New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[MLIR][LLVM] DI Expression Rewrite & Legalization #77541
Conversation
@llvm/pr-subscribers-mlir-llvm @llvm/pr-subscribers-mlir Author: Billy Zhu (zyx-billy) ChangesAdd a rewriter for DIExpressions & use it to run legalization patterns before exporting to llvm. The rewriter driver works similarly to the existing mlir rewriter drivers, except it operates on lists of DIExpressionElemAttr (i.e. DIExpressionAttr). Each rewrite pattern transforms a range of DIExpressionElemAttr into a new list of DIExpressionElemAttr. In addition, legalization patterns that are broadly applicable can be registered internally to the LLVM dialect, and they will be automatically applied prior to export. This PR adds one pattern for merging fragment operators. Full diff: https://github.com/llvm/llvm-project/pull/77541.diff 8 Files Affected:
diff --git a/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h b/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h
new file mode 100644
index 00000000000000..59ac4b9c933124
--- /dev/null
+++ b/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h
@@ -0,0 +1,44 @@
+//===- DIExpressionLegalization.h - DIExpression Legalization Patterns ----===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Declarations for known legalization patterns for DIExpressions that should
+// be performed before translation into llvm.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONLEGALIZATION_H
+#define MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONLEGALIZATION_H
+
+#include "mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h"
+
+namespace mlir {
+namespace LLVM {
+
+//===----------------------------------------------------------------------===//
+// Rewrite Patterns
+//===----------------------------------------------------------------------===//
+
+/// Adjacent DW_OP_LLVM_fragment should be merged into one.
+class MergeFragments : public DIExpressionRewriter::ExprRewritePattern {
+public:
+ OpIterT match(OpIterRange operators) const override;
+ SmallVector<OperatorT> replace(OpIterRange operators) const override;
+};
+
+//===----------------------------------------------------------------------===//
+// Runner
+//===----------------------------------------------------------------------===//
+
+/// Register all known legalization patterns declared here and apply them to
+/// all ops in `op`.
+void legalizeDIExpressionsRecursively(Operation *op);
+
+} // namespace LLVM
+} // namespace mlir
+
+#endif // MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONLEGALIZATION_H
diff --git a/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h b/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h
new file mode 100644
index 00000000000000..c1170dc49666ef
--- /dev/null
+++ b/mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h
@@ -0,0 +1,66 @@
+//===- DIExpressionRewriter.h - Rewriter for DIExpression operators -------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// A driver for running rewrite patterns on DIExpression operators.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONREWRITER_H
+#define MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONREWRITER_H
+
+#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include <deque>
+
+namespace mlir {
+namespace LLVM {
+
+/// Rewriter for DIExpressionAttr.
+///
+/// Users of this rewriter register their own rewrite patterns. Each pattern
+/// matches on a contiguous range of LLVM DIExpressionElemAttrs, and can be
+/// used to rewrite it into a new range of DIExpressionElemAttrs of any length.
+class DIExpressionRewriter {
+public:
+ using OperatorT = LLVM::DIExpressionElemAttr;
+
+ class ExprRewritePattern {
+ public:
+ using OperatorT = DIExpressionRewriter::OperatorT;
+ using OpIterT = std::deque<OperatorT>::const_iterator;
+ using OpIterRange = llvm::iterator_range<OpIterT>;
+
+ virtual ~ExprRewritePattern() = default;
+ /// Check whether a particular prefix of operators matches this pattern.
+ /// The provided argument is guaranteed non-empty.
+ /// Return the iterator after the last matched element.
+ virtual OpIterT match(OpIterRange) const = 0;
+ /// Replace the operators with a new list of operators.
+ /// The provided argument is guaranteed to be the same length as returned
+ /// by the `match` function.
+ virtual SmallVector<OperatorT> replace(OpIterRange) const = 0;
+ };
+
+ /// Register a rewrite pattern with the simplifier.
+ /// Rewriter patterns are attempted in the order of registration.
+ void addPattern(std::unique_ptr<ExprRewritePattern> pattern);
+
+ /// Simplify a DIExpression according to all the patterns registered.
+ /// A non-negative `maxNumRewrites` will limit the number of rewrites this
+ /// simplifier applies.
