.asm" /> <MASM Include="FastJmp.asm" />

9
common/common.vcxproj.filters
View File

@ -10,9 +10,6 @@
<ClCompile Include="Console.cpp"> <ClCompile Include="Console.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
<ClCompile Include="emitter\cpudetect.cpp">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="emitter\fpu.cpp"> <ClCompile Include="emitter\fpu.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
@ -28,9 +25,6 @@
<ClCompile Include="emitter\legacy_sse.cpp"> <ClCompile Include="emitter\legacy_sse.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
<ClCompile Include="emitter\LnxCpuDetect.cpp">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="emitter\avx.cpp"> <ClCompile Include="emitter\avx.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
@ -64,9 +58,6 @@
<ClCompile Include="emitter\simd.cpp"> <ClCompile Include="emitter\simd.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>
<ClCompile Include="emitter\WinCpuDetect.cpp">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="Windows\WinHostSys.cpp"> <ClCompile Include="Windows\WinHostSys.cpp">
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</ClCompile> </ClCompile>

23
common/emitter/LnxCpuDetect.cpp
View File

@ -1,23 +0,0 @@
// SPDX-FileCopyrightText: 2002-2023 PCSX2 Dev Team
// SPDX-License-Identifier: LGPL-3.0+
#ifndef _WIN32
#include "common/emitter/tools.h"
#include <unistd.h>
// Note: Apparently this solution is Linux/Solaris only.
// FreeBSD/OsX need something far more complicated (apparently)
void x86capabilities::CountLogicalCores()
{
#ifdef __linux__
// Note : GetCPUCount uses sysconf( _SC_NPROCESSORS_ONLN ) internally, which can return 1
// if sysconf info isn't available (a long standing linux bug). There are no fallbacks or
// alternatives, apparently.
LogicalCores = sysconf(_SC_NPROCESSORS_ONLN);
#else
LogicalCores = 1;
#endif
}
#endif

28
common/emitter/WinCpuDetect.cpp
View File

@ -1,28 +0,0 @@
// SPDX-FileCopyrightText: 2002-2023 PCSX2 Dev Team
// SPDX-License-Identifier: LGPL-3.0+
#if defined(_WIN32)
#include "common/Console.h"
#include "common/emitter/tools.h"
#include "common/RedtapeWindows.h"
void x86capabilities::CountLogicalCores()
{
DWORD_PTR vProcessCPUs;
DWORD_PTR vSystemCPUs;
LogicalCores = 1;
if (!GetProcessAffinityMask(GetCurrentProcess(), &vProcessCPUs, &vSystemCPUs))
return;
uint CPUs = 0;
for (DWORD_PTR bit = 1; bit != 0; bit <<= 1)
if (vSystemCPUs & bit)
CPUs++;
LogicalCores = CPUs;
}
#endif

