[clang-tools-extra] [llvm] [clang] [LLVM] Add IRNormalizer Pass (PR #68176)
michalpaszkowski wrote: [Here are the slides](https://llvm.org/devmtg/2019-10/slides/Paszkowski-LLVMCanon.pdf) that were presented at the LLVM Dev Meeting in 2019. It might be useful to include the link in the commit message for reference. @nikic Thanks for catching more things in the review! Especially the commutative ordering of operands/arguments. When it comes to including hashes in the naming algorithm (renaming is controlled using an option for already named values), the ideas is that by just looking at an instruction further down (a use) you can tell whether any inputs are different. And the other way around, by looking at initial instructions you can see how their "output footprint" has changed after transformations. The example below is a bit outdated, but reflects the general idea we had when considering naming algorithms: https://github.com/llvm/llvm-project/assets/13875746/fdb85b98-bce9-48df-91c0-35ffd115bae0";> https://github.com/llvm/llvm-project/assets/13875746/24d74cf1-ad21-47c0-9057-6065c7a4f472";> https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[llvm] [clang] [clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
https://github.com/michalpaszkowski approved this pull request. Thank you for working on this! LGTM! https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIR-V] Add Vulkan to target triple (PR #76749)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/76749 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIR-V] Add create.handle intrinsic (PR #81038)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/81038 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [Offload] Move HIP and CUDA to new driver by default (PR #84420)
michalpaszkowski wrote: Hi @jhuber6! Thank you for the pull request! > This is not intended to be landed immediately, but to allow for greater > testing. One potential issue I've discovered is the lack of SPIR-V > support or handling for `--offload`. Do you mean the SPIR-V target (backend)? I have not followed this area of work closely. What is missing or what exactly needs to be supported by the SPIR-V target? Any help or pointers would be greatly appreciated! https://github.com/llvm/llvm-project/pull/84420 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIR-V] Add SV_DispatchThreadID semantic support (PR #82536)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/82536 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] llvm-canon (PR #68176)
michalpaszkowski wrote: Thank you @justinfargnoli and @AidanGoldfarb for taking over this work and reaching out! I will review your changes in the coming week. CC @ChrisCummins FYI https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] llvm-canon (PR #68176)
michalpaszkowski wrote: Thank you @justinfargnoli and @AidanGoldfarb for taking over this work and reaching out! I will review your changes in the coming week. CC @ChrisCummins FYI https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] llvm-canon (PR #68176)
@@ -0,0 +1,638 @@
+//===--- IRCanonicalizer.cpp - IR Canonicalizer
---===//
+//
+// 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
+//
+//===--===//
+/// \file
+/// This file implements the IRCanonicalizer class which aims to transform LLVM
+/// Modules into a canonical form by reordering and renaming instructions while
+/// preserving the same semantics. The canonicalizer makes it easier to spot
+/// semantic differences while diffing two modules which have undergone
+/// different passes.
+///
+//===--===//
+
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/IR/Module.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Transforms/Utils.h"
+#include "llvm/Transforms/Utils/IRCanonicalizer.h"
+#include
+#include
+
+#define DEBUG_TYPE "canon"
+
+using namespace llvm;
+
+namespace {
+/// IRCanonicalizer aims to transform LLVM IR into canonical form.
+class IRCanonicalizer {
+public:
+ /// \name Canonicalizer flags.
+ /// @{
+ /// Preserves original order of instructions.
+ static cl::opt PreserveOrder;
+ /// Renames all instructions (including user-named).
+ static cl::opt RenameAll;
+ /// Folds all regular instructions (including pre-outputs).
+ static cl::opt FoldPreoutputs;
+ /// Sorts and reorders operands in commutative instructions.
+ static cl::opt ReorderOperands;
+ /// @}
+
+ bool runOnFunction(Function &F);
+
+private:
+ // Random constant for hashing, so the state isn't zero.
+ const uint64_t MagicHashConstant = 0x6acaa36bef8325c5ULL;
+ DenseSet NamedInstructions;
+
+ /// \name Naming.
+ /// @{
+ void nameFunctionArguments(Function &F);
+ void nameBasicBlocks(Function &F);
+ void nameInstruction(Instruction *I);
+ void nameAsInitialInstruction(Instruction *I);
+ void nameAsRegularInstruction(Instruction *I);
+ void foldInstructionName(Instruction *I);
+ /// @}
+
+ /// \name Reordering.
