[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
@@ -542,6 +542,48 @@ class OpLowerer {
});
}
+ [[nodiscard]] bool lowerRawBufferLoad(Function &F) {
+Triple TT(Triple(M.getTargetTriple()));
+VersionTuple DXILVersion = TT.getDXILVersion();
+const DataLayout &DL = F.getDataLayout();
+IRBuilder<> &IRB = OpBuilder.getIRB();
+Type *Int8Ty = IRB.getInt8Ty();
+Type *Int32Ty = IRB.getInt32Ty();
+
+return replaceFunction(F, [&](CallInst *CI) -> Error {
+ IRB.SetInsertPoint(CI);
+
+ Type *OldTy = cast(CI->getType())->getElementType(0);
+ Type *ScalarTy = OldTy->getScalarType();
+ Type *NewRetTy = OpBuilder.getResRetType(ScalarTy);
+
+ Value *Handle =
+ createTmpHandleCast(CI->getArgOperand(0), OpBuilder.getHandleType());
+ Value *Index0 = CI->getArgOperand(1);
+ Value *Index1 = CI->getArgOperand(2);
+ uint64_t NumElements =
+ DL.getTypeSizeInBits(OldTy) / DL.getTypeSizeInBits(ScalarTy);
+ Value *Mask = ConstantInt::get(Int8Ty, ~(~0U << NumElements));
+ Value *Align =
+ ConstantInt::get(Int32Ty, DL.getPrefTypeAlign(ScalarTy).value());
+
+ Expected OpCall =
+ DXILVersion >= VersionTuple(1, 2)
+ ? OpBuilder.tryCreateOp(OpCode::RawBufferLoad,
+ {Handle, Index0, Index1, Mask, Align},
+ CI->getName(), NewRetTy)
+ : OpBuilder.tryCreateOp(OpCode::BufferLoad,
+ {Handle, Index0, Index1}, CI->getName(),
+ NewRetTy);
+ if (Error E = OpCall.takeError())
+return E;
llvm-beanz wrote:
nit: personal preference, take or leave it.
```suggestion
if (!OpCall)
return OpCall.takeError();
```
https://github.com/llvm/llvm-project/pull/121012
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
https://github.com/llvm-beanz approved this pull request. LGTM. One extremely small nit that you can take or leave. https://github.com/llvm/llvm-project/pull/121012 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
https://github.com/llvm-beanz edited https://github.com/llvm/llvm-project/pull/121012 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [clang] [llvm] [Coverage][Single] Enable Branch coverage for `BinLAnd` and `BinLOr` (PR #113113)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/113113
>From 16e2bb8b73bcde1c2618bb358a905a9f463c1217 Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Sun, 20 Oct 2024 16:24:26 +0900
Subject: [PATCH 1/3] [Coverage][Single] Enable Branch coverage for `BinLAnd`
and `BinLOr`
---
clang/lib/CodeGen/CGExprScalar.cpp | 83 +++-
clang/lib/CodeGen/CGStmt.cpp | 4 --
clang/lib/CodeGen/CodeGenFunction.cpp| 43 ++--
clang/lib/CodeGen/CoverageMappingGen.cpp | 6 --
4 files changed, 104 insertions(+), 32 deletions(-)
diff --git a/clang/lib/CodeGen/CGExprScalar.cpp
b/clang/lib/CodeGen/CGExprScalar.cpp
index 11d4ec8a267605..83962ba96aa484 100644
--- a/clang/lib/CodeGen/CGExprScalar.cpp
+++ b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -4918,6 +4918,9 @@ Value *ScalarExprEmitter::VisitBinAssign(const
BinaryOperator *E) {
}
Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
+ auto HasLHSSkip = CGF.getIsCounterPair(E);
+ auto HasRHSSkip = CGF.getIsCounterPair(E->getRHS());
+
// Perform vector logical and on comparisons with zero vectors.
if (E->getType()->isVectorType()) {
CGF.incrementProfileCounter(E);
@@ -4964,11 +4967,17 @@ Value *ScalarExprEmitter::VisitBinLAnd(const
BinaryOperator *E) {
CodeGenFunction::isInstrumentedCondition(E->getRHS())) {
CGF.maybeUpdateMCDCCondBitmap(E->getRHS(), RHSCond);
llvm::BasicBlock *FBlock = CGF.createBasicBlock("land.end");
+llvm::BasicBlock *RHSSkip =
+(HasRHSSkip.second ? CGF.createBasicBlock("land.rhsskip") :
FBlock);
llvm::BasicBlock *RHSBlockCnt = CGF.createBasicBlock("land.rhscnt");
-Builder.CreateCondBr(RHSCond, RHSBlockCnt, FBlock);
+Builder.CreateCondBr(RHSCond, RHSBlockCnt, RHSSkip);
CGF.EmitBlock(RHSBlockCnt);
-CGF.incrementProfileCounter(E->getRHS());
+CGF.incrementProfileCounter(false, E->getRHS());
CGF.EmitBranch(FBlock);
+if (HasRHSSkip.second) {
+ CGF.EmitBlock(RHSSkip);
+ CGF.incrementProfileCounter(true, E->getRHS());
+}
CGF.EmitBlock(FBlock);
}
@@ -4997,12 +5006,21 @@ Value *ScalarExprEmitter::VisitBinLAnd(const
BinaryOperator *E) {
llvm::BasicBlock *ContBlock = CGF.createBasicBlock("land.end");
llvm::BasicBlock *RHSBlock = CGF.createBasicBlock("land.rhs");
+ llvm::BasicBlock *LHSFalseBlock =
+ (HasLHSSkip.second ? CGF.createBasicBlock("land.lhsskip") : ContBlock);
+
CodeGenFunction::ConditionalEvaluation eval(CGF);
// Branch on the LHS first. If it is false, go to the failure (cont) block.
- CGF.EmitBranchOnBoolExpr(E->getLHS(), RHSBlock, ContBlock,
+ CGF.EmitBranchOnBoolExpr(E->getLHS(), RHSBlock, LHSFalseBlock,
CGF.getProfileCount(E->getRHS()));
+ if (HasLHSSkip.second) {
+CGF.EmitBlock(LHSFalseBlock);
+CGF.incrementProfileCounter(true, E);
+CGF.EmitBranch(ContBlock);
+ }
+
// Any edges into the ContBlock are now from an (indeterminate number of)
// edges from this first condition. All of these values will be false.
Start
// setting up the PHI node in the Cont Block for this.
@@ -5014,7 +5032,7 @@ Value *ScalarExprEmitter::VisitBinLAnd(const
BinaryOperator *E) {
eval.begin(CGF);
CGF.EmitBlock(RHSBlock);
- CGF.incrementProfileCounter(E);
+ CGF.incrementProfileCounter(false, E);
Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
eval.end(CGF);
@@ -5024,15 +5042,24 @@ Value *ScalarExprEmitter::VisitBinLAnd(const
BinaryOperator *E) {
// If we're generating for profiling or coverage, generate a branch on the
// RHS to a block that increments the RHS true counter needed to track branch
// condition coverage.