217
common/emitter/cpudetect.cpp
View File

@ -1,217 +0,0 @@
// SPDX-FileCopyrightText: 2002-2023 PCSX2 Dev Team
// SPDX-License-Identifier: LGPL-3.0+
#include "common/General.h"
#include "common/emitter/tools.h"
#include "common/emitter/internal.h"
#include "common/VectorIntrin.h"
#include <atomic>
// CPU information support
#if defined(_WIN32)
#define cpuid __cpuid
#define cpuidex __cpuidex
#else
#include <cpuid.h>
static __inline__ __attribute__((always_inline)) void cpuidex(int CPUInfo[], const int InfoType, const int count)
{
__cpuid_count(InfoType, count, CPUInfo[0], CPUInfo[1], CPUInfo[2], CPUInfo[3]);
}
static __inline__ __attribute__((always_inline)) void cpuid(int CPUInfo[], const int InfoType)
{
__cpuid(InfoType, CPUInfo[0], CPUInfo[1], CPUInfo[2], CPUInfo[3]);
}
#endif
using namespace x86Emitter;
alignas(16) x86capabilities x86caps;
#if defined(_MSC_VER)
// We disable optimizations for this function, because we need x86capabilities for AVX
// detection, but if we keep opts on, it'll use AVX instructions for inlining memzero.
#pragma optimize("", off)
#endif
x86capabilities::x86capabilities()
: isIdentified(false)
, VendorID(x86Vendor_Unknown)
, FamilyID(0)
, Model(0)
, TypeID(0)
, StepID(0)
, Flags(0)
, Flags2(0)
, EFlags(0)
, EFlags2(0)
, SEFlag(0)
, AllCapabilities(0)
, PhysicalCores(0)
, LogicalCores(0)
{
}
#if defined(_MSC_VER)
#pragma optimize("", on)
#endif
const char* x86capabilities::GetTypeName() const
{
switch (TypeID)
{
case 0:
return "Standard OEM";
case 1:
return "Overdrive";
case 2:
return "Dual";
case 3:
return "Reserved";
default:
return "Unknown";
}
}
void x86capabilities::CountCores()
{
Identify();
// This will assign values into LogicalCores and PhysicalCores
CountLogicalCores();
}
static const char* tbl_x86vendors[] =
{
"GenuineIntel",
"AuthenticAMD",
"Unknown ",
};
// Performs all _cpuid-related activity. This fills *most* of the x86caps structure, except for
// the cpuSpeed and the mxcsr masks. Those must be completed manually.
void x86capabilities::Identify()
{
if (isIdentified)
return;
isIdentified = true;
s32 regs[4];
u32 cmds;
cpuid(regs, 0);
cmds = regs[0];
memcpy(&VendorName[0], &regs[1], 4);
memcpy(&VendorName[4], &regs[3], 4);
memcpy(&VendorName[8], &regs[2], 4);
// Determine Vendor Specifics!
// It's really not recommended that we base much (if anything) on CPU vendor names,
// however it's currently necessary in order to gain a (pseudo)reliable count of cores
// and threads used by the CPU (AMD and Intel can't agree on how to make this info available).
int vid;
for (vid = 0; vid < x86Vendor_Unknown; ++vid)
{
if (memcmp(VendorName, tbl_x86vendors[vid], 12) == 0)
break;
}
VendorID = static_cast<x86VendorType>(vid);
if (cmds >= 0x00000001)
{
cpuid(regs, 0x00000001);
StepID = regs[0] & 0xf;
Model = (regs[0] >> 4) & 0xf;
FamilyID = (regs[0] >> 8) & 0xf;
TypeID = (regs[0] >> 12) & 0x3;
//u32 x86_64_8BITBRANDID = regs[1] & 0xff;
Flags = regs[3];
Flags2 = regs[2];
}
if (cmds >= 0x00000007)
{
// Note: ECX must be 0 for AVX2 detection.
cpuidex(regs, 0x00000007, 0);
SEFlag = regs[1];
}
cpuid(regs, 0x80000000);
cmds = regs[0];
if (cmds >= 0x80000001)
{
cpuid(regs, 0x80000001);
//u32 x86_64_12BITBRANDID = regs[1] & 0xfff;
EFlags2 = regs[2];
EFlags = regs[3];
}
cpuid((int*)FamilyName, 0x80000002);
cpuid((int*)(FamilyName + 16), 0x80000003);
cpuid((int*)(FamilyName + 32), 0x80000004);
hasFloatingPointUnit = (Flags >> 0) & 1;
hasVirtual8086ModeEnhancements = (Flags >> 1) & 1;
hasDebuggingExtensions = (Flags >> 2) & 1;
hasPageSizeExtensions = (Flags >> 3) & 1;
hasTimeStampCounter = (Flags >> 4) & 1;
hasModelSpecificRegisters = (Flags >> 5) & 1;
hasPhysicalAddressExtension = (Flags >> 6) & 1;
hasMachineCheckArchitecture = (Flags >> 7) & 1;
hasCOMPXCHG8BInstruction = (Flags >> 8) & 1;
hasAdvancedProgrammableInterruptController = (Flags >> 9) & 1;
hasSEPFastSystemCall = (Flags >> 11) & 1;
hasMemoryTypeRangeRegisters = (Flags >> 12) & 1;
hasPTEGlobalFlag = (Flags >> 13) & 1;
hasMachineCheckArchitecture = (Flags >> 14) & 1;
hasConditionalMoveAndCompareInstructions = (Flags >> 15) & 1;
hasFGPageAttributeTable = (Flags >> 16) & 1;
has36bitPageSizeExtension = (Flags >> 17) & 1;
hasProcessorSerialNumber = (Flags >> 18) & 1;
hasCFLUSHInstruction = (Flags >> 19) & 1;
hasDebugStore = (Flags >> 21) & 1;
hasACPIThermalMonitorAndClockControl = (Flags >> 22) & 1;
hasFastStreamingSIMDExtensionsSaveRestore = (Flags >> 24) & 1;
hasStreamingSIMDExtensions = (Flags >> 25) & 1; //sse
hasStreamingSIMD2Extensions = (Flags >> 26) & 1; //sse2
hasSelfSnoop = (Flags >> 27) & 1;
hasMultiThreading = (Flags >> 28) & 1;
hasThermalMonitor = (Flags >> 29) & 1;
hasIntel64BitArchitecture = (Flags >> 30) & 1;
// -------------------------------------------------
// --> SSE3 / SSSE3 / SSE4.1 / SSE 4.2 detection <--
// -------------------------------------------------
hasStreamingSIMD3Extensions = (Flags2 >> 0) & 1; //sse3
hasSupplementalStreamingSIMD3Extensions = (Flags2 >> 9) & 1; //ssse3
hasStreamingSIMD4Extensions = (Flags2 >> 19) & 1; //sse4.1
hasStreamingSIMD4Extensions2 = (Flags2 >> 20) & 1; //sse4.2
if ((Flags2 >> 27) & 1) // OSXSAVE
{
// Note: In theory, we should use xgetbv to check OS support
// but all OSes we officially run under support it
// and its intrinsic requires extra compiler flags
hasAVX = (Flags2 >> 28) & 1; //avx
hasFMA = (Flags2 >> 12) & 1; //fma
hasAVX2 = (SEFlag >> 5) & 1; //avx2
}
hasBMI1 = (SEFlag >> 3) & 1;
hasBMI2 = (SEFlag >> 8) & 1;
// Ones only for AMDs:
hasAMD64BitArchitecture = (EFlags >> 29) & 1; //64bit cpu
hasStreamingSIMD4ExtensionsA = (EFlags2 >> 6) & 1; //INSERTQ / EXTRQ / MOVNT
isIdentified = true;
}