+ /// @{
+ void reorderInstructions(SmallVector &Outputs);
+ void reorderInstruction(Instruction *Used, Instruction *User,
+ SmallPtrSet &Visited);
+ void reorderInstructionOperandsByNames(Instruction *I);
+ void reorderPHIIncomingValues(PHINode *PN);
+ /// @}
+
+ /// \name Utility methods.
+ /// @{
+ SmallVector collectOutputInstructions(Function &F);
+ bool isOutput(const Instruction *I);
+ bool isInitialInstruction(const Instruction *I);
+ bool hasOnlyImmediateOperands(const Instruction *I);
+ SetVector
+ getOutputFootprint(Instruction *I,
+ SmallPtrSet &Visited);
+ /// @}
+};
+} // namespace
+
+cl::opt IRCanonicalizer::PreserveOrder(
+"preserve-order", cl::Hidden,
+cl::desc("Preserves original instruction order"));
+cl::opt IRCanonicalizer::RenameAll(
+"rename-all", cl::Hidden,
+cl::desc("Renames all instructions (including user-named)"));
+cl::opt IRCanonicalizer::FoldPreoutputs(
+"fold-all", cl::Hidden,
+cl::desc("Folds all regular instructions (including pre-outputs)"));
+cl::opt IRCanonicalizer::ReorderOperands(
+"reorder-operands", cl::Hidden,
+cl::desc("Sorts and reorders operands in commutative instructions"));
+
+/// Entry method to the IRCanonicalizer.
+///
+/// \param M Module to canonicalize.
+bool IRCanonicalizer::runOnFunction(Function &F) {
+ nameFunctionArguments(F);
+ nameBasicBlocks(F);
+
+ SmallVector Outputs = collectOutputInstructions(F);
+
+ if (!PreserveOrder)
+reorderInstructions(Outputs);
+
+ for (auto &I : Outputs)
+nameInstruction(I);
+
+ for (auto &I : instructions(F)) {
+if (!PreserveOrder) {
+ if (ReorderOperands && I.isCommutative())
+reorderInstructionOperandsByNames(&I);
+
+ if (auto *PN = dyn_cast(&I))
+reorderPHIIncomingValues(PN);
+}
+
+foldInstructionName(&I);
+ }
+
+ return true;
+}
+
+/// Numbers arguments.
+///
+/// \param F Function whose arguments will be renamed.
+void IRCanonicalizer::nameFunctionArguments(Function &F) {
+ int ArgumentCounter = 0;
+ for (auto &A : F.args()) {
+if (RenameAll || A.getName().empty()) {
+ A.setName("a" + Twine(ArgumentCounter));
+ ++ArgumentCounter;
+}
+ }
+}
+
+/// Names basic blocks using a generated hash for each basic block in
+/// a function considering the opcode and the order of output instructions.
+///
+/// \param F Function containing basic blocks to rename.
+void IRCanonicalizer::nameBas
[clang-tools-extra] llvm-canon (PR #68176)
@@ -0,0 +1,638 @@
+//===--- IRCanonicalizer.cpp - IR Canonicalizer
---===//
+//
+// 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
+//
+//===--===//
+/// \file
+/// This file implements the IRCanonicalizer class which aims to transform LLVM
+/// Modules into a canonical form by reordering and renaming instructions while
+/// preserving the same semantics. The canonicalizer makes it easier to spot
+/// semantic differences while diffing two modules which have undergone
+/// different passes.
+///
+//===--===//
+
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/IR/Module.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Transforms/Utils.h"
+#include "llvm/Transforms/Utils/IRCanonicalizer.h"
+#include
+#include
+
+#define DEBUG_TYPE "canon"
+
+using namespace llvm;
+
+namespace {
+/// IRCanonicalizer aims to transform LLVM IR into canonical form.
+class IRCanonicalizer {
+public:
+ /// \name Canonicalizer flags.
+ /// @{
+ /// Preserves original order of instructions.
+ static cl::opt PreserveOrder;
+ /// Renames all instructions (including user-named).
+ static cl::opt RenameAll;
+ /// Folds all regular instructions (including pre-outputs).
+ static cl::opt FoldPreoutputs;
+ /// Sorts and reorders operands in commutative instructions.