+ LLVM::DIExpressionAttr simplify(LLVM::DIExpressionAttr expr,
+ int64_t maxNumRewrites = -1) const;
+
+private:
+ /// The registered patterns.
+ SmallVector<std::unique_ptr<ExprRewritePattern>> patterns;
+};
+
+} // namespace LLVM
+} // namespace mlir
+
+#endif // MLIR_DIALECT_LLVMIR_TRANSFORMS_DIEXPRESSIONREWRITER_H
diff --git a/mlir/lib/Dialect/LLVMIR/Transforms/CMakeLists.txt b/mlir/lib/Dialect/LLVMIR/Transforms/CMakeLists.txt
index 47a2a251bf3e8b..c80494a440116b 100644
--- a/mlir/lib/Dialect/LLVMIR/Transforms/CMakeLists.txt
+++ b/mlir/lib/Dialect/LLVMIR/Transforms/CMakeLists.txt
@@ -1,5 +1,7 @@
add_mlir_dialect_library(MLIRLLVMIRTransforms
AddComdats.cpp
+ DIExpressionLegalization.cpp
+ DIExpressionRewriter.cpp
DIScopeForLLVMFuncOp.cpp
LegalizeForExport.cpp
OptimizeForNVVM.cpp
diff --git a/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionLegalization.cpp b/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionLegalization.cpp
new file mode 100644
index 00000000000000..0dce5d02175746
--- /dev/null
+++ b/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionLegalization.cpp
@@ -0,0 +1,61 @@
+//===- DIExpressionLegalization.cpp - DIExpression Legalization Patterns --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h"
+
+#include "llvm/BinaryFormat/Dwarf.h"
+
+using namespace mlir;
+using namespace LLVM;
+
+//===----------------------------------------------------------------------===//
+// MergeFragments
+//===----------------------------------------------------------------------===//
+
+MergeFragments::OpIterT MergeFragments::match(OpIterRange operators) const {
+ OpIterT it = operators.begin();
+ if (it == operators.end() ||
+ it->getOpcode() != llvm::dwarf::DW_OP_LLVM_fragment)
+ return operators.begin();
+
+ ++it;
+ if (it == operators.end() ||
+ it->getOpcode() != llvm::dwarf::DW_OP_LLVM_fragment)
+ return operators.begin();
+
+ return ++it;
+}
+
+SmallVector<MergeFragments::OperatorT>
+MergeFragments::replace(OpIterRange operators) const {
+ OpIterT it = operators.begin();
+ OperatorT first = *(it++);
+ OperatorT second = *it;
+ // Add offsets & select the size of the earlier operator (the one closer to
+ // the IR value).
+ uint64_t offset = first.getArguments()[0] + second.getArguments()[0];
+ uint64_t size = first.getArguments()[1];
+ OperatorT newOp = OperatorT::get(
+ first.getContext(), llvm::dwarf::DW_OP_LLVM_fragment, {offset, size});
+ return SmallVector<OperatorT>{newOp};
+}
+
+//===----------------------------------------------------------------------===//
+// Runner
+//===----------------------------------------------------------------------===//
+
+void mlir::LLVM::legalizeDIExpressionsRecursively(Operation *op) {
+ LLVM::DIExpressionRewriter rewriter;
+ rewriter.addPattern(std::make_unique<MergeFragments>());
+
+ mlir::AttrTypeReplacer replacer;
+ replacer.addReplacement([&rewriter](LLVM::DIExpressionAttr expr) {
+ return rewriter.simplify(expr);
+ });
+ replacer.recursivelyReplaceElementsIn(op);
+}
diff --git a/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionRewriter.cpp b/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionRewriter.cpp
new file mode 100644
index 00000000000000..b17b684082b13a
--- /dev/null
+++ b/mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionRewriter.cpp
@@ -0,0 +1,74 @@
+//===- DIExpressionRewriter.cpp - Rewriter for DIExpression operators -----===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h"
+#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
+#include "llvm/Support/Debug.h"
+
+using namespace mlir;
+using namespace LLVM;
+
+#define DEBUG_TYPE "llvm-di-expression-simplifier"
+
+//===----------------------------------------------------------------------===//
+// DIExpressionRewriter
+//===----------------------------------------------------------------------===//
+
+void DIExpressionRewriter::addPattern(
+ std::unique_ptr<ExprRewritePattern> pattern) {
+ patterns.emplace_back(std::move(pattern));
+}
+
+DIExpressionAttr DIExpressionRewriter::simplify(DIExpressionAttr expr,
+ int64_t maxNumRewrites) const {
+ ArrayRef<OperatorT> operators = expr.getOperations();
+
+ // `inputs` contains the unprocessed postfix of operators.
