gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[TritonIR] Convert Triton dialect's Combine pass to MLIR DRR based (#16)

Browse Source
This commit is contained in:
Yan Chunwei
2022-07-28 03:50:08 +08:00
committed by GitHub
parent 432c3df265
commit e02c82c765
4 changed files with 162 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
lib/Dialect/Triton/Transforms/CMakeLists.txt
View File

@@ -1,6 +1,11 @@
set(LLVM_TARGET_DEFINITIONS Combine.td)
mlir_tablegen(TritonCombine.inc -gen-rewriters)
add_public_tablegen_target(TritonCombineIncGen)
add_mlir_dialect_library(TritonTransforms
Combine.cpp
DEPENDS
TritonTransformsIncGen
TritonCombineIncGen
)

168
lib/Dialect/Triton/Transforms/Combine.cpp
View File

@@ -13,143 +13,43 @@
using namespace mlir;
namespace {
// dot(a, b, 0) + c => dot(a, b, c)
class CombineDotOp : public mlir::RewritePattern {
public:
CombineDotOp(mlir::MLIRContext *context)
: mlir::RewritePattern(mlir::RewritePattern::MatchAnyOpTypeTag(), 1,
context) {}
mlir::LogicalResult
matchAndRewrite(mlir::Operation *op,
mlir::PatternRewriter &rewriter) const override {
if (llvm::isa<mlir::arith::AddIOp, mlir::arith::AddFOp>(op)) {
if (isCandidate(op->getOperand(0)).succeeded()) {
auto dotOp = op->getOperand(0).getDefiningOp<mlir::triton::DotOp>();
rewriter.replaceOpWithNewOp<mlir::triton::DotOp>(
op, dotOp->getResultTypes().front(), dotOp.a(), dotOp.b(),
op->getOperand(1), dotOp.allowTF32());
return mlir::success();
} else if (isCandidate(op->getOperand(1)).succeeded()) {
auto dotOp = op->getOperand(1).getDefiningOp<mlir::triton::DotOp>();
rewriter.replaceOpWithNewOp<mlir::triton::DotOp>(
op, dotOp->getResultTypes().front(), dotOp.a(), dotOp.b(),
op->getOperand(0), dotOp.allowTF32());
return mlir::success();
}
}
return mlir::failure();
}
private:
// Is this value a dot and has 0 as `c`.
mlir::LogicalResult isCandidate(mlir::Value val) const {
if (auto dot = val.getDefiningOp<mlir::triton::DotOp>()) {
if (isZero(dot.c()))
return mlir::success();
}
return mlir::failure();
}
bool isZero(mlir::Value val) const {
if (mlir::matchPattern(val, mlir::m_Zero()) ||
mlir::matchPattern(val, mlir::m_AnyZeroFloat()))
bool isZero(mlir::Value val) {
if (mlir::matchPattern(val, mlir::m_Zero()) ||
mlir::matchPattern(val, mlir::m_AnyZeroFloat()))
return true;
// broadcast(constant_0)
if (auto bc = val.getDefiningOp<mlir::triton::BroadcastOp>()) {
if (mlir::matchPattern(bc.src(), mlir::m_Zero()) ||
mlir::matchPattern(bc.src(), mlir::m_AnyZeroFloat()))
return true;
// broadcast(constant_0)
if (auto bc = val.getDefiningOp<mlir::triton::BroadcastOp>()) {
if (mlir::matchPattern(bc.src(), mlir::m_Zero()) ||
mlir::matchPattern(bc.src(), mlir::m_AnyZeroFloat()))
return true;
}
return false;
}
};
return false;
}
// gep(gep(%ptr, %idx0), %idx1) => gep(%ptr, AddI(%idx0, %idx1))
// Note: leave (sub %c0, %c0) canceling to ArithmeticDialect
// (ref: ArithmeticCanonicalization.td)
class CombineGEPOp : public mlir::RewritePattern {
public:
CombineGEPOp(mlir::MLIRContext *context)
: mlir::RewritePattern(mlir::RewritePattern::MatchAnyOpTypeTag(), 1,
context) {}
mlir::LogicalResult
matchAndRewrite(mlir::Operation *op,
mlir::PatternRewriter &rewriter) const override {
if (llvm::isa<mlir::triton::GEPOp>(op)) {
if (auto gep2 = op->getOperand(0).getDefiningOp<mlir::triton::GEPOp>()) {
auto loc = op->getLoc();
mlir::Value newIdx = rewriter.create<mlir::arith::AddIOp>(
loc, op->getOperand(1), gep2->getOperand(1));