+ static cl::opt ReorderOperands;
+ /// @}
+
+ bool runOnFunction(Function &F);
+
+private:
+ // Random constant for hashing, so the state isn't zero.
+ const uint64_t MagicHashConstant = 0x6acaa36bef8325c5ULL;
+ DenseSet NamedInstructions;
+
+ /// \name Naming.
+ /// @{
+ void nameFunctionArguments(Function &F);
+ void nameBasicBlocks(Function &F);
+ void nameInstruction(Instruction *I);
+ void nameAsInitialInstruction(Instruction *I);
+ void nameAsRegularInstruction(Instruction *I);
+ void foldInstructionName(Instruction *I);
+ /// @}
+
+ /// \name Reordering.
+ /// @{
+ void reorderInstructions(SmallVector &Outputs);
+ void reorderInstruction(Instruction *Used, Instruction *User,
+ SmallPtrSet &Visited);
+ void reorderInstructionOperandsByNames(Instruction *I);
+ void reorderPHIIncomingValues(PHINode *PN);
+ /// @}
+
+ /// \name Utility methods.
+ /// @{
+ SmallVector collectOutputInstructions(Function &F);
+ bool isOutput(const Instruction *I);
+ bool isInitialInstruction(const Instruction *I);
+ bool hasOnlyImmediateOperands(const Instruction *I);
+ SetVector
+ getOutputFootprint(Instruction *I,
+ SmallPtrSet &Visited);
+ /// @}
+};
+} // namespace
+
+cl::opt IRCanonicalizer::PreserveOrder(
+"preserve-order", cl::Hidden,
+cl::desc("Preserves original instruction order"));
+cl::opt IRCanonicalizer::RenameAll(
+"rename-all", cl::Hidden,
+cl::desc("Renames all instructions (including user-named)"));
+cl::opt IRCanonicalizer::FoldPreoutputs(
+"fold-all", cl::Hidden,
+cl::desc("Folds all regular instructions (including pre-outputs)"));
+cl::opt IRCanonicalizer::ReorderOperands(
+"reorder-operands", cl::Hidden,
+cl::desc("Sorts and reorders operands in commutative instructions"));
+
+/// Entry method to the IRCanonicalizer.
+///
+/// \param M Module to canonicalize.
+bool IRCanonicalizer::runOnFunction(Function &F) {
+ nameFunctionArguments(F);
+ nameBasicBlocks(F);
+
+ SmallVector Outputs = collectOutputInstructions(F);
+
+ if (!PreserveOrder)
+reorderInstructions(Outputs);
+
+ for (auto &I : Outputs)
+nameInstruction(I);
+
+ for (auto &I : instructions(F)) {
+if (!PreserveOrder) {
+ if (ReorderOperands && I.isCommutative())
+reorderInstructionOperandsByNames(&I);
+
+ if (auto *PN = dyn_cast(&I))
+reorderPHIIncomingValues(PN);
+}
+
+foldInstructionName(&I);
+ }
+
+ return true;
+}
+
+/// Numbers arguments.
+///
+/// \param F Function whose arguments will be renamed.
+void IRCanonicalizer::nameFunctionArguments(Function &F) {
+ int ArgumentCounter = 0;
+ for (auto &A : F.args()) {
+if (RenameAll || A.getName().empty()) {
+ A.setName("a" + Twine(ArgumentCounter));
+ ++ArgumentCounter;
+}
+ }
+}
+
+/// Names basic blocks using a generated hash for each basic block in
+/// a function considering the opcode and the order of output instructions.
+///
+/// \param F Function containing basic blocks to rename.