+ // `result` contains the already finalized prefix of operators.
+ // Invariant: concat(result, inputs) is equivalent to `operators` after some
+ // application of the rewrite patterns.
+ // Using a deque for inputs so that we have efficient front insertion and
+ // removal. Random access is not necessary for patterns.
+ std::deque<OperatorT> inputs(operators.begin(), operators.end());
+ SmallVector<OperatorT> result;
+
+ int64_t numRewrites = 0;
+ while (!inputs.empty() &&
+ (maxNumRewrites < 0 || numRewrites < maxNumRewrites)) {
+ bool foundMatch = false;
+ for (const std::unique_ptr<ExprRewritePattern> &pattern : patterns) {
+ ExprRewritePattern::OpIterT matchEnd = pattern->match(inputs);
+ if (matchEnd == inputs.begin())
+ continue;
+
+ foundMatch = true;
+ SmallVector<OperatorT> replacement =
+ pattern->replace(llvm::make_range(inputs.cbegin(), matchEnd));
+ inputs.erase(inputs.begin(), matchEnd);
+ inputs.insert(inputs.begin(), replacement.begin(), replacement.end());
+ ++numRewrites;
+ break;
+ }
+
+ if (!foundMatch) {
+ // If no match, pass along the current operator.
+ result.push_back(inputs.front());
+ inputs.pop_front();
+ }
+ }
+
+ if (maxNumRewrites >= 0 && numRewrites >= maxNumRewrites) {
+ LLVM_DEBUG(llvm::dbgs()
+ << "LLVMDIExpressionSimplifier exceeded max num rewrites ("
+ << maxNumRewrites << ")\n");
+ // Skip rewriting the rest.
+ result.append(inputs.begin(), inputs.end());
+ }
+
+ return LLVM::DIExpressionAttr::get(expr.getContext(), result);
+}
diff --git a/mlir/lib/Dialect/LLVMIR/Transforms/LegalizeForExport.cpp b/mlir/lib/Dialect/LLVMIR/Transforms/LegalizeForExport.cpp
index 61c1378d961210..1ac994fa5fb780 100644
--- a/mlir/lib/Dialect/LLVMIR/Transforms/LegalizeForExport.cpp
+++ b/mlir/lib/Dialect/LLVMIR/Transforms/LegalizeForExport.cpp
@@ -9,6 +9,7 @@
#include "mlir/Dialect/LLVMIR/Transforms/LegalizeForExport.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h"
#include "mlir/IR/Block.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/BuiltinOps.h"
@@ -79,6 +80,7 @@ struct LegalizeForExportPass
: public LLVM::impl::LLVMLegalizeForExportBase<LegalizeForExportPass> {
void runOnOperation() override {
LLVM::ensureDistinctSuccessors(getOperation());
+ LLVM::legalizeDIExpressionsRecursively(getOperation());
}
};
} // namespace
diff --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index ce46a194ea7d9f..fbbfb5b83eb609 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -19,6 +19,7 @@
#include "mlir/Dialect/DLTI/DLTI.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "mlir/Dialect/LLVMIR/LLVMInterfaces.h"
+#include "mlir/Dialect/LLVMIR/Transforms/DIExpressionLegalization.h"
#include "mlir/Dialect/LLVMIR/Transforms/LegalizeForExport.h"
#include "mlir/Dialect/OpenMP/OpenMPDialect.h"
#include "mlir/Dialect/OpenMP/OpenMPInterfaces.h"
@@ -1568,6 +1569,7 @@ mlir::translateModuleToLLVMIR(Operation *module, llvm::LLVMContext &llvmContext,
return nullptr;
LLVM::ensureDistinctSuccessors(module);
+ LLVM::legalizeDIExpressionsRecursively(module);
ModuleTranslation translator(module, std::move(llvmModule));
llvm::IRBuilder<> llvmBuilder(llvmContext);
diff --git a/mlir/test/Dialect/LLVMIR/di-expression-legalization.mlir b/mlir/test/Dialect/LLVMIR/di-expression-legalization.mlir
new file mode 100644
index 00000000000000..60fbc8135be62d
--- /dev/null
+++ b/mlir/test/Dialect/LLVMIR/di-expression-legalization.mlir
@@ -0,0 +1,42 @@
+// RUN: mlir-opt -llvm-legalize-for-export --split-input-file %s | FileCheck %s -check-prefix=CHECK-OPT
+// RUN: mlir-translate -mlir-to-llvmir --split-input-file %s | FileCheck %s -check-prefix=CHECK-TRANSLATE
+
+#di_file = #llvm.di_file<"foo.c" in "/mlir/">
