[flang] Implement COMPLEX(10) passing and return ABI for X86-64 linux #74094
Conversation
This file contains 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
COMPLEX(10) passing by value and returning follows C complex passing/returning ABI. Cover the COMPLEX(10) case (X87 / __Complex long double on X86-64). Implements System V ABI for AMD64 version 1.0. The LLVM signatures match the one generated by clang for the __Complex long double case. Note that a FIXME is added for the COMPLEX(8) case that is incorrect in a corner case. This will be fixed when dealing with passing derived type by value in BIND(C) context.
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: None (jeanPerier) ChangesCOMPLEX(10) passing by value and returning follows C complex passing/returning ABI. Cover the COMPLEX(10) case (X87 / __Complex long double on X86-64). Implements System V ABI for AMD64 version 1.0. The LLVM signatures match the one generated by clang for the __Complex long double case. Note that a FIXME is added for the COMPLEX(8) case that is incorrect in a corner case. This will be fixed when dealing with passing derived type by value in BIND(C) context. Full diff: https://github.com/llvm/llvm-project/pull/74094.diff 2 Files Affected:
diff --git a/flang/lib/Optimizer/CodeGen/Target.cpp b/flang/lib/Optimizer/CodeGen/Target.cpp
index b136c8f7a9dd98a..112f56e268c3b90 100644
--- a/flang/lib/Optimizer/CodeGen/Target.cpp
+++ b/flang/lib/Optimizer/CodeGen/Target.cpp
@@ -261,9 +261,20 @@ struct TargetX86_64 : public GenericTarget<TargetX86_64> {
// <2 x t> vector of 2 eleTy
marshal.emplace_back(fir::VectorType::get(2, eleTy), AT{});
} else if (sem == &llvm::APFloat::IEEEdouble()) {
+ // FIXME: In case of SSE register exhaustion, the ABI here may be
+ // incorrect since LLVM may pass the real via register and the imaginary
+ // part via the stack while the ABI it should be all in register or all
+ // in memory. Register occupancy must be analyzed here.
// two distinct double arguments
marshal.emplace_back(eleTy, AT{});
marshal.emplace_back(eleTy, AT{});
+ } else if (sem == &llvm::APFloat::x87DoubleExtended()) {
+ // Use a type that will be translated into LLVM as:
+ // { x86_fp80, x86_fp80 } struct of 2 fp128, byval, align 16
+ marshal.emplace_back(
+ fir::ReferenceType::get(mlir::TupleType::get(
+ eleTy.getContext(), mlir::TypeRange{eleTy, eleTy})),
+ AT{/*align=*/16, /*byval=*/true});
} else if (sem == &llvm::APFloat::IEEEquad()) {
// Use a type that will be translated into LLVM as:
// { fp128, fp128 } struct of 2 fp128, byval, align 16
@@ -290,6 +301,11 @@ struct TargetX86_64 : public GenericTarget<TargetX86_64> {
marshal.emplace_back(mlir::TupleType::get(eleTy.getContext(),
mlir::TypeRange{eleTy, eleTy}),
AT{});
+ } else if (sem == &llvm::APFloat::x87DoubleExtended()) {
+ // { x86_fp80, x86_fp80 }
+ marshal.emplace_back(mlir::TupleType::get(eleTy.getContext(),
+ mlir::TypeRange{eleTy, eleTy}),
+ AT{});
} else if (sem == &llvm::APFloat::IEEEquad()) {
// Use a type that will be translated into LLVM as:
// { fp128, fp128 } struct of 2 fp128, sret, align 16
diff --git a/flang/test/Fir/target-rewrite-complex-10-x86.fir b/flang/test/Fir/target-rewrite-complex-10-x86.fir
new file mode 100644
index 000000000000000..8d7cc9d6ef83dae
--- /dev/null
