[flang] Add support for -mprefer-vector-width=<value> #141380
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
llvmbot
added
clang
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-clang Author: Cameron McInally (mcinally) ChangesThis patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute. Full diff: https://github.com/llvm/llvm-project/pull/141380.diff 12 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 22261621df092..b0b642796010b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5480,7 +5480,7 @@ def mrecip_EQ : CommaJoined<["-"], "mrecip=">, Group<m_Group>,
"<value> = ( ['!'] ['vec-'] ('rcp'|'sqrt') [('h'|'s'|'d')] [':'<n>] ) | 'all' | 'default' | 'none'">,
MarshallingInfoStringVector<CodeGenOpts<"Reciprocals">>;
def mprefer_vector_width_EQ : Joined<["-"], "mprefer-vector-width=">, Group<m_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Specifies preferred vector width for auto-vectorization. Defaults to 'none' which allows target specific decisions.">,
MarshallingInfoString<CodeGenOpts<"PreferVectorWidth">>;
def mstack_protector_guard_EQ : Joined<["-"], "mstack-protector-guard=">, Group<m_Group>,
diff --git a/flang/include/flang/Frontend/CodeGenOptions.h b/flang/include/flang/Frontend/CodeGenOptions.h
index 2b4e823b3fef4..61e56e51c4bbb 100644
--- a/flang/include/flang/Frontend/CodeGenOptions.h
+++ b/flang/include/flang/Frontend/CodeGenOptions.h
@@ -53,6 +53,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
/// The paths to the pass plugins that were registered using -fpass-plugin.
std::vector<std::string> LLVMPassPlugins;
+ // The prefered vector width, if requested by -mprefer-vector-width.
+ std::string PreferVectorWidth;
+
/// List of filenames passed in using the -fembed-offload-object option. These
/// are offloading binaries containing device images and metadata.
std::vector<std::string> OffloadObjects;
diff --git a/flang/include/flang/Optimizer/Transforms/Passes.td b/flang/include/flang/Optimizer/Transforms/Passes.td
index b251534e1a8f6..2e932d9ad4a26 100644
--- a/flang/include/flang/Optimizer/Transforms/Passes.td
+++ b/flang/include/flang/Optimizer/Transforms/Passes.td
@@ -418,6 +418,10 @@ def FunctionAttr : Pass<"function-attr", "mlir::func::FuncOp"> {
"module.">,
Option<"unsafeFPMath", "unsafe-fp-math", "bool", /*default=*/"false",
"Set the unsafe-fp-math attribute on functions in the module.">,
+ Option<"preferVectorWidth", "prefer-vector-width", "std::string",
+ /*default=*/"",
+ "Set the prefer-vector-width attribute on functions in the "
+ "module.">,
Option<"tuneCPU", "tune-cpu", "std::string", /*default=*/"",
"Set the tune-cpu attribute on functions in the module.">,
];
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 118695bbe2626..058024a4a04c5 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -102,6 +102,7 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
UnsafeFPMath = mathOpts.getAssociativeMath() &&
mathOpts.getReciprocalMath() && NoSignedZerosFPMath &&
ApproxFuncFPMath && mathOpts.getFPContractEnabled();
+ PreferVectorWidth = opts.PreferVectorWidth;
if (opts.InstrumentFunctions) {
InstrumentFunctionEntry = "__cyg_profile_func_enter";
InstrumentFunctionExit = "__cyg_profile_func_exit";
@@ -126,6 +127,8 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
bool NoSignedZerosFPMath =
false; ///< Set no-signed-zeros-fp-math attribute for functions.
bool UnsafeFPMath = false; ///< Set unsafe-fp-math attribute for functions.
+ std::string PreferVectorWidth = ""; ///< Set prefer-vector-width attribute for
+ ///< functions.
bool NSWOnLoopVarInc = true; ///< Add nsw flag to loop variable increments.
bool EnableOpenMP = false; ///< Enable OpenMP lowering.
std::string InstrumentFunctionEntry =
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2531819ee5e..122eb56fa0599 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -309,6 +309,20 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
for (auto *a : args.filtered(clang::driver::options::OPT_fpass_plugin_EQ))
opts.LLVMPassPlugins.push_back(a->getValue());
+ // -mprefer_vector_width option
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_mprefer_vector_width_EQ)) {
+ llvm::StringRef s = a->getValue();
+ unsigned Width;
+ if (s == "none")
+ opts.PreferVectorWidth = "none";
+ else if (s.getAsInteger(10, Width))
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ else
+ opts.PreferVectorWidth = s.str();
+ }
+
// -fembed-offload-object option
for (auto *a :
args.filtered(clang::driver::options::OPT_fembed_offload_object_EQ))
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 38dfaadf1dff9..012d0fdfe645f 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -741,6 +741,8 @@ void CodeGenAction::generateLLVMIR() {
config.VScaleMax = vsr->second;
}
+ config.PreferVectorWidth = opts.PreferVectorWidth;
+
if (ci.getInvocation().getFrontendOpts().features.IsEnabled(
Fortran::common::LanguageFeature::OpenMP))
config.EnableOpenMP = true;
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index 77751908e35be..5c1e1b9f77efb 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -354,7 +354,7 @@ void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
{framePointerKind, config.InstrumentFunctionEntry,
config.InstrumentFunctionExit, config.NoInfsFPMath, config.NoNaNsFPMath,
config.ApproxFuncFPMath, config.NoSignedZerosFPMath, config.UnsafeFPMath,
- ""}));
+ config.PreferVectorWidth, ""}));
if (config.EnableOpenMP) {
pm.addNestedPass<mlir::func::FuncOp>(
diff --git a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
index 43e4c1a7af3cd..13f447cf738b4 100644
--- a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
+++ b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
@@ -36,6 +36,7 @@ class FunctionAttrPass : public fir::impl::FunctionAttrBase<FunctionAttrPass> {
approxFuncFPMath = options.approxFuncFPMath;
noSignedZerosFPMath = options.noSignedZerosFPMath;
unsafeFPMath = options.unsafeFPMath;
+ preferVectorWidth = options.preferVectorWidth;
}
FunctionAttrPass() {}
void runOnOperation() override;
@@ -102,6 +103,10 @@ void FunctionAttrPass::runOnOperation() {
func->setAttr(
mlir::LLVM::LLVMFuncOp::getUnsafeFpMathAttrName(llvmFuncOpName),
mlir::BoolAttr::get(context, true));
+ if (!preferVectorWidth.empty())
+ func->setAttr(
+ mlir::LLVM::LLVMFuncOp::getPreferVectorWidthAttrName(llvmFuncOpName),
+ mlir::StringAttr::get(context, preferVectorWidth));
LLVM_DEBUG(llvm::dbgs() << "=== End " DEBUG_TYPE " ===\n");
}
diff --git a/flang/test/Driver/prefer-vector-width.f90 b/flang/test/Driver/prefer-vector-width.f90
new file mode 100644
index 0000000000000..d0f5fd28db826
--- /dev/null
+++ b/flang/test/Driver/prefer-vector-width.f90
@@ -0,0 +1,16 @@
+! Test that -mprefer-vector-width works as expected.
+
+! RUN: %flang_fc1 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-DEF
+! RUN: %flang_fc1 -mprefer-vector-width=none -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-NONE
+! RUN: %flang_fc1 -mprefer-vector-width=128 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-128
+! RUN: %flang_fc1 -mprefer-vector-width=256 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-256
+! RUN: not %flang_fc1 -mprefer-vector-width=xxx -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-INVALID
+
+subroutine func
+end subroutine func
+
+! CHECK-DEF-NOT: attributes #0 = { "prefer-vector-width"={{.*}} }
+! CHECK-NONE: attributes #0 = { "prefer-vector-width"="none" }
+! CHECK-128: attributes #0 = { "prefer-vector-width"="128" }
+! CHECK-256: attributes #0 = { "prefer-vector-width"="256" }
+! CHECK-INVALID:error: invalid value 'xxx' in '-mprefer-vector-width=xxx'
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 6fde45ce5c556..c0324d561b77b 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1893,6 +1893,7 @@ def LLVM_LLVMFuncOp : LLVM_Op<"func", [
OptionalAttr<FramePointerKindAttr>:$frame_pointer,
OptionalAttr<StrAttr>:$target_cpu,
OptionalAttr<StrAttr>:$tune_cpu,
+ OptionalAttr<StrAttr>:$prefer_vector_width,
OptionalAttr<LLVM_TargetFeaturesAttr>:$target_features,
OptionalAttr<BoolAttr>:$unsafe_fp_math,
OptionalAttr<BoolAttr>:$no_infs_fp_math,
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index b049064fbd31c..85417da798b22 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -2636,6 +2636,10 @@ void ModuleImport::processFunctionAttributes(llvm::Function *func,
funcOp.setTargetFeaturesAttr(
LLVM::TargetFeaturesAttr::get(context, attr.getValueAsString()));
+ if (llvm::Attribute attr = func->getFnAttribute("prefer-vector-width");
+ attr.isStringAttribute())
+ funcOp.setPreferVectorWidth(attr.getValueAsString());
+
if (llvm::Attribute attr = func->getFnAttribute("unsafe-fp-math");
attr.isStringAttribute())
funcOp.setUnsafeFpMath(attr.getValueAsBool());
diff --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index 047e870b7dcd8..2b7f0b11613aa 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -1540,6 +1540,9 @@ LogicalResult ModuleTranslation::convertOneFunction(LLVMFuncOp func) {
if (auto tuneCpu = func.getTuneCpu())
llvmFunc->addFnAttr("tune-cpu", *tuneCpu);
+ if (auto preferVectorWidth = func.getPreferVectorWidth())
+ llvmFunc->addFnAttr("prefer-vector-width", *preferVectorWidth);
+
if (auto attr = func.getVscaleRange())
llvmFunc->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
getLLVMContext(), attr->getMinRange().getInt(),
|
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Cameron McInally (mcinally) ChangesThis patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute. Full diff: https://github.com/llvm/llvm-project/pull/141380.diff 12 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 22261621df092..b0b642796010b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5480,7 +5480,7 @@ def mrecip_EQ : CommaJoined<["-"], "mrecip=">, Group<m_Group>,
"<value> = ( ['!'] ['vec-'] ('rcp'|'sqrt') [('h'|'s'|'d')] [':'<n>] ) | 'all' | 'default' | 'none'">,
MarshallingInfoStringVector<CodeGenOpts<"Reciprocals">>;
def mprefer_vector_width_EQ : Joined<["-"], "mprefer-vector-width=">, Group<m_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Specifies preferred vector width for auto-vectorization. Defaults to 'none' which allows target specific decisions.">,
MarshallingInfoString<CodeGenOpts<"PreferVectorWidth">>;
def mstack_protector_guard_EQ : Joined<["-"], "mstack-protector-guard=">, Group<m_Group>,
diff --git a/flang/include/flang/Frontend/CodeGenOptions.h b/flang/include/flang/Frontend/CodeGenOptions.h
index 2b4e823b3fef4..61e56e51c4bbb 100644
--- a/flang/include/flang/Frontend/CodeGenOptions.h
+++ b/flang/include/flang/Frontend/CodeGenOptions.h
@@ -53,6 +53,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
/// The paths to the pass plugins that were registered using -fpass-plugin.
std::vector<std::string> LLVMPassPlugins;
+ // The prefered vector width, if requested by -mprefer-vector-width.
+ std::string PreferVectorWidth;
+
/// List of filenames passed in using the -fembed-offload-object option. These
/// are offloading binaries containing device images and metadata.
std::vector<std::string> OffloadObjects;
diff --git a/flang/include/flang/Optimizer/Transforms/Passes.td b/flang/include/flang/Optimizer/Transforms/Passes.td
index b251534e1a8f6..2e932d9ad4a26 100644
--- a/flang/include/flang/Optimizer/Transforms/Passes.td
+++ b/flang/include/flang/Optimizer/Transforms/Passes.td
@@ -418,6 +418,10 @@ def FunctionAttr : Pass<"function-attr", "mlir::func::FuncOp"> {
"module.">,
Option<"unsafeFPMath", "unsafe-fp-math", "bool", /*default=*/"false",
"Set the unsafe-fp-math attribute on functions in the module.">,
+ Option<"preferVectorWidth", "prefer-vector-width", "std::string",
+ /*default=*/"",
+ "Set the prefer-vector-width attribute on functions in the "
+ "module.">,
Option<"tuneCPU", "tune-cpu", "std::string", /*default=*/"",
"Set the tune-cpu attribute on functions in the module.">,
];
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 118695bbe2626..058024a4a04c5 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -102,6 +102,7 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
UnsafeFPMath = mathOpts.getAssociativeMath() &&
mathOpts.getReciprocalMath() && NoSignedZerosFPMath &&
ApproxFuncFPMath && mathOpts.getFPContractEnabled();
+ PreferVectorWidth = opts.PreferVectorWidth;
if (opts.InstrumentFunctions) {
InstrumentFunctionEntry = "__cyg_profile_func_enter";
InstrumentFunctionExit = "__cyg_profile_func_exit";
@@ -126,6 +127,8 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
bool NoSignedZerosFPMath =
false; ///< Set no-signed-zeros-fp-math attribute for functions.
bool UnsafeFPMath = false; ///< Set unsafe-fp-math attribute for functions.
+ std::string PreferVectorWidth = ""; ///< Set prefer-vector-width attribute for
+ ///< functions.
bool NSWOnLoopVarInc = true; ///< Add nsw flag to loop variable increments.
bool EnableOpenMP = false; ///< Enable OpenMP lowering.
std::string InstrumentFunctionEntry =
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2531819ee5e..122eb56fa0599 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -309,6 +309,20 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
for (auto *a : args.filtered(clang::driver::options::OPT_fpass_plugin_EQ))
opts.LLVMPassPlugins.push_back(a->getValue());
+ // -mprefer_vector_width option
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_mprefer_vector_width_EQ)) {
+ llvm::StringRef s = a->getValue();
+ unsigned Width;
+ if (s == "none")
+ opts.PreferVectorWidth = "none";
+ else if (s.getAsInteger(10, Width))
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ else
+ opts.PreferVectorWidth = s.str();
+ }
+
// -fembed-offload-object option
for (auto *a :
args.filtered(clang::driver::options::OPT_fembed_offload_object_EQ))
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 38dfaadf1dff9..012d0fdfe645f 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -741,6 +741,8 @@ void CodeGenAction::generateLLVMIR() {
config.VScaleMax = vsr->second;
}
+ config.PreferVectorWidth = opts.PreferVectorWidth;
+
if (ci.getInvocation().getFrontendOpts().features.IsEnabled(
Fortran::common::LanguageFeature::OpenMP))
config.EnableOpenMP = true;
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index 77751908e35be..5c1e1b9f77efb 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -354,7 +354,7 @@ void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
{framePointerKind, config.InstrumentFunctionEntry,
config.InstrumentFunctionExit, config.NoInfsFPMath, config.NoNaNsFPMath,
config.ApproxFuncFPMath, config.NoSignedZerosFPMath, config.UnsafeFPMath,
- ""}));
+ config.PreferVectorWidth, ""}));
if (config.EnableOpenMP) {
pm.addNestedPass<mlir::func::FuncOp>(
diff --git a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
index 43e4c1a7af3cd..13f447cf738b4 100644
--- a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
+++ b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
@@ -36,6 +36,7 @@ class FunctionAttrPass : public fir::impl::FunctionAttrBase<FunctionAttrPass> {
approxFuncFPMath = options.approxFuncFPMath;
noSignedZerosFPMath = options.noSignedZerosFPMath;
unsafeFPMath = options.unsafeFPMath;
+ preferVectorWidth = options.preferVectorWidth;
}
FunctionAttrPass() {}
void runOnOperation() override;
@@ -102,6 +103,10 @@ void FunctionAttrPass::runOnOperation() {
func->setAttr(
mlir::LLVM::LLVMFuncOp::getUnsafeFpMathAttrName(llvmFuncOpName),
mlir::BoolAttr::get(context, true));
+ if (!preferVectorWidth.empty())
+ func->setAttr(
+ mlir::LLVM::LLVMFuncOp::getPreferVectorWidthAttrName(llvmFuncOpName),
+ mlir::StringAttr::get(context, preferVectorWidth));
LLVM_DEBUG(llvm::dbgs() << "=== End " DEBUG_TYPE " ===\n");
}
diff --git a/flang/test/Driver/prefer-vector-width.f90 b/flang/test/Driver/prefer-vector-width.f90
new file mode 100644
index 0000000000000..d0f5fd28db826
--- /dev/null
+++ b/flang/test/Driver/prefer-vector-width.f90
@@ -0,0 +1,16 @@
+! Test that -mprefer-vector-width works as expected.
+
+! RUN: %flang_fc1 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-DEF
+! RUN: %flang_fc1 -mprefer-vector-width=none -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-NONE
+! RUN: %flang_fc1 -mprefer-vector-width=128 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-128
+! RUN: %flang_fc1 -mprefer-vector-width=256 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-256
+! RUN: not %flang_fc1 -mprefer-vector-width=xxx -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-INVALID
+
+subroutine func
+end subroutine func
+
+! CHECK-DEF-NOT: attributes #0 = { "prefer-vector-width"={{.*}} }
+! CHECK-NONE: attributes #0 = { "prefer-vector-width"="none" }
+! CHECK-128: attributes #0 = { "prefer-vector-width"="128" }
+! CHECK-256: attributes #0 = { "prefer-vector-width"="256" }
+! CHECK-INVALID:error: invalid value 'xxx' in '-mprefer-vector-width=xxx'
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 6fde45ce5c556..c0324d561b77b 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1893,6 +1893,7 @@ def LLVM_LLVMFuncOp : LLVM_Op<"func", [
OptionalAttr<FramePointerKindAttr>:$frame_pointer,
OptionalAttr<StrAttr>:$target_cpu,
OptionalAttr<StrAttr>:$tune_cpu,
+ OptionalAttr<StrAttr>:$prefer_vector_width,
OptionalAttr<LLVM_TargetFeaturesAttr>:$target_features,
OptionalAttr<BoolAttr>:$unsafe_fp_math,
OptionalAttr<BoolAttr>:$no_infs_fp_math,
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index b049064fbd31c..85417da798b22 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -2636,6 +2636,10 @@ void ModuleImport::processFunctionAttributes(llvm::Function *func,
funcOp.setTargetFeaturesAttr(
LLVM::TargetFeaturesAttr::get(context, attr.getValueAsString()));
+ if (llvm::Attribute attr = func->getFnAttribute("prefer-vector-width");
+ attr.isStringAttribute())
+ funcOp.setPreferVectorWidth(attr.getValueAsString());
+
if (llvm::Attribute attr = func->getFnAttribute("unsafe-fp-math");
attr.isStringAttribute())
funcOp.setUnsafeFpMath(attr.getValueAsBool());
diff --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index 047e870b7dcd8..2b7f0b11613aa 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -1540,6 +1540,9 @@ LogicalResult ModuleTranslation::convertOneFunction(LLVMFuncOp func) {
if (auto tuneCpu = func.getTuneCpu())
llvmFunc->addFnAttr("tune-cpu", *tuneCpu);
+ if (auto preferVectorWidth = func.getPreferVectorWidth())
+ llvmFunc->addFnAttr("prefer-vector-width", *preferVectorWidth);
+
if (auto attr = func.getVscaleRange())
llvmFunc->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
getLLVMContext(), attr->getMinRange().getInt(),
|
|
@llvm/pr-subscribers-mlir-llvm Author: Cameron McInally (mcinally) ChangesThis patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute. Full diff: https://github.com/llvm/llvm-project/pull/141380.diff 12 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 22261621df092..b0b642796010b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5480,7 +5480,7 @@ def mrecip_EQ : CommaJoined<["-"], "mrecip=">, Group<m_Group>,
"<value> = ( ['!'] ['vec-'] ('rcp'|'sqrt') [('h'|'s'|'d')] [':'<n>] ) | 'all' | 'default' | 'none'">,
MarshallingInfoStringVector<CodeGenOpts<"Reciprocals">>;
def mprefer_vector_width_EQ : Joined<["-"], "mprefer-vector-width=">, Group<m_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Specifies preferred vector width for auto-vectorization. Defaults to 'none' which allows target specific decisions.">,
MarshallingInfoString<CodeGenOpts<"PreferVectorWidth">>;
def mstack_protector_guard_EQ : Joined<["-"], "mstack-protector-guard=">, Group<m_Group>,
diff --git a/flang/include/flang/Frontend/CodeGenOptions.h b/flang/include/flang/Frontend/CodeGenOptions.h
index 2b4e823b3fef4..61e56e51c4bbb 100644
--- a/flang/include/flang/Frontend/CodeGenOptions.h
+++ b/flang/include/flang/Frontend/CodeGenOptions.h
@@ -53,6 +53,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
/// The paths to the pass plugins that were registered using -fpass-plugin.
std::vector<std::string> LLVMPassPlugins;
+ // The prefered vector width, if requested by -mprefer-vector-width.
+ std::string PreferVectorWidth;
+
/// List of filenames passed in using the -fembed-offload-object option. These
/// are offloading binaries containing device images and metadata.
std::vector<std::string> OffloadObjects;
diff --git a/flang/include/flang/Optimizer/Transforms/Passes.td b/flang/include/flang/Optimizer/Transforms/Passes.td
index b251534e1a8f6..2e932d9ad4a26 100644
--- a/flang/include/flang/Optimizer/Transforms/Passes.td
+++ b/flang/include/flang/Optimizer/Transforms/Passes.td
@@ -418,6 +418,10 @@ def FunctionAttr : Pass<"function-attr", "mlir::func::FuncOp"> {
"module.">,
Option<"unsafeFPMath", "unsafe-fp-math", "bool", /*default=*/"false",
"Set the unsafe-fp-math attribute on functions in the module.">,
+ Option<"preferVectorWidth", "prefer-vector-width", "std::string",
+ /*default=*/"",
+ "Set the prefer-vector-width attribute on functions in the "
+ "module.">,
Option<"tuneCPU", "tune-cpu", "std::string", /*default=*/"",
"Set the tune-cpu attribute on functions in the module.">,
];
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 118695bbe2626..058024a4a04c5 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -102,6 +102,7 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
UnsafeFPMath = mathOpts.getAssociativeMath() &&
mathOpts.getReciprocalMath() && NoSignedZerosFPMath &&
ApproxFuncFPMath && mathOpts.getFPContractEnabled();
+ PreferVectorWidth = opts.PreferVectorWidth;
if (opts.InstrumentFunctions) {
InstrumentFunctionEntry = "__cyg_profile_func_enter";
InstrumentFunctionExit = "__cyg_profile_func_exit";
@@ -126,6 +127,8 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
bool NoSignedZerosFPMath =
false; ///< Set no-signed-zeros-fp-math attribute for functions.
bool UnsafeFPMath = false; ///< Set unsafe-fp-math attribute for functions.
+ std::string PreferVectorWidth = ""; ///< Set prefer-vector-width attribute for
+ ///< functions.
bool NSWOnLoopVarInc = true; ///< Add nsw flag to loop variable increments.
bool EnableOpenMP = false; ///< Enable OpenMP lowering.
std::string InstrumentFunctionEntry =
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2531819ee5e..122eb56fa0599 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -309,6 +309,20 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
for (auto *a : args.filtered(clang::driver::options::OPT_fpass_plugin_EQ))
opts.LLVMPassPlugins.push_back(a->getValue());
+ // -mprefer_vector_width option
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_mprefer_vector_width_EQ)) {
+ llvm::StringRef s = a->getValue();
+ unsigned Width;
+ if (s == "none")
+ opts.PreferVectorWidth = "none";
+ else if (s.getAsInteger(10, Width))
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ else
+ opts.PreferVectorWidth = s.str();
+ }
+
// -fembed-offload-object option
for (auto *a :
args.filtered(clang::driver::options::OPT_fembed_offload_object_EQ))
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 38dfaadf1dff9..012d0fdfe645f 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -741,6 +741,8 @@ void CodeGenAction::generateLLVMIR() {
config.VScaleMax = vsr->second;
}
+ config.PreferVectorWidth = opts.PreferVectorWidth;
+
if (ci.getInvocation().getFrontendOpts().features.IsEnabled(
Fortran::common::LanguageFeature::OpenMP))
config.EnableOpenMP = true;
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index 77751908e35be..5c1e1b9f77efb 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -354,7 +354,7 @@ void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
{framePointerKind, config.InstrumentFunctionEntry,
config.InstrumentFunctionExit, config.NoInfsFPMath, config.NoNaNsFPMath,
config.ApproxFuncFPMath, config.NoSignedZerosFPMath, config.UnsafeFPMath,
- ""}));
+ config.PreferVectorWidth, ""}));
if (config.EnableOpenMP) {
pm.addNestedPass<mlir::func::FuncOp>(
diff --git a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
index 43e4c1a7af3cd..13f447cf738b4 100644
--- a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
+++ b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
@@ -36,6 +36,7 @@ class FunctionAttrPass : public fir::impl::FunctionAttrBase<FunctionAttrPass> {
approxFuncFPMath = options.approxFuncFPMath;
noSignedZerosFPMath = options.noSignedZerosFPMath;
unsafeFPMath = options.unsafeFPMath;
+ preferVectorWidth = options.preferVectorWidth;
}
FunctionAttrPass() {}
void runOnOperation() override;
@@ -102,6 +103,10 @@ void FunctionAttrPass::runOnOperation() {
func->setAttr(
mlir::LLVM::LLVMFuncOp::getUnsafeFpMathAttrName(llvmFuncOpName),
mlir::BoolAttr::get(context, true));
+ if (!preferVectorWidth.empty())
+ func->setAttr(
+ mlir::LLVM::LLVMFuncOp::getPreferVectorWidthAttrName(llvmFuncOpName),
+ mlir::StringAttr::get(context, preferVectorWidth));
LLVM_DEBUG(llvm::dbgs() << "=== End " DEBUG_TYPE " ===\n");
}
diff --git a/flang/test/Driver/prefer-vector-width.f90 b/flang/test/Driver/prefer-vector-width.f90
new file mode 100644
index 0000000000000..d0f5fd28db826
--- /dev/null
+++ b/flang/test/Driver/prefer-vector-width.f90
@@ -0,0 +1,16 @@
+! Test that -mprefer-vector-width works as expected.
+
+! RUN: %flang_fc1 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-DEF
+! RUN: %flang_fc1 -mprefer-vector-width=none -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-NONE
+! RUN: %flang_fc1 -mprefer-vector-width=128 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-128
+! RUN: %flang_fc1 -mprefer-vector-width=256 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-256
+! RUN: not %flang_fc1 -mprefer-vector-width=xxx -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-INVALID
+
+subroutine func
+end subroutine func
+
+! CHECK-DEF-NOT: attributes #0 = { "prefer-vector-width"={{.*}} }
+! CHECK-NONE: attributes #0 = { "prefer-vector-width"="none" }
+! CHECK-128: attributes #0 = { "prefer-vector-width"="128" }
+! CHECK-256: attributes #0 = { "prefer-vector-width"="256" }
+! CHECK-INVALID:error: invalid value 'xxx' in '-mprefer-vector-width=xxx'
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 6fde45ce5c556..c0324d561b77b 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1893,6 +1893,7 @@ def LLVM_LLVMFuncOp : LLVM_Op<"func", [
OptionalAttr<FramePointerKindAttr>:$frame_pointer,
OptionalAttr<StrAttr>:$target_cpu,
OptionalAttr<StrAttr>:$tune_cpu,
+ OptionalAttr<StrAttr>:$prefer_vector_width,
OptionalAttr<LLVM_TargetFeaturesAttr>:$target_features,
OptionalAttr<BoolAttr>:$unsafe_fp_math,
OptionalAttr<BoolAttr>:$no_infs_fp_math,
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index b049064fbd31c..85417da798b22 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -2636,6 +2636,10 @@ void ModuleImport::processFunctionAttributes(llvm::Function *func,
funcOp.setTargetFeaturesAttr(
LLVM::TargetFeaturesAttr::get(context, attr.getValueAsString()));
+ if (llvm::Attribute attr = func->getFnAttribute("prefer-vector-width");
+ attr.isStringAttribute())
+ funcOp.setPreferVectorWidth(attr.getValueAsString());
+
if (llvm::Attribute attr = func->getFnAttribute("unsafe-fp-math");
attr.isStringAttribute())
funcOp.setUnsafeFpMath(attr.getValueAsBool());
diff --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index 047e870b7dcd8..2b7f0b11613aa 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -1540,6 +1540,9 @@ LogicalResult ModuleTranslation::convertOneFunction(LLVMFuncOp func) {
if (auto tuneCpu = func.getTuneCpu())
llvmFunc->addFnAttr("tune-cpu", *tuneCpu);
+ if (auto preferVectorWidth = func.getPreferVectorWidth())
+ llvmFunc->addFnAttr("prefer-vector-width", *preferVectorWidth);
+
if (auto attr = func.getVscaleRange())
llvmFunc->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
getLLVMContext(), attr->getMinRange().getInt(),
|
|
@llvm/pr-subscribers-flang-driver Author: Cameron McInally (mcinally) ChangesThis patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute. Full diff: https://github.com/llvm/llvm-project/pull/141380.diff 12 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 22261621df092..b0b642796010b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5480,7 +5480,7 @@ def mrecip_EQ : CommaJoined<["-"], "mrecip=">, Group<m_Group>,
"<value> = ( ['!'] ['vec-'] ('rcp'|'sqrt') [('h'|'s'|'d')] [':'<n>] ) | 'all' | 'default' | 'none'">,
MarshallingInfoStringVector<CodeGenOpts<"Reciprocals">>;
def mprefer_vector_width_EQ : Joined<["-"], "mprefer-vector-width=">, Group<m_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Specifies preferred vector width for auto-vectorization. Defaults to 'none' which allows target specific decisions.">,
MarshallingInfoString<CodeGenOpts<"PreferVectorWidth">>;
def mstack_protector_guard_EQ : Joined<["-"], "mstack-protector-guard=">, Group<m_Group>,
diff --git a/flang/include/flang/Frontend/CodeGenOptions.h b/flang/include/flang/Frontend/CodeGenOptions.h
index 2b4e823b3fef4..61e56e51c4bbb 100644
--- a/flang/include/flang/Frontend/CodeGenOptions.h
+++ b/flang/include/flang/Frontend/CodeGenOptions.h
@@ -53,6 +53,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
/// The paths to the pass plugins that were registered using -fpass-plugin.
std::vector<std::string> LLVMPassPlugins;
+ // The prefered vector width, if requested by -mprefer-vector-width.
+ std::string PreferVectorWidth;
+
/// List of filenames passed in using the -fembed-offload-object option. These
/// are offloading binaries containing device images and metadata.
std::vector<std::string> OffloadObjects;
diff --git a/flang/include/flang/Optimizer/Transforms/Passes.td b/flang/include/flang/Optimizer/Transforms/Passes.td
index b251534e1a8f6..2e932d9ad4a26 100644
--- a/flang/include/flang/Optimizer/Transforms/Passes.td
+++ b/flang/include/flang/Optimizer/Transforms/Passes.td
@@ -418,6 +418,10 @@ def FunctionAttr : Pass<"function-attr", "mlir::func::FuncOp"> {
"module.">,
Option<"unsafeFPMath", "unsafe-fp-math", "bool", /*default=*/"false",
"Set the unsafe-fp-math attribute on functions in the module.">,
+ Option<"preferVectorWidth", "prefer-vector-width", "std::string",
+ /*default=*/"",
+ "Set the prefer-vector-width attribute on functions in the "
+ "module.">,
Option<"tuneCPU", "tune-cpu", "std::string", /*default=*/"",
"Set the tune-cpu attribute on functions in the module.">,
];
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 118695bbe2626..058024a4a04c5 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -102,6 +102,7 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
UnsafeFPMath = mathOpts.getAssociativeMath() &&
mathOpts.getReciprocalMath() && NoSignedZerosFPMath &&
ApproxFuncFPMath && mathOpts.getFPContractEnabled();
+ PreferVectorWidth = opts.PreferVectorWidth;
if (opts.InstrumentFunctions) {
InstrumentFunctionEntry = "__cyg_profile_func_enter";
InstrumentFunctionExit = "__cyg_profile_func_exit";
@@ -126,6 +127,8 @@ struct MLIRToLLVMPassPipelineConfig : public FlangEPCallBacks {
bool NoSignedZerosFPMath =
false; ///< Set no-signed-zeros-fp-math attribute for functions.
bool UnsafeFPMath = false; ///< Set unsafe-fp-math attribute for functions.
+ std::string PreferVectorWidth = ""; ///< Set prefer-vector-width attribute for
+ ///< functions.
bool NSWOnLoopVarInc = true; ///< Add nsw flag to loop variable increments.
bool EnableOpenMP = false; ///< Enable OpenMP lowering.
std::string InstrumentFunctionEntry =
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2531819ee5e..122eb56fa0599 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -309,6 +309,20 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
for (auto *a : args.filtered(clang::driver::options::OPT_fpass_plugin_EQ))
opts.LLVMPassPlugins.push_back(a->getValue());
+ // -mprefer_vector_width option
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_mprefer_vector_width_EQ)) {
+ llvm::StringRef s = a->getValue();
+ unsigned Width;
+ if (s == "none")
+ opts.PreferVectorWidth = "none";
+ else if (s.getAsInteger(10, Width))
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ else
+ opts.PreferVectorWidth = s.str();
+ }
+
// -fembed-offload-object option
for (auto *a :
args.filtered(clang::driver::options::OPT_fembed_offload_object_EQ))
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 38dfaadf1dff9..012d0fdfe645f 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -741,6 +741,8 @@ void CodeGenAction::generateLLVMIR() {
config.VScaleMax = vsr->second;
}
+ config.PreferVectorWidth = opts.PreferVectorWidth;
+
if (ci.getInvocation().getFrontendOpts().features.IsEnabled(
Fortran::common::LanguageFeature::OpenMP))
config.EnableOpenMP = true;
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index 77751908e35be..5c1e1b9f77efb 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -354,7 +354,7 @@ void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
{framePointerKind, config.InstrumentFunctionEntry,
config.InstrumentFunctionExit, config.NoInfsFPMath, config.NoNaNsFPMath,
config.ApproxFuncFPMath, config.NoSignedZerosFPMath, config.UnsafeFPMath,
- ""}));
+ config.PreferVectorWidth, ""}));
if (config.EnableOpenMP) {
pm.addNestedPass<mlir::func::FuncOp>(
diff --git a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
index 43e4c1a7af3cd..13f447cf738b4 100644
--- a/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
+++ b/flang/lib/Optimizer/Transforms/FunctionAttr.cpp
@@ -36,6 +36,7 @@ class FunctionAttrPass : public fir::impl::FunctionAttrBase<FunctionAttrPass> {
approxFuncFPMath = options.approxFuncFPMath;
noSignedZerosFPMath = options.noSignedZerosFPMath;
unsafeFPMath = options.unsafeFPMath;
+ preferVectorWidth = options.preferVectorWidth;
}
FunctionAttrPass() {}
void runOnOperation() override;
@@ -102,6 +103,10 @@ void FunctionAttrPass::runOnOperation() {
func->setAttr(
mlir::LLVM::LLVMFuncOp::getUnsafeFpMathAttrName(llvmFuncOpName),
mlir::BoolAttr::get(context, true));
+ if (!preferVectorWidth.empty())
+ func->setAttr(
+ mlir::LLVM::LLVMFuncOp::getPreferVectorWidthAttrName(llvmFuncOpName),
+ mlir::StringAttr::get(context, preferVectorWidth));
LLVM_DEBUG(llvm::dbgs() << "=== End " DEBUG_TYPE " ===\n");
}
diff --git a/flang/test/Driver/prefer-vector-width.f90 b/flang/test/Driver/prefer-vector-width.f90
new file mode 100644
index 0000000000000..d0f5fd28db826
--- /dev/null
+++ b/flang/test/Driver/prefer-vector-width.f90
@@ -0,0 +1,16 @@
+! Test that -mprefer-vector-width works as expected.
+
+! RUN: %flang_fc1 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-DEF
+! RUN: %flang_fc1 -mprefer-vector-width=none -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-NONE
+! RUN: %flang_fc1 -mprefer-vector-width=128 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-128
+! RUN: %flang_fc1 -mprefer-vector-width=256 -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-256
+! RUN: not %flang_fc1 -mprefer-vector-width=xxx -emit-llvm -o - %s 2>&1| FileCheck %s --check-prefix=CHECK-INVALID
+
+subroutine func
+end subroutine func
+
+! CHECK-DEF-NOT: attributes #0 = { "prefer-vector-width"={{.*}} }
+! CHECK-NONE: attributes #0 = { "prefer-vector-width"="none" }
+! CHECK-128: attributes #0 = { "prefer-vector-width"="128" }
+! CHECK-256: attributes #0 = { "prefer-vector-width"="256" }
+! CHECK-INVALID:error: invalid value 'xxx' in '-mprefer-vector-width=xxx'
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 6fde45ce5c556..c0324d561b77b 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1893,6 +1893,7 @@ def LLVM_LLVMFuncOp : LLVM_Op<"func", [
OptionalAttr<FramePointerKindAttr>:$frame_pointer,
OptionalAttr<StrAttr>:$target_cpu,
OptionalAttr<StrAttr>:$tune_cpu,
+ OptionalAttr<StrAttr>:$prefer_vector_width,
OptionalAttr<LLVM_TargetFeaturesAttr>:$target_features,
OptionalAttr<BoolAttr>:$unsafe_fp_math,
OptionalAttr<BoolAttr>:$no_infs_fp_math,
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index b049064fbd31c..85417da798b22 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -2636,6 +2636,10 @@ void ModuleImport::processFunctionAttributes(llvm::Function *func,
funcOp.setTargetFeaturesAttr(
LLVM::TargetFeaturesAttr::get(context, attr.getValueAsString()));
+ if (llvm::Attribute attr = func->getFnAttribute("prefer-vector-width");
+ attr.isStringAttribute())
+ funcOp.setPreferVectorWidth(attr.getValueAsString());
+
if (llvm::Attribute attr = func->getFnAttribute("unsafe-fp-math");
attr.isStringAttribute())
funcOp.setUnsafeFpMath(attr.getValueAsBool());
diff --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index 047e870b7dcd8..2b7f0b11613aa 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -1540,6 +1540,9 @@ LogicalResult ModuleTranslation::convertOneFunction(LLVMFuncOp func) {
if (auto tuneCpu = func.getTuneCpu())
llvmFunc->addFnAttr("tune-cpu", *tuneCpu);
+ if (auto preferVectorWidth = func.getPreferVectorWidth())
+ llvmFunc->addFnAttr("prefer-vector-width", *preferVectorWidth);
+
if (auto attr = func.getVscaleRange())
llvmFunc->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
getLLVMContext(), attr->getMinRange().getInt(),
|
|
This is my first patch to Flang, so please be suspicious and gentle. In particular, I am not sure if |
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
klausler
reviewed
May 24, 2025
This patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute.
jeanPerier
approved these changes
May 26, 2025
tarunprabhu
requested changes
May 26, 2025
There was a problem hiding this comment.
Thanks for this. I second @jeanPerier's request for an MLIR test
This patch adds support for the -mprefer-vector-width=<value> command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute.
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.
This patch adds support for the -mprefer-vector-width= command line option. The parsing of this options is equivalent to Clang's and it is implemented by setting the "prefer-vector-width" function attribute.