+void IRCanonicalizer::nameBas
[clang] llvm-canon (PR #68176)
michalpaszkowski wrote: > Please don't use the term canon/canonicalize for this pass or tool. LLVM has > an existing notion of "canonicalization" which does not coincide with what is > being done here. @nikic I don't think the name "canon" or "canonicalizer" will lead to any confusion here. This is in fact what the pass is doing and mimics the name of MIR canonicalizer with similar assumptions as @plotfi noted . Using the term "canonicalization" for the current/existing set of passes is not really common or well defined. As for the RFC, there was a discussion thread in the LLVM Dev Mailing: https://lists.llvm.org/pipermail/llvm-dev/2019-August/134475.html https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] llvm-canon (PR #68176)
michalpaszkowski wrote: > Please don't use the term canon/canonicalize for this pass or tool. LLVM has > an existing notion of "canonicalization" which does not coincide with what is > being done here. @nikic I don't think the name "canon" or "canonicalizer" will lead to any confusion here. This is in fact what the pass is doing and mimics the name of MIR canonicalizer with similar assumptions as @plotfi noted . Using the term "canonicalization" for the current/existing set of passes is not really common or well defined. As for the RFC, there was a discussion thread in the LLVM Dev Mailing: https://lists.llvm.org/pipermail/llvm-dev/2019-August/134475.html https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -543,6 +543,15 @@ variables with initializers are marked as internal. An interprocedural variant of :ref:`Sparse Conditional Constant Propagation `. +``ir-normalizer``: Transforms IR into a canonical form that's easier to diff + + +This pass aims to transform LLVM Modules into a canonical form by reordering and +renaming instructions while preserving the same semantics. The canonicalizer makes +it easier to spot semantic differences while diffing two modules which have undergone +two different passes. + +--- michalpaszkowski wrote: Shouldn't this line be removed? and the line underscoring the "``ir-normalizer``: Transforms IR into a canonical form that's easier to diff" be made a bit longer? https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -0,0 +1,637 @@ +//===--- IRNormalizer.cpp - IR Canonicalizer ---===// michalpaszkowski wrote: Canonicalizer -> Normalizer https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -0,0 +1,637 @@
+//===--- IRNormalizer.cpp - IR Canonicalizer ---===//
+//
+// 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
+//
+//===--===//
+/// \file
+/// This file implements the IRNormalizer class which aims to transform LLVM
+/// Modules into a canonical form by reordering and renaming instructions while
+/// preserving the same semantics. The normalizer makes it easier to spot
+/// semantic differences while diffing two modules which have undergone
+/// different passes.
+///
+//===--===//
+
+#include "llvm/Transforms/Utils/IRNormalizer.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/IR/Module.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Transforms/Utils.h"
+#include
+#include
+
+#define DEBUG_TYPE "normalize"
+
+using namespace llvm;
+
+namespace {
+/// IRNormalizer aims to transform LLVM IR into canonical form.
+class IRNormalizer {
+public:
+ /// \name Canonicalizer flags.
michalpaszkowski wrote:
Canonicalizer -> Normalizer
https://github.com/llvm/llvm-project/pull/68176
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -543,6 +543,15 @@ variables with initializers are marked as internal. An interprocedural variant of :ref:`Sparse Conditional Constant Propagation `. +``ir-normalizer``: Transforms IR into a canonical form that's easier to diff + + +This pass aims to transform LLVM Modules into a canonical form by reordering and +renaming instructions while preserving the same semantics. The canonicalizer makes michalpaszkowski wrote: canonicalizer -> normalizer https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -42,7 +42,10 @@ Non-comprehensive list of changes in this release functionality, or simply have a lot to talk about), see the `NOTE` below for adding a new subsection. -* ... +* Added a new IRNormalizer pass which aims to transform LLVM modules into + a canonical form by reordering and renaming instructions while preserving the + same semantics. The canonicalizer makes it easier to spot semantic differences michalpaszkowski wrote: canonicalizer -> normalizer https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [LLVM] Add IRNormalizer Pass (PR #68176)
@@ -0,0 +1,637 @@
+//===--- IRNormalizer.cpp - IR Canonicalizer ---===//
+//
+// 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
+//
+//===--===//
+/// \file
+/// This file implements the IRNormalizer class which aims to transform LLVM
+/// Modules into a canonical form by reordering and renaming instructions while
+/// preserving the same semantics. The normalizer makes it easier to spot
+/// semantic differences while diffing two modules which have undergone
+/// different passes.
+///
+//===--===//
+
+#include "llvm/Transforms/Utils/IRNormalizer.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/IR/Module.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Transforms/Utils.h"
+#include
+#include
+
+#define DEBUG_TYPE "normalize"
+
+using namespace llvm;
+
+namespace {
+/// IRNormalizer aims to transform LLVM IR into canonical form.
+class IRNormalizer {
+public:
+ /// \name Canonicalizer flags.
+ /// @{
+ /// Preserves original order of instructions.
