[MLIR] Fix import of calls with mismatched variadic types #124286
Merged
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
|
@llvm/pr-subscribers-mlir Author: Henrich Lauko (xlauko) ChangesPreviously, an indirect call was incorrectly generated when Full diff: https://github.com/llvm/llvm-project/pull/124286.diff 2 Files Affected:
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index f6826a2362bfdf..b8f66357d9250e 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -1495,15 +1495,22 @@ LogicalResult ModuleImport::convertCallTypeAndOperands(
if (!callInst->getType()->isVoidTy())
types.push_back(convertType(callInst->getType()));
- if (!callInst->getCalledFunction()) {
- if (!allowInlineAsm ||
- !isa<llvm::InlineAsm>(callInst->getCalledOperand())) {
- FailureOr<Value> called = convertValue(callInst->getCalledOperand());
- if (failed(called))
- return failure();
- operands.push_back(*called);
- }
+ bool isInlineAsm = callInst->isInlineAsm();
+ if (isInlineAsm && !allowInlineAsm)
+ return failure();
+
+ // Cannot use isIndirectCall() here because we need to handle Constant callees
+ // that are not considered indirect calls by LLVM. However, in MLIR, they are
+ // treated as indirect calls to constant operands that need to be converted.
+ // Skip the callee operand if it's inline assembly, as it's handled separately
+ // in InlineAsmOp.
+ if (!isa<llvm::Function>(callInst->getCalledOperand()) && !isInlineAsm) {
+ FailureOr<Value> called = convertValue(callInst->getCalledOperand());
+ if (failed(called))
+ return failure();
+ operands.push_back(*called);
}
+
SmallVector<llvm::Value *> args(callInst->args());
FailureOr<SmallVector<Value>> arguments = convertValues(args);
if (failed(arguments))
@@ -1593,7 +1600,8 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
return success();
}
if (inst->getOpcode() == llvm::Instruction::Call) {
- auto *callInst = cast<llvm::CallInst>(inst);
+ auto callInst = cast<llvm::CallInst>(inst);
+ auto calledOperand = callInst->getCalledOperand();
SmallVector<Type> types;
SmallVector<Value> operands;
@@ -1601,14 +1609,12 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
/*allowInlineAsm=*/true)))
return failure();
- auto funcTy =
- dyn_cast<LLVMFunctionType>(convertType(callInst->getFunctionType()));
- if (!funcTy)
- return failure();
-
- if (auto asmI = dyn_cast<llvm::InlineAsm>(callInst->getCalledOperand())) {
+ if (auto asmI = dyn_cast<llvm::InlineAsm>(calledOperand)) {
+ auto resultTy = convertType(callInst->getType());
+ if (!resultTy)
+ return failure();
auto callOp = builder.create<InlineAsmOp>(
- loc, funcTy.getReturnType(), operands,
+ loc, resultTy, operands,
builder.getStringAttr(asmI->getAsmString()),
builder.getStringAttr(asmI->getConstraintString()),
/*has_side_effects=*/true,
@@ -1619,27 +1625,35 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
else
mapNoResultOp(inst, callOp);
} else {
- CallOp callOp;
+ auto funcTy = dyn_cast<LLVMFunctionType>([&] () -> Type {
+ // Retrieve the real function type. For direct calls, use the callee's
+ // function type, as it may differ from the operand type in the case of
+ // variadic functions. For indirect calls, use the call function type.
+ if (auto callee = dyn_cast<llvm::Function>(calledOperand))
+ return convertType(callee->getFunctionType());
+ return convertType(callInst->getFunctionType());
+ }() );
+
+ if (!funcTy)
+ return failure();
- if (llvm::Function *callee = callInst->getCalledFunction()) {
- callOp = builder.create<CallOp>(
- loc, funcTy, SymbolRefAttr::get(context, callee->getName()),
- operands);
- } else {
- callOp = builder.create<CallOp>(loc, funcTy, operands);
- }
+ auto callOp = [&]() -> CallOp {
+ if (auto callee = dyn_cast<llvm::Function>(calledOperand)) {
+ auto name = SymbolRefAttr::get(context, callee->getName());
+ return builder.create<CallOp>(loc, funcTy, name, operands);
+ }
+ return builder.create<CallOp>(loc, funcTy, operands);
+ }();
+
+ // Handle function attributes.
callOp.setCConv(convertCConvFromLLVM(callInst->getCallingConv()));
callOp.setTailCallKind(
convertTailCallKindFromLLVM(callInst->getTailCallKind()));
setFastmathFlagsAttr(inst, callOp);
- // Handle function attributes.
