AI/rocm-systems
2
0
Форкнуть 0
Код Задачи Запросы на слияние Действия Пакеты Проекты Релизы Вики Активность
Files
c7b50bb8907c00faee152dccf5be7d1f888e44f9
rocm-systems/rocclr/device/gpu/gpudevice.cpp
T
Vladislav Sytchenko c7b50bb890 SWDEV-280473 - Support HSAIL shared library build 
This change makes HSAIL usage similar to that of Comgr. By default, the
runtime will statically link against it, however if HSAIL_DYN_DLL is
defined, then the runtime will try to dynamically load HSAIL.

Currently stick to statically linking to HSAIL. In a feature patch the
dynamic loading behaviour will be enabled.

Change-Id: I6a78a4375975cf847f236b200404c8cf941d012b
2021-04-14 12:25:54 -04:00

2273 строки
76 KiB
C++
Исходник Ответственный История

/* Copyright (c) 2008-present Advanced Micro Devices, Inc.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE. */
#include "platform/program.hpp"
#include "platform/kernel.hpp"
#include "os/os.hpp"
#include "device/device.hpp"
#include "device/gpu/gpudefs.hpp"
#include "device/gpu/gpumemory.hpp"
#include "device/gpu/gpudevice.hpp"
#include "utils/flags.hpp"
#include "utils/versions.hpp"
#include "utils/options.hpp"
#include "thread/monitor.hpp"
#include "device/gpu/gpuprogram.hpp"
#include "device/gpu/gpubinary.hpp"
#include "device/gpu/gpusettings.hpp"
#include "device/gpu/gpublit.hpp"
#include "cz_id.h"
#include "hsailctx.hpp"
#include "vdi_common.hpp"
#include "CL/cl_gl.h"
#ifdef _WIN32
#include <d3d9.h>
#include <d3d10_1.h>
#include "CL/cl_d3d10.h"
#include "CL/cl_d3d11.h"
#include "CL/cl_dx9_media_sharing.h"
#endif // _WIN32
#include "os_if.h" // for osInit()
#include "gpudebugmanager.hpp"
#include <algorithm>
#include <ctype.h>
#include <cstring>
#include <fstream>
#include <iostream>
#include <sstream>
namespace {
//! Define the mapping from CAL asic enumeration values to the
//! compiler gfx major/minor/stepping version.
struct CalDevice {
uint32_t gfxipMajor_; //!< The core engine GFXIP Major version
uint32_t gfxipMinor_; //!< The core engine GFXIP Minor version
uint32_t gfxipStepping_; //!< The core engine GFXIP Stepping version
CALMachineType calMachine_; //!< CAL machine type
const char* calName_; //!< CAL device name
CALtarget calTarget_; //!< CAL target
bool preferPal_; //!< Prefer to use PAL if GPU_ENABLE_PAL=2
bool nullUseDouble_; //!< Use double precision for a NullDevice
bool nullUseOpenCL200_; //!< Use OpenCL 2.0 for a NullDevice
};
static constexpr CalDevice supportedCalDevices[] = {
// Prefer - NullDevice -
// GFX Version GSL Machine CAL Name CAL Target PAL double OCL200
{7, 0, 0, ED_ATI_CAL_MACHINE_KALINDI_ISA, "Kalindi", CAL_TARGET_KALINDI, false, true, true },
{7, 0, 0, ED_ATI_CAL_MACHINE_SPECTRE_ISA, "Spectre", CAL_TARGET_SPECTRE, false, true, true },
{7, 0, 0, ED_ATI_CAL_MACHINE_SPOOKY_ISA, "Spooky", CAL_TARGET_SPOOKY, false, true, true },
{7, 0, 2, ED_ATI_CAL_MACHINE_HAWAII_ISA, "Hawaii", CAL_TARGET_HAWAII, false, true, true }, // Also Hawaiipro (generated code is for Hawaiipro)
{7, 0, 4, ED_ATI_CAL_MACHINE_BONAIRE_ISA, "Bonaire", CAL_TARGET_BONAIRE, false, true, true },
{7, 0, 5, ED_ATI_CAL_MACHINE_GODAVARI_ISA, "Mullins", CAL_TARGET_GODAVARI, false, true, true }, // FIXME: Why is this compiled as Mullins yet reported as Godavari? Add gfx703 to support Mullins.
{8, 0, 1, ED_ATI_CAL_MACHINE_CARRIZO_ISA, "Carrizo", CAL_TARGET_CARRIZO, false, true, true }, // Also Bristol Ridge
{8, 0, 2, ED_ATI_CAL_MACHINE_ICELAND_ISA, "Iceland", CAL_TARGET_ICELAND, false, true, true },
{8, 0, 2, ED_ATI_CAL_MACHINE_TONGA_ISA, "Tonga", CAL_TARGET_TONGA, false, true, true }, // Also Tongapro (generated code is for Tonga)
{8, 0, 3, ED_ATI_CAL_MACHINE_FIJI_ISA, "Fiji", CAL_TARGET_FIJI, false, true, true },
{8, 0, 3, ED_ATI_CAL_MACHINE_ELLESMERE_ISA, "Ellesmere", CAL_TARGET_ELLESMERE, false, true, true }, // Polaris10
{8, 0, 3, ED_ATI_CAL_MACHINE_BAFFIN_ISA, "Baffin", CAL_TARGET_BAFFIN, false, true, true }, // Polaris11
{8, 0, 3, ED_ATI_CAL_MACHINE_LEXA_ISA, "gfx803", CAL_TARGET_LEXA, false, true, true }, // Polaris12
#if !defined(BRAHMA)
{8, 0, 3, ED_ATI_CAL_MACHINE_POLARIS22_ISA, "gfx803", CAL_TARGET_POLARIS22, false, true, true },
#endif
{8, 1, 0, ED_ATI_CAL_MACHINE_STONEY_ISA, "Stoney", CAL_TARGET_STONEY, false, true, true },
#if !defined(BRAHMA)
{9, 0, 0, ED_ATI_CAL_MACHINE_GREENLAND_ISA, "gfx900", CAL_TARGET_GREENLAND, true, true, true }, // Vega10
{9, 0, 2, ED_ATI_CAL_MACHINE_RAVEN_ISA, "gfx902", CAL_TARGET_RAVEN, true, true, true },
{9, 0, 4, ED_ATI_CAL_MACHINE_VEGA12_ISA, "gfx904", CAL_TARGET_VEGA12, true, true, true },
{9, 0, 6, ED_ATI_CAL_MACHINE_VEGA20_ISA, "gfx906", CAL_TARGET_VEGA20, true, true, true },
{9, 0, 9, ED_ATI_CAL_MACHINE_RAVEN2_ISA, "gfx909", CAL_TARGET_RAVEN2, true, true, true },
{9, 0, 12, ED_ATI_CAL_MACHINE_RENOIR_ISA, "gfx90c", CAL_TARGET_RENOIR, true, true, true },
#endif
};
static_assert(CAL_TARGET_LAST == CAL_TARGET_VEGA20, "Add new CAL targets to mapping");
static std::tuple<const amd::Isa*, CALMachineType, const char*, bool, bool, bool> findIsa(
CALtarget calTarget, bool sramecc, bool xnack) {
auto calDeviceIter =
std::find_if(std::begin(supportedCalDevices), std::end(supportedCalDevices),
[&](const CalDevice& calDevice) { return calDevice.calTarget_ == calTarget; });
if (calDeviceIter == std::end(supportedCalDevices)) {
return std::make_tuple(nullptr, static_cast<CALMachineType>(0), nullptr, false, false, false);
}
const amd::Isa* isa = amd::Isa::findIsa(
calDeviceIter->gfxipMajor_, calDeviceIter->gfxipMinor_, calDeviceIter->gfxipStepping_,
sramecc ? amd::Isa::Feature::Enabled : amd::Isa::Feature::Disabled,
xnack ? amd::Isa::Feature::Enabled : amd::Isa::Feature::Disabled);
return std::make_tuple(isa, calDeviceIter->calMachine_, calDeviceIter->calName_,
calDeviceIter->preferPal_, calDeviceIter->nullUseDouble_,
calDeviceIter->nullUseOpenCL200_);
}
static std::tuple<bool, CALMachineType, CALtarget, const char*, bool, bool, bool> findCal(
uint32_t gfxipMajor, uint32_t gfxipMinor, uint32_t gfxipStepping) {
auto calDeviceIter = std::find_if(std::begin(supportedCalDevices), std::end(supportedCalDevices),
[&](const CalDevice& calDevice) {
return calDevice.gfxipMajor_ == gfxipMajor &&
calDevice.gfxipMinor_ == gfxipMinor &&
calDevice.gfxipStepping_ == gfxipStepping;
});
if (calDeviceIter == std::end(supportedCalDevices)) {
return std::make_tuple(false, static_cast<CALMachineType>(0), static_cast<CALtarget>(0),
nullptr, false, false, false);
}
return std::make_tuple(true, calDeviceIter->calMachine_, calDeviceIter->calTarget_,
calDeviceIter->calName_, calDeviceIter->preferPal_,
calDeviceIter->nullUseDouble_, calDeviceIter->nullUseOpenCL200_);
}
} // namespace
bool DeviceLoad() {
bool ret = false;
// Create online devices
ret |= gpu::Device::init();
// Create offline GPU devices
ret |= gpu::NullDevice::init();
return ret;
}
void DeviceUnload() { gpu::Device::tearDown(); }
namespace gpu {
aclCompiler* NullDevice::compiler_;
aclCompiler* NullDevice::hsaCompiler_;
AppProfile Device::appProfile_;
NullDevice::NullDevice()
: amd::Device(),
calTarget_(static_cast<CALtarget>(0)),
calMachine_(static_cast<CALMachineType>(0)),
calName_(nullptr) {}
bool NullDevice::init() {
// Create offline devices for all ISAs not already associated with an online
// device. This allows code objects to be compiled for all supported ISAs.
std::vector<Device*> devices = getDevices(CL_DEVICE_TYPE_GPU, false);
for (const amd::Isa *isa = amd::Isa::begin(); isa != amd::Isa::end(); isa++) {
if (!isa->runtimeGslSupported()) {
continue;
}
bool isOnline = false;
// Check if the particular device is online
for (size_t i = 0; i < devices.size(); i++) {
if (&(devices[i]->isa()) == isa) {
isOnline = true;
break;
}
}
if (isOnline) {
continue;
}
bool found;
CALMachineType calMachine;
CALtarget calTarget;
const char* calName;
bool preferPal;
bool nullUseDouble;
bool nullUseOpenCL200;
std::tie(found, calMachine, calTarget, calName, preferPal, nullUseDouble, nullUseOpenCL200) =
findCal(isa->versionMajor(), isa->versionMinor(), isa->versionStepping());
if (!found) {
// GSL does not support this asic.
