[llvm] [IR] Allow fast math flags on calls with homogeneous FP struct types (PR #110506)
Benjamin Maxwell via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 30 06:10:18 PDT 2024
https://github.com/MacDue created https://github.com/llvm/llvm-project/pull/110506
This extends FPMathOperator to allow calls that return literal structs of homogeneous floating-point or vector-of-floating-point types.
The intended use case for this is to support FP intrinsics that return multiple values (such as `llvm.sincos`).
>From 328357f2300ebe55b8385c01f9c655f703933736 Mon Sep 17 00:00:00 2001
From: Benjamin Maxwell <benjamin.maxwell at arm.com>
Date: Mon, 30 Sep 2024 11:07:45 +0000
Subject: [PATCH] [IR] Allow fast math flags on calls with homogeneous FP
struct types
This extends FPMathOperator to allow calls that return literal structs
of homogeneous floating-point or vector-of-floating-point types.
The intended use case for this is to support FP intrinsics that return
multiple values (such as `llvm.sincos`).
---
llvm/docs/LangRef.rst | 19 ++++++------
llvm/include/llvm/IR/DerivedTypes.h | 4 +++
llvm/include/llvm/IR/Operator.h | 14 +++++++--
llvm/lib/IR/Type.cpp | 13 +++++----
llvm/test/Bitcode/compatibility.ll | 20 +++++++++++++
llvm/unittests/IR/InstructionsTest.cpp | 40 ++++++++++++++++++++++----
6 files changed, 87 insertions(+), 23 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 3f39d58b322a4f..1eb2982385fda0 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -12472,9 +12472,8 @@ instruction's return value on the same edge).
The optional ``fast-math-flags`` marker indicates that the phi has one
or more :ref:`fast-math-flags <fastmath>`. These are optimization hints
to enable otherwise unsafe floating-point optimizations. Fast-math-flags
-are only valid for phis that return a floating-point scalar or vector
-type, or an array (nested to any depth) of floating-point scalar or vector
-types.
+are only valid for phis that return a floating-point scalar or vector type,
+possibly within an array (nested to any depth), or a homogeneous struct literal.
Semantics:
""""""""""
@@ -12523,8 +12522,8 @@ class <t_firstclass>` type.
#. The optional ``fast-math flags`` marker indicates that the select has one or more
:ref:`fast-math flags <fastmath>`. These are optimization hints to enable
otherwise unsafe floating-point optimizations. Fast-math flags are only valid
- for selects that return a floating-point scalar or vector type, or an array
- (nested to any depth) of floating-point scalar or vector types.
+ for selects that return a floating-point scalar or vector type, possibly
+ within an array (nested to any depth), or a homogeneous struct literal.
Semantics:
""""""""""
@@ -12762,8 +12761,8 @@ This instruction requires several arguments:
#. The optional ``fast-math flags`` marker indicates that the call has one or more
:ref:`fast-math flags <fastmath>`, which are optimization hints to enable
otherwise unsafe floating-point optimizations. Fast-math flags are only valid
- for calls that return a floating-point scalar or vector type, or an array
- (nested to any depth) of floating-point scalar or vector types.
+ for calls that return a floating-point scalar or vector type, possibly within
+ an array (nested to any depth), or a homogeneous struct literal.
#. The optional "cconv" marker indicates which :ref:`calling
convention <callingconv>` the call should use. If none is
@@ -20528,7 +20527,8 @@ the explicit vector length.
more :ref:`fast-math flags <fastmath>`. These are optimization hints to
enable otherwise unsafe floating-point optimizations. Fast-math flags are
only valid for selects that return a floating-point scalar or vector type,
- or an array (nested to any depth) of floating-point scalar or vector types.
+ possibly within an array (nested to any depth), or a homogeneous struct
+ literal.
Semantics:
""""""""""
@@ -20586,7 +20586,8 @@ is the pivot.
more :ref:`fast-math flags <fastmath>`. These are optimization hints to
enable otherwise unsafe floating-point optimizations. Fast-math flags are
only valid for merges that return a floating-point scalar or vector type,
- or an array (nested to any depth) of floating-point scalar or vector types.
+ possibly within an array (nested to any depth), or a homogeneous struct
+ literal.