105
common/emitter/tools.h
View File

@ -5,109 +5,6 @@
#include "common/Pcsx2Defs.h" #include "common/Pcsx2Defs.h"
enum x86VendorType
{
x86Vendor_Intel = 0,
x86Vendor_AMD = 1,
x86Vendor_Unknown = 2
};
// --------------------------------------------------------------------------------------
// x86capabilities
// --------------------------------------------------------------------------------------
class x86capabilities
{
public:
bool isIdentified;
public:
x86VendorType VendorID;
uint FamilyID; // Processor Family
uint Model; // Processor Model
uint TypeID; // Processor Type
uint StepID; // Stepping ID
u32 Flags; // Feature Flags
u32 Flags2; // More Feature Flags
u32 EFlags; // Extended Feature Flags
u32 EFlags2; // Extended Feature Flags pg2
u32 SEFlag; // Structured Extended Feature Flags Enumeration
char VendorName[16] = {}; // Vendor/Creator ID
char FamilyName[50] = {}; // the original cpu name
// ----------------------------------------------------------------------------
// x86 CPU Capabilities Section (all boolean flags!)
// ----------------------------------------------------------------------------
union
{
u64 AllCapabilities = 0;
struct
{
u32 hasFloatingPointUnit : 1;
u32 hasVirtual8086ModeEnhancements : 1;
u32 hasDebuggingExtensions : 1;
u32 hasPageSizeExtensions : 1;
u32 hasTimeStampCounter : 1;
u32 hasModelSpecificRegisters : 1;
u32 hasPhysicalAddressExtension : 1;
u32 hasCOMPXCHG8BInstruction : 1;
u32 hasAdvancedProgrammableInterruptController : 1;
u32 hasSEPFastSystemCall : 1;
u32 hasMemoryTypeRangeRegisters : 1;
u32 hasPTEGlobalFlag : 1;
u32 hasMachineCheckArchitecture : 1;
u32 hasConditionalMoveAndCompareInstructions : 1;
u32 hasFGPageAttributeTable : 1;
u32 has36bitPageSizeExtension : 1;
u32 hasProcessorSerialNumber : 1;
u32 hasCFLUSHInstruction : 1;
u32 hasDebugStore : 1;
u32 hasACPIThermalMonitorAndClockControl : 1;
u32 hasFastStreamingSIMDExtensionsSaveRestore : 1;
u32 hasStreamingSIMDExtensions : 1;
u32 hasStreamingSIMD2Extensions : 1;
u32 hasSelfSnoop : 1;
// is TRUE for both multi-core and Hyperthreaded CPUs.
u32 hasMultiThreading : 1;
u32 hasThermalMonitor : 1;
u32 hasIntel64BitArchitecture : 1;
u32 hasStreamingSIMD3Extensions : 1;
u32 hasSupplementalStreamingSIMD3Extensions : 1;
u32 hasStreamingSIMD4Extensions : 1;
u32 hasStreamingSIMD4Extensions2 : 1;
u32 hasAVX : 1;
u32 hasAVX2 : 1;
u32 hasBMI1 : 1;
u32 hasBMI2 : 1;
u32 hasFMA : 1;
// AMD-specific CPU Features
u32 hasAMD64BitArchitecture : 1;
u32 hasStreamingSIMD4ExtensionsA : 1;
};
};
// Core Counts!
u32 PhysicalCores = 0;
u32 LogicalCores = 0;
public:
x86capabilities();
void Identify();
void CountCores();
const char* GetTypeName() const;
protected:
void CountLogicalCores();
};
enum SSE_RoundMode enum SSE_RoundMode
{ {
SSE_RoundMode_FIRST = 0, SSE_RoundMode_FIRST = 0,
@ -185,5 +82,3 @@ union SSE_MXCSR
operator x86Emitter::xIndirect32() const; operator x86Emitter::xIndirect32() const;
}; };
alignas(16) extern x86capabilities x86caps;

