AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

Merge pull request #1830 from asalmanp/coop_flag_define

Telusuri Sumber 
[HIP][HIPIFY] Add some missing flags for cooperative launch and occup…

[ROCm/hip commit: 2536a3093d]
This commit is contained in:
Evgeny Mankov
2020-02-12 14:07:39 +03:00
committed by GitHub
orang tua 2868c785d8 07c3e4f898
melakukan b58d79dfa3
9 mengubah file dengan 49 tambahan dan 13 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
projects/hip/bin/hipify-perl
Executable → Regular
+8
Melihat File
@@ -1172,6 +1172,7 @@ sub simpleSubstitutions {
$ft{'numeric_literal'} += s/\bCUDA_ERROR_ASSERT\b/hipErrorAssert/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_CONTEXT_ALREADY_CURRENT\b/hipErrorContextAlreadyCurrent/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_CONTEXT_ALREADY_IN_USE\b/hipErrorContextAlreadyInUse/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE\b/hipErrorCooperativeLaunchTooLarge/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_DEINITIALIZED\b/hipErrorDeinitialized/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_ECC_UNCORRECTABLE\b/hipErrorECCNotCorrectable/g;
$ft{'numeric_literal'} += s/\bCUDA_ERROR_FILE_NOT_FOUND\b/hipErrorFileNotFound/g;
@@ -1528,6 +1529,7 @@ sub simpleSubstitutions {
$ft{'numeric_literal'} += s/\bCU_MEMORYTYPE_UNIFIED\b/hipMemoryTypeUnified/g;
$ft{'numeric_literal'} += s/\bCU_MEM_ATTACH_GLOBAL\b/hipMemAttachGlobal/g;
$ft{'numeric_literal'} += s/\bCU_MEM_ATTACH_HOST\b/hipMemAttachHost/g;
$ft{'numeric_literal'} += s/\bCU_OCCUPANCY_DEFAULT\b/hipOccupancyDefault/g;
$ft{'numeric_literal'} += s/\bCU_RESOURCE_TYPE_ARRAY\b/hipResourceTypeArray/g;
$ft{'numeric_literal'} += s/\bCU_RESOURCE_TYPE_LINEAR\b/hipResourceTypeLinear/g;
$ft{'numeric_literal'} += s/\bCU_RESOURCE_TYPE_MIPMAPPED_ARRAY\b/hipResourceTypeMipmappedArray/g;
@@ -1636,6 +1638,7 @@ sub simpleSubstitutions {
$ft{'numeric_literal'} += s/\bcudaErrorAlreadyMapped\b/hipErrorAlreadyMapped/g;
$ft{'numeric_literal'} += s/\bcudaErrorArrayIsMapped\b/hipErrorArrayIsMapped/g;
$ft{'numeric_literal'} += s/\bcudaErrorAssert\b/hipErrorAssert/g;
$ft{'numeric_literal'} += s/\bcudaErrorCooperativeLaunchTooLarge\b/hipErrorCooperativeLaunchTooLarge/g;
$ft{'numeric_literal'} += s/\bcudaErrorCudartUnloading\b/hipErrorDeinitialized/g;
$ft{'numeric_literal'} += s/\bcudaErrorDeviceAlreadyInUse\b/hipErrorContextAlreadyInUse/g;
$ft{'numeric_literal'} += s/\bcudaErrorDeviceUninitialized\b/hipErrorInvalidContext/g;
@@ -1749,6 +1752,8 @@ sub simpleSubstitutions {
$ft{'define'} += s/\bCUDA_ARRAY3D_LAYERED\b/hipArrayLayered/g;
$ft{'define'} += s/\bCUDA_ARRAY3D_SURFACE_LDST\b/hipArraySurfaceLoadStore/g;
$ft{'define'} += s/\bCUDA_ARRAY3D_TEXTURE_GATHER\b/hipArrayTextureGather/g;
$ft{'define'} += s/\bCUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC\b/hipCooperativeLaunchMultiDeviceNoPostSync/g;
$ft{'define'} += s/\bCUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC\b/hipCooperativeLaunchMultiDeviceNoPreSync/g;