+ llvm::BasicBlock *ContIncoming = RHSBlock;
if (InstrumentRegions &&
CodeGenFunction::isInstrumentedCondition(E->getRHS())) {
CGF.maybeUpdateMCDCCondBitmap(E->getRHS(), RHSCond);
llvm::BasicBlock *RHSBlockCnt = CGF.createBasicBlock("land.rhscnt");
-Builder.CreateCondBr(RHSCond, RHSBlockCnt, ContBlock);
+llvm::BasicBlock *RHSBlockSkip =
+(HasRHSSkip.second ? CGF.createBasicBlock("land.rhsskip") : ContBlock);
+Builder.CreateCondBr(RHSCond, RHSBlockCnt, RHSBlockSkip);
CGF.EmitBlock(RHSBlockCnt);
-CGF.incrementProfileCounter(E->getRHS());
+CGF.incrementProfileCounter(false, E->getRHS());
CGF.EmitBranch(ContBlock);
PN->addIncoming(RHSCond, RHSBlockCnt);
+if (HasRHSSkip.second) {
+ CGF.EmitBlock(RHSBlockSkip);
+ CGF.incrementProfileCounter(true, E->getRHS());
+ CGF.EmitBranch(ContBlock);
+ ContIncoming = RHSBlockSkip;
+}
}
// Emit an unconditional branch from this block to ContBlock.
@@ -5042,7 +5069,7 @@ Value *ScalarExprEmitter::VisitBinLAnd(const
BinaryOperator *E) {
CGF.EmitBlock(ContBlock);
}
// Insert an entry into the phi node for the edge with the value of RHSCond.
[llvm-branch-commits] [clang] [llvm] [Coverage][Single] Enable Branch coverage for loop statements (PR #113109)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/113109
>From 5d19c77551c6fc585d1b15c4c2a71c3c3f99ef8a Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Fri, 18 Oct 2024 09:33:51 +0900
Subject: [PATCH 1/4] [Coverage][Single] Enable Branch coverage for loop
statements
---
clang/lib/CodeGen/CGStmt.cpp | 82 ---
clang/lib/CodeGen/CodeGenFunction.cpp | 11 +-
clang/lib/CodeGen/CodeGenPGO.cpp | 79 +--
clang/lib/CodeGen/CoverageMappingGen.cpp | 130 +-
.../CoverageMapping/single-byte-counters.cpp | 53 +++
5 files changed, 97 insertions(+), 258 deletions(-)
diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index dbc1ce9bf993cd..7d778ce58a1487 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -1039,15 +1039,11 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
SourceLocToDebugLoc(R.getEnd()),
checkIfLoopMustProgress(S.getCond(), hasEmptyLoopBody(S)));
- // When single byte coverage mode is enabled, add a counter to loop
condition.
- if (llvm::EnableSingleByteCoverage)
-incrementProfileCounter(S.getCond());
-
// As long as the condition is true, go to the loop body.
llvm::BasicBlock *LoopBody = createBasicBlock("while.body");
if (EmitBoolCondBranch) {
llvm::BasicBlock *ExitBlock = LoopExit.getBlock();
-if (ConditionScope.requiresCleanups())
+if (getIsCounterPair(&S).second || ConditionScope.requiresCleanups())
ExitBlock = createBasicBlock("while.exit");
llvm::MDNode *Weights =
createProfileWeightsForLoop(S.getCond(), getProfileCount(S.getBody()));
@@ -1058,6 +1054,7 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
if (ExitBlock != LoopExit.getBlock()) {
EmitBlock(ExitBlock);
+ incrementProfileCounter(true, &S);
EmitBranchThroughCleanup(LoopExit);
}
} else if (const Attr *A = Stmt::getLikelihoodAttr(S.getBody())) {
@@ -1075,11 +1072,7 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
{
RunCleanupsScope BodyScope(*this);
EmitBlock(LoopBody);
-// When single byte coverage mode is enabled, add a counter to the body.
-if (llvm::EnableSingleByteCoverage)
- incrementProfileCounter(S.getBody());
-else
- incrementProfileCounter(&S);
+incrementProfileCounter(false, &S);
EmitStmt(S.getBody());
}
@@ -1099,13 +1092,10 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
// The LoopHeader typically is just a branch if we skipped emitting
// a branch, try to erase it.
- if (!EmitBoolCondBranch)
+ if (!EmitBoolCondBranch) {
SimplifyForwardingBlocks(LoopHeader.getBlock());
-
- // When single byte coverage mode is enabled, add a counter to continuation
- // block.
- if (llvm::EnableSingleByteCoverage)
-incrementProfileCounter(&S);
+PGO.markStmtAsUsed(true, &S);
+ }
if (CGM.shouldEmitConvergenceTokens())
ConvergenceTokenStack.pop_back();
@@ -1124,10 +1114,7 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
// Emit the body of the loop.
llvm::BasicBlock *LoopBody = createBasicBlock("do.body");
- if (llvm::EnableSingleByteCoverage)
-EmitBlockWithFallThrough(LoopBody, S.getBody());
- else
-EmitBlockWithFallThrough(LoopBody, &S);
+ EmitBlockWithFallThrough(LoopBody, &S);
if (CGM.shouldEmitConvergenceTokens())
ConvergenceTokenStack.push_back(
@@ -1139,9 +1126,6 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
}
EmitBlock(LoopCond.getBlock());
- // When single byte coverage mode is enabled, add a counter to loop
condition.
- if (llvm::EnableSingleByteCoverage)
-incrementProfileCounter(S.getCond());
// C99 6.8.5.2: "The evaluation of the controlling expression takes place
// after each execution of the loop body."
@@ -1164,16 +1148,25 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
SourceLocToDebugLoc(R.getEnd()),
checkIfLoopMustProgress(S.getCond(), hasEmptyLoopBody(S)));
+ auto *LoopFalse =
+ (getIsCounterPair(&S).second ? createBasicBlock("do.loopfalse")
+ : LoopExit.getBlock());
+
// As long as the condition is true, iterate the loop.
if (EmitBoolCondBranch) {
uint64_t BackedgeCount = getProfileCount(S.getBody()) - ParentCount;
Builder.CreateCondBr(
-BoolCondVal, LoopBody, LoopExit.getBlock(),
+BoolCondVal, LoopBody, LoopFalse,
createProfileWeightsForLoop(S.getCond(), BackedgeCount));
}
LoopStack.pop();
+ if (LoopFalse != LoopExit.getBlock()) {
+EmitBlock(LoopFalse);
+incrementProfileCounter(true, &S, true);
+ }
+
// Emit the exit block.
EmitBlock(LoopExit.getBlock());
@@ -1182,11 +1175,6 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
if (!EmitBoolCondBranch)
SimplifyForwardingBlocks(LoopCond.getBlock());
[llvm-branch-commits] [flang] 32dc6f5 - Revert "Reland '[flang] Allow to pass an async id to allocate the descriptor …"
Author: Valentin Clement (バレンタイン クレメン)
Date: 2024-12-23T21:26:29-08:00
New Revision: 32dc6f5f1b38f09f43fa4d1b697f102723ceb3ca
URL:
https://github.com/llvm/llvm-project/commit/32dc6f5f1b38f09f43fa4d1b697f102723ceb3ca
DIFF:
https://github.com/llvm/llvm-project/commit/32dc6f5f1b38f09f43fa4d1b697f102723ceb3ca.diff
LOG: Revert "Reland '[flang] Allow to pass an async id to allocate the
descriptor …"
This reverts commit 5b74fb75d90ef6620741af45f146cefacbd9ecef.