- if (callInst->hasFnAttr(llvm::Attribute::Convergent))
- callOp.setConvergent(true);
- if (callInst->hasFnAttr(llvm::Attribute::NoUnwind))
- callOp.setNoUnwind(true);
- if (callInst->hasFnAttr(llvm::Attribute::WillReturn))
- callOp.setWillReturn(true);
+ callOp.setConvergent(callInst->isConvergent());
+ callOp.setNoUnwind(callInst->doesNotThrow());
+ callOp.setWillReturn(callInst->hasFnAttr(llvm::Attribute::WillReturn));
llvm::MemoryEffects memEffects = callInst->getMemoryEffects();
ModRefInfo othermem = convertModRefInfoFromLLVM(
diff --git a/mlir/test/Target/LLVMIR/Import/instructions.ll b/mlir/test/Target/LLVMIR/Import/instructions.ll
index 7377e2584110b5..47f821c8d29909 100644
--- a/mlir/test/Target/LLVMIR/Import/instructions.ll
+++ b/mlir/test/Target/LLVMIR/Import/instructions.ll
@@ -570,6 +570,41 @@ define void @varargs_call(i32 %0) {
; // -----
+; CHECK: @varargs(...)
+declare void @varargs(...)
+
+; CHECK-LABEL: @varargs_call
+; CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
+define void @varargs_call(i32 %0) {
+ ; CHECK: llvm.call @varargs(%[[ARG1]]) vararg(!llvm.func<void (...)>) : (i32) -> ()
+ call void @varargs(i32 %0)
+ ret void
+}
+
+; // -----
+
+; CHECK: @varargs(...)
+declare void @varargs(...)
+
+; CHECK-LABEL: @empty_varargs_call
+define void @empty_varargs_call() {
+ ; CHECK: llvm.call @varargs() vararg(!llvm.func<void (...)>) : () -> ()
+ call void @varargs()
+ ret void
+}
+
+; // -----
+
+; CHECK-LABEL: @undef_call
+define void @undef_call() {
+ ; CHECK: %[[UNDEF:[0-9]+]] = llvm.mlir.undef : !llvm.ptr
+ ; CHECK-NEXT: %[[CONST:[0-9]+]] = llvm.mlir.constant(0 : i32) : i32
+ ; CHECK-NEXT: llvm.call %[[UNDEF]](%[[CONST]]) : !llvm.ptr, (i32) -> ()
+ call void undef(i32 0)
+ ret void
+}
+; // -----
+
; CHECK: llvm.func @f()
declare void @f()
|
|
@llvm/pr-subscribers-mlir-llvm Author: Henrich Lauko (xlauko) ChangesPreviously, an indirect call was incorrectly generated when Full diff: https://github.com/llvm/llvm-project/pull/124286.diff 2 Files Affected:
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index f6826a2362bfdf..b8f66357d9250e 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -1495,15 +1495,22 @@ LogicalResult ModuleImport::convertCallTypeAndOperands(
if (!callInst->getType()->isVoidTy())
types.push_back(convertType(callInst->getType()));
- if (!callInst->getCalledFunction()) {
- if (!allowInlineAsm ||
- !isa<llvm::InlineAsm>(callInst->getCalledOperand())) {
- FailureOr<Value> called = convertValue(callInst->getCalledOperand());
- if (failed(called))
- return failure();
- operands.push_back(*called);
- }
+ bool isInlineAsm = callInst->isInlineAsm();
+ if (isInlineAsm && !allowInlineAsm)
+ return failure();
+
+ // Cannot use isIndirectCall() here because we need to handle Constant callees
+ // that are not considered indirect calls by LLVM. However, in MLIR, they are
+ // treated as indirect calls to constant operands that need to be converted.
+ // Skip the callee operand if it's inline assembly, as it's handled separately
+ // in InlineAsmOp.
