In-memory representation of kernel code properties metadata. More...
#include "llvm/Support/AMDGPUMetadata.h"
Public Member Functions | |
| Metadata ()=default | |
| Default constructor. More... | |
| bool | empty () const |
| bool | notEmpty () const |
Public Attributes | |
| uint64_t | mKernargSegmentSize = 0 |
| Size in bytes of the kernarg segment memory. More... | |
| uint32_t | mGroupSegmentFixedSize = 0 |
| Size in bytes of the group segment memory required by a workgroup. More... | |
| uint32_t | mPrivateSegmentFixedSize = 0 |
| Size in bytes of the private segment memory required by a workitem. More... | |
| uint32_t | mKernargSegmentAlign = 0 |
| Maximum byte alignment of variables used by the kernel in the kernarg memory segment. More... | |
| uint32_t | mWavefrontSize = 0 |
| Wavefront size. Required. More... | |
| uint16_t | mNumSGPRs = 0 |
| Total number of SGPRs used by a wavefront. Optional. More... | |
| uint16_t | mNumVGPRs = 0 |
| Total number of VGPRs used by a workitem. Optional. More... | |
| uint32_t | mMaxFlatWorkGroupSize = 0 |
| Maximum flat work-group size supported by the kernel. Optional. More... | |
| bool | mIsDynamicCallStack = false |
| True if the generated machine code is using a dynamically sized call stack. More... | |
| bool | mIsXNACKEnabled = false |
| True if the generated machine code is capable of supporting XNACK. More... | |
| uint16_t | mNumSpilledSGPRs = 0 |
| Number of SGPRs spilled by a wavefront. Optional. More... | |
| uint16_t | mNumSpilledVGPRs = 0 |
| Number of VGPRs spilled by a workitem. Optional. More... | |
Detailed Description
In-memory representation of kernel code properties metadata.
Definition at line 254 of file AMDGPUMetadata.h.
Constructor & Destructor Documentation
◆ Metadata()
|
default |
Default constructor.
Member Function Documentation
◆ empty()
|
inline |
- Returns
- True if kernel code properties metadata is empty, false otherwise.
Definition at line 292 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::Metadata >::mapping().
◆ notEmpty()
|
inline |
- Returns
- True if kernel code properties metadata is not empty, false otherwise.
Definition at line 298 of file AMDGPUMetadata.h.
Member Data Documentation
◆ mGroupSegmentFixedSize
| uint32_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mGroupSegmentFixedSize = 0 |
Size in bytes of the group segment memory required by a workgroup.
This value does not include any dynamically allocated group segment memory that may be added when the kernel is dispatched. Required.
Definition at line 261 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mIsDynamicCallStack
True if the generated machine code is using a dynamically sized call stack.
Definition at line 278 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mIsXNACKEnabled
True if the generated machine code is capable of supporting XNACK.
Definition at line 281 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mKernargSegmentAlign
| uint32_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mKernargSegmentAlign = 0 |
Maximum byte alignment of variables used by the kernel in the kernarg memory segment.
Required.
Definition at line 267 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mKernargSegmentSize
| uint64_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mKernargSegmentSize = 0 |
Size in bytes of the kernarg segment memory.
Kernarg segment memory holds the values of the arguments to the kernel. Required.
Definition at line 257 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mMaxFlatWorkGroupSize
| uint32_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mMaxFlatWorkGroupSize = 0 |
Maximum flat work-group size supported by the kernel. Optional.
Definition at line 275 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mNumSGPRs
| uint16_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mNumSGPRs = 0 |
Total number of SGPRs used by a wavefront. Optional.
Definition at line 271 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mNumSpilledSGPRs
| uint16_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mNumSpilledSGPRs = 0 |
Number of SGPRs spilled by a wavefront. Optional.
Definition at line 283 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mNumSpilledVGPRs
| uint16_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mNumSpilledVGPRs = 0 |
Number of VGPRs spilled by a workitem. Optional.
Definition at line 285 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mNumVGPRs
| uint16_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mNumVGPRs = 0 |
Total number of VGPRs used by a workitem. Optional.
Definition at line 273 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mPrivateSegmentFixedSize
| uint32_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mPrivateSegmentFixedSize = 0 |
Size in bytes of the private segment memory required by a workitem.
Private segment memory includes arg, spill and private segments. Required.
Definition at line 264 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
◆ mWavefrontSize
| uint32_t llvm::AMDGPU::HSAMD::Kernel::CodeProps::Metadata::mWavefrontSize = 0 |
Wavefront size. Required.
Definition at line 269 of file AMDGPUMetadata.h.
Referenced by llvm::yaml::MappingTraits< Kernel::CodeProps::Metadata >::mapping().
The documentation for this struct was generated from the following file:
- include/llvm/Support/AMDGPUMetadata.h
