From 6d340e4c448c14b7103fddfec7d429b9d535611d Mon Sep 17 00:00:00 2001 From: Vijay Kandiah Date: Wed, 19 Jun 2024 18:45:23 -0500 Subject: [flang] fixing alloca hoisting for blocks having single op. (#96009) This change fixes the issue https://github.com/llvm/llvm-project/issues/95977 due to commit c0cba5198155dba246ddd5764f57595d9bbbddef inserting allocas after the terminator op in the insertion block in the case where the block had only a single operation, its terminator, in it. With this change, the hoisted constant-sized allocas are placed at the front of the insertion block, rather than right after the first operation in it. --- flang/lib/Optimizer/CodeGen/CodeGen.cpp | 2 +- flang/test/Fir/convert-to-llvm-openmp-and-fir.fir | 13 ++++++++----- flang/test/Fir/convert-to-llvm.fir | 2 +- 3 files changed, 10 insertions(+), 7 deletions(-) diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp index 803d9e6..5f35825 100644 --- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp +++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp @@ -255,7 +255,7 @@ struct AllocaOpConversion : public fir::FIROpConversion { mlir::Region *parentRegion = rewriter.getInsertionBlock()->getParent(); mlir::Block *insertBlock = getBlockForAllocaInsert(parentOp, parentRegion); - size.getDefiningOp()->moveAfter(insertBlock, insertBlock->begin()); + size.getDefiningOp()->moveBefore(&insertBlock->front()); rewriter.setInsertionPointAfter(size.getDefiningOp()); } diff --git a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir index 45ff89b..396fbae 100644 --- a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir +++ b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir @@ -26,6 +26,7 @@ func.func @_QPsb1(%arg0: !fir.ref {fir.bindc_name = "n"}, %arg1: !fir.ref {fir.bindc_name = "n"}, %arg1: !fir.ref // CHECK-LABEL: _QPsimd1 // CHECK-SAME: %[[N_REF:.*]]: !llvm.ptr {fir.bindc_name = "n"}, %[[ARR_REF:.*]]: !llvm.ptr {fir.bindc_name = "arr"}) { +// CHECK: %[[ONE_0:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[ONE_1:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[ONE_2:.*]] = llvm.mlir.constant(1 : i32) : i32 // CHECK: omp.parallel { @@ -281,7 +283,6 @@ func.func @_QPomp_target_data() { } // CHECK-LABEL: llvm.func @_QPomp_target_data() { -// CHECK: %[[VAL_0:.*]] = llvm.mlir.constant(1024 : index) : i64 // CHECK: %[[VAL_1:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_2:.*]] = llvm.alloca %[[VAL_1]] x !llvm.array<1024 x i32> {bindc_name = "d"} : (i64) -> !llvm.ptr // CHECK: %[[VAL_3:.*]] = llvm.mlir.constant(1 : i64) : i64 @@ -290,6 +291,7 @@ func.func @_QPomp_target_data() { // CHECK: %[[VAL_6:.*]] = llvm.alloca %[[VAL_5]] x !llvm.array<1024 x i32> {bindc_name = "b"} : (i64) -> !llvm.ptr // CHECK: %[[VAL_7:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_8:.*]] = llvm.alloca %[[VAL_7]] x !llvm.array<1024 x i32> {bindc_name = "a"} : (i64) -> !llvm.ptr +// CHECK: %[[VAL_0:.*]] = llvm.mlir.constant(1024 : index) : i64 // CHECK: %[[VAL_9:.*]] = llvm.mlir.constant(1024 : index) : i64 // CHECK: %[[VAL_10:.*]] = llvm.mlir.constant(1024 : index) : i64 // CHECK: %[[VAL_11:.*]] = llvm.mlir.constant(1024 : index) : i64 @@ -373,9 +375,9 @@ func.func @_QPopenmp_target_data_region() { } // CHECK-LABEL: llvm.func @_QPopenmp_target_data_region() { -// CHECK: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_2:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_3:.*]] = llvm.alloca %[[VAL_2]] x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr +// CHECK: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_1:.*]] = llvm.alloca %[[VAL_0]] x !llvm.array<1024 x i32> {bindc_name = "a"} : (i64) -> !llvm.ptr // CHECK: %[[VAL_MAX:.