Added new device attributes (#1377)
* Added new device attributes * updated comment * updated with new device attributes supported
This commit is contained in:
ansurya
Maneesh Gupta
@@ -1443,6 +1443,7 @@ while (@ARGV) {
|
||||
$ft{'numeric_literal'} += s/\bcudaComputeModeExclusive\b/hipComputeModeExclusive/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaComputeModeExclusiveProcess\b/hipComputeModeExclusiveProcess/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaComputeModeProhibited\b/hipComputeModeProhibited/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrCanMapHostMemory\b/hipDeviceAttributeCanMapHostMemory/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrClockRate\b/hipDeviceAttributeClockRate/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrComputeCapabilityMajor\b/hipDeviceAttributeComputeCapabilityMajor/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrComputeCapabilityMinor\b/hipDeviceAttributeComputeCapabilityMinor/g;
|
||||
@@ -1450,9 +1451,11 @@ while (@ARGV) {
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrConcurrentKernels\b/hipDeviceAttributeConcurrentKernels/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrCooperativeLaunch\b/hipDeviceAttributeCooperativeLaunch/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrCooperativeMultiDeviceLaunch\b/hipDeviceAttributeCooperativeMultiDeviceLaunch/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrEccEnabled\b/hipDeviceAttributeEccEnabled/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrGlobalMemoryBusWidth\b/hipDeviceAttributeMemoryBusWidth/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrIntegrated\b/hipDeviceAttributeIntegrated/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrIsMultiGpuBoard\b/hipDeviceAttributeIsMultiGpuBoard/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrKernelExecTimeout\b/hipDeviceAttributeKernelExecTimeout/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrL2CacheSize\b/hipDeviceAttributeL2CacheSize/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxBlockDimX\b/hipDeviceAttributeMaxBlockDimX/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxBlockDimY\b/hipDeviceAttributeMaxBlockDimY/g;
|
||||
@@ -1460,6 +1463,7 @@ while (@ARGV) {
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxGridDimX\b/hipDeviceAttributeMaxGridDimX/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxGridDimY\b/hipDeviceAttributeMaxGridDimY/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxGridDimZ\b/hipDeviceAttributeMaxGridDimZ/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxPitch\b/hipDeviceAttributeMaxPitch/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxRegistersPerBlock\b/hipDeviceAttributeMaxRegistersPerBlock/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxSharedMemoryPerBlock\b/hipDeviceAttributeMaxSharedMemoryPerBlock/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMaxSharedMemoryPerMultiprocessor\b/hipDeviceAttributeMaxSharedMemoryPerMultiprocessor/g;
|
||||
@@ -1475,6 +1479,7 @@ while (@ARGV) {
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrMultiProcessorCount\b/hipDeviceAttributeMultiprocessorCount/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrPciBusId\b/hipDeviceAttributePciBusId/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrPciDeviceId\b/hipDeviceAttributePciDeviceId/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrTextureAlignment\b/hipDeviceAttributeTextureAlignment/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrTotalConstantMemory\b/hipDeviceAttributeTotalConstantMemory/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaDevAttrWarpSize\b/hipDeviceAttributeWarpSize/g;