+ if (!isa<llvm::Function>(callInst->getCalledOperand()) && !isInlineAsm) {
+ FailureOr<Value> called = convertValue(callInst->getCalledOperand());
+ if (failed(called))
+ return failure();
+ operands.push_back(*called);
}
+
SmallVector<llvm::Value *> args(callInst->args());
FailureOr<SmallVector<Value>> arguments = convertValues(args);
if (failed(arguments))
@@ -1593,7 +1600,8 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
return success();
}
if (inst->getOpcode() == llvm::Instruction::Call) {
- auto *callInst = cast<llvm::CallInst>(inst);
+ auto callInst = cast<llvm::CallInst>(inst);
+ auto calledOperand = callInst->getCalledOperand();
SmallVector<Type> types;
SmallVector<Value> operands;
@@ -1601,14 +1609,12 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
/*allowInlineAsm=*/true)))
return failure();
- auto funcTy =
- dyn_cast<LLVMFunctionType>(convertType(callInst->getFunctionType()));
- if (!funcTy)
- return failure();
-
- if (auto asmI = dyn_cast<llvm::InlineAsm>(callInst->getCalledOperand())) {
+ if (auto asmI = dyn_cast<llvm::InlineAsm>(calledOperand)) {
+ auto resultTy = convertType(callInst->getType());
+ if (!resultTy)
+ return failure();
auto callOp = builder.create<InlineAsmOp>(
- loc, funcTy.getReturnType(), operands,
+ loc, resultTy, operands,
builder.getStringAttr(asmI->getAsmString()),
builder.getStringAttr(asmI->getConstraintString()),
/*has_side_effects=*/true,
@@ -1619,27 +1625,35 @@ LogicalResult ModuleImport::convertInstruction(llvm::Instruction *inst) {
else
mapNoResultOp(inst, callOp);
} else {
- CallOp callOp;
+ auto funcTy = dyn_cast<LLVMFunctionType>([&] () -> Type {
+ // Retrieve the real function type. For direct calls, use the callee's
+ // function type, as it may differ from the operand type in the case of
+ // variadic functions. For indirect calls, use the call function type.
+ if (auto callee = dyn_cast<llvm::Function>(calledOperand))
+ return convertType(callee->getFunctionType());
+ return convertType(callInst->getFunctionType());
+ }() );
+
+ if (!funcTy)
+ return failure();
- if (llvm::Function *callee = callInst->getCalledFunction()) {
- callOp = builder.create<CallOp>(
- loc, funcTy, SymbolRefAttr::get(context, callee->getName()),
- operands);
- } else {
- callOp = builder.create<CallOp>(loc, funcTy, operands);
- }
+ auto callOp = [&]() -> CallOp {
+ if (auto callee = dyn_cast<llvm::Function>(calledOperand)) {
+ auto name = SymbolRefAttr::get(context, callee->getName());
+ return builder.create<CallOp>(loc, funcTy, name, operands);
+ }
+ return builder.create<CallOp>(loc, funcTy, operands);
+ }();
+
+ // Handle function attributes.
callOp.setCConv(convertCConvFromLLVM(callInst->getCallingConv()));
callOp.setTailCallKind(
convertTailCallKindFromLLVM(callInst->getTailCallKind()));
setFastmathFlagsAttr(inst, callOp);
- // Handle function attributes.
- if (callInst->hasFnAttr(llvm::Attribute::Convergent))
- callOp.setConvergent(true);
- if (callInst->hasFnAttr(llvm::Attribute::NoUnwind))
- callOp.setNoUnwind(true);
- if (callInst->hasFnAttr(llvm::Attribute::WillReturn))
- callOp.setWillReturn(true);
+ callOp.setConvergent(callInst->isConvergent());
+ callOp.setNoUnwind(callInst->doesNotThrow());
+ callOp.setWillReturn(callInst->hasFnAttr(llvm::Attribute::WillReturn));
llvm::MemoryEffects memEffects = callInst->getMemoryEffects();
ModRefInfo othermem = convertModRefInfoFromLLVM(
diff --git a/mlir/test/Target/LLVMIR/Import/instructions.ll b/mlir/test/Target/LLVMIR/Import/instructions.ll
index 7377e2584110b5..47f821c8d29909 100644
--- a/mlir/test/Target/LLVMIR/Import/instructions.ll
+++ b/mlir/test/Target/LLVMIR/Import/instructions.ll
@@ -570,6 +570,41 @@ define void @varargs_call(i32 %0) {
; // -----
+; CHECK: @varargs(...)