rewriter.replaceOpWithNewOp<mlir::triton::GEPOp>(
op, op->getResultTypes().front(), gep2->getOperand(0), newIdx);
return mlir::success();
}
}
return mlir::failure();
bool isBroadcastConstantCombinable(Attribute value) {
if (auto denseValue = value.dyn_cast<DenseElementsAttr>()) {
return denseValue.isSplat();
}
};
return value.isa<FloatAttr, IntegerAttr>();
}
// select(cond, load(ptrs, broadcast(cond), ???), other)
// => load(ptrs, broadcast(cond), other)
class CombineSelectMaskedLoadOp : public mlir::RewritePattern {
public:
CombineSelectMaskedLoadOp(mlir::MLIRContext *context)
: mlir::RewritePattern(mlir::RewritePattern::MatchAnyOpTypeTag(), 1,
context) {}
DenseElementsAttr getConstantValue(Builder &builder, Attribute value,
Value bcast_res) {
mlir::LogicalResult
matchAndRewrite(mlir::Operation *op,
mlir::PatternRewriter &rewriter) const override {
if (llvm::isa<mlir::SelectOp>(op)) {
if (auto load = op->getOperand(1).getDefiningOp<mlir::triton::LoadOp>()) {
mlir::Value cond = op->getOperand(0);
if (auto bc = load.mask().getDefiningOp<mlir::triton::BroadcastOp>()) {
if (bc.src().getDefiningOp() == cond.getDefiningOp()) {
rewriter.replaceOpWithNewOp<mlir::triton::LoadOp>(
op, op->getResultTypes().front(), load.ptr(), load.mask(),
op->getOperand(2), load.cache(), load.evict(),
load.isVolatile());
return mlir::success();
}
}
}
}
return mlir::failure();
Type resType = bcast_res.getType();
DenseElementsAttr res;
if (auto denseValue = value.dyn_cast<DenseElementsAttr>()) {
res =
DenseElementsAttr::get(resType, denseValue.getSplatValue<Attribute>());
} else {
res = DenseElementsAttr::get(resType, value);
}
};
return res;
}
// broadcast(cst) => cst
// TODO: move this to .td file
class CombineBroadcastConstantOp : public mlir::RewritePattern {
public:
CombineBroadcastConstantOp(mlir::MLIRContext *context)
: mlir::RewritePattern(mlir::RewritePattern::MatchAnyOpTypeTag(), 1,
context) {}
#include "TritonCombine.inc"
LogicalResult matchAndRewrite(Operation *op,
PatternRewriter &rewriter) const override {
if (auto broadcast = llvm::dyn_cast<triton::BroadcastOp>(op)) {
if (auto cst = broadcast.src().getDefiningOp<arith::ConstantOp>()) {
Attribute value = cst.getValue();
Type resType = broadcast.getResult().getType();
if (auto denseValue = value.dyn_cast<DenseElementsAttr>()) {
if (!denseValue.isSplat())
return failure();
value = DenseElementsAttr::get(resType,
denseValue.getSplatValue<Attribute>());
} else {
if (!value.isa<FloatAttr, IntegerAttr>())
return failure();
value = DenseElementsAttr::get(resType, value);
}
rewriter.replaceOpWithNewOp<arith::ConstantOp>(op, value, resType);
return success();
}
}
return failure();
}
};
} // anonymous namespace
#define GEN_PASS_CLASSES
@@ -162,11 +62,15 @@ public:
mlir::RewritePatternSet patterns(context);
mlir::ModuleOp m = getOperation();
patterns.add<CombineDotOp>(context);
patterns.add<CombineSelectMaskedLoadOp>(context);
patterns.add<CombineGEPOp>(context);
patterns.add<CombineBroadcastConstantOp>(context);
// patterns.add<CombineReduceOp>(context);
// Dot Add %{
patterns.add<CombineDotAddIPattern>(context);
patterns.add<CombineDotAddFPattern>(context);
patterns.add<CombineDotAddIRevPattern>(context);
patterns.add<CombineDotAddFRevPattern>(context);
// %}
patterns.add<CombineSelectMaskedLoadPattern>(context);
patterns.add<CombineGEPPattern>(context);
patterns.add<CombineBroadcastConstantPattern>(context);
if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())
signalPassFailure();