continue;
}
std::unique_ptr<NullDevice> nullDevice(new NullDevice());
if (!nullDevice) {
LogPrintfError("Error allocating new instance of offline CAL Device %s", isa->targetId());
return false;
}
if (!nullDevice->create(calName, *isa, calTarget, preferPal, nullUseDouble, nullUseOpenCL200)) {
// Skip over unsupported devices
LogPrintfError("Skipping creating new instance of offline CAL Device %s", isa->targetId());
continue;
}
nullDevice.release()->registerDevice();
}
return true;
}
bool NullDevice::create(const char* calName, const amd::Isa& isa, CALtarget target,
bool preferPal, bool doublePrecision, bool openCL200) {
if (!isa.runtimeGslSupported()) {
LogPrintfError("Offline CAL device %s is not supported", isa.targetId());
return false;
}
if ((GPU_ENABLE_PAL == 2) && isa.runtimePalSupported() && preferPal) {
LogPrintfError("Skipping as GPU_ENABLE_PAL=2 indicating to use PAL for offline CAL device %s",
isa.targetId());
return false;
}
online_ = false;
calTarget_ = target;
calName_ = calName;
// sets up vaCacheAccess_ and vaCacheMap_.
if (!amd::Device::create(isa)) {
LogPrintfError("Unable to setup offline device for CAL device %s", isa.targetId());
return false;
}
CALdeviceattribs calAttr = {0};
calAttr.target = calTarget();
// Force double if it could be supported
if (doublePrecision) {
calAttr.doublePrecision = CAL_TRUE;
}
// Use OpenCL 2.0 if supported
if (openCL200) {
calAttr.isOpenCL200Device = CAL_TRUE;
}
settings_ = new gpu::Settings();
gpu::Settings* gpuSettings = reinterpret_cast<gpu::Settings*>(settings_);
// Create setting for the offline target
if ((gpuSettings == NULL) || !gpuSettings->create(calAttr)) {
LogPrintfError("GPU settings failed for offline device for CAL device %s", isa.targetId());
return false;
}
if (!ValidateHsail()) {
LogPrintfError("HSAIL initialization failed for offline CAL device %s", isa.targetId());
return false;
}
gslMemInfo memInfo = {0};
// Report 512MB for all offline devices
memInfo.cardMemAvailableBytes = 512 * Mi;
memInfo.cardLargestFreeBlockBytes = 512 * Mi;
calAttr.localRAM = 512;
// Fill the device info structure
fillDeviceInfo(calAttr, memInfo, 4096, 1, 0);
// Runtime doesn't know what local size could be on the real board
info_.maxGlobalVariableSize_ = static_cast<size_t>(512 * Mi);
if (NULL == hsaCompiler_) {
const char* library = getenv("HSA_COMPILER_LIBRARY");
aclCompilerOptions opts = {
sizeof(aclCompilerOptions_0_8), library, NULL, NULL, NULL, NULL, NULL, AMD_OCL_SC_LIB};
// Initialize the compiler handle
acl_error error;
hsaCompiler_ = amd::Hsail::CompilerInit(&opts, &error);
if (error != ACL_SUCCESS) {
LogPrintfError("Error initializing the compiler for offline CAL device %s", isa.targetId());
return false;
}
}
return true;
}
bool NullDevice::isHsailProgram(amd::option::Options* options) {
bool isCIPlus = settings().ciPlus_;
bool isBlit = false;
bool isSPIRV = false;
bool isClang = false;
bool isEDG = false;
bool isLegacy = false;
bool isOCL20 = false;
std::vector<amd::option::Options*> optvec;
bool isInputOptions = false;
if (options != NULL) {
optvec.push_back(options);
isInputOptions = true;
}
amd::option::Options parsedOptions;
constexpr bool OptionChangable = true;
constexpr bool LinkOptsOnly = false;
constexpr bool IsLC = false;
if (!amd::Program::ParseAllOptions("", parsedOptions, OptionChangable, LinkOptsOnly, IsLC)) {
return NULL;
}
optvec.push_back(&parsedOptions);
for (auto const op : optvec) {
// TODO: Remove isOCL20 related code from this function along with switching HSAIL by default
if (isCIPlus && amd::Program::GetOclCVersion(op->oVariables->CLStd) >= 20) {
isOCL20 = true;
}
if (op->oVariables->clInternalKernel) {
isBlit = true;
break;
}
if (!isLegacy) {
isLegacy = op->oVariables->Legacy;
}
// Checks Frontend option only from input *options, not from Env,
// because they might be only calculated by RT based on the binaries to link.
// -frontend is being queried now instead of -cl-std=CL2.0, because the last one
// is not an indicator for HSAIL path anymore.
// TODO: Revise these binary's target checks
// and possibly remove them after switching to HSAIL by default.
if (isInputOptions) {
if (!isClang) {
isClang = op->isCStrOptionsEqual(op->oVariables->Frontend, "clang");
}
if (!isEDG) {
isEDG = op->isCStrOptionsEqual(op->oVariables->Frontend, "edg");
}
}
if (!isSPIRV) {
isSPIRV = op->oVariables->BinaryIsSpirv;
}
isInputOptions = false;
}
if (isSPIRV || (isBlit && isCIPlus) || isClang || isOCL20) {
return true;
}
if (isLegacy || isEDG) {
return false;
}
return true;
}
device::Program* NullDevice::createProgram(amd::Program& owner, amd::option::Options* options) {
if (isHsailProgram(options)) {
return new HSAILProgram(*this, owner);
}
return new NullProgram(*this, owner);
}
void NullDevice::fillDeviceInfo(const CALdeviceattribs& calAttr, const gslMemInfo& memInfo,
size_t maxTextureSize, uint numComputeRings,
uint numComputeRingsRT) {
info_.type_ = CL_DEVICE_TYPE_GPU;
info_.vendorId_ = 0x1002;
info_.maxComputeUnits_ = calAttr.numberOfSIMD;
info_.maxWorkItemDimensions_ = 3;
info_.numberOfShaderEngines = calAttr.numberOfShaderEngines;
// SI parts are scalar. Also, reads don't need to be 128-bits to get peak rates.
// For example, float4 is not faster than float as long as all threads fetch the same
// amount of data and the reads are coalesced. This is from the H/W team and confirmed
// through experimentation. May also be true on EG/NI, but no point in confusing
// developers now.
info_.nativeVectorWidthChar_ = info_.preferredVectorWidthChar_ = 4;
info_.nativeVectorWidthShort_ = info_.preferredVectorWidthShort_ = 2;
info_.nativeVectorWidthInt_ = info_.preferredVectorWidthInt_ = 1;
info_.nativeVectorWidthLong_ = info_.preferredVectorWidthLong_ = 1;
info_.nativeVectorWidthFloat_ = info_.preferredVectorWidthFloat_ = 1;
info_.nativeVectorWidthDouble_ = info_.preferredVectorWidthDouble_ =
(settings().checkExtension(ClKhrFp64)) ? 1 : 0;
info_.nativeVectorWidthHalf_ = info_.preferredVectorWidthHalf_ = 0; // no half support
info_.maxEngineClockFrequency_ = (calAttr.engineClock != 0) ? calAttr.engineClock : 555;
info_.maxMemoryClockFrequency_ = (calAttr.memoryClock != 0) ? calAttr.memoryClock : 555;
info_.timeStampFrequency_ = 1000000;
info_.vramBusBitWidth_ = calAttr.memBusWidth;
info_.l2CacheSize_ = 0;
info_.maxParameterSize_ = 1024;
info_.minDataTypeAlignSize_ = sizeof(int64_t[16]);
info_.singleFPConfig_ =
CL_FP_ROUND_TO_NEAREST | CL_FP_ROUND_TO_ZERO | CL_FP_ROUND_TO_INF | CL_FP_INF_NAN | CL_FP_FMA;
if (settings().singleFpDenorm_) {
info_.singleFPConfig_ |= CL_FP_DENORM;
}
if (settings().checkExtension(ClKhrFp64)) {
info_.doubleFPConfig_ = info_.singleFPConfig_ | CL_FP_DENORM;
}
if (settings().reportFMA_) {
info_.singleFPConfig_ |= CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT;
}
info_.globalMemCacheLineSize_ = settings().cacheLineSize_;
info_.globalMemCacheSize_ = settings().cacheSize_;
if ((settings().cacheLineSize_ != 0) || (settings().cacheSize_ != 0)) {
info_.globalMemCacheType_ = CL_READ_WRITE_CACHE;
} else {
info_.globalMemCacheType_ = CL_NONE;
}
#if defined(ATI_OS_LINUX)
info_.globalMemSize_ =
(static_cast<uint64_t>(std::min(GPU_MAX_HEAP_SIZE, 100u)) *
// globalMemSize is the actual available size for app on Linux
// Because Linux base driver doesn't support paging
static_cast<uint64_t>(memInfo.cardMemAvailableBytes + memInfo.cardExtMemAvailableBytes) /
100u);
#else
info_.globalMemSize_ = (static_cast<uint64_t>(std::min(GPU_MAX_HEAP_SIZE, 100u)) *
static_cast<uint64_t>(calAttr.localRAM) / 100u) *
Mi;
#endif
int uswcPercentAvailable = (calAttr.uncachedRemoteRAM > 1536 && IS_WINDOWS) ? 75 : 50;
if (settings().apuSystem_) {
info_.globalMemSize_ +=
(static_cast<uint64_t>(calAttr.uncachedRemoteRAM) * Mi * uswcPercentAvailable) / 100;
}
// We try to calculate the largest available memory size from
// the largest available block in either heap. In theory this
// should be the size we can actually allocate at application
// start. Note that it may not be a guarantee still as the
// application progresses.
#if defined(BRAHMA) && defined(ATI_BITS_64)
info_.maxMemAllocSize_ =
std::max(uint64_t(memInfo.cardMemAvailableBytes), uint64_t(memInfo.cardExtMemAvailableBytes));
#else
info_.maxMemAllocSize_ = std::max(uint64_t(memInfo.cardLargestFreeBlockBytes),
uint64_t(memInfo.cardExtLargestFreeBlockBytes));
#endif
if (settings().apuSystem_) {
info_.maxMemAllocSize_ = std::max(
(static_cast<uint64_t>(calAttr.uncachedRemoteRAM) * Mi * uswcPercentAvailable) / 100,
info_.maxMemAllocSize_);
}
info_.maxMemAllocSize_ =
uint64_t(info_.maxMemAllocSize_ * std::min(GPU_SINGLE_ALLOC_PERCENT, 100u) / 100u);
//! \note Force max single allocation size.