$ft{'define'} += s/\bCU_LAUNCH_PARAM_BUFFER_POINTER\b/HIP_LAUNCH_PARAM_BUFFER_POINTER/g;
$ft{'define'} += s/\bCU_LAUNCH_PARAM_BUFFER_SIZE\b/HIP_LAUNCH_PARAM_BUFFER_SIZE/g;
$ft{'define'} += s/\bCU_LAUNCH_PARAM_END\b/HIP_LAUNCH_PARAM_END/g;
@@ -1769,6 +1774,8 @@ sub simpleSubstitutions {
$ft{'define'} += s/\bcudaArrayLayered\b/hipArrayLayered/g;
$ft{'define'} += s/\bcudaArraySurfaceLoadStore\b/hipArraySurfaceLoadStore/g;
$ft{'define'} += s/\bcudaArrayTextureGather\b/hipArrayTextureGather/g;
$ft{'define'} += s/\bcudaCooperativeLaunchMultiDeviceNoPostSync\b/hipCooperativeLaunchMultiDeviceNoPostSync/g;
$ft{'define'} += s/\bcudaCooperativeLaunchMultiDeviceNoPreSync\b/hipCooperativeLaunchMultiDeviceNoPreSync/g;
$ft{'define'} += s/\bcudaDeviceBlockingSync\b/hipDeviceScheduleBlockingSync/g;
$ft{'define'} += s/\bcudaDeviceLmemResizeToMax\b/hipDeviceLmemResizeToMax/g;
$ft{'define'} += s/\bcudaDeviceMapHost\b/hipDeviceMapHost/g;
@@ -1792,6 +1799,7 @@ sub simpleSubstitutions {
$ft{'define'} += s/\bcudaIpcMemLazyEnablePeerAccess\b/hipIpcMemLazyEnablePeerAccess/g;
$ft{'define'} += s/\bcudaMemAttachGlobal\b/hipMemAttachGlobal/g;
$ft{'define'} += s/\bcudaMemAttachHost\b/hipMemAttachHost/g;
$ft{'define'} += s/\bcudaOccupancyDefault\b/hipOccupancyDefault/g;
$ft{'define'} += s/\bcudaStreamDefault\b/hipStreamDefault/g;
$ft{'define'} += s/\bcudaStreamNonBlocking\b/hipStreamNonBlocking/g;
$ft{'define'} += s/\bcudaTextureType1D\b/hipTextureType1D/g;
projects/hip/docs/markdown/CUDA_Driver_API_functions_supported_by_HIP.md
+1
Melihat File
@@ -447,6 +447,7 @@
| 717 |*`CUDA_ERROR_INVALID_ADDRESS_SPACE`* | |
| 718 |*`CUDA_ERROR_INVALID_PC`* | |
| 719 |*`CUDA_ERROR_LAUNCH_FAILED`* |*`hipErrorLaunchFailure`* |
| 720 |*`CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE`* |*`hipErrorCooperativeLaunchTooLarge`* |
| 800 |*`CUDA_ERROR_NOT_PERMITTED`* | |
| 801 |*`CUDA_ERROR_NOT_SUPPORTED`* |*`hipErrorNotSupported`* |
| 802 |*`CUDA_ERROR_SYSTEM_NOT_READY`* | | 10.0 |
projects/hip/docs/markdown/CUDA_Runtime_API_functions_supported_by_HIP.md
+2 -2
Melihat File
@@ -800,7 +800,7 @@
| 717 |*`cudaErrorInvalidAddressSpace`* | | |
| 718 |*`cudaErrorInvalidPc`* | | |
| 719 |*`cudaErrorLaunchFailure`* | |*`hipErrorLaunchFailure`* |
| 720 |*`cudaErrorCooperativeLaunchTooLarge`* | 9.0 | |
| 720 |*`cudaErrorCooperativeLaunchTooLarge`* | 9.0 |*`hipErrorCooperativeLaunchTooLarge`* |
| 800 |*`cudaErrorNotPermitted`* | | |
| 801 |*`cudaErrorNotSupported`* | |*`hipErrorNotSupported`* |
| 802 |*`cudaErrorSystemNotReady`* | 10.0 | |
@@ -1045,7 +1045,7 @@
| define |`cudaMemAttachGlobal` | |`hipMemAttachGlobal` |
| define |`cudaMemAttachHost` | |`hipMemAttachHost` |
| define |`cudaMemAttachSingle` | | |
| define |`cudaOccupancyDefault` | | |
| define |`cudaOccupancyDefault` | |`hipOccupancyDefault` |
| define |`cudaOccupancyDisableCachingOverride` | | |
| define |`cudaPeerAccessDefault` | | |
| define |`cudaStreamDefault` | |`hipStreamDefault` |
projects/hip/hipify-clang/src/CUDA2HIP_Driver_API_types.cpp
+5 -3
Melihat File