+ static cl::opt PreserveOrder;
+ /// Renames all instructions (including user-named).
+ static cl::opt RenameAll;
+ /// Folds all regular instructions (including pre-outputs).
+ static cl::opt FoldPreoutputs;
+ /// Sorts and reorders operands in commutative instructions.
+ static cl::opt ReorderOperands;
+ /// @}
+
+ bool runOnFunction(Function &F);
+
+private:
+ // Random constant for hashing, so the state isn't zero.
+ const uint64_t MagicHashConstant = 0x6acaa36bef8325c5ULL;
+ DenseSet NamedInstructions;
+
+ /// \name Naming.
+ /// @{
+ void nameFunctionArguments(Function &F);
+ void nameBasicBlocks(Function &F);
+ void nameInstruction(Instruction *I);
+ void nameAsInitialInstruction(Instruction *I);
+ void nameAsRegularInstruction(Instruction *I);
+ void foldInstructionName(Instruction *I);
+ /// @}
+
+ /// \name Reordering.
+ /// @{
+ void reorderInstructions(SmallVector &Outputs);
+ void reorderInstruction(Instruction *Used, Instruction *User,
+ SmallPtrSet &Visited);
+ void reorderInstructionOperandsByNames(Instruction *I);
+ void reorderPHIIncomingValues(PHINode *PN);
+ /// @}
+
+ /// \name Utility methods.
+ /// @{
+ SmallVector collectOutputInstructions(Function &F);
+ bool isOutput(const Instruction *I);
+ bool isInitialInstruction(const Instruction *I);
+ bool hasOnlyImmediateOperands(const Instruction *I);
+ SetVector
+ getOutputFootprint(Instruction *I,
+ SmallPtrSet &Visited);
+ /// @}
+};
+} // namespace
+
+cl::opt IRNormalizer::PreserveOrder(
+"preserve-order", cl::Hidden,
+cl::desc("Preserves original instruction order"));
+cl::opt IRNormalizer::RenameAll(
+"rename-all", cl::Hidden,
+cl::desc("Renames all instructions (including user-named)"));
+cl::opt IRNormalizer::FoldPreoutputs(
+"fold-all", cl::Hidden,
+cl::desc("Folds all regular instructions (including pre-outputs)"));
+cl::opt IRNormalizer::ReorderOperands(
+"reorder-operands", cl::Hidden,
+cl::desc("Sorts and reorders operands in commutative instructions"));
+
+/// Entry method to the IRNormalizer.
+///
+/// \param M Module to canonicalize.
michalpaszkowski wrote:
canonicalize -> normalize
https://github.com/llvm/llvm-project/pull/68176
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [LLVM] Add IRNormalizer Pass (PR #68176)
michalpaszkowski wrote: I added a couple more comments regarding the name change and formatting of the docs, but apart from these the patch looks good to me. https://github.com/llvm/llvm-project/pull/68176 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [HLSL][SPIRV] Add any intrinsic lowering (PR #88325)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/88325 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)
@@ -1,4 +1,5 @@ ; RUN: llc -O0 -mtriple=spirv64-unknown-unknown %s -o - | FileCheck %s +; RUN: llc -O0 -mtriple=spirv64-amd-amdhsa %s -o - | FileCheck %s michalpaszkowski wrote: I am okay with either approach. I think we can assume that the core functionality will stay the same for all "flavors". In case of any changes in the future to this assumption, we could create larger tests covering multiple features, which will not provide any isolation value but just ensure the features are still there. https://github.com/llvm/llvm-project/pull/89796 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)
https://github.com/michalpaszkowski edited https://github.com/llvm/llvm-project/pull/89796 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)
https://github.com/michalpaszkowski requested changes to this pull request. https://github.com/llvm/llvm-project/pull/89796 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)
@@ -56,6 +56,10 @@ static std::string computeDataLayout(const Triple &TT) {
if (Arch == Triple::spirv32)
return "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-"
"v96:128-v192:256-v256:256-v512:512-v1024:1024-G1";
+ if (TT.getVendor() == Triple::VendorType::AMD &&
michalpaszkowski wrote:
@AlexVlx Thanks for the pull request. Could you also please update the
[SPIRVUsage](https://llvm.org/docs/SPIRVUsage.html) document? Any relevant
information in the target triple section would be really appreciated.