//! 4GB limit for the blit kernels and 64 bit optimizations.
info_.maxMemAllocSize_ =
std::min(info_.maxMemAllocSize_, static_cast<uint64_t>(settings().maxAllocSize_));
if (info_.maxMemAllocSize_ < uint64_t(128 * Mi)) {
LogError(
"We are unable to get a heap large enough to support the OpenCL minimum "
"requirement for FULL_PROFILE");
}
info_.maxMemAllocSize_ = std::max(uint64_t(128 * Mi), info_.maxMemAllocSize_);
// Clamp max single alloc size to the globalMemSize since it's
// reduced by default
info_.maxMemAllocSize_ = std::min(info_.maxMemAllocSize_, info_.globalMemSize_);
// We need to verify that we are not reporting more global memory
// that 4x single alloc
info_.globalMemSize_ = std::min(4 * info_.maxMemAllocSize_, info_.globalMemSize_);
// Use 64 bit pointers
if (settings().use64BitPtr_) {
info_.addressBits_ = 64;
} else {
info_.addressBits_ = 32;
// Limit total size with 3GB for 32 bit
info_.globalMemSize_ = std::min(info_.globalMemSize_, uint64_t(3 * Gi));
}
// Alignment in BITS of the base address of any allocated memory object
static const size_t MemBaseAlignment = 256;
//! @note Force 256 bytes alignment, since currently
//! calAttr.surface_alignment returns 4KB. For pinned memory runtime
//! should be able to create a view with 256 bytes alignement
info_.memBaseAddrAlign_ = 8 * MemBaseAlignment;
info_.preferredConstantBufferSize_ = 16 * Ki;
info_.maxConstantBufferSize_ = (settings().ciPlus_) ? info_.maxMemAllocSize_ : 64 * Ki;
info_.maxConstantArgs_ = MaxConstArguments;
// Image support fields
if (settings().imageSupport_) {
info_.imageSupport_ = CL_TRUE;
info_.maxSamplers_ = MaxSamplers;
info_.maxReadImageArgs_ = MaxReadImage;
info_.maxWriteImageArgs_ = MaxWriteImage;
info_.image2DMaxWidth_ = maxTextureSize;
info_.image2DMaxHeight_ = maxTextureSize;
info_.image3DMaxWidth_ = std::min(2 * Ki, maxTextureSize);
info_.image3DMaxHeight_ = std::min(2 * Ki, maxTextureSize);
info_.image3DMaxDepth_ = std::min(2 * Ki, maxTextureSize);
info_.imagePitchAlignment_ = 256; // XXX: 256 pixel pitch alignment for now
info_.imageBaseAddressAlignment_ = 256; // XXX: 256 byte base address alignment for now
info_.bufferFromImageSupport_ = CL_TRUE;
}
info_.errorCorrectionSupport_ = CL_FALSE;
if (settings().apuSystem_) {
info_.hostUnifiedMemory_ = CL_TRUE;
}
info_.profilingTimerResolution_ = 1;
info_.profilingTimerOffset_ = amd::Os::offsetToEpochNanos();
info_.littleEndian_ = CL_TRUE;
info_.available_ = CL_TRUE;
info_.compilerAvailable_ = CL_TRUE;
info_.linkerAvailable_ = CL_TRUE;
info_.executionCapabilities_ = CL_EXEC_KERNEL;
info_.preferredPlatformAtomicAlignment_ = 0;
info_.preferredGlobalAtomicAlignment_ = 0;
info_.preferredLocalAtomicAlignment_ = 0;
info_.queueProperties_ = CL_QUEUE_PROFILING_ENABLE;
info_.platform_ = AMD_PLATFORM;
::strncpy(info_.name_, calName_, sizeof(info_.name_) - 1);
::strncpy(info_.vendor_, "Advanced Micro Devices, Inc.", sizeof(info_.vendor_) - 1);
::snprintf(info_.driverVersion_, sizeof(info_.driverVersion_) - 1, AMD_BUILD_STRING " (GSL)%s",
isOnline() ? "" : " [Offline]");
info_.profile_ = "FULL_PROFILE";
if (settings().oclVersion_ >= OpenCL20) {
info_.version_ = "OpenCL 2.0 " AMD_PLATFORM_INFO;
info_.oclcVersion_ = "OpenCL C 2.0 ";
info_.spirVersions_ = "1.2";
} else if (settings().oclVersion_ == OpenCL12) {
info_.version_ = "OpenCL 1.2 " AMD_PLATFORM_INFO;
info_.oclcVersion_ = "OpenCL C 1.2 ";
info_.spirVersions_ = "1.2";
} else {
info_.version_ = "OpenCL 1.0 " AMD_PLATFORM_INFO;
info_.oclcVersion_ = "OpenCL C 1.0 ";
info_.spirVersions_ = "";
LogError("Unknown version for support");
}
// Fill workgroup info size
info_.maxWorkGroupSize_ = settings().maxWorkGroupSize_;
info_.maxWorkItemSizes_[0] = info_.maxWorkGroupSize_;
info_.maxWorkItemSizes_[1] = info_.maxWorkGroupSize_;
info_.maxWorkItemSizes_[2] = info_.maxWorkGroupSize_;
info_.preferredWorkGroupSize_ = settings().preferredWorkGroupSize_;
if (settings().hwLDSSize_ != 0) {
info_.localMemType_ = CL_LOCAL;
info_.localMemSize_ = settings().hwLDSSize_;
} else {
info_.localMemType_ = CL_GLOBAL;
info_.localMemSize_ = 16 * Ki;
}
info_.extensions_ = getExtensionString();
::strncpy(info_.driverStore_, calAttr.driverStore, sizeof(info_.driverStore_) - 1);
// OpenCL1.2 device info fields
info_.builtInKernels_ = "";
info_.imageMaxBufferSize_ = MaxImageBufferSize;
info_.image1DMaxWidth_ = maxTextureSize;
info_.imageMaxArraySize_ = MaxImageArraySize;
info_.preferredInteropUserSync_ = true;
info_.printfBufferSize_ = PrintfDbg::WorkitemDebugSize * info().maxWorkGroupSize_;
if (settings().oclVersion_ >= OpenCL20) {
info_.svmCapabilities_ = (CL_DEVICE_SVM_COARSE_GRAIN_BUFFER | CL_DEVICE_SVM_FINE_GRAIN_BUFFER);
if (settings().svmAtomics_) {
info_.svmCapabilities_ |= CL_DEVICE_SVM_ATOMICS;
}
if (settings().svmFineGrainSystem_) {
info_.svmCapabilities_ |= CL_DEVICE_SVM_FINE_GRAIN_SYSTEM;
}
// OpenCL2.0 device info fields
info_.maxWriteImageArgs_ = MaxReadWriteImage; //!< For compatibility
info_.maxReadWriteImageArgs_ = MaxReadWriteImage;
info_.maxPipePacketSize_ = info_.maxMemAllocSize_;
info_.maxPipeActiveReservations_ = 16;
info_.maxPipeArgs_ = 16;
info_.queueOnDeviceProperties_ =
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_PROFILING_ENABLE;
info_.queueOnDevicePreferredSize_ = 256 * Ki;
info_.queueOnDeviceMaxSize_ = 8 * Mi;
info_.maxOnDeviceQueues_ = 1;
info_.maxOnDeviceEvents_ = settings().numDeviceEvents_;
info_.globalVariablePreferredTotalSize_ = static_cast<size_t>(info_.globalMemSize_);
//! \todo Remove % calculation.
//! Use 90% of max single alloc size.
//! Boards with max single alloc size around 4GB will fail allocations
info_.maxGlobalVariableSize_ =
static_cast<size_t>(amd::alignDown(info_.maxMemAllocSize_ * 9 / 10, 256));
}
if (settings().checkExtension(ClAmdDeviceAttributeQuery)) {
::strncpy(info_.boardName_, calAttr.boardName, sizeof(info_.boardName_) - 1);
info_.deviceTopology_.pcie.type = CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD;
info_.deviceTopology_.pcie.bus = (calAttr.pciTopologyInformation & (0xFF << 8)) >> 8;
info_.deviceTopology_.pcie.device = (calAttr.pciTopologyInformation & (0x1F << 3)) >> 3;
info_.deviceTopology_.pcie.function = (calAttr.pciTopologyInformation & 0x07);
info_.simdPerCU_ = isa().simdPerCU();
info_.cuPerShaderArray_ = calAttr.numberOfCUsperShaderArray;
info_.simdWidth_ = isa().simdWidth();
info_.simdInstructionWidth_ = isa().simdInstructionWidth();
info_.wavefrontWidth_ = calAttr.wavefrontSize;
info_.globalMemChannelBanks_ = calAttr.numMemBanks;
info_.globalMemChannelBankWidth_ = isa().memChannelBankWidth();
info_.localMemSizePerCU_ = isa().localMemSizePerCU();
info_.localMemBanks_ = isa().localMemBanks();
info_.numAsyncQueues_ = numComputeRings;
info_.numRTQueues_ = numComputeRingsRT;
info_.numRTCUs_ = calAttr.maxRTCUs;
info_.threadTraceEnable_ = settings().threadTraceEnable_;
info_.pcieDeviceId_ = calAttr.pcieDeviceID;
info_.pcieRevisionId_ = calAttr.pcieRevisionID;
info_.maxThreadsPerCU_ = info_.wavefrontWidth_ * isa().simdPerCU() * 10;
}
}
bool Device::Heap::create(Device& device) {
// Create global GPU heap
resource_ = new Memory(device, 0);
if (resource_ == NULL) {
return false;
}
if (!resource_->create(Resource::Heap)) {
return false;
}
baseAddress_ = resource_->gslResource()->getSurfaceAddress();
return true;
}
void Device::Engines::create(uint num, gslEngineDescriptor* desc, uint maxNumComputeRings) {
numComputeRings_ = 0;
numComputeRingsRT_ = 0;
numDmaEngines_ = 0;
for (uint i = 0; i < num; ++i) {
desc_[desc[i].id] = desc[i];
desc_[desc[i].id].priority = GSL_ENGINEPRIORITY_NEUTRAL;
if (desc[i].id >= GSL_ENGINEID_COMPUTE0 && desc[i].id <= GSL_ENGINEID_COMPUTE7) {
numComputeRings_++;
}
if (desc[i].id == GSL_ENGINEID_COMPUTE_RT) {
numComputeRingsRT_++;
}
if (desc[i].id == GSL_ENGINEID_COMPUTE_MEDIUM_PRIORITY) {
numComputeRingsRT_++;
}
if (desc[i].id >= GSL_ENGINEID_DRMDMA0 && desc[i].id <= GSL_ENGINEID_DRMDMA1) {
numDmaEngines_++;
}
}
numComputeRings_ = std::min(numComputeRings_, maxNumComputeRings);
