rocm-systems/projects/clr/rocclr/device/pal/palsettings.cpp

/* Copyright (c) 2015 - 2021 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 "top.hpp"
#include "os/os.hpp"
#include "device/device.hpp"
#include "device/pal/paldefs.hpp"
#include "device/pal/palsettings.hpp"

#include <algorithm>

#if defined(_WIN32)
#include "Windows.h"
#include "VersionHelpers.h"
#endif

namespace pal {

/*! \brief information for adjusting maximum workload time
 *
 *  This structure contains the time and OS minor version for max workload time
 *  adjustment for Windows 7 or 8.
 */
struct ModifyMaxWorkload {
  uint32_t time;          //!< max work load time (10x ms)
  uint32_t minorVersion;  //!< OS minor version
#if defined(_WIN32)
  BYTE comparisonOps;  //!< Comparison option
#endif
};

Settings::Settings() {
  // Initialize the GPU device default settings
  oclVersion_ = OpenCL12;
  debugFlags_ = 0;
  remoteAlloc_ = REMOTE_ALLOC;

  stagedXferRead_ = true;
  stagedXferWrite_ = true;
  stagedXferSize_ = GPU_STAGING_BUFFER_SIZE * Ki;

  // We will enable staged read/write if we use local memory
  disablePersistent_ = false;

  imageSupport_ = false;
  hwLDSSize_ = 0;

  // Set this to true when we drop the flag
  doublePrecision_ = ::CL_KHR_FP64;

  // Fill workgroup info size
  maxWorkGroupSize_ = 1024;
  preferredWorkGroupSize_ = 256;

  hostMemDirectAccess_ = HostMemDisable;

  libSelector_ = amd::LibraryUndefined;

  // Enable workload split by default (for 24 bit arithmetic or timeout)
  workloadSplitSize_ = 1 << GPU_WORKLOAD_SPLIT;

  // By default use host blit
  blitEngine_ = BlitEngineHost;
  constexpr size_t MaxPinnedXferSize = 64;
  pinnedXferSize_ = std::min(GPU_PINNED_XFER_SIZE, MaxPinnedXferSize) * Mi;

  constexpr size_t PinnedMinXferSize = 4 * Mi;
  pinnedMinXferSize_ = flagIsDefault(GPU_PINNED_MIN_XFER_SIZE)
      ? PinnedMinXferSize
      : GPU_PINNED_MIN_XFER_SIZE * Ki;
  pinnedMinXferSize_ = std::min(pinnedMinXferSize_, pinnedXferSize_);

  // Disable FP_FAST_FMA defines by default
  reportFMAF_ = false;
  reportFMA_ = false;

  // GPU device by default
  apuSystem_ = false;

  // Disable 64 bit pointers support by default
  use64BitPtr_ = false;

  // Max alloc size is 16GB
  maxAllocSize_ = 16 * static_cast<uint64_t>(Gi);

  // Disable memory dependency tracking by default
  numMemDependencies_ = 0;

  // By default cache isn't present
  cacheLineSize_ = 0;
  cacheSize_ = 0;

  // Initialize transfer buffer size to 1MB by default
  xferBufSize_ = 1024 * Ki;

  // Use image DMA if requested
  imageDMA_ = GPU_IMAGE_DMA;

  // Disable ASIC specific features by default
  viPlus_ = false;
  aiPlus_ = false;
  gfx10Plus_ = false;

  // Number of compute rings.
  numComputeRings_ = 0;

  minWorkloadTime_ = 1;       // 0.1 ms
  maxWorkloadTime_ = 500000;  // 500 ms

  // Controls tiled images in persistent
  //!@note IOL for Linux doesn't setup tiling aperture in CMM/QS
  linearPersistentImage_ = false;

  useSingleScratch_ = GPU_USE_SINGLE_SCRATCH;

  // Device enqueuing settings
  numDeviceEvents_ = 1024;
  numWaitEvents_ = 8;

  numScratchWavesPerCu_ = 32;

  // Don't support platform atomics by default.
  svmAtomics_ = false;

  // Use host queue for device enqueuing by default
  useDeviceQueue_ = GPU_USE_DEVICE_QUEUE;

  // Don't support Denormals for single precision by default
  singleFpDenorm_ = false;

  // Disable SDMA workaround by default
  sdamPageFaultWar_ = false;

  // SQTT buffer size in bytes
  rgpSqttDispCount_ = PAL_RGP_DISP_COUNT;
  rgpSqttWaitIdle_ = true;
  rgpSqttForceDisable_ = false;