https://github.com/llvm/llvm-project/pull/89796
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)
https://github.com/michalpaszkowski approved this pull request. @AlexVlx LGTM! Thank you! https://github.com/llvm/llvm-project/pull/89796 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][CodeGen] Add AS for Globals to SPIR & SPIRV datalayouts (PR #88455)
https://github.com/michalpaszkowski approved this pull request. LGTM (from the SPIR-V backend side)! https://github.com/llvm/llvm-project/pull/88455 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][HLSL] Add WaveIsFirstLane() intrinsic (PR #103299)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/103299 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL][SPIRV] Fix calling convention for call in entry function. (PR #110275)
michalpaszkowski wrote: LGTM! I think changing the mangling for Vulkan is the best solution here. One small comment, later on if/when you decide to use SPIRVBuiltins.cpp/.td for lowering builtins, there might be additional changes needed. The lookup function expects Itanium demangled strings. https://github.com/llvm/llvm-project/pull/110275 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [HLSL][SPIRV] Fix calling convention for call in entry function. (PR #110275)
@@ -316,6 +316,11 @@ class LLVM_LIBRARY_VISIBILITY SPIRVTargetInfo : public
BaseSPIRVTargetInfo {
SizeType = TargetInfo::UnsignedInt;
resetDataLayout("e-i64:64-v16:16-v24:32-v32:32-v48:64-"
"v96:128-v192:256-v256:256-v512:512-v1024:1024-G1");
+
+if (Triple.getOS() == llvm::Triple::Vulkan) {
+ // For Vulkan we want to same mangling as DirectX.
michalpaszkowski wrote:
Typo: `to` -> `the`?
https://github.com/llvm/llvm-project/pull/110275
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [clang][llvm][SPIR-V] Explicitly encode native integer widths for SPIR-V (PR #110695)
@@ -1,56 +0,0 @@ -; This test aims to check ability to support "Arithmetic with Overflow" intrinsics michalpaszkowski wrote: @AlexVlx I am not sure I follow the reasoning here. The idea behind those various tests in `/SPIRV/optimizations/` (including this one) is to ensure the backend can accept any IR. Either coming from the standard -O3 or some custom pipeline. I do agree that we should not expect specific optimizations (or even more sets of optimizations) to do specific things when writing a test for an independent component. Hence, after some thinking, I believe the test could be improved by already containing a specific pattern that would be coming from the optimization. However, by the same principle, we should not expect that the issue will be "optimized away" by another optimization. I would also oppose removing existing test cases. Any improvements, strengthening, or necessary changes are (of course) desirable. https://github.com/llvm/llvm-project/pull/110695 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= Message-ID: In-Reply-To: michalpaszkowski wrote: @Keenuts Apologies for the delay! The patch looks good to me. I added a couple of minor comments (above), but if the pass would be extended in subsequent commits, please feel free to address these in the next pull request or reply later. Thank you for working on this, this is really exciting work! https://github.com/llvm/llvm-project/pull/107408 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= , Nathan =?utf-8?q?Gau=C3=ABr?= Message-ID: In-Reply-To: https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/107408 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= Message-ID: In-Reply-To: https://github.com/michalpaszkowski edited https://github.com/llvm/llvm-project/pull/107408 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= , Nathan =?utf-8?q?Gauër?= Message-ID: In-Reply-To: https://github.com/michalpaszkowski edited https://github.com/llvm/llvm-project/pull/107408 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [llvm] [SPIR-V] Add SPIR-V structurizer (PR #107408)
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?= ,
Nathan =?utf-8?q?Gauër?=
Message-ID:
In-Reply-To:
@@ -0,0 +1,1242 @@
+//===-- SPIRVStructurizer.cpp --*- C++ -*-===//
+//
+// 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 "Analysis/SPIRVConvergenceRegionAnalysis.h"
+#include "SPIRV.h"
+#include "SPIRVSubtarget.h"
+#include "SPIRVTargetMachine.h"
+#include "SPIRVUtils.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/LoopSimplify.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
+#include
+#include
+#include
+
+using namespace llvm;
+using namespace SPIRV;
+
+namespace llvm {
+
+void initializeSPIRVStructurizerPass(PassRegistry &);
+
+namespace {
+
+using BlockSet = std::unordered_set;
+using Edge = std::pair;
+
+// Helper function to do a partial order visit from the block |Start|, calling
+// |Op| on each visited node.