@@ -1047,7 +1047,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
{"CUoccupancy_flags_enum", {"hipOccupancyFlags", "", CONV_TYPE, API_DRIVER, HIP_UNSUPPORTED}},
// CUoccupancy_flags enum values
// cudaOccupancyDefault
{"CU_OCCUPANCY_DEFAULT", {"hipOccupancyDefault", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 0x00
{"CU_OCCUPANCY_DEFAULT", {"hipOccupancyDefault", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 0x00
// cudaOccupancyDisableCachingOverride
{"CU_OCCUPANCY_DISABLE_CACHING_OVERRIDE", {"hipOccupancyDisableCachingOverride", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 0x01
@@ -1284,6 +1284,8 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
{"CUDA_ERROR_INVALID_PC", {"hipErrorInvalidPc", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 718
// cudaErrorLaunchFailure
{"CUDA_ERROR_LAUNCH_FAILED", {"hipErrorLaunchFailure", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 719
// cudaErrorCooperativeLaunchTooLarge
{"CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE", {"hipErrorCooperativeLaunchTooLarge", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 720
// cudaErrorNotPermitted
{"CUDA_ERROR_NOT_PERMITTED", {"hipErrorNotPermitted", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 800
// cudaErrorNotSupported
@@ -1598,9 +1600,9 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
// cudaArrayColorAttachment
{"CUDA_ARRAY3D_COLOR_ATTACHMENT", {"hipArrayColorAttachment", "", CONV_DEFINE, API_DRIVER, HIP_UNSUPPORTED}}, // 0x20
// cudaCooperativeLaunchMultiDeviceNoPreSync
{"CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC", {"hipCooperativeLaunchMultiDeviceNoPreSync", "", CONV_DEFINE, API_DRIVER, HIP_UNSUPPORTED}}, // 0x01
{"CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC", {"hipCooperativeLaunchMultiDeviceNoPreSync", "", CONV_DEFINE, API_DRIVER}}, // 0x01
// cudaCooperativeLaunchMultiDeviceNoPostSync
{"CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC", {"hipCooperativeLaunchMultiDeviceNoPostSync", "", CONV_DEFINE, API_DRIVER, HIP_UNSUPPORTED}}, // 0x02
{"CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC", {"hipCooperativeLaunchMultiDeviceNoPostSync", "", CONV_DEFINE, API_DRIVER}}, // 0x02
// cudaExternalMemoryDedicated
{"CUDA_EXTERNAL_MEMORY_DEDICATED", {"hipExternalMemoryDedicated", "", CONV_DEFINE, API_DRIVER, HIP_UNSUPPORTED}}, // 0x1
// cudaExternalSemaphoreSignalSkipNvSciBufMemSync
projects/hip/hipify-clang/src/CUDA2HIP_Runtime_API_types.cpp
+5 -5
Melihat File
@@ -787,8 +787,8 @@ const std::map<llvm::StringRef, hipCounter> CUDA_RUNTIME_TYPE_NAME_MAP {
{"cudaErrorInvalidPc", {"hipErrorInvalidPc", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 718
// CUDA_ERROR_LAUNCH_FAILED
{"cudaErrorLaunchFailure", {"hipErrorLaunchFailure", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 719
// no analogue
{"cudaErrorCooperativeLaunchTooLarge", {"hipErrorCooperativeLaunchTooLarge", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 720
// CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE
{"cudaErrorCooperativeLaunchTooLarge", {"hipErrorCooperativeLaunchTooLarge", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 720
// CUDA_ERROR_NOT_PERMITTED
{"cudaErrorNotPermitted", {"hipErrorNotPermitted", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 800
// CUDA_ERROR_NOT_SUPPORTED
@@ -1327,9 +1327,9 @@ const std::map<llvm::StringRef, hipCounter> CUDA_RUNTIME_TYPE_NAME_MAP {
// CUDA_ARRAY3D_COLOR_ATTACHMENT
{"cudaArrayColorAttachment", {"hipArrayColorAttachment", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // 0x20
// CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC
{"cudaCooperativeLaunchMultiDeviceNoPreSync", {"hipCooperativeLaunchMultiDeviceNoPreSync", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // 0x01
{"cudaCooperativeLaunchMultiDeviceNoPreSync", {"hipCooperativeLaunchMultiDeviceNoPreSync", "", CONV_DEFINE, API_RUNTIME}}, // 0x01
// CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC
{"cudaCooperativeLaunchMultiDeviceNoPostSync", {"hipCooperativeLaunchMultiDeviceNoPostSync", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // 0x02
{"cudaCooperativeLaunchMultiDeviceNoPostSync", {"hipCooperativeLaunchMultiDeviceNoPostSync", "", CONV_DEFINE, API_RUNTIME}}, // 0x02
// CU_DEVICE_CPU ((CUdevice)-1)
{"cudaCpuDeviceId", {"hipCpuDeviceId", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // ((int)-1)
// CU_DEVICE_INVALID ((CUdevice)-2)
@@ -1412,7 +1412,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_RUNTIME_TYPE_NAME_MAP {
// no analogue
{"cudaTextureTypeCubemapLayered", {"hipTextureTypeCubemapLayered", "", CONV_DEFINE, API_RUNTIME}}, // 0xFC
// CU_OCCUPANCY_DEFAULT
{"cudaOccupancyDefault", {"hipOccupancyDefault", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // 0x00
{"cudaOccupancyDefault", {"hipOccupancyDefault", "", CONV_DEFINE, API_RUNTIME}}, // 0x00
// CU_OCCUPANCY_DISABLE_CACHING_OVERRIDE
{"cudaOccupancyDisableCachingOverride", {"hipOccupancyDisableCachingOverride", "", CONV_DEFINE, API_RUNTIME, HIP_UNSUPPORTED}}, // 0x01
// CU_STREAM_DEFAULT
projects/hip/include/hip/hcc_detail/hip_runtime_api.h
+8 -3
Melihat File
@@ -212,6 +212,11 @@ enum hipLimit_t {
#define hipArrayCubemap 0x04
#define hipArrayTextureGather 0x08
#define hipOccupancyDefault 0x00
#define hipCooperativeLaunchMultiDeviceNoPreSync 0x01
#define hipCooperativeLaunchMultiDeviceNoPostSync 0x02
/*
* @brief hipJitOption
* @enum
@@ -2903,7 +2908,7 @@ hipError_t hipModuleLaunchKernel(hipFunction_t f, unsigned int gridDimX, unsigne
* @param [in] stream Stream where the kernel should be dispatched. May be 0, in which case th
* default stream is used with associated synchronization rules.
*
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue, hipErrorCooperativeLaunchTooLarge
*/
hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDimX,
void** kernelParams, unsigned int sharedMemBytes,
@@ -2917,7 +2922,7 @@ hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDim
* @param [in] numDevices Size of the launchParamsList array.
* @param [in] flags Flags to control launch behavior.