}
uint Device::Engines::getRequested(uint engines, gslEngineDescriptor* desc) const {
uint slot = 0;
for (uint i = 0; i < GSL_ENGINEID_MAX; ++i) {
if ((engines & getMask(static_cast<gslEngineID>(i))) &&
(desc_[i].id == static_cast<gslEngineID>(i))) {
desc[slot] = desc_[i];
engines &= ~getMask(static_cast<gslEngineID>(i));
slot++;
}
}
return (engines == 0) ? slot : 0;
}
Device::XferBuffers::~XferBuffers() {
// Destroy temporary buffer for reads
for (const auto& buf : freeBuffers_) {
// CPU optimization: unmap staging buffer just once
if (!buf->cal()->cardMemory_) {
buf->unmap(NULL);
}
delete buf;
}
freeBuffers_.clear();
}
bool Device::XferBuffers::create() {
Memory* xferBuf = NULL;
bool result = false;
// Create a buffer object
xferBuf = new Memory(dev(), bufSize_);
// Try to allocate memory for the transfer buffer
if ((NULL == xferBuf) || !xferBuf->create(type_)) {
delete xferBuf;
xferBuf = NULL;
LogError("Couldn't allocate a transfer buffer!");
} else {
result = true;
freeBuffers_.push_back(xferBuf);
// CPU optimization: map staging buffer just once
if (!xferBuf->cal()->cardMemory_) {
xferBuf->map(NULL);
}
}
return result;
}
Memory& Device::XferBuffers::acquire() {
Memory* xferBuf = NULL;
size_t listSize;
// Lock the operations with the staged buffer list
amd::ScopedLock l(lock_);
listSize = freeBuffers_.size();
// If the list is empty, then attempt to allocate a staged buffer
if (listSize == 0) {
// Allocate memory
xferBuf = new Memory(dev(), bufSize_);
// Allocate memory for the transfer buffer
if ((NULL == xferBuf) || !xferBuf->create(type_)) {
delete xferBuf;
xferBuf = NULL;
LogError("Couldn't allocate a transfer buffer!");
} else {
++acquiredCnt_;
// CPU optimization: map staging buffer just once
if (!xferBuf->cal()->cardMemory_) {
xferBuf->map(NULL);
}
}
}
if (xferBuf == NULL) {
xferBuf = *(freeBuffers_.begin());
freeBuffers_.erase(freeBuffers_.begin());
++acquiredCnt_;
}
return *xferBuf;
}
void Device::XferBuffers::release(VirtualGPU& gpu, Memory& buffer) {
// Make sure buffer isn't busy on the current VirtualGPU, because
// the next aquire can come from different queue
buffer.wait(gpu);
// Lock the operations with the staged buffer list
amd::ScopedLock l(lock_);
freeBuffers_.push_back(&buffer);
--acquiredCnt_;
}
Device::ScopedLockVgpus::ScopedLockVgpus(const Device& dev) : dev_(dev) {
// Lock the virtual GPU list
dev_.vgpusAccess()->lock();
// Find all available virtual GPUs and lock them
// from the execution of commands
for (uint idx = 0; idx < dev_.vgpus().size(); ++idx) {
dev_.vgpus()[idx]->execution().lock();
}
}
Device::ScopedLockVgpus::~ScopedLockVgpus() {
// Find all available virtual GPUs and unlock them
// for the execution of commands
for (uint idx = 0; idx < dev_.vgpus().size(); ++idx) {
dev_.vgpus()[idx]->execution().unlock();
}
// Unock the virtual GPU list
dev_.vgpusAccess()->unlock();
}
Device::Device()
: NullDevice(),
CALGSLDevice(),
numOfVgpus_(0),
context_(NULL),
heap_(),
dummyPage_(NULL),
lockAsyncOps_(NULL),
lockAsyncOpsForInitHeap_(NULL),
vgpusAccess_(NULL),
scratchAlloc_(NULL),
mapCacheOps_(NULL),
xferRead_(NULL),
xferWrite_(NULL),
mapCache_(NULL),
resourceCache_(NULL),
heapInitComplete_(false),
xferQueue_(NULL),
globalScratchBuf_(NULL),
srdManager_(NULL) {}
Device::~Device() {
// remove the HW debug manager
delete hwDebugMgr_;
hwDebugMgr_ = NULL;
delete srdManager_;
for (uint s = 0; s < scratch_.size(); ++s) {
delete scratch_[s];
scratch_[s] = NULL;
}
delete globalScratchBuf_;
globalScratchBuf_ = NULL;
// Destroy transfer queue
delete xferQueue_;
// Destroy blit program
delete blitProgram_;
// Release cached map targets
for (uint i = 0; mapCache_ != NULL && i < mapCache_->size(); ++i) {
if ((*mapCache_)[i] != NULL) {
(*mapCache_)[i]->release();
}
}
delete mapCache_;
// Destroy temporary buffers for read/write
delete xferRead_;
delete xferWrite_;
if (dummyPage_ != NULL) {
dummyPage_->release();
}
// Destroy resource cache
delete resourceCache_;
delete lockAsyncOps_;
delete lockAsyncOpsForInitHeap_;
delete vgpusAccess_;
delete scratchAlloc_;
delete mapCacheOps_;
if (context_ != NULL) {
context_->release();
}
// Close the active device
close();
}
extern const char* SchedulerSourceCode;
bool Device::create(CALuint ordinal, CALuint numOfDevices) {
appProfile_.init();
bool smallMemSystem = false;
if (amd::Os::hostTotalPhysicalMemory() < OCL_SYSMEM_REQUIREMENT * Gi) {
smallMemSystem = true;
}
bool noSVM = LP64_SWITCH(true, false) && !GPU_FORCE_OCL20_32BIT;
// Open GSL device
CALGSLDevice::OpenParams openData = {0};
openData.enableHighPerformanceState = appProfile_.enableHighPerformanceState();
openData.reportAsOCL12Device = (smallMemSystem ||
appProfile_.reportAsOCL12Device() ||
(OPENCL_VERSION < 200) ||
noSVM);
openData.sclkThreshold = appProfile_.GetSclkThreshold().c_str();
openData.downHysteresis = appProfile_.GetDownHysteresis().c_str();
openData.upHysteresis = appProfile_.GetUpHysteresis().c_str();
openData.powerLimit = appProfile_.GetPowerLimit().c_str();
openData.mclkThreshold = appProfile_.GetMclkThreshold().c_str();
openData.mclkUpHyst = appProfile_.GetMclkUpHyst().c_str();
openData.mclkDownHyst = appProfile_.GetMclkDownHyst().c_str();
if (!open(ordinal, openData)) {
return false;
}
// Update CAL target
calTarget_ = getAttribs().target;
// XNACK should be set for PageMigration or IOMMUv2 support.
bool isXNACKSupported = false;
// SRAMECC should be set for ecc protected GPRs.
bool isSRAMECCSupported = false;
const amd::Isa* isa;
bool preferPal;
std::tie(isa, calMachine_, calName_, preferPal, std::ignore, std::ignore) =
findIsa(calTarget(), isSRAMECCSupported, isXNACKSupported);
if ((calTarget() == CAL_TARGET_CARRIZO) && ASICREV_IS_CARRIZO_BRISTOL(getAttribs().asicRevision)) {
calName_ = "Bristol Ridge";
}
if (!isa) {
LogPrintfError("Unsupported CAL device #%d", calTarget());
return false;
}
if (!isa->runtimeGslSupported()) {
LogPrintfError("Unsupported CAL device with ISA %s", isa->targetId());
return false;
}
if ((GPU_ENABLE_PAL == 2) && isa->runtimePalSupported() && preferPal) {
LogPrintfError("Skipping as GPU_ENABLE_PAL=2 indicating to use PAL for CAL device %s",
isa->targetId());
return false;
}
if (!amd::Device::create(*isa)) {
LogPrintfError("Unable to setup device for CAL device %s", isa->targetId());
return false;
}
// Creates device settings
settings_ = new gpu::Settings();
gpu::Settings* gpuSettings = reinterpret_cast<gpu::Settings*>(settings_);
if ((gpuSettings == NULL) ||
!gpuSettings->create(getAttribs(), appProfile_.reportAsOCL12Device(), smallMemSystem)) {
return false;
}
if (!ValidateHsail()) {
LogError("Hsail initialization failed!");
return false;
}
engines_.create(m_nEngines, m_engines, settings().numComputeRings_);
amd::Context::Info info = {0};
std::vector<amd::Device*> devices;
devices.push_back(this);
// Create a dummy context
context_ = new amd::Context(devices, info);
if (context_ == NULL) {
return false;
}
// Create the locks
lockAsyncOps_ = new amd::Monitor("Device Async Ops Lock", true);
if (NULL == lockAsyncOps_) {
return false;
}
lockAsyncOpsForInitHeap_ =
new amd::Monitor("Async Ops Lock For Initialization of Heap Resource", true);
if (NULL == lockAsyncOpsForInitHeap_) {
return false;
}
vgpusAccess_ = new amd::Monitor("Virtual GPU List Ops Lock", true);
if (NULL == vgpusAccess_) {
return false;
}
scratchAlloc_ = new amd::Monitor("Scratch Allocation Lock", true);
if (NULL == scratchAlloc_) {
return false;
}
mapCacheOps_ = new amd::Monitor("Map Cache Lock", true);
if (NULL == mapCacheOps_) {
return false;
}
mapCache_ = new std::vector<amd::Memory*>();
if (mapCache_ == NULL) {
return false;
}
// Use just 1 entry by default for the map cache
mapCache_->push_back(NULL);
size_t resourceCacheSize = settings().resourceCacheSize_;
#ifdef DEBUG
std::stringstream message;
if (settings().remoteAlloc_) {
message << "Using *Remote* memory";
} else {
message << "Using *Local* memory";
}
message << std::endl;
LogInfo(message.str().c_str());
#endif // DEBUG
// Create resource cache.