2
common/emitter/x86emitter.cpp
View File

@ -563,7 +563,7 @@ const xRegister32
// Core2/i7 CPUs prefer unaligned addresses. Checking for SSSE3 is a decent filter. // Core2/i7 CPUs prefer unaligned addresses. Checking for SSSE3 is a decent filter.
// (also align in debug modes for disasm convenience) // (also align in debug modes for disasm convenience)
if (IsDebugBuild || !x86caps.hasSupplementalStreamingSIMD3Extensions) if constexpr (IsDebugBuild)
{ {
// - P4's and earlier prefer 16 byte alignment. // - P4's and earlier prefer 16 byte alignment.
// - AMD Athlons and Phenoms prefer 8 byte alignment, but I don't have an easy // - AMD Athlons and Phenoms prefer 8 byte alignment, but I don't have an easy

5
pcsx2/System.cpp
View File

@ -106,7 +106,6 @@ void SysLogMachineCaps()
GetOSVersionString().c_str(), GetOSVersionString().c_str(),
(u32)(GetPhysicalMemory() / _1mb)); (u32)(GetPhysicalMemory() / _1mb));
cpuinfo_initialize();
Console.Indent().WriteLn("Processor = %s", cpuinfo_get_package(0)->name); Console.Indent().WriteLn("Processor = %s", cpuinfo_get_package(0)->name);
Console.Indent().WriteLn("Core Count = %u cores", cpuinfo_get_cores_count()); Console.Indent().WriteLn("Core Count = %u cores", cpuinfo_get_cores_count());
Console.Indent().WriteLn("Thread Count = %u threads", cpuinfo_get_processors_count()); Console.Indent().WriteLn("Thread Count = %u threads", cpuinfo_get_processors_count());
@ -115,9 +114,9 @@ void SysLogMachineCaps()
std::string features; std::string features;
if (x86caps.hasAVX) if (cpuinfo_has_x86_avx())
features += "AVX "; features += "AVX ";
if (x86caps.hasAVX2) if (cpuinfo_has_x86_avx2())
features += "AVX2 "; features += "AVX2 ";
StringUtil::StripWhitespace(&features); StringUtil::StripWhitespace(&features);