  // Sub allocation parameters
  subAllocationMinSize_ = 4 * Ki;
  subAllocationChunkSize_ = 64 * Mi;
  subAllocationMaxSize_ =
      std::min(static_cast<uint64_t>(GPU_MAX_SUBALLOC_SIZE) * Ki, subAllocationChunkSize_);

  maxCmdBuffers_ = 12;
  useLightning_ = amd::IS_HIP ? true : ((!flagIsDefault(GPU_ENABLE_LC)) ? GPU_ENABLE_LC : false);
  enableWgpMode_ = false;
  enableWave32Mode_ = false;
  hsailExplicitXnack_ = false;
  lcWavefrontSize64_ = true;
  enableHwP2P_ = false;
  imageBufferWar_ = false;
  disableSdma_ = PAL_DISABLE_SDMA;
  mallPolicy_ = 0;
  alwaysResident_ = amd::IS_HIP ? true : false;
  prepinnedMinSize_ = 0;
}

bool Settings::create(const Pal::DeviceProperties& palProp,
                      const Pal::GpuMemoryHeapProperties* heaps, const Pal::WorkStationCaps& wscaps,
                      bool enableXNACK, bool reportAsOCL12Device) {
  uint32_t osVer = 0x0;

  // Disable thread trace by default for all devices
  threadTraceEnable_ = false;
  bool doublePrecision = true;

  // Update GPU specific settings and info structure if we have any
#if defined(_WIN32)
  ModifyMaxWorkload modifyMaxWorkload = {0, 1, VER_EQUAL};
#else
  ModifyMaxWorkload modifyMaxWorkload = {0};
#endif

  // APU systems
  if (palProp.gpuType == Pal::GpuType::Integrated) {
    apuSystem_ = true;
  }

  enableXNACK_ = enableXNACK;
  hsailExplicitXnack_ = enableXNACK;

  switch (palProp.revision) {
    case Pal::AsicRevision::Navi14:
    case Pal::AsicRevision::Navi12:
    case Pal::AsicRevision::Navi10:
    case Pal::AsicRevision::Navi10_A0:
    case Pal::AsicRevision::Navi24:
    case Pal::AsicRevision::Navi23:
    case Pal::AsicRevision::Navi22:
    case Pal::AsicRevision::Navi21:
      gfx10Plus_ = true;
      // Force luxmark to use HSAIL
      {
        std::string appName = {};
        std::string appPathAndName = {};
        amd::Os::getAppPathAndFileName(appName, appPathAndName);
        if ((appName == "luxmark.exe") ||
            (appName == "luxmark")) {
          useLightning_ = flagIsDefault(GPU_ENABLE_LC) ? false : GPU_ENABLE_LC;
        }
        else {
          useLightning_ = GPU_ENABLE_LC;
        }
      }
      enableWgpMode_ = GPU_ENABLE_WGP_MODE;
      if (useLightning_) {
        enableWave32Mode_ = true;
      }
      if (!flagIsDefault(GPU_ENABLE_WAVE32_MODE)) {
        enableWave32Mode_ = GPU_ENABLE_WAVE32_MODE;
      }
      lcWavefrontSize64_ = !enableWave32Mode_;
      if (palProp.gfxLevel == Pal::GfxIpLevel::GfxIp10_1) {
        // GFX10.1 HW doesn't support custom pitch. Enable double copy workaround
        imageBufferWar_ = GPU_IMAGE_BUFFER_WAR;
      }
      if (false) {
        // UnknownDevice0 HW doesn't have SDMA engine
        disableSdma_ = true;
        // And LDS is limited to 32KB
        hwLDSSize_ = 32 * Ki;
        // No fp64 support
        doublePrecision = false;
      }
      // Fall through to AI (gfx9) ...
    case Pal::AsicRevision::Vega20:
      // Enable HW P2P path for Vega20+. Runtime still relies on KMD/PAL for support
      enableHwP2P_ = true;
    case Pal::AsicRevision::Vega12:
    case Pal::AsicRevision::Vega10:
    case Pal::AsicRevision::Raven:
    case Pal::AsicRevision::Raven2:
    case Pal::AsicRevision::Renoir:
      aiPlus_ = true;
      enableCoopGroups_ = IS_LINUX;
      enableCoopMultiDeviceGroups_ = IS_LINUX;
    // Fall through to VI ...
    case Pal::AsicRevision::Carrizo:
    case Pal::AsicRevision::Bristol:
    case Pal::AsicRevision::Stoney:
      if (!aiPlus_) {
        // Fix BSOD/TDR issues observed on Stoney Win7/8.1/10
        minWorkloadTime_ = 1000;
        modifyMaxWorkload.time = 1000;       // Decided by experiment
        modifyMaxWorkload.minorVersion = 1;  // Win 7
#if defined(_WIN32)
        modifyMaxWorkload.comparisonOps = VER_EQUAL;  // Limit to Win 7 only
#endif
      }
    case Pal::AsicRevision::Iceland:
    case Pal::AsicRevision::Tonga:
    case Pal::AsicRevision::Fiji:
    case Pal::AsicRevision::Polaris10:
    case Pal::AsicRevision::Polaris11:
    case Pal::AsicRevision::Polaris12:
    case Pal::AsicRevision::Polaris22:
      // Disable tiling aperture on VI+
      linearPersistentImage_ = true;
      // Keep this false even though we have support
      // singleFpDenorm_ = true;
      viPlus_ = true;
      // SDMA may have memory access outside of
      // the valid buffer range and cause a page fault
      sdamPageFaultWar_ = true;
      enableExtension(ClKhrFp16);
    // Fall through to CI ...
    case Pal::AsicRevision::Kalindi:
    case Pal::AsicRevision::Godavari:
    case Pal::AsicRevision::Spectre:
    case Pal::AsicRevision::Spooky:
      if (!viPlus_) {
        // Fix BSOD/TDR issues observed on Kaveri Win7 (EPR#416903)
        modifyMaxWorkload.time = 250000;     // 250ms
        modifyMaxWorkload.minorVersion = 1;  // Win 7
#if defined(_WIN32)
        modifyMaxWorkload.comparisonOps = VER_EQUAL;  // limit to Win 7
#endif
      }
    // Fall through ...
    case Pal::AsicRevision::Bonaire:
    case Pal::AsicRevision::Hawaii:
    case Pal::AsicRevision::HawaiiPro:
      threadTraceEnable_ = AMD_THREAD_TRACE_ENABLE;
      reportFMAF_ = false;
      if ((palProp.revision == Pal::AsicRevision::Hawaii) || aiPlus_) {
        reportFMAF_ = true;
      }
      // Cache line size is 64 bytes
      cacheLineSize_ = 64;
      // L1 cache size is 16KB
      cacheSize_ = 16 * Ki;