+void partialOrderVisit(BasicBlock &Start,
+ std::function Op) {
+ PartialOrderingVisitor V(*Start.getParent());
+ V.partialOrderVisit(Start, Op);
+}
+
+// Returns the exact convergence region in the tree defined by `Node` for which
+// `BB` is the header, nullptr otherwise.
+const ConvergenceRegion *getRegionForHeader(const ConvergenceRegion *Node,
+BasicBlock *BB) {
+ if (Node->Entry == BB)
+return Node;
+
+ for (auto *Child : Node->Children) {
+const auto *CR = getRegionForHeader(Child, BB);
+if (CR != nullptr)
+ return CR;
+ }
+ return nullptr;
+}
+
+// Returns the single BasicBlock exiting the convergence region `CR`,
+// nullptr if no such exit exists.
+BasicBlock *getExitFor(const ConvergenceRegion *CR) {
+ std::unordered_set ExitTargets;
+ for (BasicBlock *Exit : CR->Exits) {
+for (BasicBlock *Successor : successors(Exit)) {
+ if (CR->Blocks.count(Successor) == 0)
+ExitTargets.insert(Successor);
+}
+ }
+
+ assert(ExitTargets.size() <= 1);
+ if (ExitTargets.size() == 0)
+return nullptr;
+
+ return *ExitTargets.begin();
+}
+
+// Returns the merge block designated by I if I is a merge instruction, nullptr
+// otherwise.
+BasicBlock *getDesignatedMergeBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge &&
+ II->getIntrinsicID() != Intrinsic::spv_selection_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(0));
+ return BA->getBasicBlock();
+}
+
+// Returns the continue block designated by I if I is an OpLoopMerge, nullptr
+// otherwise.
+BasicBlock *getDesignatedContinueBlock(Instruction *I) {
+ IntrinsicInst *II = dyn_cast(I);
+ if (II == nullptr)
+return nullptr;
+
+ if (II->getIntrinsicID() != Intrinsic::spv_loop_merge)
+return nullptr;
+
+ BlockAddress *BA = cast(II->getOperand(1));
+ return BA->getBasicBlock();
+}
+
+// Returns true if Header has one merge instruction which designated Merge as
+// merge block.
+bool isDefinedAsSelectionMergeBy(BasicBlock &Header, BasicBlock &Merge) {
+ for (auto &I : Header) {
+BasicBlock *MB = getDesignatedMergeBlock(&I);
+if (MB == &Merge)
+ return true;
+ }
+ return false;
+}
+
+// Returns true if the BB has one OpLoopMerge instruction.
+bool hasLoopMergeInstruction(BasicBlock &BB) {
+ for (auto &I : BB)
+if (getDesignatedContinueBlock(&I))
+ return true;
+ return false;
+}
+
+// Returns true is I is an OpSelectionMerge or OpLoopMerge instruction, false
+// otherwise.
+bool isMergeInstruction(Instruction *I) {
+ return getDesignatedMergeBlock(I) != nullptr;
+}
+
+// Returns all blocks in F having at least one OpLoopMerge or OpSelectionMerge
+// instruction.