Semantics:
""""""""""
diff --git a/llvm/include/llvm/IR/DerivedTypes.h b/llvm/include/llvm/IR/DerivedTypes.h
index 975c142f1a4572..a24801d8bdf834 100644
--- a/llvm/include/llvm/IR/DerivedTypes.h
+++ b/llvm/include/llvm/IR/DerivedTypes.h
@@ -301,6 +301,10 @@ class StructType : public Type {
/// {<vscale x 2 x i32>, <vscale x 4 x i64>}}
bool containsHomogeneousScalableVectorTypes() const;
+ /// Return true if this struct is non-empty and all element types are the
+ /// same.
+ bool containsHomogeneousTypes() const;
+
/// Return true if this is a named struct that has a non-empty name.
bool hasName() const { return SymbolTableEntry != nullptr; }
diff --git a/llvm/include/llvm/IR/Operator.h b/llvm/include/llvm/IR/Operator.h
index 88b9bfc0be4b15..22ffcc730e7b68 100644
--- a/llvm/include/llvm/IR/Operator.h
+++ b/llvm/include/llvm/IR/Operator.h
@@ -15,6 +15,7 @@
#define LLVM_IR_OPERATOR_H
#include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/TypeSwitch.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/FMF.h"
#include "llvm/IR/GEPNoWrapFlags.h"
@@ -351,8 +352,17 @@ class FPMathOperator : public Operator {
case Instruction::Select:
case Instruction::Call: {
Type *Ty = V->getType();
- while (ArrayType *ArrTy = dyn_cast<ArrayType>(Ty))
- Ty = ArrTy->getElementType();
+ TypeSwitch<Type *>(Ty)
+ .Case([&](StructType *StructTy) {
+ if (!StructTy->isLiteral() || !StructTy->containsHomogeneousTypes())
+ return;
+ Ty = StructTy->elements().front();
+ })
+ .Case([&](ArrayType *ArrTy) {
+ do {
+ Ty = ArrTy->getElementType();
+ } while ((ArrTy = dyn_cast<ArrayType>(Ty)));
+ });
return Ty->isFPOrFPVectorTy();
}
default:
diff --git a/llvm/lib/IR/Type.cpp b/llvm/lib/IR/Type.cpp
index 3784ad28d7219d..f618263f79c313 100644
--- a/llvm/lib/IR/Type.cpp
+++ b/llvm/lib/IR/Type.cpp
@@ -430,13 +430,14 @@ bool StructType::containsScalableVectorType(
}
bool StructType::containsHomogeneousScalableVectorTypes() const {
- Type *FirstTy = getNumElements() > 0 ? elements()[0] : nullptr;
- if (!FirstTy || !isa<ScalableVectorType>(FirstTy))
+ if (getNumElements() <= 0 || !isa<ScalableVectorType>(elements().front()))
return false;
- for (Type *Ty : elements())
- if (Ty != FirstTy)
- return false;
- return true;
+ return containsHomogeneousTypes();
+}
+
+bool StructType::containsHomogeneousTypes() const {
+ ArrayRef<Type *> ElementTys = elements();
+ return !ElementTys.empty() && all_equal(ElementTys);
}
void StructType::setBody(ArrayRef<Type*> Elements, bool isPacked) {
diff --git a/llvm/test/Bitcode/compatibility.ll b/llvm/test/Bitcode/compatibility.ll
index ea29ff634a43bb..4fe9d9b11f8831 100644
--- a/llvm/test/Bitcode/compatibility.ll
+++ b/llvm/test/Bitcode/compatibility.ll
@@ -1122,6 +1122,26 @@ define void @fastMathFlagsForArrayCalls([2 x float] %f, [2 x double] %d1, [2 x <
ret void
}
+declare { float, float } @fmf_struct_f32()
+declare { double, double } @fmf_struct_f64()
+declare { <4 x double>, <4 x double> } @fmf_struct_v4f64()
+
+; CHECK-LABEL: fastMathFlagsForStructCalls(
+define void @fastMathFlagsForStructCalls({ float, float } %f, { double, double } %d1, { <4 x double>, <4 x double> } %d2) {
+ %call.fast = call fast { float, float } @fmf_struct_f32()
+ ; CHECK: %call.fast = call fast { float, float } @fmf_struct_f32()
+
+ ; Throw in some other attributes to make sure those stay in the right places.
