replica
/
hanchenye-Polygeist
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Bump LLVM

This commit is contained in:
William S. Moses 2021-12-24 18:06:02 -05:00 committed by William Moses
parent a311f76736
commit 41966050a6
11 changed files with 78 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
lib/polygeist/Ops.cpp
View File

@ -291,7 +291,7 @@ public:
auto constIdx = op.index().getDefiningOp<arith::ConstantOp>();
if (!constIdx)
return failure();
auto constValue = constIdx.value().dyn_cast<IntegerAttr>();
auto constValue = constIdx.getValue().dyn_cast<IntegerAttr>();
if (!constValue || !constValue.getType().isa<IndexType>() ||
constValue.getValue().getZExtValue() != 0)
return failure();
@ -672,7 +672,7 @@ struct SelectOfCast : public OpRewritePattern<SelectOp> {
if (cst1.source().getType() != cst2.source().getType())
return failure();
auto newSel = rewriter.create<SelectOp>(op.getLoc(), op.condition(),
auto newSel = rewriter.create<SelectOp>(op.getLoc(), op.getCondition(),
cst1.source(), cst2.source());
rewriter.replaceOpWithNewOp<memref::CastOp>(op, op.getType(), newSel);
@ -697,9 +697,9 @@ struct SelectOfSubIndex : public OpRewritePattern<SelectOp> {
if (cst1.source().getType() != cst2.source().getType())
return failure();
auto newSel = rewriter.create<SelectOp>(op.getLoc(), op.condition(),
auto newSel = rewriter.create<SelectOp>(op.getLoc(), op.getCondition(),
cst1.source(), cst2.source());
auto newIdx = rewriter.create<SelectOp>(op.getLoc(), op.condition(),
auto newIdx = rewriter.create<SelectOp>(op.getLoc(), op.getCondition(),
cst1.index(), cst2.index());
rewriter.replaceOpWithNewOp<SubIndexOp>(op, op.getType(), newSel, newIdx);
return success();
@ -874,7 +874,8 @@ public:
if (!src)
return failure();
rewriter.replaceOpWithNewOp<Pointer2MemrefOp>(op, op.getType(), src.arg());
rewriter.replaceOpWithNewOp<Pointer2MemrefOp>(op, op.getType(),
src.getArg());
return success();
}
};

12
lib/polygeist/Passes/AffineCFG.cpp
View File

@ -714,17 +714,17 @@ bool handle(OpBuilder &b, CmpIOp cmpi, SmallVectorImpl<AffineExpr> &exprs,
SmallVectorImpl<bool> &eqflags, SmallVectorImpl<Value> &applies) {
AffineMap lhsmap =
AffineMap::get(0, 1, getAffineSymbolExpr(0, cmpi.getContext()));
if (!isValidIndex(cmpi.lhs())) {
if (!isValidIndex(cmpi.getLhs())) {
LLVM_DEBUG(llvm::dbgs()
<< "illegal lhs: " << cmpi.lhs() << " - " << cmpi << "\n");
<< "illegal lhs: " << cmpi.getLhs() << " - " << cmpi << "\n");
return false;
}
if (!isValidIndex(cmpi.rhs())) {
if (!isValidIndex(cmpi.getRhs())) {
LLVM_DEBUG(llvm::dbgs()
<< "illegal rhs: " << cmpi.rhs() << " - " << cmpi << "\n");
<< "illegal rhs: " << cmpi.getRhs() << " - " << cmpi << "\n");
return false;
}
SmallVector<Value, 4> lhspack = {cmpi.lhs()};
SmallVector<Value, 4> lhspack = {cmpi.getLhs()};
if (!lhspack[0].getType().isa<IndexType>()) {
auto op = b.create<IndexCastOp>(cmpi.getLoc(), lhspack[0],
IndexType::get(cmpi.getContext()));
@ -733,7 +733,7 @@ bool handle(OpBuilder &b, CmpIOp cmpi, SmallVectorImpl<AffineExpr> &exprs,
AffineMap rhsmap =
AffineMap::get(0, 1, getAffineSymbolExpr(0, cmpi.getContext()));
SmallVector<Value, 4> rhspack = {cmpi.rhs()};
SmallVector<Value, 4> rhspack = {cmpi.getRhs()};
if (!rhspack[0].getType().isa<IndexType>()) {
auto op = b.create<IndexCastOp>(cmpi.getLoc(), rhspack[0],
IndexType::get(cmpi.getContext()));

2
lib/polygeist/Passes/BarrierRemovalContinuation.cpp
View File

@ -247,7 +247,7 @@ replicateIntoRegion(Region &region, Value storage, ValueRange ivs,
if (auto branch = dyn_cast<BranchOp>(&op)) {
// if (!blocks.contains(branch.dest())) {
if (isa_and_nonnull<polygeist::BarrierOp>(branch->getPrevNode())) {
auto it = llvm::find(subgraphEntryPoints, branch.dest());
auto it = llvm::find(subgraphEntryPoints, branch.getDest());
assert(it != subgraphEntryPoints.end());
emitStoreContinuationID(
branch.getLoc(), std::distance(subgraphEntryPoints.begin(), it),

31
lib/polygeist/Passes/CanonicalizeFor.cpp
View File

@ -557,7 +557,7 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
return failure();
}
BlockArgument indVar = cmpIOp.lhs().dyn_cast<BlockArgument>();
BlockArgument indVar = cmpIOp.getLhs().dyn_cast<BlockArgument>();
if (!indVar)
return failure();
if (indVar.getOwner() != &loop.before().front())
@ -608,11 +608,11 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
loop.getLoc(), loop.getOperand(indVar.getArgNumber()),
rewriter.create<ConstantIntOp>(loop.getLoc(), 1, indVar.getType()));
if (isBlockArg(cmpIOp.rhs()) || dominateWhile(cmpIOp.rhs(), loop)) {
if (isBlockArg(cmpIOp.getRhs()) || dominateWhile(cmpIOp.getRhs(), loop)) {
switch (cmpIOp.getPredicate()) {
case CmpIPredicate::slt:
case CmpIPredicate::ult: {
loopInfo.ub = cmpIOp.rhs();
loopInfo.ub = cmpIOp.getRhs();
break;
}
case CmpIPredicate::ule:
@ -620,13 +620,14 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
// TODO: f32 likely not always true.
auto one =
rewriter.create<ConstantIntOp>(loop.getLoc(), 1, indVar.getType());
auto addIOp = rewriter.create<AddIOp>(loop.getLoc(), cmpIOp.rhs(), one);
auto addIOp =
rewriter.create<AddIOp>(loop.getLoc(), cmpIOp.getRhs(), one);
loopInfo.ub = addIOp.getResult();
break;
}
case CmpIPredicate::uge:
case CmpIPredicate::sge: {
loopInfo.lb = cmpIOp.rhs();
loopInfo.lb = cmpIOp.getRhs();
break;
}
@ -635,7 +636,8 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
// TODO: f32 likely not always true.
auto one =
rewriter.create<ConstantIntOp>(loop.getLoc(), 1, indVar.getType());
auto addIOp = rewriter.create<AddIOp>(loop.getLoc(), cmpIOp.rhs(), one);
auto addIOp =
rewriter.create<AddIOp>(loop.getLoc(), cmpIOp.getRhs(), one);
loopInfo.lb = addIOp.getResult();
break;
}
@ -647,12 +649,12 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
} else {
if (negativeStep)
return failure();
auto *op = cmpIOp.rhs().getDefiningOp();
auto *op = cmpIOp.getRhs().getDefiningOp();
if (!op || !canMoveOpOutsideWhile(op, loop) || (op->getNumResults() != 1))
return failure();
auto newOp = rewriter.clone(*op);
loopInfo.ub = newOp->getResult(0);
cmpIOp.rhs().replaceAllUsesWith(newOp->getResult(0));
cmpIOp.getRhs().replaceAllUsesWith(newOp->getResult(0));
}
if ((!loopInfo.ub) || (!loopInfo.lb) || (!step))
@ -683,7 +685,7 @@ struct MoveWhileToFor : public OpRewritePattern<WhileOp> {
Type cst = nullptr;
if (auto idx = arg.getDefiningOp<IndexCastOp>()) {
cst = idx.getType();
arg = idx.in();
arg = idx.getIn();
}
Value res;
if (isTopLevelArgValue(arg, &loop.before())) {
@ -1048,13 +1050,12 @@ struct WhileLogicalNegation : public OpRewritePattern<WhileOp> {
for (auto pair : llvm::zip(op.getResults(), term.args(), origAfterArgs)) {
if (!std::get<0>(pair).use_empty()) {
if (auto termCmp = std::get<1>(pair).getDefiningOp<CmpIOp>()) {
if (termCmp.lhs() == condCmp.lhs() &&
termCmp.rhs() == condCmp.rhs()) {
if (termCmp.getLhs() == condCmp.getLhs() &&
termCmp.getRhs() == condCmp.getRhs()) {
// TODO generalize to logical negation of
if (condCmp.getPredicate() == CmpIPredicate::slt &&
termCmp.getPredicate() == CmpIPredicate::sge) {
auto i1Ty = rewriter.getIntegerType(1);
rewriter.updateRootInPlace(op, [&] {
rewriter.setInsertionPoint(op);
auto truev =
@ -1086,7 +1087,7 @@ struct WhileCmpOffset : public OpRewritePattern<WhileOp> {
assert(origAfterArgs.size() == term.args().size());
if (auto condCmp = term.condition().getDefiningOp<CmpIOp>()) {
if (auto addI = condCmp.lhs().getDefiningOp<AddIOp>()) {
if (auto addI = condCmp.getLhs().getDefiningOp<AddIOp>()) {
if (addI.getOperand(1).getDefiningOp() &&
!op.before().isAncestor(
addI.getOperand(1).getDefiningOp()->getParentRegion()))
@ -1368,8 +1369,8 @@ struct MoveSideEffectFreeWhile : public OpRewritePattern<WhileOp> {
auto rep =
op.after().front().addArgument(IC->getOperand(0).getType());
IC->moveBefore(&op.after().front(), op.after().front().begin());
conds.push_back(IC.in());
IC.inMutable().assign(rep);
conds.push_back(IC.getIn());
IC.getInMutable().assign(rep);
op.after().front().getArgument(i).replaceAllUsesWith(
IC->getResult(0));
changed = true;

45
lib/polygeist/Passes/Mem2Reg.cpp
View File

@ -297,7 +297,7 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
}
}
if (auto storeOp = dyn_cast<LLVM::StoreOp>(user)) {
if (storeOp.addr() == val) {
if (storeOp.getAddr() == val) {
LLVM_DEBUG(llvm::dbgs() << "Matching Store: " << storeOp << "\n");
allStoreOps.insert(storeOp);
continue;
@ -319,14 +319,14 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
}
}
if (auto callOp = dyn_cast<mlir::CallOp>(user)) {
if (callOp.callee() != "free") {
if (callOp.getCallee() != "free") {
LLVM_DEBUG(llvm::dbgs() << "Aliasing Store: " << callOp << "\n");
AliasingStoreOperations.insert(callOp);
captured = true;
}
}
if (auto callOp = dyn_cast<mlir::LLVM::CallOp>(user)) {
if (*callOp.callee() != "free") {
if (*callOp.getCallee() != "free") {
LLVM_DEBUG(llvm::dbgs() << "Aliasing Store: " << callOp << "\n");
AliasingStoreOperations.insert(callOp);
captured = true;
@ -722,7 +722,7 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
}
} else if (auto storeOp = dyn_cast<LLVM::StoreOp>(a)) {
if (allStoreOps.count(storeOp)) {
lastVal = storeOp.value();
lastVal = storeOp.getValue();
seenSubStore = false;
}
} else if (auto storeOp = dyn_cast<AffineStoreOp>(a)) {
@ -975,15 +975,15 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
op.erase();
} else if (auto op = dyn_cast<SwitchOp>(pred->getTerminator())) {
mlir::OpBuilder builder(op.getOperation());
SmallVector<Value> defaultOps(op.defaultOperands().begin(),
op.defaultOperands().end());
SmallVector<Value> defaultOps(op.getDefaultOperands().begin(),
op.getDefaultOperands().end());
if (op.defaultDestination() == block)
if (op.getDefaultDestination() == block)
defaultOps.push_back(pval);
SmallVector<SmallVector<Value>> cases;
SmallVector<ValueRange> vrange;
for (auto pair : llvm::enumerate(op.caseDestinations())) {
for (auto pair : llvm::enumerate(op.getCaseDestinations())) {
cases.emplace_back(op.getCaseOperands(pair.index()).begin(),
op.getCaseOperands(pair.index()).end());
if (pair.value() == block) {
@ -992,8 +992,8 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
vrange.push_back(cases.back());
}
builder.create<mlir::SwitchOp>(
op.getLoc(), op.flag(), op.defaultDestination(), defaultOps,
op.case_valuesAttr(), op.caseDestinations(), vrange);
op.getLoc(), op.getFlag(), op.getDefaultDestination(), defaultOps,
op.getCaseValuesAttr(), op.getCaseDestinations(), vrange);
op.erase();
} else {
llvm_unreachable("unknown pred branch");
@ -1064,12 +1064,12 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
}
} else if (auto op = dyn_cast<SwitchOp>(pred->getTerminator())) {
mlir::OpBuilder subbuilder(op.getOperation());
if (op.defaultDestination() == block) {
pval = op.defaultOperands()[blockArg.getArgNumber()];
if (op.getDefaultDestination() == block) {
pval = op.getDefaultOperands()[blockArg.getArgNumber()];
if (pval == blockArg)
pval = nullptr;
}
for (auto pair : llvm::enumerate(op.caseDestinations())) {
for (auto pair : llvm::enumerate(op.getCaseDestinations())) {
if (pair.value() == block) {
auto pval2 =
op.getCaseOperands(pair.index())[blockArg.getArgNumber()];
@ -1210,14 +1210,14 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
op.erase();
} else if (auto op = dyn_cast<SwitchOp>(pred->getTerminator())) {
mlir::OpBuilder builder(op.getOperation());
SmallVector<Value> defaultOps(op.defaultOperands().begin(),
op.defaultOperands().end());
if (op.defaultDestination() == block)
SmallVector<Value> defaultOps(op.getDefaultOperands().begin(),
op.getDefaultOperands().end());
if (op.getDefaultDestination() == block)
defaultOps.erase(defaultOps.begin() + blockArg.getArgNumber());
SmallVector<SmallVector<Value>> cases;
SmallVector<ValueRange> vrange;
for (auto pair : llvm::enumerate(op.caseDestinations())) {
for (auto pair : llvm::enumerate(op.getCaseDestinations())) {
cases.emplace_back(op.getCaseOperands(pair.index()).begin(),
op.getCaseOperands(pair.index()).end());
if (pair.value() == block) {
@ -1226,9 +1226,10 @@ bool Mem2Reg::forwardStoreToLoad(mlir::Value AI, std::vector<ssize_t> idx,
}
vrange.push_back(cases.back());
}
builder.create<mlir::SwitchOp>(
op.getLoc(), op.flag(), op.defaultDestination(), defaultOps,
op.case_valuesAttr(), op.caseDestinations(), vrange);
builder.create<mlir::SwitchOp>(op.getLoc(), op.getFlag(),
op.getDefaultDestination(),
defaultOps, op.getCaseValuesAttr(),
op.getCaseDestinations(), vrange);
op.erase();
}
}
@ -1295,7 +1296,7 @@ bool isPromotable(mlir::Value AI) {
continue;
} else if (isa<memref::DeallocOp>(U)) {
continue;
} else if (isa<CallOp>(U) && cast<CallOp>(U).callee() == "free") {
} else if (isa<CallOp>(U) && cast<CallOp>(U).getCallee() == "free") {
continue;
} else if (isa<CallOp>(U)) {
// TODO check "no capture", currently assume as a fallback always
@ -1466,7 +1467,7 @@ void Mem2Reg::runOnFunction() {
if (isa<LLVM::StoreOp, memref::StoreOp, AffineStoreOp,
memref::DeallocOp>(U)) {
toErase.push_back(U);
} else if (isa<CallOp>(U) && cast<CallOp>(U).callee() == "free") {
} else if (isa<CallOp>(U) && cast<CallOp>(U).getCallee() == "free") {
toErase.push_back(U);
} else if (auto CO = dyn_cast<memref::CastOp>(U)) {
toErase.push_back(U);

10
lib/polygeist/Passes/ParallelLower.cpp
View File

@ -171,7 +171,7 @@ void ParallelLower::runOnOperation() {
symbolTable.getSymbolTable(getOperation());
getOperation()->walk([&](mlir::CallOp bidx) {
if (bidx.callee() == "cudaThreadSynchronize")
if (bidx.getCallee() == "cudaThreadSynchronize")
bidx.erase();
});
@ -300,7 +300,7 @@ void ParallelLower::runOnOperation() {
bz.setInsertionPointToStart(blockB);
auto newAlloca = bz.create<LLVM::AllocaOp>(
alop.getLoc(), LLVM::LLVMPointerType::get(PT.getElementType(), 0),
alop.arraySize());
alop.getArraySize());
alop.replaceAllUsesWith((mlir::Value)bz.create<LLVM::AddrSpaceCastOp>(
alop.getLoc(), PT, newAlloca));
alop.erase();
@ -394,7 +394,7 @@ void ParallelLower::runOnOperation() {
});
container.walk([&](LLVM::CallOp call) {
if (call.callee().getValue() == "cudaMemcpy") {
if (call.getCallee().getValue() == "cudaMemcpy") {
OpBuilder bz(call);
auto falsev = bz.create<ConstantIntOp>(call.getLoc(), false, 1);
bz.create<LLVM::MemcpyOp>(call.getLoc(), call.getOperand(0),
@ -404,7 +404,7 @@ void ParallelLower::runOnOperation() {
bz.create<ConstantIntOp>(call.getLoc(), 0, call.getType(0)));
call.erase();
}
if (call.callee().getValue() == "cudaMemset") {
if (call.getCallee().getValue() == "cudaMemset") {
OpBuilder bz(call);
auto falsev = bz.create<ConstantIntOp>(call.getLoc(), false, 1);
bz.create<LLVM::MemsetOp>(call.getLoc(), call.getOperand(0),
@ -414,7 +414,7 @@ void ParallelLower::runOnOperation() {
call.replaceAllUsesWith(ArrayRef<Value>(vals));
call.erase();
}
if (call.callee().getValue() == "cudaMalloc") {
if (call.getCallee().getValue() == "cudaMalloc") {
Value vals[] = {call.getOperand(0)};
call.replaceAllUsesWith(ArrayRef<Value>(vals));

2
llvm-project

@ -1 +1 @@
Subproject commit 30d87d4a5d02f00ef58ebc24a0ee5c6c370b8b4c
Subproject commit 2709fd1520bca98667db9c10b3156cac892949bc

11
tools/mlir-clang/Lib/clang-mlir.cc
View File

@ -2145,6 +2145,17 @@ ValueCategory MLIRScanner::VisitCallExpr(clang::CallExpr *expr) {
if (auto ic = dyn_cast<ImplicitCastExpr>(expr->getCallee()))
if (auto sr = dyn_cast<DeclRefExpr>(ic->getSubExpr())) {
if (sr->getDecl()->getIdentifier() &&
(sr->getDecl()->getName() == "memmove" ||
sr->getDecl()->getName() == "__builtin_memmove")) {
std::vector<mlir::Value> args = {
getLLVM(expr->getArg(0)), getLLVM(expr->getArg(1)),
getLLVM(expr->getArg(2)), /*isVolatile*/
builder.create<ConstantIntOp>(loc, false, 1)};
builder.create<LLVM::MemmoveOp>(loc, args[0], args[1], args[2],
args[3]);
return ValueCategory(args[0], /*isReference*/ false);
}
if (sr->getDecl()->getIdentifier() &&
(sr->getDecl()->getName() == "cudaMemcpy" ||
sr->getDecl()->getName() == "cudaMemcpyToSymbol" ||

15
tools/mlir-clang/Lib/utils.cc
View File

@ -19,7 +19,7 @@
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringRef.h"
#include "mlir/Dialect/Linalg/IR/LinalgOps.h"
#include "mlir/Dialect/Linalg/IR/Linalg.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
#include "clang/AST/Expr.h"
@ -28,10 +28,9 @@ using namespace mlir;
using namespace llvm;
using namespace clang;
Operation *mlirclang::buildLinalgOp(const AbstractOperation *op, OpBuilder &b,
SmallVectorImpl<mlir::Value> &input,
SmallVectorImpl<mlir::Value> &output) {
StringRef name = op->name;
Operation *buildLinalgOp(StringRef name, OpBuilder &b,
SmallVectorImpl<mlir::Value> &input,
SmallVectorImpl<mlir::Value> &output) {
if (name.compare("linalg.copy") == 0) {
assert(input.size() == 1 && "linalg::copyOp requires 1 input");
assert(output.size() == 1 && "linalg::CopyOp requires 1 output");
@ -50,13 +49,11 @@ mlirclang::replaceFuncByOperation(FuncOp f, StringRef opName, OpBuilder &b,
assert(ctx->isOperationRegistered(opName) &&
"Provided lower_to opName should be registered.");
const AbstractOperation *op = AbstractOperation::lookup(opName, ctx);
if (opName.startswith("linalg"))
return buildLinalgOp(op, b, input, output);
return buildLinalgOp(opName, b, input, output);
// NOTE: The attributes of the provided FuncOp is ignored.
OperationState opState(b.getUnknownLoc(), op->name, input,
OperationState opState(b.getUnknownLoc(), opName, input,
f.getCallableResults(), {});
return b.createOperation(opState);
}

5
tools/mlir-clang/Lib/utils.h
View File

@ -42,11 +42,6 @@ replaceFuncByOperation(mlir::FuncOp f, llvm::StringRef opName,
mlir::OpBuilder &b,
llvm::SmallVectorImpl<mlir::Value> &input,
llvm::SmallVectorImpl<mlir::Value> &output);
mlir::Operation *buildLinalgOp(const mlir::AbstractOperation *op,
mlir::OpBuilder &b,
llvm::SmallVectorImpl<mlir::Value> &input,
llvm::SmallVectorImpl<mlir::Value> &output);
} // namespace mlirclang
#endif

4
tools/mlir-clang/mlir-clang.cc
View File

@ -28,7 +28,7 @@
#include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVMPass.h"
#include "mlir/Dialect/Affine/Passes.h"
#include "mlir/Dialect/GPU/GPUDialect.h"
#include "mlir/Dialect/Linalg/IR/LinalgOps.h"
#include "mlir/Dialect/Linalg/IR/Linalg.h"
#include "mlir/Dialect/OpenMP/OpenMPDialect.h"
#include "mlir/Dialect/SCF/Passes.h"
#include "mlir/Dialect/SCF/SCF.h"
@ -388,6 +388,8 @@ int main(int argc, char **argv) {
LLVM::LLVMPointerType::attachInterface<MemRefInsider>(context);
LLVM::LLVMStructType::attachInterface<MemRefInsider>(context);
MemRefType::attachInterface<PtrElementModel<MemRefType>>(context);
LLVM::LLVMStructType::attachInterface<PtrElementModel<LLVM::LLVMStructType>>(
context);
if (showDialects) {
outs() << "Registered Dialects:\n";