+#di_compile_unit = #llvm.di_compile_unit<id = distinct[0]<>, sourceLanguage = DW_LANG_C, file = #di_file, producer = "MLIR", isOptimized = true, emissionKind = Full>
+#di_subprogram = #llvm.di_subprogram<compileUnit = #di_compile_unit, scope = #di_file, name = "simplify", file = #di_file, subprogramFlags = Definition>
+#i32_type = #llvm.di_basic_type<tag = DW_TAG_base_type, name = "i32", sizeInBits = 32, encoding = DW_ATE_unsigned>
+#i8_type = #llvm.di_basic_type<tag = DW_TAG_base_type, name = "i8", sizeInBits = 8, encoding = DW_ATE_unsigned>
+
+// struct0: {i8, i32}
+#struct0_first = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct0_first", baseType = #i8_type, sizeInBits = 8, alignInBits = 8>
+#struct0_second = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct0_second", baseType = #i32_type, sizeInBits = 32, alignInBits = 32, offsetInBits = 32>
+#struct0 = #llvm.di_composite_type<tag = DW_TAG_structure_type, name = "struct0", sizeInBits = 64, alignInBits = 32, elements = #struct0_first, #struct0_second>
+
+// struct1: {i8, struct0}
+#struct1_first = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct1_first", baseType = #i8_type, sizeInBits = 8, alignInBits = 8>
+#struct1_second = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct1_second", baseType = #struct0, sizeInBits = 64, alignInBits = 32>
+#struct1 = #llvm.di_composite_type<tag = DW_TAG_structure_type, name = "struct1", sizeInBits = 96, alignInBits = 32, elements = #struct1_first, #struct1_second>
+
+// struct2: {i32, struct1}
+#struct2_first = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct2_first", baseType = #i32_type, sizeInBits = 32, alignInBits = 32>
+#struct2_second = #llvm.di_derived_type<tag = DW_TAG_member, name = "struct2_second", baseType = #struct1, sizeInBits = 96, alignInBits = 32>
+#struct2 = #llvm.di_composite_type<tag = DW_TAG_structure_type, name = "struct2", sizeInBits = 128, alignInBits = 32, elements = #struct2_first, #struct2_second>
+
+#var0 = #llvm.di_local_variable<scope = #di_subprogram, name = "struct0_var", file = #di_file, line = 10, alignInBits = 32, type = #struct0>
+#var1 = #llvm.di_local_variable<scope = #di_subprogram, name = "struct1_var", file = #di_file, line = 10, alignInBits = 32, type = #struct1>
+#var2 = #llvm.di_local_variable<scope = #di_subprogram, name = "struct2_var", file = #di_file, line = 10, alignInBits = 32, type = #struct2>
+
+#loc = loc("test.mlir":0:0)
+
+llvm.func @merge_fragments(%arg0: !llvm.ptr, %arg1: !llvm.ptr, %arg2: !llvm.ptr) {
+ // CHECK-OPT: #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(32, 32)]>
+ // CHECK-TRANSLATE: !DIExpression(DW_OP_deref, DW_OP_LLVM_fragment, 32, 32))
+ llvm.intr.dbg.value #var0 #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(32, 32)]> = %arg0 : !llvm.ptr loc(fused<#di_subprogram>[#loc])
+ // CHECK-OPT: #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(64, 32)]>
+ // CHECK-TRANSLATE: !DIExpression(DW_OP_deref, DW_OP_LLVM_fragment, 64, 32))
+ llvm.intr.dbg.value #var1 #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(32, 32), DW_OP_LLVM_fragment(32, 64)]> = %arg1 : !llvm.ptr loc(fused<#di_subprogram>[#loc])
+ // CHECK-OPT: #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(96, 32)]>
+ // CHECK-TRANSLATE: !DIExpression(DW_OP_deref, DW_OP_LLVM_fragment, 96, 32))
+ llvm.intr.dbg.value #var2 #llvm.di_expression<[DW_OP_deref, DW_OP_LLVM_fragment(32, 32), DW_OP_LLVM_fragment(32, 64), DW_OP_LLVM_fragment(32, 96)]> = %arg2 : !llvm.ptr loc(fused<#di_subprogram>[#loc])
+ llvm.return
+}
|
Awesome! Please clarify in the motivation that the LLVM dialect allows DI expressions to be valid that may not be valid in LLVM, hence why we need the legalization pass |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice!