// \note Cache must be created before any resource creation to avoid NULL check
resourceCache_ = new ResourceCache(resourceCacheSize);
if (NULL == resourceCache_) {
return false;
}
// Fill the device info structure
fillDeviceInfo(getAttribs(), getMemInfo(), static_cast<size_t>(getMaxTextureSize()),
engines().numComputeRings(), engines().numComputeRingsRT());
if (NULL == hsaCompiler_) {
const char* library = getenv("HSA_COMPILER_LIBRARY");
aclCompilerOptions opts = {
sizeof(aclCompilerOptions_0_8), library, NULL, NULL, NULL, NULL, NULL, AMD_OCL_SC_LIB};
// Initialize the compiler handle
acl_error error;
hsaCompiler_ = amd::Hsail::CompilerInit(&opts, &error);
if (error != ACL_SUCCESS) {
LogError("Error initializing the compiler");
return false;
}
}
// Allocate SRD manager
srdManager_ = new SrdManager(*this, std::max(HsaImageObjectSize, HsaSamplerObjectSize), 64 * Ki);
if (srdManager_ == NULL) {
return false;
}
// create the HW debug manager if needed
if (settings().enableHwDebug_) {
hwDebugMgr_ = new GpuDebugManager(this);
}
return true;
}
bool Device::initializeHeapResources() {
amd::ScopedLock k(lockAsyncOpsForInitHeap_);
if (!heapInitComplete_) {
heapInitComplete_ = true;
PerformFullInitialization();
uint numComputeRings = engines_.numComputeRings() + engines_.numComputeRingsRT();
scratch_.resize((settings().useSingleScratch_) ? 1 : (numComputeRings ? numComputeRings : 1));
// Initialize the number of mem object for the scratch buffer
for (uint s = 0; s < scratch_.size(); ++s) {
scratch_[s] = new ScratchBuffer();
if (NULL == scratch_[s]) {
return false;
}
}
// Complete initialization of the heap and other buffers
if (!heap_.create(*this)) {
LogError("Failed GPU heap creation");
return false;
}
size_t dummySize = amd::Os::pageSize();
// Allocate a dummy page for NULL pointer processing
dummyPage_ = new (*context_) amd::Buffer(*context_, 0, dummySize);
if ((dummyPage_ != NULL) && !dummyPage_->create()) {
dummyPage_->release();
return false;
}
Memory* devMemory = reinterpret_cast<Memory*>(dummyPage_->getDeviceMemory(*this));
if (devMemory == NULL) {
// Release memory
dummyPage_->release();
dummyPage_ = NULL;
return false;
}
if (settings().stagedXferSize_ != 0) {
// Initialize staged write buffers
if (settings().stagedXferWrite_) {
Resource::MemoryType type = Resource::RemoteUSWC;
xferWrite_ = new XferBuffers(*this, type, amd::alignUp(settings().stagedXferSize_, 4 * Ki));
if ((xferWrite_ == NULL) || !xferWrite_->create()) {
LogError("Couldn't allocate transfer buffer objects for read");
return false;
}
}
// Initialize staged read buffers
if (settings().stagedXferRead_) {
xferRead_ = new XferBuffers(*this, Resource::Remote,
amd::alignUp(settings().stagedXferSize_, 4 * Ki));
if ((xferRead_ == NULL) || !xferRead_->create()) {
LogError("Couldn't allocate transfer buffer objects for write");
return false;
}
}
}
// Delay compilation due to brig_loader memory allocation
if (settings().ciPlus_) {
const char* CL20extraBlits = NULL;
const char* ocl20 = NULL;
if (settings().oclVersion_ >= OpenCL20) {
CL20extraBlits = SchedulerSourceCode;
ocl20 = "-cl-std=CL2.0";
}
blitProgram_ = new BlitProgram(context_);
// Create blit programs
if (blitProgram_ == NULL || !blitProgram_->create(this, CL20extraBlits, ocl20)) {
delete blitProgram_;
blitProgram_ = NULL;
LogError("Couldn't create blit kernels!");
return false;
}
}
// Create a synchronized transfer queue
xferQueue_ = new VirtualGPU(*this);
if (!(xferQueue_ && xferQueue_->create(false))) {
delete xferQueue_;
xferQueue_ = NULL;
}
if (NULL == xferQueue_) {
LogError("Couldn't create the device transfer manager!");
return false;
}
xferQueue_->enableSyncedBlit();
}
return true;
}
device::VirtualDevice* Device::createVirtualDevice(amd::CommandQueue* queue) {
bool profiling = false;
bool interopQueue = false;
uint rtCUs = amd::CommandQueue::RealTimeDisabled;
uint deviceQueueSize = 0;
if (queue != NULL) {
profiling = queue->properties().test(CL_QUEUE_PROFILING_ENABLE);
if (queue->asHostQueue() != NULL) {
interopQueue = (0 != (queue->context().info().flags_ &
(amd::Context::GLDeviceKhr | amd::Context::D3D10DeviceKhr |
amd::Context::D3D11DeviceKhr)));
rtCUs = queue->rtCUs();
} else if (queue->asDeviceQueue() != NULL) {
deviceQueueSize = queue->asDeviceQueue()->size();
}
}
// Not safe to add a queue. So lock the device
amd::ScopedLock k(lockAsyncOps());
amd::ScopedLock lock(vgpusAccess());
// Initialization of heap and other resources occur during the command queue creation time.
if (!initializeHeapResources()) {
return NULL;
}
VirtualGPU* vgpu = new VirtualGPU(*this);
if (vgpu && vgpu->create(profiling, rtCUs, deviceQueueSize, queue->priority())) {
return vgpu;
} else {
delete vgpu;
return NULL;
}
}
device::Program* Device::createProgram(amd::Program& owner, amd::option::Options* options) {
if (isHsailProgram(options)) {
return new HSAILProgram(*this, owner);
}
return new Program(*this, owner);
}
//! Requested devices list as configured by the GPU_DEVICE_ORDINAL
typedef std::unordered_map<int, bool> requestedDevices_t;
//! Parses the requested list of devices to be exposed to the user.
static void parseRequestedDeviceList(requestedDevices_t& requestedDevices) {
char* pch = NULL;
int requestedDeviceCount = 0;
const char* requestedDeviceList = GPU_DEVICE_ORDINAL;
pch = strtok(const_cast<char*>(requestedDeviceList), ",");
while (pch != NULL) {
bool deviceIdValid = true;
int currentDeviceIndex = atoi(pch);
// Validate device index.
for (size_t i = 0; i < strlen(pch); i++) {
if (!isdigit(pch[i])) {
deviceIdValid = false;
break;
}
}
if (currentDeviceIndex < 0) {
deviceIdValid = false;
}
// Get next token.
pch = strtok(NULL, ",");
if (!deviceIdValid) {
continue;
}
// Requested device is valid.
requestedDevices[currentDeviceIndex] = true;
}
}
bool Device::init() {
CALuint numDevices = 0;
bool useDeviceList = false;
requestedDevices_t requestedDevices;
hsaCompiler_ = NULL;
compiler_ = NULL;
#if defined(_WIN32) && !defined(_WIN64)
// @toto: FIXME: remove this when CAL is fixed!!!
unsigned int old, ignored;
_controlfp_s(&old, 0, 0);
#endif // _WIN32 && !_WIN64
// FIXME_lmoriche: needs cleanup
osInit();
#if defined(_WIN32)
// osAssertSetStyle(OSASSERT_STYLE_LOGANDEXIT);
#endif // WIN32
gslInit();
#if defined(_WIN32) && !defined(_WIN64)
_controlfp_s(&ignored, old, _MCW_RC | _MCW_PC);
#endif // _WIN32 && !_WIN64
// Get the total number of active devices
// Count up all the devices in the system.
numDevices = gsAdaptor::enumerateAdaptors();
const char* selectDeviceByName = NULL;
if (!flagIsDefault(GPU_DEVICE_ORDINAL)) {
useDeviceList = true;
parseRequestedDeviceList(requestedDevices);
}
// Loop through all active devices and initialize the device info structure
for (CALuint ordinal = 0; ordinal < numDevices; ++ordinal) {
// Create the GPU device object
Device* d = new Device();
bool result = (NULL != d) && d->create(ordinal, numDevices);
if (useDeviceList) {
result &= (requestedDevices.find(ordinal) != requestedDevices.end());
}
if (result) {
d->registerDevice();
} else {
delete d;
}
}
return true;
}
void Device::tearDown() {
osExit();
gslExit();
amd::Hsail::CompilerFini(compiler_);
if (hsaCompiler_ != NULL) {
amd::Hsail::CompilerFini(hsaCompiler_);
}
}
gpu::Memory* Device::getGpuMemory(amd::Memory* mem) const {
return static_cast<gpu::Memory*>(mem->getDeviceMemory(*this));
}
const device::BlitManager& Device::xferMgr() const { return xferQueue_->blitMgr(); }
CalFormat Device::getCalFormat(const amd::Image::Format& format) const {
// Find CAL format
for (uint i = 0; i < sizeof(MemoryFormatMap) / sizeof(MemoryFormat); ++i) {
if ((format.image_channel_data_type == MemoryFormatMap[i].clFormat_.image_channel_data_type) &&
(format.image_channel_order == MemoryFormatMap[i].clFormat_.image_channel_order)) {
return MemoryFormatMap[i].calFormat_;
}
}
osAssert(0 && "We didn't find CAL resource format!");
return MemoryFormatMap[0].calFormat_;
}
amd::Image::Format Device::getOclFormat(const CalFormat& format) const {
// Find CL format
for (uint i = 0; i < sizeof(MemoryFormatMap) / sizeof(MemoryFormat); ++i) {
if ((format.type_ == MemoryFormatMap[i].calFormat_.type_) &&
(format.channelOrder_ == MemoryFormatMap[i].calFormat_.channelOrder_)) {
return MemoryFormatMap[i].clFormat_;
}
}
osAssert(0 && "We didn't find OCL resource format!");
return MemoryFormatMap[0].clFormat_;
}
// Create buffer without an owner (merge common code with createBuffer() ?)