      libSelector_ = amd::GPU_Library_CI;
      if (LP64_SWITCH(false, true)) {
        oclVersion_ = !reportAsOCL12Device /*&& calAttr.isOpenCL200Device*/
            ? XCONCAT(OpenCL, XCONCAT(OPENCL_MAJOR, OPENCL_MINOR))
            : OpenCL12;
      }
      if (GPU_FORCE_OCL20_32BIT) {
        force32BitOcl20_ = true;
        oclVersion_ = !reportAsOCL12Device /*&& calAttr.isOpenCL200Device*/
            ? XCONCAT(OpenCL, XCONCAT(OPENCL_MAJOR, OPENCL_MINOR))
            : OpenCL12;
      }
      if (OPENCL_VERSION < 200) {
        oclVersion_ = OpenCL12;
      }
      numComputeRings_ = 8;

      // Cap at OpenCL20 for now
      if (oclVersion_ > OpenCL20) oclVersion_ = OpenCL20;

      // This needs to be cleaned once 64bit addressing is stable
      if (oclVersion_ < OpenCL20) {
        use64BitPtr_ = flagIsDefault(GPU_FORCE_64BIT_PTR)
            ? LP64_SWITCH(false,
                          /*calAttr.isWorkstation ||*/ true)
            : GPU_FORCE_64BIT_PTR;
      } else {
        if (GPU_FORCE_64BIT_PTR || LP64_SWITCH(false, true)) {
          use64BitPtr_ = true;
        }
      }

      if (oclVersion_ >= OpenCL20) {
        supportDepthsRGB_ = true;
      }
      if (use64BitPtr_) {
        if (GPU_ENABLE_LARGE_ALLOCATION) {
          maxAllocSize_ = 64ULL * Gi;
        } else {
          maxAllocSize_ = 4048 * Mi;
        }
      } else {
        maxAllocSize_ = 3ULL * Gi;
      }

      // Note: More than 4 command buffers may cause a HW hang
      // with HWSC on pre-gfx9 devices in OCLPerfKernelArguments
      if (!aiPlus_) {
        maxCmdBuffers_ = 4;
      }

      supportRA_ = false;
      numMemDependencies_ = GPU_NUM_MEM_DEPENDENCY;
      break;
    default:
      assert(0 && "Unknown ASIC type!");
      return false;
  }

  if (0 == palProp.engineProperties[Pal::EngineTypeDma].engineCount) {
    disableSdma_ = true;
  }