LGTM modulo some minor comments.
mlir/include/mlir/Dialect/LLVMIR/Transforms/DIExpressionRewriter.h
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/DIExpressionLegalization.cpp
Outdated
Show resolved
Hide resolved
|
||
int64_t numRewrites = 0; | ||
while (!inputs.empty() && | ||
(maxNumRewrites < 0 || numRewrites < maxNumRewrites)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(maxNumRewrites < 0 || numRewrites < maxNumRewrites)) { | |
(maxNumRewrites < 0 || numRewrites < maxNumRewrites)) { |
Wouldn't std::optional<uint64_t> be much nicer than using -1 in this case?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
+1
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
sounds good 👍 was a bit hesitant to wrap an int since the mlir rewriter was not, but I like this 😂 .
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks!
// Rewrite Patterns | ||
//===----------------------------------------------------------------------===// | ||
|
||
/// Adjacent DW_OP_LLVM_fragment should be merged into one. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
suggestion: please elaborate with a short psuedocode example, like expr(frag, frag) -> expr(frag)
or something.
|
||
int64_t numRewrites = 0; | ||
while (!inputs.empty() && | ||
(maxNumRewrites < 0 || numRewrites < maxNumRewrites)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
+1
Co-authored-by: Tobias Gysi <tobias.gysi@nextsilicon.com>
} | ||
} | ||
|
||
if (maxNumRewrites.has_value() && numRewrites >= maxNumRewrites.value()) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
if (maxNumRewrites.has_value() && numRewrites >= maxNumRewrites.value()) { | |
if (maxNumRewrites && numRewrites >= *maxNumRewrites) { |
Drive-by comment :). The optional evaluates to a boolean so you can drop has_value() and if you want to access the value you can use the star syntax *maxNumRewrites. I think MLIR prefers these short-forms.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh awesome. Just in time. Thanks 🙏
Add a rewriter for DIExpressions & use it to run legalization patterns before exporting to llvm (because LLVM dialect allows DI Expressions that may not be valid in LLVM IR). The rewriter driver works similarly to the existing mlir rewriter drivers, except it operates on lists of DIExpressionElemAttr (i.e. DIExpressionAttr). Each rewrite pattern transforms a range of DIExpressionElemAttr into a new list of DIExpressionElemAttr. In addition, this PR sets up a place to add legalization patterns that are broadly applicable internally to the LLVM dialect, and they will always be applied prior to export. This PR adds one pattern for merging fragment operators. --------- Co-authored-by: Tobias Gysi <tobias.gysi@nextsilicon.com>
Add a rewriter for DIExpressions & use it to run legalization patterns before exporting to llvm (because LLVM dialect allows DI Expressions that may not be valid in LLVM IR).
The rewriter driver works similarly to the existing mlir rewriter drivers, except it operates on lists of DIExpressionElemAttr (i.e. DIExpressionAttr). Each rewrite pattern transforms a range of DIExpressionElemAttr into a new list of DIExpressionElemAttr.
In addition, this PR sets up a place to add legalization patterns that are broadly applicable internally to the LLVM dialect, and they will always be applied prior to export. This PR adds one pattern for merging fragment operators.