*]] = llvm.mlir.constant(1024 : index) : i64 // CHECK: %[[VAL_ONE:.*]] = llvm.mlir.constant(1 : index) : i64 @@ -459,15 +461,15 @@ func.func @_QPomp_target() { } // CHECK-LABEL: llvm.func @_QPomp_target() { -// CHECK: %[[EXTENT:.*]] = llvm.mlir.constant(512 : index) : i64 // CHECK: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[VAL_1:.*]] = llvm.alloca %[[VAL_0]] x !llvm.array<512 x i32> {bindc_name = "a"} : (i64) -> !llvm.ptr +// CHECK: %[[EXTENT:.*]] = llvm.mlir.constant(512 : index) : i64 // CHECK: %[[VAL_2:.*]] = llvm.mlir.constant(64 : i32) : i32 // CHECK: %[[STRIDE:.*]] = llvm.mlir.constant(1 : index) : i64 // CHECK: %[[LOWER:.*]] = llvm.mlir.constant(0 : index) : i64 // CHECK: %[[UPPER:.*]] = llvm.mlir.constant(511 : index) : i64 // CHECK: %[[BOUNDS:.*]] = omp.map.bounds lower_bound(%[[LOWER]] : i64) upper_bound(%[[UPPER]] : i64) extent(%[[EXTENT]] : i64) stride(%[[STRIDE]] : i64) start_idx(%[[STRIDE]] : i64) -// CHECK: %[[MAP:.*]] = omp.map.info var_ptr(%2 : !llvm.ptr, !llvm.array<512 x i32>) map_clauses(tofrom) capture(ByRef) bounds(%[[BOUNDS]]) -> !llvm.ptr {name = "a"} +// CHECK: %[[MAP:.*]] = omp.map.info var_ptr(%[[VAL_1]] : !llvm.ptr, !llvm.array<512 x i32>) map_clauses(tofrom) capture(ByRef) bounds(%[[BOUNDS]]) -> !llvm.ptr {name = "a"} // CHECK: omp.target thread_limit(%[[VAL_2]] : i32) map_entries(%[[MAP]] -> %[[ARG_0:.*]] : !llvm.ptr) { // CHECK: ^bb0(%[[ARG_0]]: !llvm.ptr): // CHECK: %[[VAL_3:.*]] = llvm.mlir.constant(10 : i32) : i32 @@ -715,7 +717,8 @@ func.func @_QPsb() { // CHECK: } // CHECK-LABEL: @_QPsimple_reduction // CHECK-SAME: %[[ARRAY_REF:.*]]: !llvm.ptr -// CHECK: %[[RED_ACCUMULATOR:.*]] = llvm.alloca %1 x i32 {bindc_name = "x"} : (i64) -> !llvm.ptr +// CHECK: %[[VAL_1:.*]] = llvm.mlir.constant(1 : i64) : i64 +// CHECK: %[[RED_ACCUMULATOR:.*]] = llvm.alloca %[[VAL_1]] x i32 {bindc_name = "x"} : (i64) -> !llvm.ptr // CHECK: omp.parallel { // CHECK: omp.wsloop reduction(@[[EQV_REDUCTION]] %[[RED_ACCUMULATOR]] -> %[[PRV:.+]] : !llvm.ptr) { // CHECK-NEXT: omp.loop_nest diff --git a/flang/test/Fir/convert-to-llvm.fir b/flang/test/Fir/convert-to-llvm.fir index c951445..c7f3160 100644 --- a/flang/test/Fir/convert-to-llvm.fir +++ b/flang/test/Fir/convert-to-llvm.fir @@ -2047,7 +2047,6 @@ func.func private @_QPtest_dt_callee(%arg0: !fir.box>) // GENERIC: %[[ALLOCA:.*]] = llvm.alloca %[[ALLOCA_SIZE]] x !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr // AMDGPU: %[[AA:.*]] = llvm.alloca %[[ALLOCA_SIZE]] x !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr<5> // AMDGPU: %[[ALLOCA:.*]] = llvm.addrspacecast %[[AA]] : !llvm.ptr<5> to !llvm.ptr -// CHECK: %[[C20:.*]] = llvm.mlir.constant(20 : index) : i64 // CHECK: %[[ALLOCA_SIZE_X:.*]] = llvm.mlir.constant(1 : i64) : i64 // GENERIC: %[[X:.*]] = llvm.alloca %[[ALLOCA_SIZE_X]] x !llvm.array<20 x struct<"_QFtest_dt_sliceTt", (i32, i32)>> {bindc_name = "x"} : (i64) -> !llvm.ptr // AMDGPU: %[[AC:.*]] = llvm.alloca %[[ALLOCA_SIZE_X]] x !llvm.array<20 x struct<"_QFtest_dt_sliceTt", (i32, i32)>> {bindc_name = "x"} : (i64) -> !llvm.ptr<5> @@ -2056,6 +2055,7 @@ func.func private @_QPtest_dt_callee(%arg0: !fir.box>) // GENERIC: %[[V:.*]] = llvm.alloca %[[ALLOCA_SIZE_V]] x i32 {bindc_name = "v"} : (i64) -> !llvm.ptr // AMDGPU: %[[AB:.*]] = llvm.alloca %[[ALLOCA_SIZE_V]] x i32 {bindc_name = "v"} : (i64) -> !llvm.ptr<5> // AMDGPU: %[[V:.*]] = llvm.addrspacecast %[[AB]] : !llvm.ptr<5> to !llvm.ptr +// CHECK: %[[C20:.*]] = llvm.mlir.constant(20 : index) : i64 // CHECK: %[[C1:.*]] = llvm.mlir.constant(1 : i64) : i64 // CHECK: %[[C10:.*]] = llvm.mlir.constant(10 : i64) : i64 // CHECK: %[[C2:.*]] = llvm.mlir.constant(2 : i64) : i64 -- cgit v1.1