53
lib/Dialect/Triton/Transforms/Combine.td Normal file
View File

@@ -0,0 +1,53 @@
#ifndef TRITON_PATTERNS
#define TRITON_PATTERNS
include "mlir/Dialect/StandardOps/IR/Ops.td"
include "mlir/Dialect/Arithmetic/IR/ArithmeticOps.td"
include "triton/Dialect/Triton/IR/TritonOps.td"
// AddIOp(DotOp(a, b, c), d) and c==0 => DotOp(a, b, d)
// AddFOp(DotOp(a, b, c), d) and c==0 => DotOp(a, b, d)
// AddIOp(d, DotOp(a, b, c)) and c==0 => DotOp(a, b, d)
// AddFOp(d, DotOp(a, b, c)) and c==0 => DotOp(a, b, d)
def CombineDotAddIPattern : Pat<
(Arith_AddIOp $d, (TT_DotOp:$res $a, $b, $c, $allowTF32)),
(TT_DotOp $a, $b, $d, $allowTF32),
[(Constraint<CPred<"isZero($0)">> $c)]>;
def CombineDotAddFPattern : Pat<
(Arith_AddFOp $d, (TT_DotOp:$res $a, $b, $c, $allowTF32)),
(TT_DotOp $a, $b, $d, $allowTF32),
[(Constraint<CPred<"isZero($0)">> $c)]>;
def CombineDotAddIRevPattern : Pat<
(Arith_AddIOp (TT_DotOp:$res $a, $b, $c, $allowTF32), $d),
(TT_DotOp $a, $b, $d, $allowTF32),
[(Constraint<CPred<"isZero($0)">> $c)]>;
def CombineDotAddFRevPattern : Pat<
(Arith_AddFOp (TT_DotOp:$res $a, $b, $c, $allowTF32), $d),
(TT_DotOp $a, $b, $d, $allowTF32),
[(Constraint<CPred<"isZero($0)">> $c)]>;
// gep(gep(%ptr, %idx0), %idx1) => gep(%ptr, AddI(%idx0, %idx1))
// Note: leave (sub %c0, %c0) canceling to ArithmeticDialect
// (ref: ArithmeticCanonicalization.td)
def CombineGEPPattern : Pat<
(TT_GEPOp (TT_GEPOp $ptr, $idx0), $idx1),
(TT_GEPOp $ptr, (Arith_AddIOp $idx0, $idx1))>;
// select(cond, load(ptrs, broadcast(cond), ???), other)
// => load(ptrs, broadcast(cond), other)
def CombineSelectMaskedLoadPattern : Pat<
(SelectOp $cond, (TT_LoadOp $ptrs, (TT_BroadcastOp:$bcast_res $cond), $other, $cache, $evict, $isVolatile), $falseValue),
(TT_LoadOp $ptrs, $bcast_res, $falseValue, $cache, $evict, $isVolatile)>;
// broadcast(cst) => cst
def getConstantValue : NativeCodeCall<"getConstantValue($_builder, $0, $1)">;
def CombineBroadcastConstantPattern : Pat<
(TT_BroadcastOp:$bcast_res (Arith_ConstantOp $value)),
(Arith_ConstantOp (getConstantValue $value, $bcast_res)),
[(Constraint<CPred<"isBroadcastConstantCombinable($0)">> $value)]>;
#endif