From 7c188abb4fe28d9784b12125e6924df2ce7de078 Mon Sep 17 00:00:00 2001 From: Tom Eccles Date: Tue, 27 Aug 2024 11:18:55 +0100 Subject: [PATCH] [mlir][LLVMIR][OpenMP] Move reduction allocas before stores (#105683) Delay implicit reduction var stores until after all allocas. This fixes a couple of cases in existing unit tests. Follow up to https://github.com/llvm/llvm-project/pull/102524 --- .../Lower/OpenMP/parallel-reduction-mixed.f90 | 4 +- .../OpenMP/OpenMPToLLVMIRTranslation.cpp | 47 ++++++++++++++----- .../openmp-reduction-array-sections.mlir | 2 +- 3 files changed, 38 insertions(+), 15 deletions(-) diff --git a/flang/test/Lower/OpenMP/parallel-reduction-mixed.f90 b/flang/test/Lower/OpenMP/parallel-reduction-mixed.f90 index 1457be05ca1025..262075ec9b25d0 100644 --- a/flang/test/Lower/OpenMP/parallel-reduction-mixed.f90 +++ b/flang/test/Lower/OpenMP/parallel-reduction-mixed.f90 @@ -26,10 +26,10 @@ end subroutine proc !CHECK: store i32 %[[TID]], ptr %[[TID_LOCAL]] !CHECK: %[[F_priv:.*]] = alloca ptr !CHECK: %[[I_priv:.*]] = alloca i32 -!CHECK: store ptr %{{.*}}, ptr %[[F_priv]] !CHECK: omp.reduction.init: -!CHECK: store i32 0, ptr %[[I_priv]] +!CHECK: store ptr %{{.*}}, ptr %[[F_priv]] +!CHECK: store i32 0, ptr %[[I_priv]] !CHECK: omp.par.region: !CHECK: br label %[[MALLOC_BB:.*]] diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp index 83a14290bcf64b..6a32aeb4441406 100644 --- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp +++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp @@ -592,7 +592,20 @@ convertOmpOrderedRegion(Operation &opInst, llvm::IRBuilderBase &builder, return bodyGenStatus; } +namespace { +/// Contains the arguments for an LLVM store operation +struct DeferredStore { + DeferredStore(llvm::Value *value, llvm::Value *address) + : value(value), address(address) {} + + llvm::Value *value; + llvm::Value *address; +}; +} // namespace + /// Allocate space for privatized reduction variables. +/// `deferredStores` contains information to create store operations which needs +/// to be inserted after all allocas template static LogicalResult allocReductionVars(T loop, ArrayRef reductionArgs, @@ -602,13 +615,13 @@ allocReductionVars(T loop, ArrayRef reductionArgs, SmallVectorImpl &reductionDecls, SmallVectorImpl &privateReductionVariables, DenseMap &reductionVariableMap, + SmallVectorImpl &deferredStores, llvm::ArrayRef isByRefs) { llvm::IRBuilderBase::InsertPointGuard guard(builder); builder.SetInsertPoint(allocaIP.getBlock()->getTerminator()); // delay creating stores until after all allocas - SmallVector> storesToCreate; - storesToCreate.reserve(loop.getNumReductionVars()); + deferredStores.reserve(loop.getNumReductionVars()); for (std::size_t i = 0; i < loop.getNumReductionVars(); ++i) { Region &allocRegion = reductionDecls[i].getAllocRegion(); @@ -628,7 +641,7 @@ allocReductionVars(T loop, ArrayRef reductionArgs, // variable allocated in the inlined region) llvm::Value *var = builder.CreateAlloca( moduleTranslation.convertType(reductionDecls[i].getType())); - storesToCreate.emplace_back(phis[0], var); + deferredStores.emplace_back(phis[0], var); privateReductionVariables[i] = var; moduleTranslation.mapValue(reductionArgs[i], phis[0]); @@ -644,10 +657,6 @@ allocReductionVars(T loop, ArrayRef reductionArgs, } } - // TODO: further delay this so it doesn't come in the entry block at all - for (auto [data, addr] : storesToCreate) - builder.CreateStore(data, addr); - return success(); } @@ -819,12 +828,19 @@ static LogicalResult allocAndInitializeReductionVars( if (op.getNumReductionVars() == 0) return success(); + SmallVector deferredStores; + if (failed(allocReductionVars(op, reductionArgs, builder, moduleTranslation, allocaIP, reductionDecls, privateReductionVariables, reductionVariableMap, - isByRef))) + deferredStores, isByRef))) return failure(); + // store result of the alloc region to the allocated pointer to the real + // reduction variable + for (auto [data, addr] : deferredStores) + builder.CreateStore(data, addr); + // Before the loop, store the initial values of reductions into reduction // variables. Although this could be done after allocas, we don't want to mess // up with the alloca insertion point. @@ -1359,6 +1375,7 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder, collectReductionDecls(opInst, reductionDecls); SmallVector privateReductionVariables( opInst.getNumReductionVars()); + SmallVector deferredStores; auto bodyGenCB = [&](InsertPointTy allocaIP, InsertPointTy codeGenIP) { // Allocate reduction vars @@ -1373,10 +1390,10 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder, InsertPointTy(allocaIP.getBlock(), allocaIP.getBlock()->getTerminator()->getIterator()); - if (failed(allocReductionVars(opInst, reductionArgs, builder, - moduleTranslation, allocaIP, reductionDecls, - privateReductionVariables, - reductionVariableMap, isByRef))) + if (failed(allocReductionVars( + opInst, reductionArgs, builder, moduleTranslation, allocaIP, + reductionDecls, privateReductionVariables, reductionVariableMap, + deferredStores, isByRef))) bodyGenStatus = failure(); // Initialize reduction vars @@ -1401,6 +1418,12 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder, builder.SetInsertPoint(initBlock->getFirstNonPHIOrDbgOrAlloca()); + // insert stores deferred until after all allocas + // these store the results of the alloc region into the allocation for the + // pointer to the reduction variable + for (auto [data, addr] : deferredStores) + builder.CreateStore(data, addr); + for (unsigned i = 0; i < opInst.getNumReductionVars(); ++i) { SmallVector phis; diff --git a/mlir/test/Target/LLVMIR/openmp-reduction-array-sections.mlir b/mlir/test/Target/LLVMIR/openmp-reduction-array-sections.mlir index da6f9430046123..2d8a13ccd2a1f5 100644 --- a/mlir/test/Target/LLVMIR/openmp-reduction-array-sections.mlir +++ b/mlir/test/Target/LLVMIR/openmp-reduction-array-sections.mlir @@ -89,7 +89,6 @@ llvm.func @sectionsreduction_(%arg0: !llvm.ptr {fir.bindc_name = "x"}) attribute // CHECK: %[[VAL_13:.*]] = load i32, ptr %[[VAL_10]], align 4 // CHECK: %[[VAL_20:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 // CHECK: %[[VAL_21:.*]] = alloca ptr, align 8 -// CHECK: store ptr %[[VAL_20]], ptr %[[VAL_21]], align 8 // CHECK: %[[VAL_14:.*]] = alloca [1 x ptr], align 8 // CHECK: br label %[[VAL_15:.*]] // CHECK: omp.reduction.init: ; preds = %[[VAL_16:.*]] @@ -98,6 +97,7 @@ llvm.func @sectionsreduction_(%arg0: !llvm.ptr {fir.bindc_name = "x"}) attribute // CHECK: br label %[[VAL_18:.*]] // CHECK: omp.par.region1: ; preds = %[[VAL_17]] // CHECK: %[[VAL_19:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 +// CHECK: store ptr %[[VAL_20]], ptr %[[VAL_21]], align 8 // CHECK: br label %[[VAL_22:.*]] // CHECK: omp_section_loop.preheader: ; preds = %[[VAL_18]] // CHECK: store i32 0, ptr %[[VAL_7]], align 4