29
pcsx2/VMManager.cpp
View File

@ -51,6 +51,7 @@
#include "common/Timer.h" #include "common/Timer.h"
#include "IconsFontAwesome5.h" #include "IconsFontAwesome5.h"
#include "cpuinfo.h"
#include "discord_rpc.h" #include "discord_rpc.h"
#include "fmt/core.h" #include "fmt/core.h"
@ -186,10 +187,9 @@ bool VMManager::PerformEarlyHardwareChecks(const char** error)
#if defined(_M_X86) #if defined(_M_X86)
// On Windows, this gets called as a global object constructor, before any of our objects are constructed. // On Windows, this gets called as a global object constructor, before any of our objects are constructed.
// So, we have to put it on the stack instead. // So, we have to put it on the stack instead.
x86capabilities temp_x86_caps; cpuinfo_initialize();
temp_x86_caps.Identify();
if (!temp_x86_caps.hasStreamingSIMD4Extensions) if (!cpuinfo_has_x86_sse4_1())
{ {
*error = *error =
"PCSX2 requires the Streaming SIMD 4.1 Extensions instruction set, which your CPU does not support.\n\n" "PCSX2 requires the Streaming SIMD 4.1 Extensions instruction set, which your CPU does not support.\n\n"
@ -199,7 +199,7 @@ bool VMManager::PerformEarlyHardwareChecks(const char** error)
} }
#if _M_SSE >= 0x0501 #if _M_SSE >= 0x0501
if (!temp_x86_caps.hasAVX || !temp_x86_caps.hasAVX2) if (!cpuinfo_has_x86_avx2())
{ {
*error = "This build of PCSX2 requires the Advanced Vector Extensions 2 instruction set, which your CPU does " *error = "This build of PCSX2 requires the Advanced Vector Extensions 2 instruction set, which your CPU does "
"not support.\n\n" "not support.\n\n"
@ -342,8 +342,9 @@ bool VMManager::Internal::CPUThreadInitialize()
} }
#endif #endif
x86caps.Identify(); if (!cpuinfo_initialize())
x86caps.CountCores(); Console.Error("cpuinfo_initialize() failed.");
SysLogMachineCaps(); SysLogMachineCaps();
if (!SysMemory::Allocate()) if (!SysMemory::Allocate())
@ -2803,8 +2804,6 @@ static std::once_flag s_processor_list_initialized;
#if defined(__linux__) || defined(_WIN32) #if defined(__linux__) || defined(_WIN32)
#include "cpuinfo.h"
static u32 GetProcessorIdForProcessor(const cpuinfo_processor* proc) static u32 GetProcessorIdForProcessor(const cpuinfo_processor* proc)
{ {
#if defined(__linux__) #if defined(__linux__)
@ -2816,14 +2815,8 @@ static u32 GetProcessorIdForProcessor(const cpuinfo_processor* proc)
#endif #endif
} }
static void InitializeCPUInfo() static void InitializeProcessorList()
{ {
if (!cpuinfo_initialize())
{
Console.Error("Failed to initialize cpuinfo");
return;
}
const u32 cluster_count = cpuinfo_get_clusters_count(); const u32 cluster_count = cpuinfo_get_clusters_count();
if (cluster_count == 0) if (cluster_count == 0)
{ {
@ -2922,7 +2915,7 @@ static void SetMTVUAndAffinityControlDefault(SettingsInterface& si)
static u32 s_big_cores; static u32 s_big_cores;
static u32 s_small_cores; static u32 s_small_cores;
static void InitializeCPUInfo() static void InitializeProcessorList()
{ {
s_big_cores = 0; s_big_cores = 0;
s_small_cores = 0; s_small_cores = 0;
@ -2957,7 +2950,7 @@ static void SetMTVUAndAffinityControlDefault(SettingsInterface& si)
#else #else
static void InitializeCPUInfo() static void InitializeProcessorList()
{ {
DevCon.WriteLn("(VMManager) InitializeCPUInfo() not implemented."); DevCon.WriteLn("(VMManager) InitializeCPUInfo() not implemented.");
} }
@ -2970,7 +2963,7 @@ static void SetMTVUAndAffinityControlDefault(SettingsInterface& si)
void VMManager::EnsureCPUInfoInitialized() void VMManager::EnsureCPUInfoInitialized()
{ {
std::call_once(s_processor_list_initialized, InitializeCPUInfo); std::call_once(s_processor_list_initialized, InitializeProcessorList);
} }
void VMManager::SetEmuThreadAffinities() void VMManager::SetEmuThreadAffinities()