+
+ %call.nsz.arcp = notail call nsz arcp { double, double } @fmf_struct_f64()
+ ; CHECK: %call.nsz.arcp = notail call nsz arcp { double, double } @fmf_struct_f64()
+
+ %call.nnan.ninf = tail call nnan ninf fastcc { <4 x double>, <4 x double> } @fmf_struct_v4f64()
+ ; CHECK: %call.nnan.ninf = tail call nnan ninf fastcc { <4 x double>, <4 x double> } @fmf_struct_v4f64()
+
+ ret void
+}
+
;; Type System
%opaquety = type opaque
define void @typesystem() {
diff --git a/llvm/unittests/IR/InstructionsTest.cpp b/llvm/unittests/IR/InstructionsTest.cpp
index 481fe96607e48e..9d8056a768af2f 100644
--- a/llvm/unittests/IR/InstructionsTest.cpp
+++ b/llvm/unittests/IR/InstructionsTest.cpp
@@ -1559,12 +1559,40 @@ TEST(InstructionsTest, FPCallIsFPMathOperator) {
CallInst::Create(AVFFnTy, AVFCallee, {}, ""));
EXPECT_TRUE(isa<FPMathOperator>(AVFCall));
- Type *AAVFTy = ArrayType::get(AVFTy, 2);
- FunctionType *AAVFFnTy = FunctionType::get(AAVFTy, {});
- Value *AAVFCallee = Constant::getNullValue(PtrTy);
- std::unique_ptr<CallInst> AAVFCall(
- CallInst::Create(AAVFFnTy, AAVFCallee, {}, ""));
- EXPECT_TRUE(isa<FPMathOperator>(AAVFCall));
+ Type *StructITy = StructType::get(ITy, ITy);
+ FunctionType *StructIFnTy = FunctionType::get(StructITy, {});
+ Value *StructICallee = Constant::getNullValue(PtrTy);
+ std::unique_ptr<CallInst> StructICall(
+ CallInst::Create(StructIFnTy, StructICallee, {}, ""));
+ EXPECT_FALSE(isa<FPMathOperator>(StructICall));
+
+ Type *NamedStructFTy = StructType::create({FTy, FTy}, "AStruct");
+ FunctionType *NamedStructFFnTy = FunctionType::get(NamedStructFTy, {});
+ Value *NamedStructFCallee = Constant::getNullValue(PtrTy);
+ std::unique_ptr<CallInst> NamedStructFCall(
+ CallInst::Create(NamedStructFFnTy, NamedStructFCallee, {}, ""));
+ EXPECT_FALSE(isa<FPMathOperator>(NamedStructFCall));
+
+ Type *MixedStructTy = StructType::get(FTy, ITy);
+ FunctionType *MixedStructFnTy = FunctionType::get(MixedStructTy, {});
+ Value *MixedStructCallee = Constant::getNullValue(PtrTy);
+ std::unique_ptr<CallInst> MixedStructCall(
+ CallInst::Create(MixedStructFnTy, MixedStructCallee, {}, ""));
+ EXPECT_FALSE(isa<FPMathOperator>(MixedStructCall));
+
+ Type *StructFTy = StructType::get(FTy, FTy);
+ FunctionType *StructFFnTy = FunctionType::get(StructFTy, {});
+ Value *StructFCallee = Constant::getNullValue(PtrTy);
+ std::unique_ptr<CallInst> StructFCall(
+ CallInst::Create(StructFFnTy, StructFCallee, {}, ""));
+ EXPECT_TRUE(isa<FPMathOperator>(StructFCall));
+
+ Type *StructVFTy = StructType::get(VFTy, VFTy);
+ FunctionType *StructVFFnTy = FunctionType::get(StructVFTy, {});
+ Value *StructVFCallee = Constant::getNullValue(PtrTy);
+ std::unique_ptr<CallInst> StructVFCall(
+ CallInst::Create(StructVFFnTy, StructVFCallee, {}, ""));
+ EXPECT_TRUE(isa<FPMathOperator>(StructVFCall));
}
TEST(InstructionsTest, FNegInstruction) {
More information about the llvm-commits
mailing list