+SmallPtrSet getHeaderBlocks(Function &F) {
+ SmallPtrSet Output;
+ for (BasicBlock &BB : F) {
+for (Instruction &I : BB) {
+ if (getDesignatedMergeBlock(&I) != nullptr)
+
[clang] [Clang][TableGen] Add missing __bf16 type to the builtins parser (PR #120662)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/120662 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [Clang][TableGen] Add missing __bf16 type to the builtins parser (PR #120662)
https://github.com/michalpaszkowski closed https://github.com/llvm/llvm-project/pull/120662 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] add pre legalization instruction combine (PR #122839)
@@ -0,0 +1,26 @@ +//=- SPIRVCombine.td - Define SPIRV Combine Rules -*-tablegen -*-=// +// +// 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 +// +//===--===// +// michalpaszkowski wrote: Should the file include some comment? Otherwise, I think this part could be removed. https://github.com/llvm/llvm-project/pull/122839 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][SPIR-V] Use the SPIR-V backend by default (PR #129545)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/129545 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [Clang] Add support for missing OpenCL extensions (PR #129777)
https://github.com/michalpaszkowski approved this pull request. https://github.com/llvm/llvm-project/pull/129777 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
https://github.com/michalpaszkowski requested changes to this pull request. https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
@@ -0,0 +1,501 @@ +//===- LowerGPUIntrinsic.cpp --===// +// +// 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 +// +//===--===// +// +// Lower the llvm.gpu intrinsics to target specific code sequences. +// Can be called from clang if building for a specific GPU or from the backend +// as part of a SPIRV lowering pipeline. Initial pass can lower to amdgcn or michalpaszkowski wrote: I don't understand how this pass fits into the SPIR-V lowering pipeline. Is this confusing [SPIR and SPIR-V](https://en.wikipedia.org/wiki/Standard_Portable_Intermediate_Representation)? SPIR being the format based on LLVM IR (although on an old version of LLVM 3.4)? SPIR-V being a custom [Khronos-defined format](https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html). [While SPIR-V easily maps to LLVM IR](https://github.com/KhronosGroup/SPIRV-LLVM-Translator/blob/main/docs/SPIRVRepresentationInLLVM.rst), it is far from being a standardized mapping and the actual translation is usually handled by either [SPIRV-LLVM-Translator](https://github.com/KhronosGroup/SPIRV-LLVM-Translator) or the in-tree SPIR-V backend. Could you please comment on how this pass and the changes would fit into the current SPIR-V code generation flow through the SPIR-V backend? https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
https://github.com/michalpaszkowski edited https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
michalpaszkowski wrote: I need to read through the RFC and the changes in the PR yet but I think `llvm/test/CodeGen/SPIRV/` is not really the best place for this test as it is not testing the SPIR-V backend. https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
@@ -0,0 +1,501 @@ +//===- LowerGPUIntrinsic.cpp --===// +// +// 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 +// +//===--===// +// +// Lower the llvm.gpu intrinsics to target specific code sequences. +// Can be called from clang if building for a specific GPU or from the backend +// as part of a SPIRV lowering pipeline. Initial pass can lower to amdgcn or michalpaszkowski wrote: I would like to better understand your goal here with this patch and potential future patches. How is this written in terms of the spirv64 triple? What is specific about the intrinsics that they only "work" for SPIR-V triple or the other way around? The SPIR-V triple in a module does describe characteristics about the module and obviously the architecture -- including a calling convention, vendor, address spaces, environment, etc. and most importantly indicates the target used -- in this case the in-tree SPIR-V backend. However, I do not see how the SPIR-V triple in case of this patch is relevant to the SPIR-V backend. If the goal is to define new intrinsics which could be handled by AMDGPU and NVPTX, then I don't think those should be associated with the SPIR-V triple in LLVM. Is your goal to have SPIR-V backend implement these intrinsics (lower into proper spec defined SPIR-V builtin calls or primitive instructions)? Or would you see the backend just keep these as function calls? If the latter, then I am not sure this is the right use of `llvm.` intrinicis [according to the guidelines](https://llvm.org/docs/ExtendingLLVM.html#adding-a-new-intrinsic-function)? In theory, I would expect all `llvm.` intrinsics to be handled by the backends. Having SPIR-V backend emit abstract function calls for intrinsics which could be lowered into the SPIR-V spec defined instructions or builtins is a very dangerous path and opposes the idea of the SPIR-V spec altogether. https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
@@ -0,0 +1,501 @@ +//===- LowerGPUIntrinsic.cpp --===// +// +// 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 +// +//===--===// +// +// Lower the llvm.gpu intrinsics to target specific code sequences. +// Can be called from clang if building for a specific GPU or from the backend +// as part of a SPIRV lowering pipeline. Initial pass can lower to amdgcn or michalpaszkowski wrote: The problem is that SPIR-V in itself is an architecture (although a synthetic one) and the SPIR-V backend is what lowers LLVM IR to this architecture. There are proprietary SPIR-V consumers which are not LLVM-based. For this reason it might be unreasonable to expect those to accept `llvm.` calls and it would be good to plan early how the new `llvm.gpu` intrinsics would be handled in the SPIR-V backend case by case. https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [llvm] [SPIRV] GPU intrinsics (PR #131190)
https://github.com/michalpaszkowski edited https://github.com/llvm/llvm-project/pull/131190 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