*
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue, hipErrorCooperativeLaunchTooLarge
*/
hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices, unsigned int flags);
@@ -2960,7 +2965,7 @@ hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
* @param [in] flags Extra flags for occupancy calculation (currently ignored)
*/
hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags);
uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags __dparm(hipOccupancyDefault));
#if __HIP_VDI__ && !defined(__HCC__)
/**
projects/hip/include/hip/hip_runtime_api.h
+4
Melihat File
@@ -255,6 +255,10 @@ typedef enum __HIP_NODISCARD hipError_t {
713, ///< Produced when trying to unlock a non-page-locked memory.
hipErrorLaunchFailure =
719, ///< An exception occurred on the device while executing a kernel.
hipErrorCooperativeLaunchTooLarge =
720, ///< This error indicates that the number of blocks launched per grid for a kernel
///< that was launched via cooperative launch APIs exceeds the maximum number of
///< allowed blocks for the current device
hipErrorNotSupported = 801, ///< Produced when the hip API is not supported/implemented
hipErrorUnknown = 999, //< Unknown error.
// HSA Runtime Error Codes start here.
projects/hip/include/hip/nvcc_detail/hip_runtime_api.h
+14
Melihat File
@@ -140,6 +140,14 @@ typedef enum cudaChannelFormatKind hipChannelFormatKind;
#define hipLimitMallocHeapSize cudaLimitMallocHeapSize
#define hipIpcMemLazyEnablePeerAccess cudaIpcMemLazyEnablePeerAccess
#define hipOccupancyDefault cudaOccupancyDefault
#define hipCooperativeLaunchMultiDeviceNoPreSync \
cudaCooperativeLaunchMultiDeviceNoPreSync
#define hipCooperativeLaunchMultiDeviceNoPostSync \
cudaCooperativeLaunchMultiDeviceNoPostSync
// enum CUjit_option redefines
#define hipJitOptionMaxRegisters CU_JIT_MAX_REGISTERS
#define hipJitOptionThreadsPerBlock CU_JIT_THREADS_PER_BLOCK
@@ -267,6 +275,8 @@ inline static hipError_t hipCUDAErrorTohipError(cudaError_t cuError) {
return hipErrorNotInitialized;
case cudaErrorLaunchFailure:
return hipErrorLaunchFailure;
case cudaErrorCooperativeLaunchTooLarge:
return hipErrorCooperativeLaunchTooLarge;
case cudaErrorPriorLaunchFailure:
return hipErrorPriorLaunchFailure;
case cudaErrorLaunchOutOfResources:
@@ -445,6 +455,8 @@ inline static hipError_t hipCUResultTohipError(CUresult cuError) {
return hipErrorHostMemoryNotRegistered;
case CUDA_ERROR_LAUNCH_FAILED:
return hipErrorLaunchFailure;
case CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE:
return hipErrorCooperativeLaunchTooLarge;
case CUDA_ERROR_NOT_SUPPORTED:
return hipErrorNotSupported;
case CUDA_ERROR_UNKNOWN:
@@ -601,6 +613,8 @@ inline static cudaError_t hipErrorToCudaError(hipError_t hError) {
return cudaErrorSetOnActiveProcess;
case hipErrorLaunchFailure:
return cudaErrorLaunchFailure;
case hipErrorCooperativeLaunchTooLarge:
return cudaErrorCooperativeLaunchTooLarge;
case hipErrorNotSupported:
return cudaErrorNotSupported;
// HSA: does not exist in CUDA
projects/hip/src/hip_hcc.cpp
+2
Melihat File
@@ -1796,6 +1796,8 @@ const char* ihipErrorString(hipError_t hip_error) {
return "hipErrorMissingConfiguration";
case hipErrorLaunchFailure:
return "hipErrorLaunchFailure";
case hipErrorCooperativeLaunchTooLarge:
return "hipErrorCooperativeLaunchTooLarge";
case hipErrorPriorLaunchFailure:
return "hipErrorPriorLaunchFailure";
case hipErrorLaunchTimeOut: