[mlir][gpu] Adding ELF section option to the gpu-module-to-binary pass #119440
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-llvm Author: Renaud Kauffmann (Renaud-K) ChangesThis is a follow-up of #117246. I thought then it would be easy to edit a DictionaryAttr but it turns out that these attributes are immutable and need to be passed during the construction of the gpu.binary Op. This requires the section to be part of the NVVMTargetAttr and to be an option to the GpuNVVMAttachTarget pass. Full diff: https://github.com/llvm/llvm-project/pull/119440.diff 6 Files Affected:
diff --git a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
index 4a9ddafdd177d2..784721d19d0ccd 100644
--- a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
+++ b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
@@ -139,6 +139,9 @@ def GpuNVVMAttachTarget: Pass<"nvvm-attach-target", ""> {
Option<"ftzFlag", "ftz", "bool",
/*default=*/"false",
"Enable flush to zero for denormals.">,
+ Option<"elfSection", "section", "std::string",
+ /*default=*/"\"\"",
+ "ELF section where the module needs to be created.">,
ListOption<"linkLibs", "l", "std::string",
"Extra bitcode libraries paths to link to.">,
];
diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index 296a3c305e5bf4..03c03d5f66b9af 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -2266,20 +2266,22 @@ def NVVM_TargettAttr : NVVM_Attr<"NVVMTarget", "target"> {
StringRefParameter<"Target triple.", "\"nvptx64-nvidia-cuda\"">:$triple,
StringRefParameter<"Target chip.", "\"sm_50\"">:$chip,
StringRefParameter<"Target chip features.", "\"+ptx60\"">:$features,
+ OptionalParameter<"StringAttr", "ELF section.">:$section,
OptionalParameter<"DictionaryAttr", "Target specific flags.">:$flags,
OptionalParameter<"ArrayAttr", "Files to link to the LLVM module.">:$link
);
let assemblyFormat = [{
- (`<` struct($O, $triple, $chip, $features, $flags, $link)^ `>`)?
+ (`<` struct($O, $triple, $chip, $features, $section, $flags, $link)^ `>`)?
}];
let builders = [
AttrBuilder<(ins CArg<"int", "2">:$optLevel,
CArg<"StringRef", "\"nvptx64-nvidia-cuda\"">:$triple,
CArg<"StringRef", "\"sm_50\"">:$chip,
CArg<"StringRef", "\"+ptx60\"">:$features,
+ CArg<"StringAttr", "nullptr">:$section,
CArg<"DictionaryAttr", "nullptr">:$targetFlags,
CArg<"ArrayAttr", "nullptr">:$linkFiles), [{
- return Base::get($_ctxt, optLevel, triple, chip, features, targetFlags, linkFiles);
+ return Base::get($_ctxt, optLevel, triple, chip, features, section, targetFlags, linkFiles);
}]>
];
let skipDefaultBuilders = 1;
diff --git a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
index dd705cd3383121..5628361c6a843c 100644
--- a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
@@ -63,7 +63,9 @@ void NVVMAttachTarget::runOnOperation() {
ArrayRef<std::string> libs(linkLibs);
SmallVector<StringRef> filesToLink(libs);
auto target = builder.getAttr<NVVMTargetAttr>(
- optLevel, triple, chip, features, getFlags(builder),
+ optLevel, triple, chip, features,
+ elfSection.empty() ? nullptr : builder.getStringAttr(elfSection),
+ getFlags(builder),
filesToLink.empty() ? nullptr : builder.getStrArrayAttr(filesToLink));
llvm::Regex matcher(moduleMatcher);
for (Region ®ion : getOperation()->getRegions())
diff --git a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
index ca04af0b060b4f..696cab52d41900 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
@@ -1185,7 +1185,7 @@ LogicalResult NVVMDialect::verifyRegionArgAttribute(Operation *op,
LogicalResult
NVVMTargetAttr::verify(function_ref<InFlightDiagnostic()> emitError,
int optLevel, StringRef triple, StringRef chip,
- StringRef features, DictionaryAttr flags,
+ StringRef features, StringAttr elfSection, DictionaryAttr flags,
ArrayAttr files) {
if (optLevel < 0 || optLevel > 3) {
emitError() << "The optimization level must be a number between 0 and 3.";
diff --git a/mlir/lib/Target/LLVM/NVVM/Target.cpp b/mlir/lib/Target/LLVM/NVVM/Target.cpp