Added:
Modified:
flang/include/flang/Runtime/CUDA/allocator.h
flang/include/flang/Runtime/CUDA/common.h
flang/include/flang/Runtime/allocatable.h
flang/include/flang/Runtime/allocator-registry.h
flang/include/flang/Runtime/descriptor.h
flang/lib/Lower/Allocatable.cpp
flang/lib/Optimizer/Builder/Runtime/Allocatable.cpp
flang/runtime/CUDA/allocatable.cpp
flang/runtime/CUDA/allocator.cpp
flang/runtime/CUDA/descriptor.cpp
flang/runtime/allocatable.cpp
flang/runtime/array-constructor.cpp
flang/runtime/descriptor.cpp
flang/test/HLFIR/elemental-codegen.fir
flang/test/Lower/OpenACC/acc-declare.f90
flang/test/Lower/allocatable-polymorphic.f90
flang/test/Lower/allocatable-runtime.f90
flang/test/Lower/allocate-mold.f90
flang/test/Lower/polymorphic.f90
flang/unittests/Runtime/CUDA/Allocatable.cpp
flang/unittests/Runtime/CUDA/AllocatorCUF.cpp
flang/unittests/Runtime/CUDA/Memory.cpp
Removed:
diff --git a/flang/include/flang/Runtime/CUDA/allocator.h
b/flang/include/flang/Runtime/CUDA/allocator.h
index b6f0e7f303176c..4fb4c94c5e9b0a 100644
--- a/flang/include/flang/Runtime/CUDA/allocator.h
+++ b/flang/include/flang/Runtime/CUDA/allocator.h
@@ -20,16 +20,16 @@ extern "C" {
void RTDECL(CUFRegisterAllocator)();
}
-void *CUFAllocPinned(std::size_t, std::int64_t = kCudaNoStream);
+void *CUFAllocPinned(std::size_t);
void CUFFreePinned(void *);
-void *CUFAllocDevice(std::size_t, std::int64_t);
+void *CUFAllocDevice(std::size_t);
void CUFFreeDevice(void *);
-void *CUFAllocManaged(std::size_t, std::int64_t = kCudaNoStream);
+void *CUFAllocManaged(std::size_t);
void CUFFreeManaged(void *);
-void *CUFAllocUnified(std::size_t, std::int64_t = kCudaNoStream);
+void *CUFAllocUnified(std::size_t);
void CUFFreeUnified(void *);
} // namespace Fortran::runtime::cuda
diff --git a/flang/include/flang/Runtime/CUDA/common.h
b/flang/include/flang/Runtime/CUDA/common.h
index 9c95f727ee6734..474f8e6578b891 100644
--- a/flang/include/flang/Runtime/CUDA/common.h
+++ b/flang/include/flang/Runtime/CUDA/common.h
@@ -23,9 +23,6 @@ static constexpr unsigned kHostToDevice = 0;
static constexpr unsigned kDeviceToHost = 1;
static constexpr unsigned kDeviceToDevice = 2;
-/// Value used for asyncId when no specific stream is specified.
-static constexpr std::int64_t kCudaNoStream = -1;
-
#define CUDA_REPORT_IF_ERROR(expr) \
[](cudaError_t err) { \
if (err == cudaSuccess) \
diff --git a/flang/include/flang/Runtime/allocatable.h
b/flang/include/flang/Runtime/allocatable.h
index 121c31af963aa0..58061d9862095e 100644
--- a/flang/include/flang/Runtime/allocatable.h
+++ b/flang/include/flang/Runtime/allocatable.h
@@ -94,9 +94,9 @@ int RTDECL(AllocatableCheckLengthParameter)(Descriptor &,
// Successfully allocated memory is initialized if the allocatable has a
// derived type, and is always initialized by AllocatableAllocateSource().
// Performs all necessary coarray synchronization and validation actions.
-int RTDECL(AllocatableAllocate)(Descriptor &, std::int64_t asyncId = -1,
-bool hasStat = false, const Descriptor *errMsg = nullptr,
-const char *sourceFile = nullptr, int sourceLine = 0);
+int RTDECL(AllocatableAllocate)(Descriptor &, bool hasStat = false,
+const Descriptor *errMsg = nullptr, const char *sourceFile = nullptr,
+int sourceLine = 0);
int RTDECL(AllocatableAllocateSource)(Descriptor &, const Descriptor &source,
bool hasStat = false, const Descriptor *errMsg = nullptr,
const char *sourceFile = nullptr, int sourceLine = 0);
diff --git a/flang/include/flang/Runtime/allocator-registry.h
b/flang/include/flang/Runtime/allocator-registry.h
index 4c3295edf13d9a..29302c5d825bc9 100644
--- a/flang/include/flang/Runtime/allocator-registry.h
+++ b/flang/include/flang/Runtime/allocator-registry.h
@@ -11,7 +11,6 @@
#include "flang/Common/api-attrs.h"
#include "flang/Runtime/allocator-registry-consts.h"
-#include
#include
#include
@@ -19,7 +18,7 @@
namespace Fortran::runtime {
-using AllocFct = void *(*)(std::size_t, std::int64_t);
+using AllocFct = void *(*)(std::size_t);
using FreeFct = void (*)(void *);
typedef struct Allocator_t {
@@ -27,11 +26,10 @@ typedef struct Allocator_t {
FreeFct free{nullptr};
} Allocator_t;
-static RT_API_ATTRS void *MallocWrapper(
-std::size_t size, [[maybe_unused]] std::int64_t) {
+#ifdef RT_DE
[llvm-branch-commits] [clang] [llvm] [Coverage][Single] Enable Branch coverage for CondOp (PR #113110)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/113110
>From 744c5b634de08f9214c82d6fcfde7179bc4edfb0 Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Sun, 20 Oct 2024 14:46:07 +0900
Subject: [PATCH 1/4] [Coverage][Single] Enable Branch coverage for CondOp
---
clang/lib/CodeGen/CGExpr.cpp | 6 +--
clang/lib/CodeGen/CGExprAgg.cpp | 14 +--
clang/lib/CodeGen/CGExprComplex.cpp | 15 +---
clang/lib/CodeGen/CGExprScalar.cpp| 37 +++
clang/lib/CodeGen/CodeGenFunction.cpp | 3 +-
clang/lib/CodeGen/CodeGenPGO.cpp | 8
clang/lib/CodeGen/CoverageMappingGen.cpp | 16 ++--
.../CoverageMapping/single-byte-counters.cpp | 11 +++---
8 files changed, 25 insertions(+), 85 deletions(-)
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index cc85f05ad9f70c..67e3a1de17e679 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -5137,8 +5137,7 @@ std::optional
HandleConditionalOperatorLValueSimpleCase(
if (!CGF.ContainsLabel(Dead)) {
// If the true case is live, we need to track its region.