|
||||
$ft{'numeric_literal'} += s/\bcudaErrorAssert\b/hipErrorAssert/g;
|
||||
|
||||
@@ -78,16 +78,16 @@
|
||||
| 8 |*`CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK`* |*`hipDeviceAttributeMaxSharedMemoryPerBlock`* |
|
||||
| 9 |*`CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY`* |*`hipDeviceAttributeTotalConstantMemory`* |
|
||||
| 10 |*`CU_DEVICE_ATTRIBUTE_WARP_SIZE`* |*`hipDeviceAttributeWarpSize`* |
|
||||
| 11 |*`CU_DEVICE_ATTRIBUTE_MAX_PITCH`* | |
|
||||
| 11 |*`CU_DEVICE_ATTRIBUTE_MAX_PITCH`* |*`hipDeviceAttributeMaxPitch`* |
|
||||
| 12 |*`CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK`* |*`hipDeviceAttributeMaxRegistersPerBlock`* |
|
||||
| 12 |*`CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK`* |*`hipDeviceAttributeMaxRegistersPerBlock`* |
|
||||
| 13 |*`CU_DEVICE_ATTRIBUTE_CLOCK_RATE`* |*`hipDeviceAttributeClockRate`* |
|
||||
| 14 |*`CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT`* | |
|
||||
| 14 |*`CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT`* |*`hipDeviceAttributeTextureAlignment`* |
|
||||
| 15 |*`CU_DEVICE_ATTRIBUTE_GPU_OVERLAP`* | |
|
||||
| 16 |*`CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT`* |*`hipDeviceAttributeMultiprocessorCount`* |
|
||||
| 17 |*`CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT`* | |
|
||||
| 17 |*`CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT`* |*`hipDeviceAttributeKernelExecTimeout`* |
|
||||
| 18 |*`CU_DEVICE_ATTRIBUTE_INTEGRATED`* |*`hipDeviceAttributeIntegrated`* |
|
||||
| 19 |*`CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY`* | |
|
||||
| 19 |*`CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY`* |*`hipDeviceAttributeCanMapHostMemory`* |
|
||||
| 20 |*`CU_DEVICE_ATTRIBUTE_COMPUTE_MODE`* |*`hipDeviceAttributeComputeMode`* |
|
||||
| 21 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH`* | |
|
||||
| 22 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH`* | |
|
||||
@@ -103,7 +103,7 @@
|
||||
| 29 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_NUMSLICES`* | |
|
||||
| 30 |*`CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT`* | |
|
||||
| 31 |*`CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS`* |*`hipDeviceAttributeConcurrentKernels`* |
|
||||
| 32 |*`CU_DEVICE_ATTRIBUTE_ECC_ENABLED`* | |
|
||||
| 32 |*`CU_DEVICE_ATTRIBUTE_ECC_ENABLED`* |*`hipDeviceAttributeEccEnabled`* |
|
||||
| 33 |*`CU_DEVICE_ATTRIBUTE_PCI_BUS_ID`* |*`hipDeviceAttributePciBusId`* |
|
||||
| 34 |*`CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID`* |*`hipDeviceAttributePciDeviceId`* |
|
||||
| 35 |*`CU_DEVICE_ATTRIBUTE_TCC_DRIVER`* | |
|
||||
|
||||
@@ -510,15 +510,15 @@
|
||||
| 8 |*`cudaDevAttrMaxSharedMemoryPerBlock`* | |*`hipDeviceAttributeMaxSharedMemoryPerBlock`* |
|
||||
| 9 |*`cudaDevAttrTotalConstantMemory`* | |*`hipDeviceAttributeTotalConstantMemory`* |
|
||||
| 10 |*`cudaDevAttrWarpSize`* | |*`hipDeviceAttributeWarpSize`* |
|
||||
| 11 |*`cudaDevAttrMaxPitch`* | | |
|
||||
| 11 |*`cudaDevAttrMaxPitch`* | |*`hipDeviceAttributeMaxPitch`* |
|
||||
| 12 |*`cudaDevAttrMaxRegistersPerBlock`* | |*`hipDeviceAttributeMaxRegistersPerBlock`* |
|
||||
| 13 |*`cudaDevAttrClockRate`* | |*`hipDeviceAttributeClockRate`* |
|
||||
| 14 |*`cudaDevAttrTextureAlignment`* | | |
|
||||
| 14 |*`cudaDevAttrTextureAlignment`* | |*`hipDeviceAttributeTextureAlignment`* |
|
||||
| 15 |*`cudaDevAttrGpuOverlap`* | | |
|
||||
| 16 |*`cudaDevAttrMultiProcessorCount`* | |*`hipDeviceAttributeMultiprocessorCount`* |
|
||||
| 17 |*`cudaDevAttrKernelExecTimeout`* | | |
|
||||