index 3c92359915ded4..82ead374420d9e 100644
--- a/mlir/lib/Target/LLVM/NVVM/Target.cpp
+++ b/mlir/lib/Target/LLVM/NVVM/Target.cpp
@@ -664,9 +664,18 @@ NVVMTargetAttrImpl::createObject(Attribute attribute, Operation *module,
gpu::CompilationTarget format = options.getCompilationTarget();
DictionaryAttr objectProps;
Builder builder(attribute.getContext());
+ SmallVector<NamedAttribute, 2> properties;
if (format == gpu::CompilationTarget::Assembly)
- objectProps = builder.getDictionaryAttr(
- {builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO()))});
+ properties.push_back(
+ builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO())));
+
+ if (StringAttr section = target.getSection())
+ properties.push_back(
+ builder.getNamedAttr("section", section));
+
+ if (!properties.empty())
+ objectProps = builder.getDictionaryAttr(properties);
+
return builder.getAttr<gpu::ObjectAttr>(
attribute, format,
builder.getStringAttr(StringRef(object.data(), object.size())),
diff --git a/mlir/test/Dialect/LLVMIR/attach-targets.mlir b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
index 83733db400798e..832e0443e720f7 100644
--- a/mlir/test/Dialect/LLVMIR/attach-targets.mlir
+++ b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
@@ -1,13 +1,16 @@
// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* O=3 chip=sm_90' --rocdl-attach-target='module=rocdl.* O=3 chip=gfx90a' | FileCheck %s
// RUN: mlir-opt %s --nvvm-attach-target='module=options.* O=1 chip=sm_70 fast=true ftz=true' --rocdl-attach-target='module=options.* l=file1.bc,file2.bc wave64=false finite-only=true' | FileCheck %s --check-prefix=CHECK_OPTS
+// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* section=__fatbin' | FileCheck %s --check-prefix=CHECK_SECTION
module attributes {gpu.container_module} {
// Verify the target is appended.
// CHECK: @nvvm_module_1 [#nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: @nvvm_module_1 [#nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_1 {
}
// Verify the target is appended.
// CHECK: @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">] {
}
// Verify the target is not added multiple times.
|
|
@llvm/pr-subscribers-mlir Author: Renaud Kauffmann (Renaud-K) ChangesThis is a follow-up of #117246. I thought then it would be easy to edit a DictionaryAttr but it turns out that these attributes are immutable and need to be passed during the construction of the gpu.binary Op. This requires the section to be part of the NVVMTargetAttr and to be an option to the GpuNVVMAttachTarget pass. Full diff: https://github.com/llvm/llvm-project/pull/119440.diff 6 Files Affected:
diff --git a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
index 4a9ddafdd177d2..784721d19d0ccd 100644
--- a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
+++ b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
@@ -139,6 +139,9 @@ def GpuNVVMAttachTarget: Pass<"nvvm-attach-target", ""> {
Option<"ftzFlag", "ftz", "bool",
/*default=*/"false",
"Enable flush to zero for denormals.">,
+ Option<"elfSection", "section", "std::string",
+ /*default=*/"\"\"",
+ "ELF section where the module needs to be created.">,
ListOption<"linkLibs", "l", "std::string",
"Extra bitcode libraries paths to link to.">,
];
diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index 296a3c305e5bf4..03c03d5f66b9af 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -2266,20 +2266,22 @@ def NVVM_TargettAttr : NVVM_Attr<"NVVMTarget", "target"> {
StringRefParameter<"Target triple.", "\"nvptx64-nvidia-cuda\"">:$triple,
StringRefParameter<"Target chip.", "\"sm_50\"">:$chip,
StringRefParameter<"Target chip features.", "\"+ptx60\"">:$features,
+ OptionalParameter<"StringAttr", "ELF section.">:$section,
OptionalParameter<"DictionaryAttr", "Target specific flags.">:$flags,
OptionalParameter<"ArrayAttr", "Files to link to the LLVM module.">:$link
);
let assemblyFormat = [{
- (`<` struct($O, $triple, $chip, $features, $flags, $link)^ `>`)?