- if (CondExprBool)
-CGF.incrementProfileCounter(E);
+ CGF.incrementProfileCounter(!CondExprBool, E, true);
CGF.markStmtMaybeUsed(Dead);
// If a throw expression we emit it and return an undefined lvalue
// because it can't be used.
@@ -5177,7 +5176,7 @@ ConditionalInfo EmitConditionalBlocks(CodeGenFunction
&CGF,
// Any temporaries created here are conditional.
CGF.EmitBlock(Info.lhsBlock);
- CGF.incrementProfileCounter(E);
+ CGF.incrementProfileCounter(false, E);
eval.begin(CGF);
Info.LHS = BranchGenFunc(CGF, E->getTrueExpr());
eval.end(CGF);
@@ -5188,6 +5187,7 @@ ConditionalInfo EmitConditionalBlocks(CodeGenFunction
&CGF,
// Any temporaries created here are conditional.
CGF.EmitBlock(Info.rhsBlock);
+ CGF.incrementProfileCounter(true, E);
eval.begin(CGF);
Info.RHS = BranchGenFunc(CGF, E->getFalseExpr());
eval.end(CGF);
diff --git a/clang/lib/CodeGen/CGExprAgg.cpp b/clang/lib/CodeGen/CGExprAgg.cpp
index 2ad6587089f101..0c778ef185532f 100644
--- a/clang/lib/CodeGen/CGExprAgg.cpp
+++ b/clang/lib/CodeGen/CGExprAgg.cpp
@@ -36,10 +36,6 @@ using namespace CodeGen;
//Aggregate Expression Emitter
//===--===//
-namespace llvm {
-extern cl::opt EnableSingleByteCoverage;
-} // namespace llvm
-
namespace {
class AggExprEmitter : public StmtVisitor {
CodeGenFunction &CGF;
@@ -1293,10 +1289,7 @@ VisitAbstractConditionalOperator(const
AbstractConditionalOperator *E) {
eval.begin(CGF);
CGF.EmitBlock(LHSBlock);
- if (llvm::EnableSingleByteCoverage)
-CGF.incrementProfileCounter(E->getTrueExpr());
- else
-CGF.incrementProfileCounter(E);
+ CGF.incrementProfileCounter(false, E);
Visit(E->getTrueExpr());
eval.end(CGF);
@@ -1311,8 +1304,7 @@ VisitAbstractConditionalOperator(const
AbstractConditionalOperator *E) {
eval.begin(CGF);
CGF.EmitBlock(RHSBlock);
- if (llvm::EnableSingleByteCoverage)
-CGF.incrementProfileCounter(E->getFalseExpr());
+ CGF.incrementProfileCounter(true, E);
Visit(E->getFalseExpr());
eval.end(CGF);
@@ -1321,8 +1313,6 @@ VisitAbstractConditionalOperator(const
AbstractConditionalOperator *E) {
E->getType());
CGF.EmitBlock(ContBlock);
- if (llvm::EnableSingleByteCoverage)
-CGF.incrementProfileCounter(E);
}
void AggExprEmitter::VisitChooseExpr(const ChooseExpr *CE) {
diff --git a/clang/lib/CodeGen/CGExprComplex.cpp
b/clang/lib/CodeGen/CGExprComplex.cpp
index fef26e7b4ccdbd..bcece9431de764 100644
--- a/clang/lib/CodeGen/CGExprComplex.cpp
+++ b/clang/lib/CodeGen/CGExprComplex.cpp
@@ -28,10 +28,6 @@ using namespace CodeGen;
//Complex Expression Emitter
//===--===//
-namespace llvm {
-extern cl::opt EnableSingleByteCoverage;
-} // namespace llvm
-
typedef CodeGenFunction::ComplexPairTy ComplexPairTy;
/// Return the complex type that we are meant to emit.
@@ -1381,11 +1377,7 @@ VisitAbstractConditionalOperator(const
AbstractConditionalOperator *E) {
eval.begin(CGF);
CGF.EmitBlock(LHSBlock);
- if (llvm::EnableSingleByteCoverage)
-CGF.incrementProfileCounter(E->getTrueExpr());
- else
-CGF.incrementProfileCounter(E);
-
+ CGF.incrementProfileCounter(false, E);
ComplexPairTy LHS = Visit(E->getTrueExpr());
LHSBlock = Builder.GetInsertBlock();
CGF.EmitBranch(ContBlock);
@@ -1393,13 +1385,10 @@ VisitAbstractConditionalOperator(const
AbstractConditionalOperator *E) {
eval.begin(CGF);
CGF.EmitBlock(RHSBlock);
- if (llvm::EnableSingleByteCoverage)
-CGF.incrementProfileCounter(E->getFalseExpr());
+ CGF.incrementProfileCounter(true, E);
Com
[llvm-branch-commits] [clang] [llvm] [Coverage][Single] Enable Branch coverage for IfStmt (PR #113111)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/113111
>From 3ea6383e2142889550f37389dfaaee81e5ae7d9c Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Sun, 20 Oct 2024 15:15:03 +0900
Subject: [PATCH 1/3] [Coverage][Single] Enable Branch coverage for IfStmt
---
clang/lib/CodeGen/CGStmt.cpp | 31 +++-
clang/lib/CodeGen/CodeGenPGO.cpp | 12 ---
clang/lib/CodeGen/CoverageMappingGen.cpp | 21 +++
.../CoverageMapping/single-byte-counters.cpp | 36 ++-
4 files changed, 38 insertions(+), 62 deletions(-)
diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index dbc1ce9bf993cd..c511e5f4f4213a 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -840,8 +840,7 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
// If the skipped block has no labels in it, just emit the executed block.
// This avoids emitting dead code and simplifies the CFG substantially.
if (S.isConstexpr() || !ContainsLabel(Skipped)) {
- if (CondConstant)
-incrementProfileCounter(&S);
+ incrementProfileCounter(!CondConstant, &S, true);
if (Executed) {
RunCleanupsScope ExecutedScope(*this);
EmitStmt(Executed);
@@ -851,14 +850,14 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
}
}
+ auto HasSkip = getIsCounterPair(&S);
+
// Otherwise, the condition did not fold, or we couldn't elide it. Just emit
// the conditional branch.
llvm::BasicBlock *ThenBlock = createBasicBlock("if.then");
llvm::BasicBlock *ContBlock = createBasicBlock("if.end");
- llvm::BasicBlock *ElseBlock = ContBlock;
- if (Else)
-ElseBlock = createBasicBlock("if.else");
-
+ llvm::BasicBlock *ElseBlock =
+ (Else || HasSkip.second ? createBasicBlock("if.else") : ContBlock);
// Prefer the PGO based weights over the likelihood attribute.
// When the build isn't optimized the metadata isn't used, so don't generate
// it.
@@ -891,10 +890,7 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
// Emit the 'then' code.