+++ b/flang/test/Fir/target-rewrite-complex-10-x86.fir
@@ -0,0 +1,32 @@
+// Test COMPLEX(10) passing and returning on X86
+// REQUIRES: x86-registered-target
+// RUN: fir-opt --target-rewrite="target=x86_64-unknown-linux-gnu" %s | FileCheck %s --check-prefix=AMD64
+// RUN: tco -target="x86_64-unknown-linux-gnu" %s | FileCheck %s --check-prefix=AMD64_LLVM
+
+func.func @returncomplex10() -> !fir.complex<10> {
+ %1 = fir.zero_bits !fir.complex<10>
+ return %1 : !fir.complex<10>
+}
+// AMD64-LABEL: func.func @returncomplex10() -> tuple<!fir.real<10>, !fir.real<10>> {
+// AMD64: %[[VAL_0:.*]] = fir.zero_bits !fir.complex<10>
+// AMD64: %[[VAL_1:.*]] = fir.alloca tuple<!fir.real<10>, !fir.real<10>>
+// AMD64: %[[VAL_2:.*]] = fir.convert %[[VAL_1]] : (!fir.ref<tuple<!fir.real<10>, !fir.real<10>>>) -> !fir.ref<!fir.complex<10>>
+// AMD64: fir.store %[[VAL_0]] to %[[VAL_2]] : !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_3:.*]] = fir.load %[[VAL_1]] : !fir.ref<tuple<!fir.real<10>, !fir.real<10>>>
+// AMD64: return %[[VAL_3]] : tuple<!fir.real<10>, !fir.real<10>>
+
+// AMD64_LLVM: define { x86_fp80, x86_fp80 } @returncomplex10()
+
+func.func @takecomplex10(%z: !fir.complex<10>) {
+ %0 = fir.alloca !fir.complex<10>
+ fir.store %z to %0 : !fir.ref<!fir.complex<10>>
+ return
+}
+// AMD64-LABEL: func.func @takecomplex10(
+// AMD64-SAME: %[[VAL_0:.*]]: !fir.ref<tuple<!fir.real<10>, !fir.real<10>>> {llvm.align = 16 : i32, llvm.byval = tuple<!fir.real<10>, !fir.real<10>>}) {
+// AMD64: %[[VAL_1:.*]] = fir.convert %[[VAL_0]] : (!fir.ref<tuple<!fir.real<10>, !fir.real<10>>>) -> !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_2:.*]] = fir.load %[[VAL_1]] : !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_3:.*]] = fir.alloca !fir.complex<10>
+// AMD64: fir.store %[[VAL_2]] to %[[VAL_3]] : !fir.ref<!fir.complex<10>>
+
+// AMD64_LLVM: define void @takecomplex10(ptr byval({ x86_fp80, x86_fp80 }) align 16 %0)
|
|
@llvm/pr-subscribers-flang-codegen Author: None (jeanPerier) ChangesCOMPLEX(10) passing by value and returning follows C complex passing/returning ABI. Cover the COMPLEX(10) case (X87 / __Complex long double on X86-64). Implements System V ABI for AMD64 version 1.0. The LLVM signatures match the one generated by clang for the __Complex long double case. Note that a FIXME is added for the COMPLEX(8) case that is incorrect in a corner case. This will be fixed when dealing with passing derived type by value in BIND(C) context. Full diff: https://github.com/llvm/llvm-project/pull/74094.diff 2 Files Affected:
diff --git a/flang/lib/Optimizer/CodeGen/Target.cpp b/flang/lib/Optimizer/CodeGen/Target.cpp
index b136c8f7a9dd98a..112f56e268c3b90 100644
--- a/flang/lib/Optimizer/CodeGen/Target.cpp
+++ b/flang/lib/Optimizer/CodeGen/Target.cpp
@@ -261,9 +261,20 @@ struct TargetX86_64 : public GenericTarget<TargetX86_64> {
// <2 x t> vector of 2 eleTy
marshal.emplace_back(fir::VectorType::get(2, eleTy), AT{});
} else if (sem == &llvm::APFloat::IEEEdouble()) {
+ // FIXME: In case of SSE register exhaustion, the ABI here may be
+ // incorrect since LLVM may pass the real via register and the imaginary
+ // part via the stack while the ABI it should be all in register or all
+ // in memory. Register occupancy must be analyzed here.
// two distinct double arguments
marshal.emplace_back(eleTy, AT{});
marshal.emplace_back(eleTy, AT{});
+ } else if (sem == &llvm::APFloat::x87DoubleExtended()) {
+ // Use a type that will be translated into LLVM as:
+ // { x86_fp80, x86_fp80 } struct of 2 fp128, byval, align 16
+ marshal.emplace_back(
+ fir::ReferenceType::get(mlir::TupleType::get(
+ eleTy.getContext(), mlir::TypeRange{eleTy, eleTy})),
+ AT{/*align=*/16, /*byval=*/true});
} else if (sem == &llvm::APFloat::IEEEquad()) {
// Use a type that will be translated into LLVM as:
// { fp128, fp128 } struct of 2 fp128, byval, align 16
@@ -290,6 +301,11 @@ struct TargetX86_64 : public GenericTarget<TargetX86_64> {
marshal.emplace_back(mlir::TupleType::get(eleTy.getContext(),
mlir::TypeRange{eleTy, eleTy}),
AT{});
+ } else if (sem == &llvm::APFloat::x87DoubleExtended()) {
+ // { x86_fp80, x86_fp80 }
+ marshal.emplace_back(mlir::TupleType::get(eleTy.getContext(),
+ mlir::TypeRange{eleTy, eleTy}),
+ AT{});
} else if (sem == &llvm::APFloat::IEEEquad()) {
// Use a type that will be translated into LLVM as:
// { fp128, fp128 } struct of 2 fp128, sret, align 16
diff --git a/flang/test/Fir/target-rewrite-complex-10-x86.fir b/flang/test/Fir/target-rewrite-complex-10-x86.fir
new file mode 100644
index 000000000000000..8d7cc9d6ef83dae
--- /dev/null
+++ b/flang/test/Fir/target-rewrite-complex-10-x86.fir
@@ -0,0 +1,32 @@
+// Test COMPLEX(10) passing and returning on X86
+// REQUIRES: x86-registered-target
+// RUN: fir-opt --target-rewrite="target=x86_64-unknown-linux-gnu" %s | FileCheck %s --check-prefix=AMD64
+// RUN: tco -target="x86_64-unknown-linux-gnu" %s | FileCheck %s --check-prefix=AMD64_LLVM
+
+func.func @returncomplex10() -> !fir.complex<10> {
+ %1 = fir.zero_bits !fir.complex<10>
+ return %1 : !fir.complex<10>
+}
+// AMD64-LABEL: func.func @returncomplex10() -> tuple<!fir.real<10>, !fir.real<10>> {
+// AMD64: %[[VAL_0:.*]] = fir.zero_bits !fir.complex<10>
+// AMD64: %[[VAL_1:.*]] = fir.alloca tuple<!fir.real<10>, !fir.real<10>>
+// AMD64: %[[VAL_2:.*]] = fir.convert %[[VAL_1]] : (!fir.ref<tuple<!fir.real<10>, !fir.real<10>>>) -> !fir.ref<!fir.complex<10>>
+// AMD64: fir.store %[[VAL_0]] to %[[VAL_2]] : !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_3:.*]] = fir.load %[[VAL_1]] : !fir.ref<tuple<!fir.real<10>, !fir.real<10>>>
+// AMD64: return %[[VAL_3]] : tuple<!fir.real<10>, !fir.real<10>>
+
+// AMD64_LLVM: define { x86_fp80, x86_fp80 } @returncomplex10()
+
+func.func @takecomplex10(%z: !fir.complex<10>) {
+ %0 = fir.alloca !fir.complex<10>
+ fir.store %z to %0 : !fir.ref<!fir.complex<10>>
+ return
+}
+// AMD64-LABEL: func.func @takecomplex10(
+// AMD64-SAME: %[[VAL_0:.*]]: !fir.ref<tuple<!fir.real<10>, !fir.real<10>>> {llvm.align = 16 : i32, llvm.byval = tuple<!fir.real<10>, !fir.real<10>>}) {
+// AMD64: %[[VAL_1:.*]] = fir.convert %[[VAL_0]] : (!fir.ref<tuple<!fir.real<10>, !fir.real<10>>>) -> !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_2:.*]] = fir.load %[[VAL_1]] : !fir.ref<!fir.complex<10>>
+// AMD64: %[[VAL_3:.*]] = fir.alloca !fir.complex<10>
+// AMD64: fir.store %[[VAL_2]] to %[[VAL_3]] : !fir.ref<!fir.complex<10>>
+
+// AMD64_LLVM: define void @takecomplex10(ptr byval({ x86_fp80, x86_fp80 }) align 16 %0)
|
vzakhari
approved these changes
Dec 1, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
COMPLEX(10) passing by value and returning follows C complex passing/returning ABI.
Cover the COMPLEX(10) case (X87 / __Complex long double on X86-64).
Implements System V ABI for AMD64 version 1.0.
The LLVM signatures match the one generated by clang for the __Complex long double case.
Note that a FIXME is added for the COMPLEX(8) case that is incorrect in a corner case. This will be fixed when dealing with passing derived type by value in BIND(C) context.