+declare void @varargs(...)
+
+; CHECK-LABEL: @varargs_call
+; CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
+define void @varargs_call(i32 %0) {
+ ; CHECK: llvm.call @varargs(%[[ARG1]]) vararg(!llvm.func<void (...)>) : (i32) -> ()
+ call void @varargs(i32 %0)
+ ret void
+}
+
+; // -----
+
+; CHECK: @varargs(...)
+declare void @varargs(...)
+
+; CHECK-LABEL: @empty_varargs_call
+define void @empty_varargs_call() {
+ ; CHECK: llvm.call @varargs() vararg(!llvm.func<void (...)>) : () -> ()
+ call void @varargs()
+ ret void
+}
+
+; // -----
+
+; CHECK-LABEL: @undef_call
+define void @undef_call() {
+ ; CHECK: %[[UNDEF:[0-9]+]] = llvm.mlir.undef : !llvm.ptr
+ ; CHECK-NEXT: %[[CONST:[0-9]+]] = llvm.mlir.constant(0 : i32) : i32
+ ; CHECK-NEXT: llvm.call %[[UNDEF]](%[[CONST]]) : !llvm.ptr, (i32) -> ()
+ call void undef(i32 0)
+ ret void
+}
+; // -----
+
; CHECK: llvm.func @f()
declare void @f()
|
|
✅ With the latest revision this PR passed the undef deprecator. |
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
|
Previously, the translation for the following LLVM IR: define i32 @bar() {
entry:
%call = call i32 @foo()
ret i32 %call
}
declare i32 @foo(...)was generating an indirect call like this in MLIR: llvm.func @bar() -> i32 {
%0 = llvm.mlir.addressof @foo : !llvm.ptr
%1 = llvm.call %0() : !llvm.ptr, () -> i32
llvm.return %1 : i32
}
llvm.func @foo(...) -> i32However, this translation was incorrect. The fixed translation now generates the correct MLIR: llvm.func @bar() -> i32 {
%0 = llvm.call @foo() vararg(!llvm.func<i32 (...)>) : () -> i32
llvm.return %0 : i32
}
llvm.func @foo(...) -> i32 |
xlauko
force-pushed
the
main
branch
3 times, most recently
from
adc300e to
3fce659
Compare
Dinistro
approved these changes
Jan 24, 2025
gysit
approved these changes
Jan 24, 2025
Previously, an indirect call was incorrectly generated when `llvm::CallBase::getCalledFunction` returned null due to a type mismatch between the call and the function. This patch updates the code to use `llvm::CallBase::getCalledOperand` instead.
xlauko
added a commit
to xlauko/llvm-project
that referenced
this pull request
Jan 28, 2025
This resolves the same issue addressed in llvm#124286 for call operations and refactors the common conversion code for both call and invoke instructions.
xlauko
added a commit
to xlauko/llvm-project
that referenced
this pull request
Jan 29, 2025
This resolves the same issue addressed in llvm#124286 for call operations and refactors the common conversion code for both call and invoke instructions.
xlauko
added a commit
to xlauko/llvm-project
that referenced
this pull request
Jan 29, 2025
This resolves the same issue addressed in llvm#124286, but for invoke operations. The issue arose from duplicated logic for both imports. This PR also refactors the common import code for call and invoke instructions to mitigate issues in the future.
gysit
pushed a commit
that referenced
this pull request
Jan 29, 2025
This resolves the same issue addressed in #124286, but for invoke operations. The issue arose from duplicated logic for both imports. This PR also refactors the common import code for call and invoke instructions to mitigate issues in the future.
github-actions bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Jan 29, 2025
…s (#124828) This resolves the same issue addressed in llvm/llvm-project#124286, but for invoke operations. The issue arose from duplicated logic for both imports. This PR also refactors the common import code for call and invoke instructions to mitigate issues in the future.
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.
Previously, an indirect call was incorrectly generated when
llvm::CallBase::getCalledFunctionreturned null due to a type mismatch between the call and the function. This patch updates the code to usellvm::CallBase::getCalledOperandinstead.