[flang] also set llvm ABI argument attributes on direct calls #130736
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
jeanPerier requested review from ashermancinelli, clementval, kkwli, tblah and vzakhari 
llvmbot added flang 
Member
| @llvm/pr-subscribers-flang-fir-hlfir @llvm/pr-subscribers-flang-codegen Author: None (jeanPerier) ChangesSo far, flang was not setting argument attributes on direct calls assuming that putting them on the function operation was enough. It was clarified in 38565da that they must be set on both call and functions, even for direct calls. Crashes have been observed because of the lack of the attribute when compiling Patch is 36.76 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/130736.diff 17 Files Affected:
diff --git a/flang/lib/Optimizer/CodeGen/TargetRewrite.cpp b/flang/lib/Optimizer/CodeGen/TargetRewrite.cpp index 5c9da0321bcc4..43ed60fc0c292 100644 --- a/flang/lib/Optimizer/CodeGen/TargetRewrite.cpp +++ b/flang/lib/Optimizer/CodeGen/TargetRewrite.cpp @@ -541,37 +541,37 @@ class TargetRewrite : public fir::impl::TargetRewritePassBase<TargetRewrite> { callOp.getContext(), mlir::TypeRange{newInTypes}.drop_front(dropFront), newResTys)); newCall = rewriter->create<fir::CallOp>(loc, newResTys, newOpers); - // Set ABI argument attributes on call operation since they are not - // accessible via a FuncOp in indirect calls. - if (hasByValOrSRetArgs(newInTyAndAttrs)) { - llvm::SmallVector<mlir::Attribute> argAttrsArray; - for (const auto &arg : - llvm::ArrayRef<fir::CodeGenSpecifics::TypeAndAttr>( - newInTyAndAttrs) - .drop_front(dropFront)) { - mlir::NamedAttrList argAttrs; - const auto &attr = std::get<fir::CodeGenSpecifics::Attributes>(arg); - if (attr.isByVal()) { - mlir::Type elemType = - fir::dyn_cast_ptrOrBoxEleTy(std::get<mlir::Type>(arg)); - argAttrs.set(mlir::LLVM::LLVMDialect::getByValAttrName(), - mlir::TypeAttr::get(elemType)); - } else if (attr.isSRet()) { - mlir::Type elemType = - fir::dyn_cast_ptrOrBoxEleTy(std::get<mlir::Type>(arg)); - argAttrs.set(mlir::LLVM::LLVMDialect::getStructRetAttrName(), - mlir::TypeAttr::get(elemType)); - if (auto align = attr.getAlignment()) { - argAttrs.set(mlir::LLVM::LLVMDialect::getAlignAttrName(), - rewriter->getIntegerAttr( - rewriter->getIntegerType(32), align)); - } - } - argAttrsArray.emplace_back( - argAttrs.getDictionary(rewriter->getContext())); + } + // Always set ABI argument attributes on call operations, even when + // direct, as required by + // https://llvm.org/docs/LangRef.html#parameter-attributes. + if (hasByValOrSRetArgs(newInTyAndAttrs)) { + llvm::SmallVector<mlir::Attribute> argAttrsArray; + for (const auto &arg : + llvm::ArrayRef<fir::CodeGenSpecifics::TypeAndAttr>(newInTyAndAttrs) + .drop_front(dropFront)) { + mlir::NamedAttrList argAttrs; + const auto &attr = std::get<fir::CodeGenSpecifics::Attributes>(arg); + if (attr.isByVal()) { + mlir::Type elemType = + fir::dyn_cast_ptrOrBoxEleTy(std::get<mlir::Type>(arg)); + argAttrs.set(mlir::LLVM::LLVMDialect::getByValAttrName(), + mlir::TypeAttr::get(elemType)); + } else if (attr.isSRet()) { + mlir::Type elemType = + fir::dyn_cast_ptrOrBoxEleTy(std::get<mlir::Type>(arg)); + argAttrs.set(mlir::LLVM::LLVMDialect::getStructRetAttrName(), + mlir::TypeAttr::get(elemType)); + } + if (auto align = attr.getAlignment()) { + argAttrs.set( + mlir::LLVM::LLVMDialect::getAlignAttrName(), + rewriter->getIntegerAttr(rewriter->getIntegerType(32), align)); } - newCall.setArgAttrsAttr(rewriter->getArrayAttr(argAttrsArray)); + argAttrsArray.emplace_back( + argAttrs.getDictionary(rewriter->getContext())); } + newCall.setArgAttrsAttr(rewriter->getArrayAttr(argAttrsArray)); } LLVM_DEBUG(llvm::dbgs() << "replacing call with " << newCall << '\n'); if (wrap) diff --git a/flang/test/Fir/CUDA/cuda-target-rewrite.mlir b/flang/test/Fir/CUDA/cuda-target-rewrite.mlir index 0e7534e06c89c..831b5a0048008 100644 --- a/flang/test/Fir/CUDA/cuda-target-rewrite.mlir +++ b/flang/test/Fir/CUDA/cuda-target-rewrite.mlir @@ -1,6 +1,7 @@ -// REQUIRES: x86-registered-target // RUN: fir-opt --split-input-file --target-rewrite="target=x86_64-unknown-linux-gnu" %s | FileCheck %s +module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} { + gpu.module @testmod { gpu.func @_QPvcpowdk(%arg0: !fir.ref<complex<f64>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "a"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} { %0 = fir.alloca i64 @@ -15,22 +16,25 @@ gpu.module @testmod { // CHECK-LABEL: gpu.func @_QPvcpowdk // CHECK: %{{.*}} = fir.call @_FortranAzpowk(%{{.*}}, %{{.*}}, %{{.*}}) : (f64, f64, i64) -> tuple<f64, f64> // CHECK: func.func private @_FortranAzpowk(f64, f64, i64) -> tuple<f64, f64> attributes {fir.bindc_name = "_FortranAzpowk", fir.runtime} +} // ----- +module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} { + gpu.module @testmod { gpu.func @_QPtest(%arg0: complex<f64>) -> (complex<f64>) { gpu.return %arg0 : complex<f64> } } +} // CHECK-LABEL: gpu.func @_QPtest // CHECK-SAME: (%arg0: f64, %arg1: f64) -> tuple<f64, f64> { // CHECK: gpu.return %{{.*}} : tuple<f64, f64> - // ----- -module attributes {gpu.container_module} { +module attributes {gpu.container_module, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} { gpu.module @testmod { gpu.func @_QPtest(%arg0: complex<f64>) -> () kernel { diff --git a/flang/test/Fir/struct-passing-return-loongarch64-bystack.fir b/flang/test/Fir/struct-passing-return-loongarch64-bystack.fir index 5041a39e69798..91ac345d2d7af 100644 --- a/flang/test/Fir/struct-passing-return-loongarch64-bystack.fir +++ b/flang/test/Fir/struct-passing-return-loongarch64-bystack.fir @@ -1,7 +1,6 @@ /// Test LoongArch64 ABI rewrite of struct passed and returned by value (BIND(C), VALUE derived types). /// This test test cases where the struct must be passed or returned on the stack. -// REQUIRES: loongarch-registered-target // RUN: tco --target=loongarch64-unknown-linux-gnu %s | FileCheck %s !ty_int_toobig = !fir.type<int_toobig{i:!fir.array<5xi32>}> diff --git a/flang/test/Fir/struct-passing-x86-64-byval.fir b/flang/test/Fir/struct-passing-x86-64-byval.fir index 13cf52de581f7..8451c26095226 100644 --- a/flang/test/Fir/struct-passing-x86-64-byval.fir +++ b/flang/test/Fir/struct-passing-x86-64-byval.fir @@ -1,7 +1,6 @@ // Test X86-64 ABI rewrite of struct passed by value (BIND(C), VALUE derived types). // This test test cases where the struct must be passed on the stack according // to the System V ABI. -// REQUIRES: x86-registered-target // RUN: tco --target=x86_64-unknown-linux-gnu %s | FileCheck %s module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} { diff --git a/flang/test/Fir/struct-passing-x86-64-one-field-inreg.fir b/flang/test/Fir/struct-passing-x86-64-one-field-inreg.fir index e37e8dd4481d0..7feb698e0d5c5 100644 --- a/flang/test/Fir/struct-passing-x86-64-one-field-inreg.fir +++ b/flang/test/Fir/struct-passing-x86-64-one-field-inreg.fir @@ -1,6 +1,5 @@ // Test X86-64 passing ABI of struct in registers for the simple case // where the struct has a single intrinsic component that is not a complex. -// REQUIRES: x86-registered-target // RUN: fir-opt -target-rewrite="target=x86_64-unknown-linux-gnu" %s -o - | FileCheck %s module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} { diff --git a/flang/test/Fir/struct-passing-x86-64-several-fields-inreg.fir b/flang/test/Fir/struct-passing-x86-64-several-fields-inreg.fir index 9a0a41e1da542..afc8c2de85019 100644 --- a/flang/test/Fir/struct-passing-x86-64-several-fields-inreg.fir +++ b/flang/test/Fir/struct-passing-x86-64-several-fields-inreg.fir @@ -1,6 +1,5 @@ // Test X86-64 passing ABI of struct in registers for the cases where the // struct has more than one field. -// REQUIRES: x86-registered-target // RUN: fir-opt -target-rewrite="target=x86_64-unknown-linux-gnu" %s -o - | FileCheck %s diff --git a/flang/test/Fir/struct-return-aarch64.fir b/flang/test/Fir/struct-return-aarch64.fir index 8b75c2cac7b6b..57be5a9828284 100644 --- a/flang/test/Fir/struct-return-aarch64.fir +++ b/flang/test/Fir/struct-return-aarch64.fir @@ -125,7 +125,7 @@ func.func private @test_too_big() -> !too_big func.func @test_call_too_big(%arg0 : !fir.ref<!too_big>) { // CHECK: %[[STACK:.*]] = llvm.intr.stacksave : !llvm.ptr // CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t7{x:i64,y:i64,z:i64}> - // CHECK: fir.call @test_too_big(%[[ARG]]) : (!fir.ref<!fir.type<t7{x:i64,y:i64,z:i64}>>) -> () + // CHECK: fir.call @test_too_big(%[[ARG]]) : (!fir.ref<!fir.type<t7{x:i64,y:i64,z:i64}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t7{x:i64,y:i64,z:i64}>}) -> () // CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t7{x:i64,y:i64,z:i64}>>) -> !fir.ref<!fir.type<t7{x:i64,y:i64,z:i64}>> // CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t7{x:i64,y:i64,z:i64}>> // CHECK: llvm.intr.stackrestore %[[STACK]] : !llvm.ptr @@ -145,7 +145,7 @@ func.func private @test_too_big_hfa() -> !too_big_hfa func.func @test_call_too_big_hfa(%arg0 : !fir.ref<!too_big_hfa>) { // CHECK: %[[STACK:.*]] = llvm.intr.stacksave : !llvm.ptr // CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t8{i:!fir.array<5xf32>}> - // CHECK: fir.call @test_too_big_hfa(%[[ARG]]) : (!fir.ref<!fir.type<t8{i:!fir.array<5xf32>}>>) -> () + // CHECK: fir.call @test_too_big_hfa(%[[ARG]]) : (!fir.ref<!fir.type<t8{i:!fir.array<5xf32>}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t8{i:!fir.array<5xf32>}>}) -> () // CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t8{i:!fir.array<5xf32>}>>) -> !fir.ref<!fir.type<t8{i:!fir.array<5xf32>}>> // CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t8{i:!fir.array<5xf32>}>> // CHECK: llvm.intr.stackrestore %[[STACK]] : !llvm.ptr diff --git a/flang/test/Fir/struct-return-loongarch64-byreg.fir b/flang/test/Fir/struct-return-loongarch64-byreg.fir index eb3d4d50d8866..baf9c74d6231a 100644 --- a/flang/test/Fir/struct-return-loongarch64-byreg.fir +++ b/flang/test/Fir/struct-return-loongarch64-byreg.fir @@ -7,7 +7,6 @@ /// only the first example in each category checks the entire invocation process, /// while the other examples only check the signatures. -// REQUIRES: loongarch-registered-target // RUN: fir-opt --split-input-file --target-rewrite="target=loongarch64-unknown-linux-gnu" %s | FileCheck %s diff --git a/flang/test/Fir/struct-return-powerpc64-aix.fir b/flang/test/Fir/struct-return-powerpc64-aix.fir index 3058ef6b7e965..5037ffca0c79d 100644 --- a/flang/test/Fir/struct-return-powerpc64-aix.fir +++ b/flang/test/Fir/struct-return-powerpc64-aix.fir @@ -15,7 +15,7 @@ func.func @test_call_t1(%arg0 : !fir.ref<!fir.type<t1<{c:!fir.char<1>}>>>) { return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t1<{c:!fir.char<1>}>> - //CHECK: fir.call @test_t1(%[[ARG]]) : (!fir.ref<!fir.type<t1<{c:!fir.char<1>}>>>) -> () + //CHECK: fir.call @test_t1(%[[ARG]]) : (!fir.ref<!fir.type<t1<{c:!fir.char<1>}>>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t1<{c:!fir.char<1>}>>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t1<{c:!fir.char<1>}>>>) -> !fir.ref<!fir.type<t1<{c:!fir.char<1>}>>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t1<{c:!fir.char<1>}>>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -34,7 +34,7 @@ func.func @test_call_t2(%arg0 : !fir.ref<!fir.type<t2<{i:i32}>>>) { return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t2<{i:i32}>> - //CHECK: fir.call @test_t2(%[[ARG]]) : (!fir.ref<!fir.type<t2<{i:i32}>>>) -> () + //CHECK: fir.call @test_t2(%[[ARG]]) : (!fir.ref<!fir.type<t2<{i:i32}>>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t2<{i:i32}>>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t2<{i:i32}>>>) -> !fir.ref<!fir.type<t2<{i:i32}>>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t2<{i:i32}>>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -53,7 +53,7 @@ func.func @test_call_t3(%arg0 : !fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>>) return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t3<{r1:f32,r2:f32,r3:f32}>> - //CHECK: fir.call @test_t3(%[[ARG]]) : (!fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>>) -> () + //CHECK: fir.call @test_t3(%[[ARG]]) : (!fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t3<{r1:f32,r2:f32,r3:f32}>>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>>) -> !fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t3<{r1:f32,r2:f32,r3:f32}>>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -72,7 +72,7 @@ func.func @test_call_t4(%arg0 : !fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>>) return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t4<{r:!fir.array<8xf32>}>> - //CHECK: fir.call @test_t4(%[[ARG]]) : (!fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>>) -> () + //CHECK: fir.call @test_t4(%[[ARG]]) : (!fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t4<{r:!fir.array<8xf32>}>>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>>) -> !fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t4<{r:!fir.array<8xf32>}>>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -91,7 +91,7 @@ func.func @test_call_t5(%arg0 : !fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i6 return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>> - //CHECK: fir.call @test_t5(%[[ARG]]) : (!fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>>) -> () + //CHECK: fir.call @test_t5(%[[ARG]]) : (!fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>>) -> !fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t5<{c:!fir.char<1>,r:f32,i:i64}>>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr diff --git a/flang/test/Fir/struct-return-ppc64le.fir b/flang/test/Fir/struct-return-ppc64le.fir index a5906280a0ef8..a19f8885ee0c5 100644 --- a/flang/test/Fir/struct-return-ppc64le.fir +++ b/flang/test/Fir/struct-return-ppc64le.fir @@ -15,7 +15,7 @@ func.func @test_call_t1(%arg0 : !fir.ref<!fir.type<t1{c:!fir.char<1>}>>) { return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t1{c:!fir.char<1>}> - //CHECK: fir.call @test_t1(%[[ARG]]) : (!fir.ref<!fir.type<t1{c:!fir.char<1>}>>) -> () + //CHECK: fir.call @test_t1(%[[ARG]]) : (!fir.ref<!fir.type<t1{c:!fir.char<1>}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t1{c:!fir.char<1>}>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t1{c:!fir.char<1>}>>) -> !fir.ref<!fir.type<t1{c:!fir.char<1>}>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t1{c:!fir.char<1>}>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -34,7 +34,7 @@ func.func @test_call_t2(%arg0 : !fir.ref<!fir.type<t2{i:i32}>>) { return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t2{i:i32}> - //CHECK: fir.call @test_t2(%[[ARG]]) : (!fir.ref<!fir.type<t2{i:i32}>>) -> () + //CHECK: fir.call @test_t2(%[[ARG]]) : (!fir.ref<!fir.type<t2{i:i32}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t2{i:i32}>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t2{i:i32}>>) -> !fir.ref<!fir.type<t2{i:i32}>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t2{i:i32}>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -73,7 +73,7 @@ func.func @test_call_t4(%arg0 : !fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>>) { return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t4{r:!fir.array<9xf32>}> - //CHECK: fir.call @test_t4(%[[ARG]]) : (!fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>>) -> () + //CHECK: fir.call @test_t4(%[[ARG]]) : (!fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t4{r:!fir.array<9xf32>}>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>>) -> !fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t4{r:!fir.array<9xf32>}>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr @@ -92,7 +92,7 @@ func.func @test_call_t5(%arg0 : !fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64 return //CHECK: %[[STCK:.*]] = llvm.intr.stacksave : !llvm.ptr //CHECK: %[[ARG:.*]] = fir.alloca !fir.type<t5{c:!fir.char<1>,r:f32,i:i64}> - //CHECK: fir.call @test_t5(%[[ARG]]) : (!fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>>) -> () + //CHECK: fir.call @test_t5(%[[ARG]]) : (!fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>}) -> () //CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (!fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>>) -> !fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>> //CHECK: %[[LD:.*]] = fir.load %[[CVT]] : !fir.ref<!fir.type<t5{c:!fir.char<1>,r:f32,i:i64}>> //CHECK: llvm.intr.stackrestore %[[STCK]] : !llvm.ptr diff --git a/flang/test/Fir/struct-return-x86-64.fir b/flang/test/Fir/struct-return-x86-64.fir index f4c2add69ff7e..5d1e6129d8f69 100644 --- a/flang/test/Fir/struct-return-x86-64.fir +++ b/flang/test/Fir/struct-return-x86-64.fir @@ -1,5 +1,4 @@ // Test X86-64 ABI rewrite of struct returned by value (BIND(C), VALUE derived types). -// REQUIRES: x86-registered-target // RUN: fir-opt --target-rewrite %s | FileCheck %s !fits_in_reg = !fir.type<t1{i:f32,j:i32,k:f32}> @@ -86,7 +85,7 @@ module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.data // CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>>) { // CHECK: %[[VAL_1:.*]] = llvm.intr.stacksave : !llvm.ptr // CHECK: %[[VAL_2:.*]] = fir.alloca !fir.type<t2{i:!fir.array<5xf32>}> -// CHECK: fir.call @test_sret(%[[VAL_2]]) : (!fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>>) -> () +// CHECK: fir.call @test_sret(%[[VAL_2]]) : (!fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>> {llvm.align = 8 : i32, llvm.sret = !fir.type<t2{i:!fir.array<5xf32>}>}) -> () // CHECK: %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>>) -> !fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>> // CHECK: %[[VAL_4:.*]] = fir.load %[[VAL_3]] : !fir.ref<!fir.type<t2{i:!fir.array<5xf32>}>> // CHECK: llvm.intr.stackrestore %[[VAL_1]] : !llvm.ptr diff --git a/flang/test/Fir/target-rewrite-complex-10-x86.fir b/flang/test/Fir/target-rewrite-complex-10-x86.fir index a6dd178cfe70a..6404b4f766d39 100644 --- a/flang/test/Fir/target-rewrite-complex-10-x86.fir +++ b/flang/test/Fir/target-rewrite-complex-10-x86.fir @@ -1,8 +1,9 @@ // Test COMPLEX... [truncated] |
tblah approved these changes Mar 11, 2025
Contributor
tblah left a comment
tblah left a comment There was a problem hiding this comment.
Nice catch. LGTM and tests pass fine on aarch64
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
So far, flang was not setting argument attributes on direct calls assuming that putting them on the function operation was enough.
It was clarified in 38565da that they must be set on both call and functions, even for direct calls.
Crashes have been observed because of the lack of the attribute when compiling
abs(x)atO2and above on X86-64 for complex(16).