  // Image DMA must be disabled if SDMA is disabled
  imageDMA_ &= !disableSdma_;

  splitSizeForWin7_ = false;

#if defined(_WIN32)
  OSVERSIONINFOEX versionInfo = {0};
  versionInfo.dwOSVersionInfoSize = sizeof(OSVERSIONINFOEX);
  versionInfo.dwMajorVersion = 6;
  versionInfo.dwMinorVersion = modifyMaxWorkload.minorVersion;

  DWORDLONG conditionMask = 0;
  VER_SET_CONDITION(conditionMask, VER_MAJORVERSION, modifyMaxWorkload.comparisonOps);
  VER_SET_CONDITION(conditionMask, VER_MINORVERSION, modifyMaxWorkload.comparisonOps);

  if (VerifyVersionInfo(&versionInfo, VER_MAJORVERSION | VER_MINORVERSION, conditionMask)) {
    splitSizeForWin7_ = true;  // Update flag of DMA flush split size for Win 7
    if (modifyMaxWorkload.time > 0) {
      maxWorkloadTime_ = modifyMaxWorkload.time;  // Update max workload time
    }
  }
#endif  // defined(_WIN32)

  // Enable atomics support
  enableExtension(ClKhrInt64BaseAtomics);
  enableExtension(ClKhrInt64ExtendedAtomics);
  enableExtension(ClKhrGlobalInt32BaseAtomics);
  enableExtension(ClKhrGlobalInt32ExtendedAtomics);
  enableExtension(ClKhrLocalInt32BaseAtomics);
  enableExtension(ClKhrLocalInt32ExtendedAtomics);
  enableExtension(ClKhrByteAddressableStore);
  enableExtension(ClKhr3DImageWrites);
  enableExtension(ClKhrImage2dFromBuffer);
  enableExtension(ClAmdMediaOps);
  enableExtension(ClAmdMediaOps2);

  {
    // Not supported by Unknown device
    enableExtension(ClKhrGlSharing);
    enableExtension(ClKhrGlEvent);
    enableExtension(ClAmdCopyBufferP2P);
  }

  if (!useLightning_) {
    enableExtension(ClAmdPopcnt);
    enableExtension(ClAmdVec3);
    enableExtension(ClAmdPrintf);
  }
  // Enable some platform extensions
  enableExtension(ClAmdDeviceAttributeQuery);


#ifdef ATI_OS_LINUX
  if (palProp.gpuMemoryProperties.busAddressableMemSize > 0)
#endif
  {
    enableExtension(ClAMDLiquidFlash);
  }

  if (hwLDSSize_ == 0) {
    // Use hardcoded values for now, since PAL properties aren't available with offline devices
    hwLDSSize_ = (IS_LINUX || amd::IS_HIP || gfx10Plus_) ? 64 * Ki: 32 * Ki;
  }

  imageSupport_ = true;

  // Use kernels for blit if appropriate
  blitEngine_ = BlitEngineKernel;

  hostMemDirectAccess_ |= HostMemBuffer;
  // HW doesn't support untiled image writes
  // hostMemDirectAccess_ |= HostMemImage;

  if (doublePrecision) {
    // Report FP_FAST_FMA define if double precision HW
    reportFMA_ = true;
    // FMA is 1/4 speed on Pitcairn, Cape Verde, Devastator and Scrapper
    // Bonaire, Kalindi, Spectre and Spooky so disable
    // FP_FMA_FMAF for those parts in switch below
    reportFMAF_ = true;
  }

  // Make sure device actually supports double precision
  doublePrecision_ = (doublePrecision) ? doublePrecision_ : false;
  if (doublePrecision_) {
    // Enable KHR double precision extension
    enableExtension(ClKhrFp64);
  }

  if (!useLightning_ && doublePrecision) {
    // Enable AMD double precision extension
    doublePrecision_ = true;
    enableExtension(ClAmdFp64);
  }

  if (palProp.gpuMemoryProperties.busAddressableMemSize > 0) {
    // Enable bus addressable memory extension
    enableExtension(ClAMDBusAddressableMemory);
  }

  svmFineGrainSystem_ = palProp.gpuMemoryProperties.flags.iommuv2Support;
  svmAtomics_ = svmFineGrainSystem_;

// SVM is not currently supported for DX Interop
#if defined(_WIN32)
  enableExtension(ClKhrD3d9Sharing);
  enableExtension(ClKhrD3d10Sharing);
  enableExtension(ClKhrD3d11Sharing);
#endif  // _WIN32