EmitBlock(ThenBlock);
- if (llvm::EnableSingleByteCoverage)
-incrementProfileCounter(S.getThen());
- else
-incrementProfileCounter(&S);
+ incrementProfileCounter(false, &S);
{
RunCleanupsScope ThenScope(*this);
EmitStmt(S.getThen());
@@ -908,9 +904,9 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
auto NL = ApplyDebugLocation::CreateEmpty(*this);
EmitBlock(ElseBlock);
}
-// When single byte coverage mode is enabled, add a counter to else block.
-if (llvm::EnableSingleByteCoverage)
- incrementProfileCounter(Else);
+// Add a counter to else block unless it has CounterExpr.
+if (HasSkip.second)
+ incrementProfileCounter(true, &S);
{
RunCleanupsScope ElseScope(*this);
EmitStmt(Else);
@@ -920,15 +916,14 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
auto NL = ApplyDebugLocation::CreateEmpty(*this);
EmitBranch(ContBlock);
}
+ } else if (HasSkip.second) {
+EmitBlock(ElseBlock);
+incrementProfileCounter(true, &S);
+EmitBranch(ContBlock);
}
// Emit the continuation block for code after the if.
EmitBlock(ContBlock, true);
-
- // When single byte coverage mode is enabled, add a counter to continuation
- // block.
- if (llvm::EnableSingleByteCoverage)
-incrementProfileCounter(&S);
}
bool CodeGenFunction::checkIfLoopMustProgress(const Expr
*ControllingExpression,
diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp
index 0f2090da47a374..f6b9b5c82952c4 100644
--- a/clang/lib/CodeGen/CodeGenPGO.cpp
+++ b/clang/lib/CodeGen/CodeGenPGO.cpp
@@ -366,18 +366,6 @@ struct MapRegionCounters : public
RecursiveASTVisitor {
if (Hash.getHashVersion() == PGO_HASH_V1)
return Base::TraverseIfStmt(If);
-// When single byte coverage mode is enabled, add a counter to then and
-// else.
-bool NoSingleByteCoverage = !llvm::EnableSingleByteCoverage;
-for (Stmt *CS : If->children()) {
- if (!CS || NoSingleByteCoverage)
-continue;
- if (CS == If->getThen())
-CounterMap[If->getThen()] = NextCounter++;
- else if (CS == If->getElse())
-CounterMap[If->getElse()] = NextCounter++;
-}
-
// Otherwise, keep track of which branch we're in while traversing.
VisitStmt(If);
diff --git a/clang/lib/CodeGen/CoverageMappingGen.cpp
b/clang/lib/CodeGen/CoverageMappingGen.cpp
index a331d5bc68286b..6c6aecb9994c6b 100644
--- a/clang/lib/CodeGen/CoverageMappingGen.cpp
+++ b/clang/lib/CodeGen/CoverageMappingGen.cpp
@@ -2050,12 +2050,7 @@ struct CounterCoverageMappingBuilder
extendRegion(S->getCond());
Counter ParentCount = getRegion().getCounter();
-auto [ThenCount, ElseCount] =
-(llvm::EnableSingleByteCoverage
- ? std::make_pair(getRegionCounter(S->getThen())
[llvm-branch-commits] [clang] [llvm] [Coverage][Single] Enable Branch coverage for SwitchStmt (PR #113112)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/113112
>From ec05cc37e1177f06c9a44a1e39dadc9306cc5c68 Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Mon, 21 Oct 2024 08:09:31 +0900
Subject: [PATCH 1/2] [Coverage][Single] Enable Branch coverage for SwitchStmt
---
clang/lib/CodeGen/CGStmt.cpp | 12 ++
clang/lib/CodeGen/CoverageMappingGen.cpp | 22 ++-
.../CoverageMapping/single-byte-counters.cpp | 13 ++-
3 files changed, 32 insertions(+), 15 deletions(-)
diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index dbc1ce9bf993cd..80fe5cf183de16 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -2259,6 +2259,18 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt
&S) {
ConditionScope.ForceCleanup();
+ // Close the last case (or DefaultBlock).
+ EmitBranch(SwitchExit.getBlock());
+
+ // Insert a False Counter if SwitchStmt doesn't have DefaultStmt.
+ if (getIsCounterPair(S.getCond()).second) {
+auto *ImplicitDefaultBlock = createBasicBlock("sw.false");
+EmitBlock(ImplicitDefaultBlock);
+incrementProfileCounter(true, S.getCond());
+Builder.CreateBr(SwitchInsn->getDefaultDest());
+SwitchInsn->setDefaultDest(ImplicitDefaultBlock);
+ }
+
// Emit continuation.
EmitBlock(SwitchExit.getBlock(), true);
incrementProfileCounter(&S);
diff --git a/clang/lib/CodeGen/CoverageMappingGen.cpp
b/clang/lib/CodeGen/CoverageMappingGen.cpp
index a331d5bc68286b..c5fdf23299e4e9 100644
--- a/clang/lib/CodeGen/CoverageMappingGen.cpp
+++ b/clang/lib/CodeGen/CoverageMappingGen.cpp
@@ -958,6 +958,14 @@ struct CounterCoverageMappingBuilder
return {ExecCnt, SkipCnt};
}
+ Counter getSwitchImplicitDefaultCounter(const Stmt *Cond, Counter
ParentCount,
+ Counter CaseCountSum) {
+return (
+llvm::EnableSingleByteCoverage
+? Counter::getCounter(CounterMap[Cond].second = NextCounterNum++)
+: subtractCounters(ParentCount, CaseCountSum));
+ }
+
bool IsCounterEqual(Counter OutCount, Counter ParentCount) {
if (OutCount == ParentCount)
return true;
@@ -1885,7 +1893,7 @@ struct CounterCoverageMappingBuilder
propagateCounts(Counter::getZero(), Body);
BreakContinue BC = BreakContinueStack.pop_back_val();
-if (!BreakContinueStack.empty() && !llvm::EnableSingleByteCoverage)
+if (!BreakContinueStack.empty())
BreakContinueStack.back().ContinueCount = addCounters(
BreakContinueStack.back().ContinueCount, BC.ContinueCount);
@@ -1900,11 +1908,6 @@ struct CounterCoverageMappingBuilder
MostRecentLocation = getStart(S);
handleFileExit(ExitLoc);
-// When single byte coverage mode is enabled, do not create branch region
by
-// early returning.
-if (llvm::EnableSingleByteCoverage)
- return;
-
// Create a Branch Region around each Case. Subtract the case's
// counter from the Parent counter to track the "False" branch count.