gpu::Memory* Device::createScratchBuffer(size_t size) const {
Memory* gpuMemory = NULL;
// Create a memory object
gpuMemory = new gpu::Memory(*this, size);
if (NULL == gpuMemory || !gpuMemory->create(Resource::Local)) {
delete gpuMemory;
gpuMemory = NULL;
}
return gpuMemory;
}
gpu::Memory* Device::createBuffer(amd::Memory& owner, bool directAccess) const {
size_t size = owner.getSize();
gpu::Memory* gpuMemory;
// Create resource
bool result = false;
if (owner.getType() == CL_MEM_OBJECT_PIPE) {
// directAccess isnt needed as Pipes shouldnt be host accessible for GPU
directAccess = false;
}
if (NULL != owner.parent()) {
gpu::Memory* gpuParent = getGpuMemory(owner.parent());
if (NULL == gpuParent) {
LogError("Can't get the owner object for subbuffer allocation");
return NULL;
}
if (nullptr != owner.parent()->getSvmPtr()) {
amd::Memory* amdParent = owner.parent();
{
// Lock memory object, so only one commitment will occur
amd::ScopedLock lock(amdParent->lockMemoryOps());
amdParent->commitSvmMemory();
amdParent->setHostMem(amdParent->getSvmPtr());
}
// Ignore a possible pinning error. Runtime will fallback to SW emulation
// bool ok = gpuParent->pinSystemMemory(
// amdParent->getHostMem(), amdParent->getSize());
}
return gpuParent->createBufferView(owner);
}
Resource::MemoryType type =
(owner.forceSysMemAlloc() || (owner.getMemFlags() & CL_MEM_SVM_FINE_GRAIN_BUFFER))
? Resource::Remote
: Resource::Local;
// Check if runtime can force a tiny buffer into USWC memory
if ((size <= (GPU_MAX_REMOTE_MEM_SIZE * Ki)) && (type == Resource::Local) &&
(owner.getMemFlags() & CL_MEM_READ_ONLY)) {
type = Resource::RemoteUSWC;
}
if (owner.getMemFlags() & CL_MEM_BUS_ADDRESSABLE_AMD) {
type = Resource::BusAddressable;
} else if (owner.getMemFlags() & CL_MEM_EXTERNAL_PHYSICAL_AMD) {
type = Resource::ExternalPhysical;
}
// Use direct access if it's possible
bool remoteAlloc = false;
// Internal means VirtualDevice!=NULL
bool internalAlloc =
((owner.getMemFlags() & CL_MEM_USE_HOST_PTR) && (owner.getVirtualDevice() != NULL)) ? true
: false;
// Create a memory object
gpuMemory = new gpu::Buffer(*this, owner, owner.getSize());
if (NULL == gpuMemory) {
return NULL;
}
// Check if owner is interop memory
if (owner.isInterop()) {
result = gpuMemory->createInterop(Memory::InteropDirectAccess);
} else if (owner.getMemFlags() & CL_MEM_USE_PERSISTENT_MEM_AMD) {
// Attempt to allocate from persistent heap
result = gpuMemory->create(Resource::Persistent);
} else if (directAccess || (type == Resource::Remote)) {
// Check for system memory allocations
if ((owner.getMemFlags() & (CL_MEM_ALLOC_HOST_PTR | CL_MEM_USE_HOST_PTR)) ||
(settings().remoteAlloc_)) {
// Allocate remote memory if AHP allocation and context has just 1 device
if ((owner.getMemFlags() & CL_MEM_ALLOC_HOST_PTR) &&
(owner.getContext().devices().size() == 1)) {
if (owner.getMemFlags() &
(CL_MEM_READ_ONLY | CL_MEM_HOST_WRITE_ONLY | CL_MEM_HOST_NO_ACCESS)) {
// GPU will be reading from this host memory buffer,
// so assume Host write into it
type = Resource::RemoteUSWC;
remoteAlloc = true;
}
}
// Make sure owner has a valid hostmem pointer and it's not COPY
if (!remoteAlloc && (owner.getHostMem() != NULL)) {
Resource::PinnedParams params;
params.owner_ = &owner;
params.gpu_ = reinterpret_cast<VirtualGPU*>(owner.getVirtualDevice());
params.hostMemRef_ = owner.getHostMemRef();
params.size_ = owner.getHostMemRef()->size();
if (0 == params.size_) {
params.size_ = owner.getSize();
}
// Create memory object
result = gpuMemory->create(Resource::Pinned, &params);
// If direct access failed
if (!result) {
// Don't use cached allocation
// if size is biger than max single alloc
if (owner.getSize() > info().maxMemAllocSize_) {
delete gpuMemory;
return NULL;
}
}
}
}
}
if (!result &&
// Make sure it's not internal alloc
!internalAlloc) {
Resource::CreateParams params;
params.owner_ = &owner;
params.gpu_ = static_cast<VirtualGPU*>(owner.getVirtualDevice());
// Create memory object
result = gpuMemory->create(type, &params);
// If allocation was successful
if (result) {
// Initialize if the memory is a pipe object
if (owner.getType() == CL_MEM_OBJECT_PIPE) {
// Pipe initialize in order read_idx, write_idx, end_idx. Refer clk_pipe_t structure.
// Init with 3 DWORDS for 32bit addressing and 6 DWORDS for 64bit
size_t pipeInit[3] = {0, 0, owner.asPipe()->getMaxNumPackets()};
gpuMemory->writeRawData(*xferQueue_, sizeof(pipeInit), pipeInit, true);
}
// If memory has direct access from host, then get CPU address
if (gpuMemory->isHostMemDirectAccess() && (type != Resource::ExternalPhysical)) {
void* address = gpuMemory->map(NULL);
if (address != NULL) {
// Copy saved memory
// Note: UHP is an optional check if pinning failed and sysmem alloc was forced
if (owner.getMemFlags() & (CL_MEM_COPY_HOST_PTR | CL_MEM_USE_HOST_PTR)) {
memcpy(address, owner.getHostMem(), owner.getSize());
}
// It should be safe to change the host memory pointer,
// because it's lock protected from the upper caller
owner.setHostMem(address);
} else {
result = false;
}
}
// An optimization for CHP. Copy memory and destroy sysmem allocation
else if ((gpuMemory->memoryType() != Resource::Pinned) &&
(owner.getMemFlags() & CL_MEM_COPY_HOST_PTR) &&
(owner.getContext().devices().size() == 1)) {
amd::Coord3D origin(0, 0, 0);
amd::Coord3D region(owner.getSize());
static const bool Entire = true;
if (xferMgr().writeBuffer(owner.getHostMem(), *gpuMemory, origin, region, Entire)) {
// Clear CHP memory
owner.setHostMem(NULL);
}
}
}
}
if (!result) {
delete gpuMemory;
return NULL;
}
return gpuMemory;
}
gpu::Memory* Device::createImage(amd::Memory& owner, bool directAccess) const {
size_t size = owner.getSize();
amd::Image& image = *owner.asImage();
gpu::Memory* gpuImage = NULL;
CalFormat format = getCalFormat(image.getImageFormat());
if ((NULL != owner.parent()) && (owner.parent()->asImage() != NULL)) {
device::Memory* devParent = owner.parent()->getDeviceMemory(*this);
if (NULL == devParent) {
LogError("Can't get the owner object for image view allocation");
return NULL;
}
// Create a view on the specified device
gpuImage = (gpu::Memory*)createView(owner, *devParent);
if ((NULL != gpuImage) && (gpuImage->owner() != NULL)) {
gpuImage->owner()->setHostMem((address)(owner.parent()->getHostMem()) +
gpuImage->owner()->getOrigin());
}
return gpuImage;
}
gpuImage =
new gpu::Image(*this, owner, image.getWidth(), image.getHeight(), image.getDepth(),
format.type_, format.channelOrder_, image.getType(), image.getMipLevels());
// Create resource
if (NULL != gpuImage) {
const bool imageBuffer = ((owner.getType() == CL_MEM_OBJECT_IMAGE1D_BUFFER) ||
((owner.getType() == CL_MEM_OBJECT_IMAGE2D) &&
(owner.parent() != NULL) && (owner.parent()->asBuffer() != NULL)));
bool result = false;
// Check if owner is interop memory
if (owner.isInterop()) {
result = gpuImage->createInterop(Memory::InteropDirectAccess);
} else if (imageBuffer) {
Resource::ImageBufferParams params;
gpu::Memory* buffer = reinterpret_cast<gpu::Memory*>(image.parent()->getDeviceMemory(*this));
if (buffer == NULL) {
LogError("Buffer creation for ImageBuffer failed!");
delete gpuImage;
return NULL;
}
params.owner_ = &owner;
params.resource_ = buffer;
params.memory_ = buffer;
// Create memory object
result = gpuImage->create(Resource::ImageBuffer, &params);
} else if (directAccess && (owner.getMemFlags() & CL_MEM_ALLOC_HOST_PTR)) {
Resource::PinnedParams params;
params.owner_ = &owner;
params.hostMemRef_ = owner.getHostMemRef();
params.size_ = owner.getHostMemRef()->size();
// Create memory object
result = gpuImage->create(Resource::Pinned, &params);
}
if (!result && !owner.isInterop()) {
if (owner.getMemFlags() & CL_MEM_USE_PERSISTENT_MEM_AMD) {
// Attempt to allocate from persistent heap
result = gpuImage->create(Resource::Persistent);
} else {
Resource::MemoryType type =
(owner.forceSysMemAlloc()) ? Resource::RemoteUSWC : Resource::Local;
// Create memory object
result = gpuImage->create(type);
}
}
if (!result) {
delete gpuImage;
return NULL;
} else if ((gpuImage->memoryType() != Resource::Pinned) &&
(owner.getMemFlags() & CL_MEM_COPY_HOST_PTR) &&
(owner.getContext().devices().size() == 1)) {
// Ignore copy for image1D_buffer, since it was already done for buffer
if (imageBuffer) {
// Clear CHP memory
owner.setHostMem(NULL);
} else {
amd::Coord3D origin(0, 0, 0);
static const bool Entire = true;
if (xferMgr().writeImage(owner.getHostMem(), *gpuImage, origin, image.getRegion(), 0, 0,
Entire)) {
// Clear CHP memory
owner.setHostMem(NULL);
}
}
}
if (result) {
gslMemObject temp = gpuImage->gslResource();
size_t bytePitch = gpuImage->elementSize() * temp->getPitch();
image.setBytePitch(bytePitch);
}
}
return gpuImage;
}
//! Allocates cache memory on the card
device::Memory* Device::createMemory(amd::Memory& owner) const {
bool directAccess = false;
gpu::Memory* memory = NULL;
if (owner.asBuffer()) {
directAccess = (settings().hostMemDirectAccess_ & Settings::HostMemBuffer) ? true : false;
memory = createBuffer(owner, directAccess);
} else if (owner.asImage()) {
directAccess = (settings().hostMemDirectAccess_ & Settings::HostMemImage) ? true : false;
memory = createImage(owner, directAccess);
} else {
LogError("Unknown memory type!");
}
// Attempt to pin system memory if runtime didn't use direct access
if ((memory != NULL) && (memory->memoryType() != Resource::Pinned) &&
(memory->memoryType() != Resource::Remote) &&
(memory->memoryType() != Resource::RemoteUSWC) &&
(memory->memoryType() != Resource::ExternalPhysical) &&
((owner.getHostMem() != NULL) ||
((NULL != owner.parent()) && (owner.getHostMem() != NULL)))) {
bool ok = memory->pinSystemMemory(owner.getHostMem(), (owner.getHostMemRef()->size())
? owner.getHostMemRef()->size()
: owner.getSize());
//! \note: Ignore the pinning result for now
}
return memory;
}
bool Device::createSampler(const amd::Sampler& owner, device::Sampler** sampler) const {
*sampler = NULL;
Sampler* gpuSampler = new Sampler(*this);
if ((NULL == gpuSampler) || !gpuSampler->create(owner)) {
delete gpuSampler;
return false;
}
*sampler = gpuSampler;
return true;
}
device::Memory* Device::createView(amd::Memory& owner, const device::Memory& parent) const {
size_t size = owner.getSize();
assert((owner.asImage() != NULL) && "View supports images only");
const amd::Image& image = *owner.asImage();
gpu::Memory* gpuImage = NULL;
CalFormat format = getCalFormat(image.getImageFormat());
gpuImage =
new gpu::Image(*this, owner, image.getWidth(), image.getHeight(), image.getDepth(),
format.type_, format.channelOrder_, image.getType(), image.getMipLevels());
// Create resource
if (NULL != gpuImage) {
bool result = false;
Resource::ImageViewParams params;
const gpu::Memory& gpuMem = static_cast<const gpu::Memory&>(parent);
params.owner_ = &owner;
params.level_ = image.getBaseMipLevel();
params.layer_ = 0;
params.resource_ = &gpuMem;
params.gpu_ = reinterpret_cast<VirtualGPU*>(owner.getVirtualDevice());
params.memory_ = &gpuMem;
// Create memory object
result = gpuImage->create(Resource::ImageView, &params);
if (!result) {
delete gpuImage;
return NULL;
}
}
return gpuImage;
}
//! Attempt to bind with external graphics API's device/context
bool Device::bindExternalDevice(uint flags, void* const pDevice[], void* pContext,
bool validateOnly) {
assert(pDevice);
if (flags & amd::Context::Flags::GLDeviceKhr) {
// There is no need to perform full initialization here
// if the GSLDevice is still uninitialized.