  // Enable some OpenCL 2.0 extensions
  if (oclVersion_ >= OpenCL20) {
    {
      // Not supported by Unknown device
      enableExtension(ClKhrGLDepthImages);
    }
    enableExtension(ClKhrSubGroups);
    enableExtension(ClKhrDepthImages);

    if (GPU_MIPMAP) {
      enableExtension(ClKhrMipMapImage);
      enableExtension(ClKhrMipMapImageWrites);
    }

    // Enable HW debug
    if (GPU_ENABLE_HW_DEBUG) {
      enableHwDebug_ = true;
    }

#if defined(_WIN32)
    enableExtension(ClAmdPlanarYuv);
#endif
  }

  if (apuSystem_ &&
      ((heaps[Pal::GpuHeapLocal].heapSize + heaps[Pal::GpuHeapInvisible].heapSize) < (150 * Mi))) {
    remoteAlloc_ = true;
  }

  // Update resource cache size
  if (remoteAlloc_) {
    resourceCacheSize_ = std::max((heaps[Pal::GpuHeapGartUswc].heapSize / 8),
                                  (uint64_t)GPU_RESOURCE_CACHE_SIZE * Mi);
  } else {
    resourceCacheSize_ =
        std::max(((heaps[Pal::GpuHeapLocal].heapSize + heaps[Pal::GpuHeapInvisible].heapSize) / 8),
                 (uint64_t)GPU_RESOURCE_CACHE_SIZE * Mi);
#if !defined(_LP64)
    resourceCacheSize_ = std::min(resourceCacheSize_, 1 * Gi);
#endif
  }

  if (useLightning_) {
    switch (palProp.gfxLevel) {
      case Pal::GfxIpLevel::GfxIp10_3:
      case Pal::GfxIpLevel::GfxIp10_1:
      case Pal::GfxIpLevel::GfxIp9:
        singleFpDenorm_ = true;
        break;
      default:
        break;
    }
  }
  // If is Rebar, override prepinned memory size.
  if ((heaps[Pal::GpuHeapInvisible].heapSize == 0) &&
      (heaps[Pal::GpuHeapLocal].heapSize > 256 * Mi)) {
    prepinnedMinSize_ = PAL_PREPINNED_MEMORY_SIZE * Ki;
  }

  // Override current device settings
  override();

  return true;
}

void Settings::override() {
  // Limit reported workgroup size
  if (GPU_MAX_WORKGROUP_SIZE != 0) {
    preferredWorkGroupSize_ = GPU_MAX_WORKGROUP_SIZE;
  }

  // Override blit engine type
  if (GPU_BLIT_ENGINE_TYPE != BlitEngineDefault) {
    blitEngine_ = GPU_BLIT_ENGINE_TYPE;
  }

  if (!flagIsDefault(DEBUG_GPU_FLAGS)) {
    debugFlags_ = DEBUG_GPU_FLAGS;
  }

  if (!flagIsDefault(GPU_XFER_BUFFER_SIZE)) {
    xferBufSize_ = GPU_XFER_BUFFER_SIZE * Ki;
  }

  if (!flagIsDefault(GPU_NUM_COMPUTE_RINGS)) {
    numComputeRings_ = GPU_NUM_COMPUTE_RINGS;
  }

  if (!flagIsDefault(GPU_RESOURCE_CACHE_SIZE)) {
    resourceCacheSize_ = GPU_RESOURCE_CACHE_SIZE * Mi;
  }

  if (!flagIsDefault(GPU_ENABLE_HW_P2P)) {
    enableHwP2P_ = GPU_ENABLE_HW_P2P;
  }

  if (!flagIsDefault(AMD_GPU_FORCE_SINGLE_FP_DENORM)) {
    switch (AMD_GPU_FORCE_SINGLE_FP_DENORM) {
      case 0:
        singleFpDenorm_ = false;
        break;
      case 1:
        singleFpDenorm_ = true;
        break;
      default:
        break;
    }
  }

  if (!flagIsDefault(GPU_MAX_COMMAND_BUFFERS)) {
    maxCmdBuffers_ = GPU_MAX_COMMAND_BUFFERS;
  }

  if (!flagIsDefault(GPU_ENABLE_COOP_GROUPS)) {
    enableCoopGroups_ = GPU_ENABLE_COOP_GROUPS;
    enableCoopMultiDeviceGroups_ = GPU_ENABLE_COOP_GROUPS;
  }

  if (!flagIsDefault(PAL_MALL_POLICY)) {
    mallPolicy_ = PAL_MALL_POLICY;
  }

  if (!flagIsDefault(PAL_ALWAYS_RESIDENT)) {
    alwaysResident_ = PAL_ALWAYS_RESIDENT;
  }
}

}  // namespace pal