Counter CaseCountSum;
@@ -1920,7 +1923,8 @@ struct CounterCoverageMappingBuilder
// the hidden branch, which will be added later by the CodeGen. This region
// will be associated with the switch statement's condition.
if (!HasDefaultCase) {
- Counter DefaultCount = subtractCounters(ParentCount, CaseCountSum);
+ Counter DefaultCount = getSwitchImplicitDefaultCounter(
+ S->getCond(), ParentCount, CaseCountSum);
createBranchRegion(S->getCond(), Counter::getZero(), DefaultCount);
}
}
@@ -1929,9 +1933,7 @@ struct CounterCoverageMappingBuilder
extendRegion(S);
SourceMappingRegion &Parent = getRegion();
-Counter Count = llvm::EnableSingleByteCoverage
-? getRegionCounter(S)
-: addCounters(Parent.getCounter(),
getRegionCounter(S));
+Counter Count = addCounters(Parent.getCounter(), getRegionCounter(S));
// Reuse the existing region if it starts at our label. This is typical of
// the first case in a switch.
diff --git a/clang/test/CoverageMapping/single-byte-counters.cpp
b/clang/test/CoverageMapping/single-byte-counters.cpp
index d20b695bc2636a..464fa370d86f09 100644
--- a/clang/test/CoverageMapping/single-byte-counters.cpp
+++ b/clang/test/CoverageMapping/single-byte-counters.cpp
@@ -34,19 +34,22 @@ int testIfElseReturn(int x) { // CHECK-NEXT: File 0,
[[@LINE]]:29 -> [[@LINE+9]]
}
// CHECK-NEXT: testSwitch
-int testSwitch(int x) { // CHECK-NEXT: File 0, [[@LINE]]:23 -> [[@LINE+17]]:2
= [[C30:#0]]
+int testSwitch(int x) { // CHECK-NEXT: File 0, [[@LINE]]:23 -> [[@LINE+20]]:2
= [[C30:#0]]
int result;
switch (x) {
-// CHECK-NEXT: Gap,File 0, [[@LINE-1]]:14 ->
[[@LINE+10]]:15 = 0
- case 1: // CHECK-NEXT: File 0, [[@LINE]]:3 -> [[@LINE+2]]:10 =
[[C31:#2]]
+// CHECK-NEXT: Gap,File
[llvm-branch-commits] [clang] [compiler-rt] [llvm] [Coverage] Move SingleByteCoverage out of CountedRegion (PR #110966)
https://github.com/chapuni edited https://github.com/llvm/llvm-project/pull/110966 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [clang] [compiler-rt] [llvm] [Coverage] Move SingleByteCoverage out of CountedRegion (PR #110966)
https://github.com/chapuni edited https://github.com/llvm/llvm-project/pull/110966 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -56,20 +58,80 @@ class DefaultPriorityAdvisor : public
RegAllocPriorityAdvisor {
unsigned getPriority(const LiveInterval &LI) const override;
};
-class RegAllocPriorityAdvisorAnalysis : public ImmutablePass {
+/// Common provider for getting the priority advisor and logging rewards.
+/// Legacy analysis forwards all calls to this provider.
+/// New analysis serves the provider as the analysis result.
+/// Expensive setup is done in the constructor, so that the advisor can be
+/// created quickly for every machine function.
+/// TODO: Remove once legacy PM support is dropped.
+class RegAllocPriorityAdvisorProvider {
public:
enum class AdvisorMode : int { Default, Release, Development };
- RegAllocPriorityAdvisorAnalysis(AdvisorMode Mode)
- : ImmutablePass(ID), Mode(Mode){};
- static char ID;
+ RegAllocPriorityAdvisorProvider(AdvisorMode Mode) : Mode(Mode) {}
+
+ virtual ~RegAllocPriorityAdvisorProvider() = default;
+
+ virtual void logRewardIfNeeded(const MachineFunction &MF,
+ llvm::function_ref GetReward) {};
- /// Get an advisor for the given context (i.e. machine function, etc)
virtual std::unique_ptr
getAdvisor(const MachineFunction &MF, const RAGreedy &RA) = 0;
+
+ void setAnalyses(SlotIndexes *SI) { this->SI = SI; }
optimisan wrote:
This provider is initialized only once for the pipeline and we just set the new
analysis results while running Greedy.
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [clang] [llvm] [Coverage] Make additional counters available for BranchRegion. NFC. (PR #120930)
https://github.com/chapuni updated
https://github.com/llvm/llvm-project/pull/120930
>From 5e460594c8a2550c38c759b2e6f1c5dc4152f820 Mon Sep 17 00:00:00 2001
From: NAKAMURA Takumi
Date: Thu, 17 Oct 2024 22:15:12 +0900
Subject: [PATCH 1/8] [Coverage] Make additional counters available for
BranchRegion. NFC.
`getBranchCounterPair()` allocates an additional Counter to SkipPath
in `SingleByteCoverage`.
`IsCounterEqual()` calculates the comparison with rewinding counter
replacements.
`NumRegionCounters` is updated to take additional counters in account.
`incrementProfileCounter()` has a few additiona arguments.
- `UseSkipPath=true`, to specify setting counters for SkipPath. It
assumes `UseSkipPath=false` is used together.
- `UseBoth` may be specified for marking another path. It introduces
the same effect as issueing `markStmtAsUsed(!SkipPath, S)`.
`llvm-cov` discovers counters in `FalseCount` to allocate
`MaxCounterID` for empty profile data.
---
clang/lib/CodeGen/CodeGenFunction.h | 8 -
clang/lib/CodeGen/CodeGenPGO.cpp | 31 +--
clang/lib/CodeGen/CodeGenPGO.h| 1 +
clang/lib/CodeGen/CoverageMappingGen.cpp | 31 ++-
.../ProfileData/Coverage/CoverageMapping.cpp | 4 +++
5 files changed, 65 insertions(+), 10 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenFunction.h
b/clang/lib/CodeGen/CodeGenFunction.h
index 89ac3b342d0a7c..cb1192bf6e11fe 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -1629,11 +1629,17 @@ class CodeGenFunction : public CodeGenTypeCache {
/// Increment the profiler's counter for the given statement by \p StepV.
/// If \p StepV is null, the default increment is 1.
void incrementProfileCounter(const Stmt *S, llvm::Value *StepV = nullptr) {
+incrementProfileCounter(false, S, false, StepV);
+ }
+
+ void incrementProfileCounter(bool UseSkipPath, const Stmt *S,
+ bool UseBoth = false,
+ llvm::Value *StepV = nullptr) {
if (CGM.getCodeGenOpts().hasProfileClangInstr() &&
!CurFn->hasFnAttribute(llvm::Attribute::NoProfile) &&
!CurFn->hasFnAttribute(llvm::Attribute::SkipProfile)) {
auto AL = ApplyDebugLocation::CreateArtificial(*this);
- PGO.emitCounterSetOrIncrement(Builder, S, StepV);
+ PGO.emitCounterSetOrIncrement(Builder, S, UseSkipPath, UseBoth, StepV);
}
PGO.setCurrentStmt(S);
}
diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp
index 069469e3de856b..aefd53e12088b4 100644
--- a/clang/lib/CodeGen/CodeGenPGO.cpp
+++ b/clang/lib/CodeGen/CodeGenPGO.cpp
@@ -1138,6 +1138,19 @@ void CodeGenPGO::emitCounterRegionMapping(const Decl *D)
{
if (CoverageMapping.empty())
return;
+ // Scan max(FalseCnt) and update NumRegionCounters.