+ (`<` struct($O, $triple, $chip, $features, $section, $flags, $link)^ `>`)?
}];
let builders = [
AttrBuilder<(ins CArg<"int", "2">:$optLevel,
CArg<"StringRef", "\"nvptx64-nvidia-cuda\"">:$triple,
CArg<"StringRef", "\"sm_50\"">:$chip,
CArg<"StringRef", "\"+ptx60\"">:$features,
+ CArg<"StringAttr", "nullptr">:$section,
CArg<"DictionaryAttr", "nullptr">:$targetFlags,
CArg<"ArrayAttr", "nullptr">:$linkFiles), [{
- return Base::get($_ctxt, optLevel, triple, chip, features, targetFlags, linkFiles);
+ return Base::get($_ctxt, optLevel, triple, chip, features, section, targetFlags, linkFiles);
}]>
];
let skipDefaultBuilders = 1;
diff --git a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
index dd705cd3383121..5628361c6a843c 100644
--- a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
@@ -63,7 +63,9 @@ void NVVMAttachTarget::runOnOperation() {
ArrayRef<std::string> libs(linkLibs);
SmallVector<StringRef> filesToLink(libs);
auto target = builder.getAttr<NVVMTargetAttr>(
- optLevel, triple, chip, features, getFlags(builder),
+ optLevel, triple, chip, features,
+ elfSection.empty() ? nullptr : builder.getStringAttr(elfSection),
+ getFlags(builder),
filesToLink.empty() ? nullptr : builder.getStrArrayAttr(filesToLink));
llvm::Regex matcher(moduleMatcher);
for (Region ®ion : getOperation()->getRegions())
diff --git a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
index ca04af0b060b4f..696cab52d41900 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
@@ -1185,7 +1185,7 @@ LogicalResult NVVMDialect::verifyRegionArgAttribute(Operation *op,
LogicalResult
NVVMTargetAttr::verify(function_ref<InFlightDiagnostic()> emitError,
int optLevel, StringRef triple, StringRef chip,
- StringRef features, DictionaryAttr flags,
+ StringRef features, StringAttr elfSection, DictionaryAttr flags,
ArrayAttr files) {
if (optLevel < 0 || optLevel > 3) {
emitError() << "The optimization level must be a number between 0 and 3.";
diff --git a/mlir/lib/Target/LLVM/NVVM/Target.cpp b/mlir/lib/Target/LLVM/NVVM/Target.cpp
index 3c92359915ded4..82ead374420d9e 100644
--- a/mlir/lib/Target/LLVM/NVVM/Target.cpp
+++ b/mlir/lib/Target/LLVM/NVVM/Target.cpp
@@ -664,9 +664,18 @@ NVVMTargetAttrImpl::createObject(Attribute attribute, Operation *module,
gpu::CompilationTarget format = options.getCompilationTarget();
DictionaryAttr objectProps;
Builder builder(attribute.getContext());
+ SmallVector<NamedAttribute, 2> properties;
if (format == gpu::CompilationTarget::Assembly)
- objectProps = builder.getDictionaryAttr(
- {builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO()))});
+ properties.push_back(
+ builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO())));
+
+ if (StringAttr section = target.getSection())
+ properties.push_back(
+ builder.getNamedAttr("section", section));
+
+ if (!properties.empty())
+ objectProps = builder.getDictionaryAttr(properties);
+
return builder.getAttr<gpu::ObjectAttr>(
attribute, format,
builder.getStringAttr(StringRef(object.data(), object.size())),
diff --git a/mlir/test/Dialect/LLVMIR/attach-targets.mlir b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
index 83733db400798e..832e0443e720f7 100644
--- a/mlir/test/Dialect/LLVMIR/attach-targets.mlir
+++ b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
@@ -1,13 +1,16 @@
// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* O=3 chip=sm_90' --rocdl-attach-target='module=rocdl.* O=3 chip=gfx90a' | FileCheck %s
// RUN: mlir-opt %s --nvvm-attach-target='module=options.* O=1 chip=sm_70 fast=true ftz=true' --rocdl-attach-target='module=options.* l=file1.bc,file2.bc wave64=false finite-only=true' | FileCheck %s --check-prefix=CHECK_OPTS
+// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* section=__fatbin' | FileCheck %s --check-prefix=CHECK_SECTION
module attributes {gpu.container_module} {
// Verify the target is appended.
