GPUHeuristics.cpp File Reference

#include "mlir/Dialect/Linalg/TransformOps/GPUHeuristics.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <cmath>
#include <numeric>

Go to the source code of this file.

Macros
#define	DEBUG_TYPE   "linalg-transforms"
#define	DBGS()   (llvm::dbgs() << "[" DEBUG_TYPE "]: ")
#define	LDBG(X)   LLVM_DEBUG(DBGS() << X << "\n")

Functions
static Attribute	linearId0 (MLIRContext *ctx)
static Attribute	linearId1 (MLIRContext *ctx)
static Attribute	linearId2 (MLIRContext *ctx)
static SmallVector< int64_t >	getFactors (int64_t val)
	Get the list of all factors that divide val, not just the prime factors. More...
static int64_t	product (ArrayRef< int64_t > vals)
static SmallVector< int64_t >	maximizeNumThreads (ArrayRef< int64_t > sizes, int64_t currentIndex, int64_t maxNumThreads)
	Extract result from sizes with the following constraints: More...

Macro Definition Documentation

DBGS

#define DBGS

(

)

(llvm::dbgs() << "[" DEBUG_TYPE "]: ")

Definition at line 24 of file GPUHeuristics.cpp.

DEBUG_TYPE

#define DEBUG_TYPE   "linalg-transforms"

Definition at line 23 of file GPUHeuristics.cpp.

LDBG

#define LDBG

(

X

)

LLVM_DEBUG(DBGS() << X << "\n")

Definition at line 25 of file GPUHeuristics.cpp.

Function Documentation

getFactors()

static SmallVector<int64_t> getFactors ( int64_t  val )

static

Get the list of all factors that divide val, not just the prime factors.

Definition at line 104 of file GPUHeuristics.cpp.

Referenced by maximizeNumThreads().

linearId0()

static Attribute linearId0 ( MLIRContext *  ctx )

static

Definition at line 27 of file GPUHeuristics.cpp.

References mlir::get().

Referenced by mlir::transform::gpu::CopyMappingInfo::CopyMappingInfo().

linearId1()

static Attribute linearId1 ( MLIRContext *  ctx )

static

Definition at line 30 of file GPUHeuristics.cpp.

References mlir::get().

Referenced by mlir::transform::gpu::CopyMappingInfo::CopyMappingInfo().

linearId2()

static Attribute linearId2 ( MLIRContext *  ctx )

static

Definition at line 33 of file GPUHeuristics.cpp.

References mlir::get().

Referenced by mlir::transform::gpu::CopyMappingInfo::CopyMappingInfo().

maximizeNumThreads()

static SmallVector<int64_t> maximizeNumThreads ( ArrayRef< int64_t >  sizes, int64_t  currentIndex, int64_t  maxNumThreads  )

static

Extract result from sizes with the following constraints:

1. sizes[i] % result[i] for all i
2. product_of_threadsPerDim <= maxNumThreads
3. if currentIndex is sizes.size() - 1, then threadsPerDim[currentIndex] must be sizes[currentIndex]. This is used to greedily extract the maximum number of threads usable for mapping a copy of size sizes, while being bounded by totalNumThreads and ensuring coalesced access along the most minor dimension. Return the number of threads used in the range: threadsPerDim[currentIndex .. sizes.end()]

Definition at line 137 of file GPUHeuristics.cpp.

References DBGS, getFactors(), LDBG, and product().

product()

static int64_t product ( ArrayRef< int64_t >  vals )

static

Definition at line 116 of file GPUHeuristics.cpp.

Referenced by mlir::sparse_tensor::genReshapeDstShape(), getProductOfIntsOrIndexes(), maximizeNumThreads(), mlir::presburger::QuasiPolynomial::operator*(), and simplifyMul().