+ unsigned MaxNumCounters = NumRegionCounters;
+ for (const auto [_, V] : *RegionCounterMap) {
+auto HasCounters = V.getIsCounterPair();
+assert((!HasCounters.first ||
+MaxNumCounters > (V.first & CounterPair::Mask)) &&
+ "TrueCnt should not be reassigned");
+if (HasCounters.second)
+ MaxNumCounters =
+ std::max(MaxNumCounters, (V.second & CounterPair::Mask) + 1);
+ }
+ NumRegionCounters = MaxNumCounters;
+
CGM.getCoverageMapping()->addFunctionMappingRecord(
FuncNameVar, FuncName, FunctionHash, CoverageMapping);
}
@@ -1193,11 +1206,25 @@ std::pair
CodeGenPGO::getIsCounterPair(const Stmt *S) const {
}
void CodeGenPGO::emitCounterSetOrIncrement(CGBuilderTy &Builder, const Stmt *S,
+ bool UseSkipPath, bool UseBoth,
llvm::Value *StepV) {
- if (!RegionCounterMap || !Builder.GetInsertBlock())
+ if (!RegionCounterMap)
return;
- unsigned Counter = (*RegionCounterMap)[S].first;
+ unsigned Counter;
+ auto &TheMap = (*RegionCounterMap)[S];
+ auto IsCounter = TheMap.getIsCounterPair();
+ if (!UseSkipPath) {
+assert(IsCounter.first);
+Counter = (TheMap.first & CounterPair::Mask);
+ } else {
+if (!IsCounter.second)
+ return;
+Counter = (TheMap.second & CounterPair::Mask);
+ }
+
+ if (!Builder.GetInsertBlock())
+return;
// Make sure that pointer to global is passed in with zero addrspace
// This is relevant during GPU profiling
diff --git a/clang/lib/CodeGen/CodeGenPGO.h b/clang/lib/CodeGen/CodeGenPGO.h
index 83f35785e5327d..8b769dd88d7f1e 100644
--- a/clang/lib/CodeGen/CodeGenPGO.h
+++ b/clang/lib/CodeGen/CodeGenPGO.h
@@ -112,6 +112,7 @@ class CodeGenPGO {
public:
std::pair getIsCounterPair(const Stmt *S) const;
void emitCounterSetOrIncrement(CGBuilderTy &Builder, const Stmt *S,
+ bool UseFalsePath, bool UseBoth,
llvm::Value *StepV);
void emitMCDCTestVectorBitmapUpdate(CGBuilderTy &Builder, const Expr *S,
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -150,12 +138,48 @@ class ReleaseModePriorityAdvisorAnalysis final
InteractiveChannelBaseName + ".out",
InteractiveChannelBaseName + ".in");
}
-return std::make_unique(
-MF, RA, &getAnalysis().getSI(), Runner.get());
+assert(SI && "SlotIndexes result must be set");
+return std::make_unique(MF, RA, SI, Runner.get());
}
+
+private:
std::unique_ptr Runner;
};
+RegAllocPriorityAdvisorProvider *createReleaseModePriorityAdvisorProvider() {
+ return new ReleaseModePriorityAdvisorProvider();
+}
+
+class ReleaseModePriorityAdvisorAnalysisLegacy final
+: public RegAllocPriorityAdvisorAnalysisLegacy {
+public:
+ ReleaseModePriorityAdvisorAnalysisLegacy()
+ : RegAllocPriorityAdvisorAnalysisLegacy(AdvisorMode::Release) {}
+ // support for isa<> and dyn_cast.
+ static bool classof(const RegAllocPriorityAdvisorAnalysisLegacy *R) {
+return R->getAdvisorMode() == AdvisorMode::Release;
+ }
+
+private:
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+AU.setPreservesAll();
+AU.addRequired();
+RegAllocPriorityAdvisorAnalysisLegacy::getAnalysisUsage(AU);
+ }
+
+ std::unique_ptr &getProvider() override {
+Provider->setAnalyses(&getAnalysis().getSI());
+return Provider;
+ }
+
+ bool doInitialization(Module &M) override {
+Provider = std::make_unique();
+return false;
+ }
+
+ // std::unique_ptr Provider;
arsenm wrote:
Commented out code
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -20,76 +20,138 @@
using namespace llvm;
-static cl::opt Mode(
+static cl::opt Mode(
"regalloc-enable-priority-advisor", cl::Hidden,
-cl::init(RegAllocPriorityAdvisorAnalysis::AdvisorMode::Default),
+cl::init(RegAllocPriorityAdvisorProvider::AdvisorMode::Default),
cl::desc("Enable regalloc advisor mode"),
cl::values(
-clEnumValN(RegAllocPriorityAdvisorAnalysis::AdvisorMode::Default,
+clEnumValN(RegAllocPriorityAdvisorProvider::AdvisorMode::Default,
"default", "Default"),
-clEnumValN(RegAllocPriorityAdvisorAnalysis::AdvisorMode::Release,
+clEnumValN(RegAllocPriorityAdvisorProvider::AdvisorMode::Release,
"release", "precompiled"),
-clEnumValN(RegAllocPriorityAdvisorAnalysis::AdvisorMode::Development,
+clEnumValN(RegAllocPriorityAdvisorProvider::AdvisorMode::Development,
"development", "for training")));
-char RegAllocPriorityAdvisorAnalysis::ID = 0;
-INITIALIZE_PASS(RegAllocPriorityAdvisorAnalysis, "regalloc-priority",
+char RegAllocPriorityAdvisorAnalysisLegacy::ID = 0;
+INITIALIZE_PASS(RegAllocPriorityAdvisorAnalysisLegacy, "regalloc-priority",
"Regalloc priority policy", false, true)
namespace {
-class DefaultPriorityAdvisorAnalysis final
-: public RegAllocPriorityAdvisorAnalysis {
+
+class DefaultPriorityAdvisorProvider final
+: public RegAllocPriorityAdvisorProvider {
public:
- DefaultPriorityAdvisorAnalysis(bool NotAsRequested)
- : RegAllocPriorityAdvisorAnalysis(AdvisorMode::Default),
+ DefaultPriorityAdvisorProvider(bool NotAsRequested, LLVMContext &Ctx)
+ : RegAllocPriorityAdvisorProvider(AdvisorMode::Default) {
+if (NotAsRequested)
+ Ctx.emitError("Requested regalloc priority advisor analysis "
+"could be created. Using default");
+ }
+
+ // support for isa<> and dyn_cast.
+ static bool classof(const RegAllocPriorityAdvisorProvider *R) {
+return R->getAdvisorMode() == AdvisorMode::Default;
+ }
+
+ std::unique_ptr
+ getAdvisor(const MachineFunction &MF, const RAGreedy &RA) override {
+assert(SI && "SlotIndexes result must be set");
+return std::make_unique(MF, RA, SI);
+ }
+};
+
+class DefaultPriorityAdvisorAnalysisLegacy final
+: public RegAllocPriorityAdvisorAnalysisLegacy {
+public:
+ DefaultPriorityAdvisorAnalysisLegacy(bool NotAsRequested)
+ : RegAllocPriorityAdvisorAnalysisLegacy(AdvisorMode::Default),
NotAsRequested(NotAsRequested) {}
// support for isa<> and dyn_cast.