// CHECK: @nvvm_module_1 [#nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: @nvvm_module_1 [#nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_1 {
}
// Verify the target is appended.
// CHECK: @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">] {
}
// Verify the target is not added multiple times.
|
|
@llvm/pr-subscribers-mlir-gpu Author: Renaud Kauffmann (Renaud-K) ChangesThis is a follow-up of #117246. I thought then it would be easy to edit a DictionaryAttr but it turns out that these attributes are immutable and need to be passed during the construction of the gpu.binary Op. This requires the section to be part of the NVVMTargetAttr and to be an option to the GpuNVVMAttachTarget pass. Full diff: https://github.com/llvm/llvm-project/pull/119440.diff 6 Files Affected:
diff --git a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
index 4a9ddafdd177d2..784721d19d0ccd 100644
--- a/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
+++ b/mlir/include/mlir/Dialect/GPU/Transforms/Passes.td
@@ -139,6 +139,9 @@ def GpuNVVMAttachTarget: Pass<"nvvm-attach-target", ""> {
Option<"ftzFlag", "ftz", "bool",
/*default=*/"false",
"Enable flush to zero for denormals.">,
+ Option<"elfSection", "section", "std::string",
+ /*default=*/"\"\"",
+ "ELF section where the module needs to be created.">,
ListOption<"linkLibs", "l", "std::string",
"Extra bitcode libraries paths to link to.">,
];
diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index 296a3c305e5bf4..03c03d5f66b9af 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -2266,20 +2266,22 @@ def NVVM_TargettAttr : NVVM_Attr<"NVVMTarget", "target"> {
StringRefParameter<"Target triple.", "\"nvptx64-nvidia-cuda\"">:$triple,
StringRefParameter<"Target chip.", "\"sm_50\"">:$chip,
StringRefParameter<"Target chip features.", "\"+ptx60\"">:$features,
+ OptionalParameter<"StringAttr", "ELF section.">:$section,
OptionalParameter<"DictionaryAttr", "Target specific flags.">:$flags,
OptionalParameter<"ArrayAttr", "Files to link to the LLVM module.">:$link
);
let assemblyFormat = [{
- (`<` struct($O, $triple, $chip, $features, $flags, $link)^ `>`)?
+ (`<` struct($O, $triple, $chip, $features, $section, $flags, $link)^ `>`)?