| 17 |*`cudaDevAttrKernelExecTimeout`* | |*`hipDeviceAttributeKernelExecTimeout`* |
|
||||
| 18 |*`cudaDevAttrIntegrated`* | |*`hipDeviceAttributeIntegrated`* |
|
||||
| 19 |*`cudaDevAttrCanMapHostMemory`* | | |
|
||||
| 19 |*`cudaDevAttrCanMapHostMemory`* | |*`hipDeviceAttributeCanMapHostMemory`* |
|
||||
| 20 |*`cudaDevAttrComputeMode`* | |*`hipDeviceAttributeComputeMode`* |
|
||||
| 21 |*`cudaDevAttrMaxTexture1DWidth`* | |*`hipDeviceAttributeMaxTexture1DWidth`* |
|
||||
| 22 |*`cudaDevAttrMaxTexture2DWidth`* | |*`hipDeviceAttributeMaxTexture2DWidth`* |
|
||||
@@ -531,7 +531,7 @@
|
||||
| 29 |*`cudaDevAttrMaxTexture2DLayeredLayers`* | | |
|
||||
| 30 |*`cudaDevAttrSurfaceAlignment`* | | |
|
||||
| 31 |*`cudaDevAttrConcurrentKernels`* | |*`hipDeviceAttributeConcurrentKernels`* |
|
||||
| 32 |*`cudaDevAttrEccEnabled`* | | |
|
||||
| 32 |*`cudaDevAttrEccEnabled`* | |*`hipDeviceAttributeEccEnabled`* |
|
||||
| 33 |*`cudaDevAttrPciBusId`* | |*`hipDeviceAttributePciBusId`* |
|
||||
| 34 |*`cudaDevAttrPciDeviceId`* | |*`hipDeviceAttributePciDeviceId`* |
|
||||
| 35 |*`cudaDevAttrTccDriver`* | | |
|
||||
|
||||
@@ -305,7 +305,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
|
||||
// cudaDevAttrWarpSize
|
||||
{"CU_DEVICE_ATTRIBUTE_WARP_SIZE", {"hipDeviceAttributeWarpSize", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 10
|
||||
// cudaDevAttrMaxPitch
|
||||
{"CU_DEVICE_ATTRIBUTE_MAX_PITCH", {"hipDeviceAttributeMaxPitch", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 11
|
||||
{"CU_DEVICE_ATTRIBUTE_MAX_PITCH", {"hipDeviceAttributeMaxPitch", "", CONV_NUMERIC_LR}}, // 11
|
||||
// cudaDevAttrMaxRegistersPerBlock
|
||||
{"CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK", {"hipDeviceAttributeMaxRegistersPerBlock", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 12
|
||||
// no analogue
|
||||
@@ -313,18 +313,18 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
|
||||
// cudaDevAttrClockRate
|
||||
{"CU_DEVICE_ATTRIBUTE_CLOCK_RATE", {"hipDeviceAttributeClockRate", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 13
|
||||
// cudaDevAttrTextureAlignment
|
||||
{"CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT", {"hipDeviceAttributeTextureAlignment", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 14
|
||||
{"CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT", {"hipDeviceAttributeTextureAlignment", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 14
|
||||
// cudaDevAttrGpuOverlap
|
||||
// NOTE: Deprecated, use instead CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT
|
||||
{"CU_DEVICE_ATTRIBUTE_GPU_OVERLAP", {"hipDeviceAttributeAsyncEngineCount", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 15
|
||||
// cudaDevAttrMultiProcessorCount
|
||||
{"CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT", {"hipDeviceAttributeMultiprocessorCount", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 16
|
||||
// cudaDevAttrKernelExecTimeout
|
||||
{"CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT", {"hipDeviceAttributeKernelExecTimeout", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 17
|
||||
{"CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT", {"hipDeviceAttributeKernelExecTimeout", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 17
|
||||
// cudaDevAttrIntegrated
|
||||
{"CU_DEVICE_ATTRIBUTE_INTEGRATED", {"hipDeviceAttributeIntegrated", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 18
|
||||
// cudaDevAttrCanMapHostMemory
|
||||
{"CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY", {"hipDeviceAttributeCanMapHostMemory", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 19
