|
MLIR 22.0.0git
|
#include "mlir/Dialect/LLVMIR/NVVMDialect.h"#include "mlir/Dialect/Vector/IR/VectorOps.h"#include "mlir/IR/PatternMatch.h"#include "mlir/IR/Types.h"Go to the source code of this file.
Classes | |
| struct | mlir::nvgpu::WarpMatrixInfo |
| Collects information about a warp-level matrix operand represented by a VectorType. More... | |
Namespaces | |
| namespace | mlir |
| Include the generated interface declarations. | |
| namespace | mlir::nvgpu |
Enumerations | |
| enum class | mlir::nvgpu::MatMulOperandRole : int32_t { mlir::nvgpu::A = 0 , mlir::nvgpu::B , mlir::nvgpu::C } |
| Represents the role of an operand in an MMA instruction: result := matmul(A, B) + C More... | |
Functions | |
| FailureOr< vector::ContractionOp > | mlir::nvgpu::getUserContract (Operation *op) |
| Returns the first user of the op that is vector.contract. | |
| FailureOr< WarpMatrixInfo > | mlir::nvgpu::getWarpMatrixInfo (Operation *op) |
| If op is a vector.transfer_write, return the WarpMatrixInfo for the vector operand. | |
| bool | mlir::nvgpu::canLowerToWarpMatrixOperation (vector::TransferWriteOp op) |
| Returns the number of bits in a single tile row. | |