}];
let builders = [
AttrBuilder<(ins CArg<"int", "2">:$optLevel,
CArg<"StringRef", "\"nvptx64-nvidia-cuda\"">:$triple,
CArg<"StringRef", "\"sm_50\"">:$chip,
CArg<"StringRef", "\"+ptx60\"">:$features,
+ CArg<"StringAttr", "nullptr">:$section,
CArg<"DictionaryAttr", "nullptr">:$targetFlags,
CArg<"ArrayAttr", "nullptr">:$linkFiles), [{
- return Base::get($_ctxt, optLevel, triple, chip, features, targetFlags, linkFiles);
+ return Base::get($_ctxt, optLevel, triple, chip, features, section, targetFlags, linkFiles);
}]>
];
let skipDefaultBuilders = 1;
diff --git a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
index dd705cd3383121..5628361c6a843c 100644
--- a/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/NVVMAttachTarget.cpp
@@ -63,7 +63,9 @@ void NVVMAttachTarget::runOnOperation() {
ArrayRef<std::string> libs(linkLibs);
SmallVector<StringRef> filesToLink(libs);
auto target = builder.getAttr<NVVMTargetAttr>(
- optLevel, triple, chip, features, getFlags(builder),
+ optLevel, triple, chip, features,
+ elfSection.empty() ? nullptr : builder.getStringAttr(elfSection),
+ getFlags(builder),
filesToLink.empty() ? nullptr : builder.getStrArrayAttr(filesToLink));
llvm::Regex matcher(moduleMatcher);
for (Region ®ion : getOperation()->getRegions())
diff --git a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
index ca04af0b060b4f..696cab52d41900 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
@@ -1185,7 +1185,7 @@ LogicalResult NVVMDialect::verifyRegionArgAttribute(Operation *op,
LogicalResult
NVVMTargetAttr::verify(function_ref<InFlightDiagnostic()> emitError,
int optLevel, StringRef triple, StringRef chip,
- StringRef features, DictionaryAttr flags,
+ StringRef features, StringAttr elfSection, DictionaryAttr flags,
ArrayAttr files) {
if (optLevel < 0 || optLevel > 3) {
emitError() << "The optimization level must be a number between 0 and 3.";
diff --git a/mlir/lib/Target/LLVM/NVVM/Target.cpp b/mlir/lib/Target/LLVM/NVVM/Target.cpp
index 3c92359915ded4..82ead374420d9e 100644
--- a/mlir/lib/Target/LLVM/NVVM/Target.cpp
+++ b/mlir/lib/Target/LLVM/NVVM/Target.cpp
@@ -664,9 +664,18 @@ NVVMTargetAttrImpl::createObject(Attribute attribute, Operation *module,
gpu::CompilationTarget format = options.getCompilationTarget();
DictionaryAttr objectProps;
Builder builder(attribute.getContext());
+ SmallVector<NamedAttribute, 2> properties;
if (format == gpu::CompilationTarget::Assembly)
- objectProps = builder.getDictionaryAttr(
- {builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO()))});
+ properties.push_back(
+ builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO())));
+
+ if (StringAttr section = target.getSection())
+ properties.push_back(
+ builder.getNamedAttr("section", section));
+
+ if (!properties.empty())
+ objectProps = builder.getDictionaryAttr(properties);
+
return builder.getAttr<gpu::ObjectAttr>(
attribute, format,
builder.getStringAttr(StringRef(object.data(), object.size())),
diff --git a/mlir/test/Dialect/LLVMIR/attach-targets.mlir b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
index 83733db400798e..832e0443e720f7 100644
--- a/mlir/test/Dialect/LLVMIR/attach-targets.mlir
+++ b/mlir/test/Dialect/LLVMIR/attach-targets.mlir
@@ -1,13 +1,16 @@
// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* O=3 chip=sm_90' --rocdl-attach-target='module=rocdl.* O=3 chip=gfx90a' | FileCheck %s
// RUN: mlir-opt %s --nvvm-attach-target='module=options.* O=1 chip=sm_70 fast=true ftz=true' --rocdl-attach-target='module=options.* l=file1.bc,file2.bc wave64=false finite-only=true' | FileCheck %s --check-prefix=CHECK_OPTS
+// RUN: mlir-opt %s --nvvm-attach-target='module=nvvm.* section=__fatbin' | FileCheck %s --check-prefix=CHECK_SECTION
module attributes {gpu.container_module} {
// Verify the target is appended.
// CHECK: @nvvm_module_1 [#nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: @nvvm_module_1 [#nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_1 {
}
// Verify the target is appended.
// CHECK: @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<O = 3, chip = "sm_90">] {
+// CHECK_SECTION: gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">, #nvvm.target<section = "__fatbin">]
gpu.module @nvvm_module_2 [#nvvm.target<chip = "sm_60">] {
}
// Verify the target is not added multiple times.
|
Renaud-K
changed the title
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Renaud-K
force-pushed
the
nvvm-attr-section
branch
from
d62fa8d to
13ba4b1
Compare
fabianmcg
requested changes
Dec 10, 2024
There was a problem hiding this comment.
Thank you for pushing for this, overall supportive of the idea. However, the section doesn't belong in a target attribute but rather the object attribute.
I thought then it would be easy to edit a DictionaryAttr but it turns out that these attributes are immutable and need to be passed during the construction of the gpu.binary Op.
It's possible to update the attribute, you just need to construct it again.
The other option is adding a pass option gpu-module-to-binary to create gpu.object with the correct section from the start.
Also, if this is for flang, I would suggest adding a OffloadingLLVMTranslationAttrInterface attribute instead of using SelectObjectAttr, see: https://github.com/llvm/llvm-project/blob/main/mlir/include/mlir/Dialect/GPU/IR/CompilationAttrInterfaces.td#L100-L142
|
The My goal is to leverage as much upstream code as possible.
Would this mean that I would have to replace the gpu.binary op with a new one? |
I like this approach. Thank you. I will look into it. |
The reason I mention a new attribute, is because a new attribute would be able to handle cuda binary registration for using CUDA RT.
No, you would only need to update the attributes.
It shouldn't |
Renaud-K
changed the title
Renaud-K
force-pushed
the
nvvm-attr-section
branch
from
5443ad9 to
be02039
Compare
This was referenced Dec 12, 2024
|
Can I proceed with these changes? |
fabianmcg
reviewed
Dec 12, 2024
| ]; | ||
| let options = | ||
| [Option<"toolkitPath", "toolkit", "std::string", [{""}], "Toolkit path.">, | ||
| ListOption<"linkFiles", "l", "std::string", "Extra files to link to.">, |
There was a problem hiding this comment.
Don't change the format, just add a new option.
There was a problem hiding this comment.
git-clang-format does that. Will the checks be happy if I undo it? I am afraid not since the diff is now on the radar. Shall I try? The version of clang, I am using, is the one from the build. Should I be using a different one?
There was a problem hiding this comment.
I'm surprised git clang-format actually touches .td files!
There was a problem hiding this comment.
I think it does since the latest version
There was a problem hiding this comment.
The checks are still happy with the old format but it is probably just a matter of time until it fails there too.
Renaud-K
force-pushed
the
nvvm-attr-section
branch
from
d0a59be to
08adf38
Compare
grypp
reviewed
Dec 13, 2024
mlir/lib/Target/LLVM/NVVM/Target.cpp
Outdated
| properties.push_back( | ||
| builder.getNamedAttr("O", builder.getI32IntegerAttr(target.getO()))); | ||
|
|
||
| if (auto section = options.getELFSection(); !section.empty()) |
There was a problem hiding this comment.
not: we need explicit type instead of auto when type isn't obvious
mlir/lib/Target/LLVM/NVVM/Target.cpp
Outdated
|
|
||
| if (auto section = options.getELFSection(); !section.empty()) | ||
| properties.push_back( | ||
| builder.getNamedAttr("section", builder.getStringAttr(section))); |
There was a problem hiding this comment.
maybe create a constant constexpr kSectionName = "section"
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 is a follow-up of #117246.
I thought then it would be easy to edit a DictionaryAttr but it turns out that these attributes are immutable and need to be passed during the construction of the gpu.binary Op.
The first commit was using the NVVMTargetAttr to pass the information. After feedback from @fabianmcg, this PR now passes the information through a new option of the gpu-module-to-binary pass.
Please add reviewers, as you see fit.