|
||||
{"CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY", {"hipDeviceAttributeCanMapHostMemory", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 19
|
||||
// cudaDevAttrComputeMode
|
||||
{"CU_DEVICE_ATTRIBUTE_COMPUTE_MODE", {"hipDeviceAttributeComputeMode", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 20
|
||||
// cudaDevAttrMaxTexture1DWidth
|
||||
@@ -359,7 +359,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DRIVER_TYPE_NAME_MAP{
|
||||
// cudaDevAttrConcurrentKernels
|
||||
{"CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS", {"hipDeviceAttributeConcurrentKernels", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 31
|
||||
// cudaDevAttrEccEnabled
|
||||
{"CU_DEVICE_ATTRIBUTE_ECC_ENABLED", {"hipDeviceAttributeEccEnabled", "", CONV_NUMERIC_LITERAL, API_DRIVER, HIP_UNSUPPORTED}}, // 32
|
||||
{"CU_DEVICE_ATTRIBUTE_ECC_ENABLED", {"hipDeviceAttributeEccEnabled", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 32
|
||||
// cudaDevAttrPciBusId
|
||||
{"CU_DEVICE_ATTRIBUTE_PCI_BUS_ID", {"hipDeviceAttributePciBusId", "", CONV_NUMERIC_LITERAL, API_DRIVER}}, // 33
|
||||
// cudaDevAttrPciDeviceId
|
||||
|
||||
@@ -234,24 +234,24 @@ const std::map<llvm::StringRef, hipCounter> CUDA_RUNTIME_TYPE_NAME_MAP {
|
||||
// CU_DEVICE_ATTRIBUTE_WARP_SIZE
|
||||
{"cudaDevAttrWarpSize", {"hipDeviceAttributeWarpSize", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 10
|
||||
// CU_DEVICE_ATTRIBUTE_MAX_PITCH
|
||||
{"cudaDevAttrMaxPitch", {"hipDeviceAttributeMaxPitch", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 11
|
||||
{"cudaDevAttrMaxPitch", {"hipDeviceAttributeMaxPitch", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 11
|
||||
// CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK
|
||||
{"cudaDevAttrMaxRegistersPerBlock", {"hipDeviceAttributeMaxRegistersPerBlock", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 12
|
||||
// CU_DEVICE_ATTRIBUTE_CLOCK_RATE
|
||||
{"cudaDevAttrClockRate", {"hipDeviceAttributeClockRate", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 13
|
||||
// CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT
|
||||
{"cudaDevAttrTextureAlignment", {"hipDeviceAttributeTextureAlignment", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 14
|
||||
{"cudaDevAttrTextureAlignment", {"hipDeviceAttributeTextureAlignment", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 14
|
||||
// CU_DEVICE_ATTRIBUTE_GPU_OVERLAP
|
||||
// NOTE: Is not deprecated as CUDA Driver's API analogue CU_DEVICE_ATTRIBUTE_GPU_OVERLAP
|
||||
{"cudaDevAttrGpuOverlap", {"hipDeviceAttributeGpuOverlap", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 15
|
||||
// CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT
|
||||
{"cudaDevAttrMultiProcessorCount", {"hipDeviceAttributeMultiprocessorCount", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 16
|
||||
// CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT
|
||||
{"cudaDevAttrKernelExecTimeout", {"hipDeviceAttributeKernelExecTimeout", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 17
|
||||
{"cudaDevAttrKernelExecTimeout", {"hipDeviceAttributeKernelExecTimeout", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 17
|
||||
// CU_DEVICE_ATTRIBUTE_INTEGRATED
|
||||
{"cudaDevAttrIntegrated", {"hipDeviceAttributeIntegrated", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 18
|
||||
// CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY
|
||||
{"cudaDevAttrCanMapHostMemory", {"hipDeviceAttributeCanMapHostMemory", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 19
|
||||