// Only adapter initialization is required to validate
// GL interoperability.
PerformAdapterInitialization(validateOnly);
// Attempt to associate GSL-OGL
if (!glAssociate((CALvoid*)pContext, pDevice[amd::Context::DeviceFlagIdx::GLDeviceKhrIdx])) {
CloseInitializedAdapter(validateOnly);
LogError("Failed gslGLAssociate()");
return false;
}
CloseInitializedAdapter(validateOnly);
}
#ifdef _WIN32
if (flags & amd::Context::Flags::D3D10DeviceKhr) {
// There is no need to perform full initialization here
// if the GSLDevice is still uninitialized.
// Only adapter initialization is required
// to validate D3D10 interoperability.
PerformAdapterInitialization(validateOnly);
// Associate GSL-D3D
if (!associateD3D10Device(reinterpret_cast<ID3D10Device*>(
pDevice[amd::Context::DeviceFlagIdx::D3D10DeviceKhrIdx]))) {
CloseInitializedAdapter(validateOnly);
LogError("Failed gslD3D10Associate()");
return false;
}
CloseInitializedAdapter(validateOnly);
}
if (flags & amd::Context::Flags::D3D11DeviceKhr) {
// There is no need to perform full initialization here
// if the GSLDevice is still uninitialized.
// Only adapter initialization is required to validate
// D3D11 interoperability.
PerformAdapterInitialization(validateOnly);
// Associate GSL-D3D
if (!associateD3D11Device(reinterpret_cast<ID3D11Device*>(
pDevice[amd::Context::DeviceFlagIdx::D3D11DeviceKhrIdx]))) {
CloseInitializedAdapter(validateOnly);
LogError("Failed gslD3D11Associate()");
return false;
}
CloseInitializedAdapter(validateOnly);
}
if (flags & amd::Context::Flags::D3D9DeviceKhr) {
PerformAdapterInitialization(validateOnly);
// Associate GSL-D3D
if (!associateD3D9Device(reinterpret_cast<IDirect3DDevice9*>(
pDevice[amd::Context::DeviceFlagIdx::D3D9DeviceKhrIdx]))) {
CloseInitializedAdapter(validateOnly);
LogWarning("D3D9<->OpenCL adapter mismatch or D3D9Associate() failure");
return false;
}
CloseInitializedAdapter(validateOnly);
}
if (flags & amd::Context::Flags::D3D9DeviceEXKhr) {
PerformAdapterInitialization(validateOnly);
// Associate GSL-D3D
if (!associateD3D9Device(reinterpret_cast<IDirect3DDevice9Ex*>(
pDevice[amd::Context::DeviceFlagIdx::D3D9DeviceEXKhrIdx]))) {
CloseInitializedAdapter(validateOnly);
LogWarning("D3D9<->OpenCL adapter mismatch or D3D9Associate() failure");
return false;
}
CloseInitializedAdapter(validateOnly);
}
if (flags & amd::Context::Flags::D3D9DeviceVAKhr) {
}
#endif //_WIN32
return true;
}
bool Device::unbindExternalDevice(uint flags, void* const pDevice[], void* pContext,
bool validateOnly) {
if ((flags & amd::Context::Flags::GLDeviceKhr) == 0) {
return true;
}
void* glDevice = pDevice[amd::Context::DeviceFlagIdx::GLDeviceKhrIdx];
if (glDevice != NULL) {
// Dissociate GSL-OGL
if (true != glDissociate(pContext, glDevice)) {
if (validateOnly) {
LogWarning("Failed gslGLDiassociate()");
}
return false;
}
}
return true;
}
bool Device::globalFreeMemory(size_t* freeMemory) const {
const uint TotalFreeMemory = 0;
const uint LargestFreeBlock = 1;
// Initialization of heap and other resources because getMemInfo needs it.
if (!(const_cast<Device*>(this)->initializeHeapResources())) {
return false;
}
gslMemInfo memInfo = { 0 };
gslCtx()->getMemInfo(&memInfo, GSL_MEMINFO_BASIC);
// Fill free memory info
freeMemory[TotalFreeMemory] = (memInfo.cardMemAvailableBytes + memInfo.cardExtMemAvailableBytes +
resourceCache().lclCacheSize()) / Ki;
freeMemory[LargestFreeBlock] =
std::max(memInfo.cardLargestFreeBlockBytes, memInfo.cardExtLargestFreeBlockBytes) / Ki;
if (settings().apuSystem_) {
uint64_t sysMem = 0;
if ((memInfo.agpMemAvailableBytes + resourceCache().cacheSize()) > resourceCache().lclCacheSize()) {
sysMem = (memInfo.agpMemAvailableBytes + resourceCache().cacheSize()) - resourceCache().lclCacheSize();
}
sysMem /= Ki;
freeMemory[TotalFreeMemory] += sysMem;
if (settings().viPlus_) {
// for viPlus_, OCL is using remote instead remoteUSWC to avoid extra copy
freeMemory[LargestFreeBlock] += memInfo.agpCacheableLargestFreeBlockBytes / Ki;
} else {
freeMemory[LargestFreeBlock] += memInfo.agpLargestFreeBlockBytes / Ki;
}
}
return true;
}
amd::Memory* Device::findMapTarget(size_t size) const {
// Must be serialised for access
amd::ScopedLock lk(*mapCacheOps_);
amd::Memory* map = NULL;
size_t minSize = 0;
size_t maxSize = 0;
uint mapId = mapCache_->size();
uint releaseId = mapCache_->size();
// Find if the list has a map target of appropriate size
for (uint i = 0; i < mapCache_->size(); i++) {
if ((*mapCache_)[i] != NULL) {
// Requested size is smaller than the entry size
if (size < (*mapCache_)[i]->getSize()) {
if ((minSize == 0) || (minSize > (*mapCache_)[i]->getSize())) {
minSize = (*mapCache_)[i]->getSize();
mapId = i;
}
}
// Requeted size matches the entry size
else if (size == (*mapCache_)[i]->getSize()) {
mapId = i;
break;
} else {
// Find the biggest map target in the list
if (maxSize < (*mapCache_)[i]->getSize()) {
maxSize = (*mapCache_)[i]->getSize();
releaseId = i;
}
}
}
}
// Check if we found any map target
if (mapId < mapCache_->size()) {
map = (*mapCache_)[mapId];
(*mapCache_)[mapId] = NULL;
Memory* gpuMemory = reinterpret_cast<Memory*>(map->getDeviceMemory(*this));
// Get the base pointer for the map resource
if ((gpuMemory == NULL) || (NULL == gpuMemory->map(NULL))) {
(*mapCache_)[mapId]->release();
map = NULL;
}
}
// If cache is full, then release the biggest map target
else if (releaseId < mapCache_->size()) {
(*mapCache_)[releaseId]->release();
(*mapCache_)[releaseId] = NULL;
}
return map;
}
bool Device::addMapTarget(amd::Memory* memory) const {
// Must be serialised for access
amd::ScopedLock lk(*mapCacheOps_);
// the svm memory shouldn't be cached
if (!memory->canBeCached()) {
return false;
}
// Find if the list has a map target of appropriate size
for (uint i = 0; i < mapCache_->size(); ++i) {
if ((*mapCache_)[i] == NULL) {
(*mapCache_)[i] = memory;
return true;
}
}
// Add a new entry
mapCache_->push_back(memory);
return true;
}
Device::ScratchBuffer::~ScratchBuffer() { destroyMemory(); }
void Device::ScratchBuffer::destroyMemory() {
// Release memory object
delete memObj_;
memObj_ = NULL;
}
bool Device::allocScratch(uint regNum, const VirtualGPU* vgpu) {
if (regNum > 0) {
// Serialize the scratch buffer allocation code
amd::ScopedLock lk(*scratchAlloc_);
uint sb = vgpu->hwRing();
static const uint WaveSizeLimit = ((1 << 21) - 256);
const uint threadSizeLimit = WaveSizeLimit / getAttribs().wavefrontSize;
if (regNum > threadSizeLimit) {
LogError("Requested private memory is bigger than HW supports!");
regNum = threadSizeLimit;
}
// Check if the current buffer isn't big enough
if (regNum > scratch_[sb]->regNum_) {
// Stall all command queues, since runtime will reallocate memory
ScopedLockVgpus lock(*this);
scratch_[sb]->regNum_ = regNum;
uint64_t size = 0;
uint64_t offset = 0;
// Destroy all views
for (uint s = 0; s < scratch_.size(); ++s) {
ScratchBuffer* scratchBuf = scratch_[s];
if (scratchBuf->regNum_ > 0) {
scratchBuf->destroyMemory();
// Calculate the size of the scratch buffer for a queue
scratchBuf->size_ = calcScratchBufferSize(scratchBuf->regNum_);
scratchBuf->size_ = std::min(scratchBuf->size_, info().maxMemAllocSize_);
scratchBuf->size_ = std::min(scratchBuf->size_, uint64_t(3 * Gi));
scratchBuf->size_ = amd::alignUp(scratchBuf->size_, 0xFFFF);
scratchBuf->offset_ = offset;
size += scratchBuf->size_;
offset += scratchBuf->size_;
}
}
delete globalScratchBuf_;
// Allocate new buffer.