59
pcsx2/x86/iMMI.cpp
View File

@ -2618,18 +2618,8 @@ void recPSRAVW()
xPSRA.D(xRegisterSSE(t1reg), xRegisterSSE(t0reg)); xPSRA.D(xRegisterSSE(t1reg), xRegisterSSE(t0reg));
// merge & sign extend // merge & sign extend
if (x86caps.hasStreamingSIMD4Extensions) xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
{ xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else
{
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31); // get the signs
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
}
_freeXMMreg(t0reg); _freeXMMreg(t0reg);
_freeXMMreg(t1reg); _freeXMMreg(t1reg);
@ -2739,25 +2729,10 @@ void recPMULTUW()
} }
// interleave & sign extend // interleave & sign extend
if (x86caps.hasStreamingSIMD4Extensions) xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88);
{ xPSHUF.D(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0xdd);
xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88); xPMOVSX.DQ(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO));
xPSHUF.D(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0xdd); xPMOVSX.DQ(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI));
xPMOVSX.DQ(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO));
xPMOVSX.DQ(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI));
}
else
{
int t0reg = _allocTempXMMreg(XMMT_INT);
xPSHUF.D(xRegisterSSE(t0reg), xRegisterSSE(EEREC_HI), 0xd8);
xMOVDQA(xRegisterSSE(EEREC_LO), xRegisterSSE(t0reg));
xMOVDQA(xRegisterSSE(EEREC_HI), xRegisterSSE(t0reg));
xPSRA.D(xRegisterSSE(t0reg), 31); // get the signs
xPUNPCK.LDQ(xRegisterSSE(EEREC_LO), xRegisterSSE(t0reg));
xPUNPCK.HDQ(xRegisterSSE(EEREC_HI), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
} }
_clearNeededXMMregs(); _clearNeededXMMregs();
} }
@ -2805,25 +2780,11 @@ void recPMADDUW()
xPADD.Q(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_LO)); xPADD.Q(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_LO));
// interleave & sign extend // interleave & sign extend
if (x86caps.hasStreamingSIMD4Extensions) xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88);
{ xPSHUF.D(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0xdd);
xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88); xPMOVSX.DQ(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO));
xPSHUF.D(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0xdd); xPMOVSX.DQ(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI));
xPMOVSX.DQ(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO));
xPMOVSX.DQ(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI));
}
else
{
int t0reg = _allocTempXMMreg(XMMT_INT);
xPSHUF.D(xRegisterSSE(t0reg), xRegisterSSE(EEREC_HI), 0xd8);
xMOVDQA(xRegisterSSE(EEREC_LO), xRegisterSSE(t0reg));
xMOVDQA(xRegisterSSE(EEREC_HI), xRegisterSSE(t0reg));
xPSRA.D(xRegisterSSE(t0reg), 31); // get the signs
xPUNPCK.LDQ(xRegisterSSE(EEREC_LO), xRegisterSSE(t0reg));
xPUNPCK.HDQ(xRegisterSSE(EEREC_HI), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
_clearNeededXMMregs(); _clearNeededXMMregs();
} }

6
pcsx2/x86/microVU_Execute.inl
View File

@ -3,6 +3,8 @@
#pragma once #pragma once
#include "cpuinfo.h"
//------------------------------------------------------------------ //------------------------------------------------------------------
// Dispatcher Functions // Dispatcher Functions
//------------------------------------------------------------------ //------------------------------------------------------------------
@ -204,7 +206,7 @@ static void mVUGenerateCopyPipelineState(mV)
{ {
mVU.copyPLState = xGetAlignedCallTarget(); mVU.copyPLState = xGetAlignedCallTarget();
if (x86caps.hasAVX2) if (cpuinfo_has_x86_avx())
{ {
xVMOVAPS(ymm0, ptr[rax]); xVMOVAPS(ymm0, ptr[rax]);
xVMOVAPS(ymm1, ptr[rax + 32u]); xVMOVAPS(ymm1, ptr[rax + 32u]);
@ -249,7 +251,7 @@ static void mVUGenerateCompareState(mV)
{ {
mVU.compareStateF = xGetAlignedCallTarget(); mVU.compareStateF = xGetAlignedCallTarget();
if (!x86caps.hasAVX2) if (!cpuinfo_has_x86_avx2())
{ {
xMOVAPS (xmm0, ptr32[arg1reg]); xMOVAPS (xmm0, ptr32[arg1reg]);
xPCMP.EQD(xmm0, ptr32[arg2reg]); xPCMP.EQD(xmm0, ptr32[arg2reg]);

8
tests/ctest/core/GS/swizzle_test_main.cpp
View File

@ -8,6 +8,8 @@
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include <string.h> #include <string.h>
#include "cpuinfo.h"
#ifdef MULTI_ISA_UNSHARED_COMPILATION #ifdef MULTI_ISA_UNSHARED_COMPILATION
#include "common/emitter/tools.h" #include "common/emitter/tools.h"
@ -22,10 +24,10 @@ enum class TestISA
static bool CheckCapabilities(TestISA required_caps) static bool CheckCapabilities(TestISA required_caps)
{ {
x86caps.Identify(); cpuinfo_initialize();
if (required_caps == TestISA::isa_avx && !x86caps.hasAVX) if (required_caps == TestISA::isa_avx && !cpuinfo_has_x86_avx())
return false; return false;
if (required_caps == TestISA::isa_avx2 && !x86caps.hasAVX2) if (required_caps == TestISA::isa_avx2 && !cpuinfo_has_x86_avx2())
return false; return false;
return true; return true;