{"cudaDevAttrCanMapHostMemory", {"hipDeviceAttributeCanMapHostMemory", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 19
|
||||
// CU_DEVICE_ATTRIBUTE_COMPUTE_MODE
|
||||
{"cudaDevAttrComputeMode", {"hipDeviceAttributeComputeMode", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 20
|
||||
// CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH
|
||||
@@ -277,7 +277,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_RUNTIME_TYPE_NAME_MAP {
|
||||
// CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS
|
||||
{"cudaDevAttrConcurrentKernels", {"hipDeviceAttributeConcurrentKernels", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 31
|
||||
// CU_DEVICE_ATTRIBUTE_ECC_ENABLED
|
||||
{"cudaDevAttrEccEnabled", {"hipDeviceAttributeEccEnabled", "", CONV_NUMERIC_LITERAL, API_RUNTIME, HIP_UNSUPPORTED}}, // 32
|
||||
{"cudaDevAttrEccEnabled", {"hipDeviceAttributeEccEnabled", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 32
|
||||
// CU_DEVICE_ATTRIBUTE_PCI_BUS_ID
|
||||
{"cudaDevAttrPciBusId", {"hipDeviceAttributePciBusId", "", CONV_NUMERIC_LITERAL, API_RUNTIME}}, // 33
|
||||
// CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID
|
||||
|
||||
@@ -122,6 +122,10 @@ typedef struct hipDeviceProp_t {
|
||||
int maxTexture3D[3]; ///< Maximum dimensions (width, height, depth) of 3D images, in image elements
|
||||
unsigned int* hdpMemFlushCntl; ///< Addres of HDP_MEM_COHERENCY_FLUSH_CNTL register
|
||||
unsigned int* hdpRegFlushCntl; ///< Addres of HDP_REG_COHERENCY_FLUSH_CNTL register
|
||||
size_t memPitch; ///<Maximum pitch in bytes allowed by memory copies
|
||||
size_t textureAlignment; ///<Alignment requirement for textures
|
||||
int kernelExecTimeoutEnabled; ///<Run time limit for kernels executed on the device
|
||||
int ECCEnabled; ///<Device has ECC support enabled
|
||||
} hipDeviceProp_t;
|
||||
|
||||
|
||||
@@ -311,7 +315,14 @@ typedef enum hipDeviceAttribute_t {
|
||||
hipDeviceAttributeMaxTexture3DDepth, ///< Maximum dimensions depth of 3D images in image elements
|
||||
|
||||
hipDeviceAttributeHdpMemFlushCntl, ///< Address of the HDP_MEM_COHERENCY_FLUSH_CNTL register
|
||||
hipDeviceAttributeHdpRegFlushCntl ///< Address of the HDP_REG_COHERENCY_FLUSH_CNTL register
|
||||
hipDeviceAttributeHdpRegFlushCntl, ///< Address of the HDP_REG_COHERENCY_FLUSH_CNTL register
|
||||
|
||||
hipDeviceAttributeMaxPitch, ///< Maximum pitch in bytes allowed by memory copies
|
||||
hipDeviceAttributeTextureAlignment, ///<Alignment requirement for textures
|
||||
hipDeviceAttributeKernelExecTimeout, ///<Run time limit for kernels executed on the device
|
||||
hipDeviceAttributeCanMapHostMemory, ///<Device can map host memory into device address space
|
||||
hipDeviceAttributeEccEnabled ///<Device has ECC support enabled
|
||||
|
||||
} hipDeviceAttribute_t;
|
||||
|
||||
enum hipComputeMode {
|
||||
|
||||
@@ -842,6 +842,11 @@ inline static hipError_t hipGetDeviceProperties(hipDeviceProp_t* p_prop, int dev
|
||||
p_prop->maxTexture3D[1] = cdprop.maxTexture3D[1];
|
||||
p_prop->maxTexture3D[2] = cdprop.maxTexture3D[2];
|
||||
|
||||
p_prop->memPitch = cdprop.memPitch;
|
||||
p_prop->textureAlignment = cdprop.textureAlignment;
|
||||
p_prop->kernelExecTimeoutEnabled = cdprop.kernelExecTimeoutEnabled;
|
||||
p_prop->ECCEnabled = cdprop.ECCEnabled;
|
||||
|
||||
return hipCUDAErrorTohipError(cerror);
|
||||
}
|
||||
|
||||
@@ -946,6 +951,21 @@ inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t att
|
||||
case hipDeviceAttributeMaxTexture3DDepth:
|
||||
cdattr = cudaDevAttrMaxTexture3DDepth;
|
||||
break;
|
||||
case hipDeviceAttributeMaxPitch:
|
||||
cdattr = cudaDevAttrMaxPitch;
|
||||
break;
|
||||
case hipDeviceAttributeTextureAlignment:
|
||||
cdattr = cudaDevAttrTextureAlignment;
|
||||
break;
|
||||
case hipDeviceAttributeKernelExecTimeout:
|
||||
cdattr = cudaDevAttrKernelExecTimeout;
|
||||
break;
|
||||
case hipDeviceAttributeCanMapHostMemory:
|
||||
cdattr = cudaDevAttrCanMapHostMemory;
|
||||
break;
|
||||
case hipDeviceAttributeEccEnabled:
|
||||
cdattr = cudaDevAttrEccEnabled;
|
||||
break;
|
||||
default:
|
||||
cerror = cudaErrorInvalidValue;
|
||||
break;
|
||||
|
||||
@@ -311,6 +311,21 @@ hipError_t ihipDeviceGetAttribute(int* pi, hipDeviceAttribute_t attr, int device
|
||||
case hipDeviceAttributeCooperativeMultiDeviceLaunch:
|
||||
*pi = prop->cooperativeMultiDeviceLaunch;
|
||||
break;
|
||||
case hipDeviceAttributeMaxPitch:
|
||||
*pi = prop->memPitch;
|
||||
break;
|
||||
case hipDeviceAttributeTextureAlignment:
|
||||
*pi = prop->textureAlignment;
|
||||
break;
|
||||
case hipDeviceAttributeKernelExecTimeout:
|
||||
*pi = prop->kernelExecTimeoutEnabled;
|
||||
break;
|
||||
case hipDeviceAttributeCanMapHostMemory:
|
||||
*pi = prop->canMapHostMemory;
|
||||
break;
|
||||
case hipDeviceAttributeEccEnabled:
|
||||
*pi = prop->ECCEnabled;
|
||||
break;
|
||||
default:
|
||||
e = hipErrorInvalidValue;
|
||||
break;
|
||||
|
||||
+16
-1
@@ -734,7 +734,6 @@ hipError_t ihipDevice_t::initProperties(hipDeviceProp_t* prop) {
|
||||
err = hsa_agent_get_info(_hsaAgent, HSA_AGENT_INFO_NAME, &archName);
|
||||
|
||||
prop->gcnArch = atoi(archName + 3);
|
||||
|
||||
DeviceErrorCheck(err);
|
||||
|
||||
// Get agent node
|
||||
@@ -912,6 +911,22 @@ hipError_t ihipDevice_t::initProperties(hipDeviceProp_t* prop) {
|
||||
prop->hdpMemFlushCntl = hdpinfo.HDP_MEM_FLUSH_CNTL;
|
||||
prop->hdpRegFlushCntl = hdpinfo.HDP_REG_FLUSH_CNTL;
|
||||
|
||||
prop->memPitch = INT_MAX; //Maximum pitch in bytes allowed by memory copies (hardcoded 128 bytes in hipMallocPitch)
|
||||
prop->textureAlignment = 0; //Alignment requirement for textures
|
||||
prop->kernelExecTimeoutEnabled = 0; //no run time limit for running kernels on device
|
||||
|
||||
hsa_isa_t isa;
|
||||
err = hsa_agent_get_info(_hsaAgent, (hsa_agent_info_t)HSA_AGENT_INFO_ISA, &isa);
|
||||
DeviceErrorCheck(err);
|
||||
std::size_t isa_sz = 0u;
|
||||
hsa_isa_get_info_alt(isa, HSA_ISA_INFO_NAME_LENGTH, &isa_sz);
|
||||
std::string isa_name(isa_sz, '\0');
|
||||
hsa_isa_get_info_alt(isa, HSA_ISA_INFO_NAME, &isa_name.front());
|
||||
if (isa_name.find("sram-ecc") != std::string::npos)
|
||||
prop->ECCEnabled = 1; //Device has ECC support Enabled
|
||||
else
|
||||
prop->ECCEnabled = 0; //Device has ECC support disabled
|
||||
|
||||
return e;
|
||||
}
|
||||
|
||||
|
||||
@@ -140,5 +140,10 @@ int main(int argc, char* argv[]) {
|
||||
CHECK(test_hipDeviceGetHdpAddress(deviceId, hipDeviceAttributeHdpMemFlushCntl, props.hdpMemFlushCntl));
|
||||
CHECK(test_hipDeviceGetHdpAddress(deviceId, hipDeviceAttributeHdpRegFlushCntl, props.hdpRegFlushCntl));
|
||||
#endif
|
||||
CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeMaxPitch, props.memPitch));
|
||||
CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeTextureAlignment, props.textureAlignment));
|
||||
CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeKernelExecTimeout, props.kernelExecTimeoutEnabled));
|
||||
CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeCanMapHostMemory, props.canMapHostMemory));
|
||||
CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeEccEnabled, props.ECCEnabled));
|
||||
passed();
|
||||
};
|
||||
|
||||
مرجع در شماره جدید
Block a user