globalScratchBuf_ = new gpu::Memory(*this, static_cast<size_t>(size));
if ((globalScratchBuf_ == NULL) || !globalScratchBuf_->create(Resource::Scratch)) {
LogError("Couldn't allocate scratch memory");
for (uint s = 0; s < scratch_.size(); ++s) {
scratch_[s]->regNum_ = 0;
}
return false;
}
for (uint s = 0; s < scratch_.size(); ++s) {
// Loop through all memory objects and reallocate them
if (scratch_[s]->regNum_ > 0) {
// Allocate new buffer
scratch_[s]->memObj_ = new gpu::Memory(*this, scratch_[s]->size_);
Resource::ViewParams view;
view.resource_ = globalScratchBuf_;
view.offset_ = scratch_[s]->offset_;
view.size_ = scratch_[s]->size_;
if ((scratch_[s]->memObj_ == NULL) ||
!scratch_[s]->memObj_->create(Resource::View, &view)) {
LogError("Couldn't allocate a scratch view");
delete scratch_[s]->memObj_;
scratch_[s]->regNum_ = 0;
return false;
}
}
}
}
}
return true;
}
bool Device::validateKernel(
const amd::Kernel& kernel, const device::VirtualDevice* vdev, bool coop_groups) {
// Find the number of scratch registers used in the kernel
const device::Kernel* devKernel = kernel.getDeviceKernel(*this);
uint regNum = static_cast<uint>(devKernel->workGroupInfo()->scratchRegs_);
const VirtualGPU* vgpu = static_cast<const VirtualGPU*>(vdev);
if (!allocScratch(regNum, vgpu)) {
return false;
}
if (devKernel->hsa()) {
const HSAILKernel* hsaKernel = static_cast<const HSAILKernel*>(devKernel);
if (hsaKernel->dynamicParallelism()) {
amd::DeviceQueue* defQueue = kernel.program().context().defDeviceQueue(*this);
if (defQueue != NULL) {
vgpu = static_cast<VirtualGPU*>(defQueue->vDev());
if (!allocScratch(hsaKernel->prog().maxScratchRegs(), vgpu)) {
return false;
}
} else {
return false;
}
}
}
return true;
}
void Device::destroyScratchBuffers() {
if (globalScratchBuf_ != NULL) {
for (uint s = 0; s < scratch_.size(); ++s) {
scratch_[s]->destroyMemory();
scratch_[s]->regNum_ = 0;
}
delete globalScratchBuf_;
globalScratchBuf_ = NULL;
}
}
void Device::fillHwSampler(uint32_t state, void* hwState, uint32_t hwStateSize, uint32_t mipFilter,
float minLod, float maxLod) const {
// All GSL sampler's parameters are in floats
uint32_t gslAddress = GSL_CLAMP_TO_BORDER;
uint32_t gslMinFilter = GSL_MIN_NEAREST;
uint32_t gslMagFilter = GSL_MAG_NEAREST;
bool unnorm = !(state & amd::Sampler::StateNormalizedCoordsMask);
state &= ~amd::Sampler::StateNormalizedCoordsMask;
// Program the sampler address mode
switch (state & amd::Sampler::StateAddressMask) {
case amd::Sampler::StateAddressRepeat:
gslAddress = GSL_REPEAT;
break;
case amd::Sampler::StateAddressClampToEdge:
gslAddress = GSL_CLAMP_TO_EDGE;
break;
case amd::Sampler::StateAddressMirroredRepeat:
gslAddress = GSL_MIRRORED_REPEAT;
break;
case amd::Sampler::StateAddressClamp:
case amd::Sampler::StateAddressNone:
default:
break;
}
state &= ~amd::Sampler::StateAddressMask;
// Program texture filter mode
if (state == amd::Sampler::StateFilterLinear) {
gslMinFilter = GSL_MIN_LINEAR;
gslMagFilter = GSL_MAG_LINEAR;
}
if (mipFilter == CL_FILTER_NEAREST) {
if (gslMinFilter == GSL_MIN_NEAREST) {
gslMinFilter = GSL_MIN_NEAREST_MIPMAP_NEAREST;
} else {
gslMinFilter = GSL_MIN_LINEAR_MIPMAP_NEAREST;
}
} else if (mipFilter == CL_FILTER_LINEAR) {
if (gslMinFilter == GSL_MIN_NEAREST) {
gslMinFilter = GSL_MIN_NEAREST_MIPMAP_LINEAR;
} else {
gslMinFilter = GSL_MIN_LINEAR_MIPMAP_LINEAR;
}
}
fillSamplerHwState(unnorm, gslMinFilter, gslMagFilter, gslAddress, minLod, maxLod, hwState,
hwStateSize);
}
void* Device::hostAlloc(size_t size, size_t alignment, MemorySegment mem_seg) const {
// for discrete gpu, we only reserve,no commit yet.
return amd::Os::reserveMemory(NULL, size, alignment, amd::Os::MEM_PROT_NONE);
}
void Device::hostFree(void* ptr, size_t size) const {
// If we allocate the host memory, we need free, or we have to release
amd::Os::releaseMemory(ptr, size);
}
void* Device::svmAlloc(amd::Context& context, size_t size, size_t alignment, cl_svm_mem_flags flags,
void* svmPtr) const {
alignment = std::max(alignment, static_cast<size_t>(info_.memBaseAddrAlign_));
amd::Memory* mem = NULL;
if (NULL == svmPtr) {
if (isFineGrainedSystem()) {
return amd::Os::alignedMalloc(size, alignment);
}
// create a hidden buffer, which will allocated on the device later
mem = new (context) amd::Buffer(context, flags, size, reinterpret_cast<void*>(1));
if (mem == NULL) {
LogError("failed to create a svm mem object!");
return NULL;
}
if (!mem->create(NULL, false)) {
LogError("failed to create a svm hidden buffer!");
mem->release();
return NULL;
}
// if the device supports SVM FGS, return the committed CPU address directly.
gpu::Memory* gpuMem = getGpuMemory(mem);
// add the information to context so that we can use it later.
amd::MemObjMap::AddMemObj(mem->getSvmPtr(), mem);
svmPtr = mem->getSvmPtr();
} else {
// find the existing amd::mem object
mem = amd::MemObjMap::FindMemObj(svmPtr);
if (NULL == mem) {
return NULL;
}
// commit the CPU memory for FGS device.
if (isFineGrainedSystem()) {
mem->commitSvmMemory();
} else {
gpu::Memory* gpuMem = getGpuMemory(mem);
}
svmPtr = mem->getSvmPtr();
}
return svmPtr;
}
void Device::svmFree(void* ptr) const {
if (isFineGrainedSystem()) {
amd::Os::alignedFree(ptr);
} else {
amd::Memory* svmMem = NULL;
svmMem = amd::MemObjMap::FindMemObj(ptr);
if (NULL != svmMem) {
svmMem->release();
amd::MemObjMap::RemoveMemObj(ptr);
}
}
}
Device::SrdManager::~SrdManager() {
for (uint i = 0; i < pool_.size(); ++i) {
pool_[i].buf_->unmap(NULL);
delete pool_[i].buf_;
delete pool_[i].flags_;
}
}
bool Sampler::create(uint32_t oclSamplerState) {
hwSrd_ = dev_.srds().allocSrdSlot(&hwState_);
if (0 == hwSrd_) {
return false;
}
dev_.fillHwSampler(oclSamplerState, hwState_, HsaSamplerObjectSize);
return true;
}
bool Sampler::create(const amd::Sampler& owner) {
hwSrd_ = dev_.srds().allocSrdSlot(&hwState_);
if (0 == hwSrd_) {
return false;
}
dev_.fillHwSampler(owner.state(), hwState_, HsaSamplerObjectSize, owner.mipFilter(),
owner.minLod(), owner.maxLod());
return true;
}
Sampler::~Sampler() { dev_.srds().freeSrdSlot(hwSrd_); }
uint64_t Device::SrdManager::allocSrdSlot(address* cpuAddr) {
amd::ScopedLock lock(ml_);
// Check all buffers in the pool of chunks
for (uint i = 0; i < pool_.size(); ++i) {
const Chunk& ch = pool_[i];
// Search for an empty slot
for (uint s = 0; s < numFlags_; ++s) {
uint mask = ch.flags_[s];
// Check if there is an empty slot in this group
if (mask != 0) {
uint idx;
// Find the first empty index
for (idx = 0; (mask & 0x1) == 0; mask >>= 1, ++idx)
;
// Mark the slot as busy
ch.flags_[s] &= ~(1 << idx);
// Calculate SRD offset in the buffer
uint offset = (s * MaskBits + idx) * srdSize_;
*cpuAddr = ch.buf_->data() + offset;
return ch.buf_->vmAddress() + offset;
}
}
}
// At this point the manager doesn't have empty slots
// and has to allocate a new chunk
Chunk chunk;
chunk.flags_ = new uint[numFlags_];
if (chunk.flags_ == NULL) {
return 0;
}
chunk.buf_ = new Memory(dev_, bufSize_);
if (chunk.buf_ == NULL || !chunk.buf_->create(Resource::Remote) ||
(NULL == chunk.buf_->map(NULL))) {
delete[] chunk.flags_;
delete chunk.buf_;
return 0;
}
// All slots in the chunk are in "free" state
memset(chunk.flags_, 0xff, numFlags_ * sizeof(uint));
// Take the first one...
chunk.flags_[0] &= ~0x1;
pool_.push_back(chunk);
*cpuAddr = chunk.buf_->data();
return chunk.buf_->vmAddress();
}
void Device::SrdManager::freeSrdSlot(uint64_t addr) {
amd::ScopedLock lock(ml_);
// Check all buffers in the pool of chunks
for (uint i = 0; i < pool_.size(); ++i) {
Chunk* ch = &pool_[i];
// Find the offset
int64_t offs = static_cast<int64_t>(addr) - static_cast<int64_t>(ch->buf_->vmAddress());
// Check if the offset inside the chunk buffer
if ((offs >= 0) && (offs < bufSize_)) {
// Find the index in the chunk
uint idx = offs / srdSize_;
uint s = idx / MaskBits;
// Free the slot
ch->flags_[s] |= 1 << (idx % MaskBits);
return;
}
}
assert(false && "Wrong slot address!");
}
void Device::SrdManager::fillResourceList(std::vector<const Memory*>& memList) {
for (uint i = 0; i < pool_.size(); ++i) {
memList.push_back(pool_[i].buf_);
}
}
int32_t Device::hwDebugManagerInit(amd::Context* context, uintptr_t messageStorage) {
int32_t status = hwDebugMgr_->registerDebugger(context, messageStorage);
if (CL_SUCCESS != status) {
delete hwDebugMgr_;
hwDebugMgr_ = NULL;
}
return status;
}
bool Device::SetClockMode(const cl_set_device_clock_mode_input_amd setClockModeInput, cl_set_device_clock_mode_output_amd* pSetClockModeOutput) {
bool result = true;
static const bool bValidate = true;
PerformAdapterInitialization(bValidate);
GSLClockModeInfo clockModeInfo = {};
clockModeInfo.clockmode = static_cast<GSLClockMode>(setClockModeInput.clock_mode);
result = gslSetClockMode(&clockModeInfo);
CloseInitializedAdapter(bValidate);
return result;
}
} // namespace gpu
Ссылка в новой задаче Показать git blame Копировать постоянную ссылку