- static bool classof(const RegAllocPriorityAdvisorAnalysis *R) {
+ static bool classof(const RegAllocPriorityAdvisorAnalysisLegacy *R) {
return R->getAdvisorMode() == AdvisorMode::Default;
}
private:
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired();
-RegAllocPriorityAdvisorAnalysis::getAnalysisUsage(AU);
+RegAllocPriorityAdvisorAnalysisLegacy::getAnalysisUsage(AU);
}
- std::unique_ptr
- getAdvisor(const MachineFunction &MF, const RAGreedy &RA) override {
-return std::make_unique(
-MF, RA, &getAnalysis().getSI());
+
+ std::unique_ptr &getProvider() override {
+Provider->setAnalyses(&getAnalysis().getSI());
+return Provider;
}
+
bool doInitialization(Module &M) override {
-if (NotAsRequested)
- M.getContext().emitError("Requested regalloc priority advisor analysis "
- "could be created. Using default");
-return RegAllocPriorityAdvisorAnalysis::doInitialization(M);
+Provider.reset(
+new DefaultPriorityAdvisorProvider(NotAsRequested, M.getContext()));
+return false;
}
+
const bool NotAsRequested;
+ // std::unique_ptr Provider;
};
} // namespace
-template <> Pass *llvm::callDefaultCtor() {
+void RegAllocPriorityAdvisorAnalysis::initializeProvider(LLVMContext &Ctx) {
+ if (Provider)
+return;
+
+ switch (Mode) {
+ case RegAllocPriorityAdvisorProvider::AdvisorMode::Default:
+Provider.reset(
+new DefaultPriorityAdvisorProvider(/*NotAsRequested*/ false, Ctx));
+break;
+ case RegAllocPriorityAdvisorProvider::AdvisorMode::Development:
+#if defined(LLVM_HAVE_TFLITE)
+Provider.reset(createDevelopmentModePriorityAdvisorProvider(Ctx));
+#endif
+break;
+ case RegAllocPriorityAdvisorProvider::AdvisorMode::Release:
+Provider.reset(createReleaseModePriorityAdvisorProvider());
+break;
+ }
+ if (!Provider)
+Provider.reset(
+new DefaultPriorityAdvisorProvider(/*NotAsRequested*/ true, Ctx));
arsenm wrote:
```suggestion
new DefaultPriorityAdvisorProvider(/*NotAsRequested=*/ true, Ctx));
```
This "NotAsRequested" name is bad, I don't know what it means
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailma
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -56,20 +58,80 @@ class DefaultPriorityAdvisor : public
RegAllocPriorityAdvisor {
unsigned getPriority(const LiveInterval &LI) const override;
};
-class RegAllocPriorityAdvisorAnalysis : public ImmutablePass {
+/// Common provider for getting the priority advisor and logging rewards.
+/// Legacy analysis forwards all calls to this provider.
+/// New analysis serves the provider as the analysis result.
+/// Expensive setup is done in the constructor, so that the advisor can be
+/// created quickly for every machine function.
+/// TODO: Remove once legacy PM support is dropped.
+class RegAllocPriorityAdvisorProvider {
public:
enum class AdvisorMode : int { Default, Release, Development };
- RegAllocPriorityAdvisorAnalysis(AdvisorMode Mode)
- : ImmutablePass(ID), Mode(Mode){};
- static char ID;
+ RegAllocPriorityAdvisorProvider(AdvisorMode Mode) : Mode(Mode) {}
+
+ virtual ~RegAllocPriorityAdvisorProvider() = default;
+
+ virtual void logRewardIfNeeded(const MachineFunction &MF,
+ llvm::function_ref GetReward) {};
arsenm wrote:
```suggestion
function_ref GetReward) {};
```
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -56,20 +58,80 @@ class DefaultPriorityAdvisor : public
RegAllocPriorityAdvisor {
unsigned getPriority(const LiveInterval &LI) const override;
};
-class RegAllocPriorityAdvisorAnalysis : public ImmutablePass {
+/// Common provider for getting the priority advisor and logging rewards.
+/// Legacy analysis forwards all calls to this provider.
+/// New analysis serves the provider as the analysis result.
+/// Expensive setup is done in the constructor, so that the advisor can be
+/// created quickly for every machine function.
+/// TODO: Remove once legacy PM support is dropped.
+class RegAllocPriorityAdvisorProvider {
public:
enum class AdvisorMode : int { Default, Release, Development };
- RegAllocPriorityAdvisorAnalysis(AdvisorMode Mode)
- : ImmutablePass(ID), Mode(Mode){};
- static char ID;
+ RegAllocPriorityAdvisorProvider(AdvisorMode Mode) : Mode(Mode) {}
+
+ virtual ~RegAllocPriorityAdvisorProvider() = default;
+
+ virtual void logRewardIfNeeded(const MachineFunction &MF,
+ llvm::function_ref GetReward) {};
- /// Get an advisor for the given context (i.e. machine function, etc)
virtual std::unique_ptr
getAdvisor(const MachineFunction &MF, const RAGreedy &RA) = 0;
+
+ void setAnalyses(SlotIndexes *SI) { this->SI = SI; }
arsenm wrote:
Why can't this just be set at construction time?
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [CodeGen][NewPM] Port RegAllocPriorityAdvisor analysis to NPM (PR #118462)
@@ -254,7 +249,22 @@ class DevelopmentModePriorityAdvisorAnalysis final
Log = std::make_unique(std::move(OS), LFS, Reward,
/*IncludeReward*/ true);
-return false;
+ }
+
+ void logRewardIfNeeded(const MachineFunction &MF,
+ llvm::function_ref GetReward) override {
+if (!Log || !Log->hasAnyObservationForContext(MF.getName()))
+ return;
+// The function pass manager would run all the function passes for a
+// function, so we assume the last context belongs to this function. If
+// this invariant ever changes, we can implement at that time switching
+// contexts. At this point, it'd be an error
+if (Log->currentContext() != MF.getName()) {
+ MF.getFunction().getContext().emitError(
+ "The training log context shouldn't have had changed.");
arsenm wrote:
Errors should start with lowercase and should not use contractions
https://github.com/llvm/llvm-project/pull/118462
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] RegAllocGreedy: Fix subrange based instruction split logic (PR #120199)
@@ -1387,18 +1357,34 @@ static bool readsLaneSubset(const MachineRegisterInfo
&MRI,
DestSrc->Destination->getSubReg() == DestSrc->Source->getSubReg())
return false;
+ Register Reg = VirtReg.reg();
+
// FIXME: We're only considering uses, but should be consider defs too?
- LaneBitmask ReadMask = getInstReadLaneMask(MRI, *TRI, *MI, VirtReg.reg());
+ LaneBitmask UseMask;
+ SmallVector, 8> Ops;
+ (void)AnalyzeVirtRegInBundle(const_cast(*MI), Reg, &Ops);
- LaneBitmask LiveAtMask;
- for (const LiveInterval::SubRange &S : VirtReg.subranges()) {
-if (S.liveAt(Use))
- LiveAtMask |= S.LaneMask;
+ for (auto [MI, OpIdx] : Ops) {
+const MachineOperand &MO = MI->getOperand(OpIdx);
+assert(MO.isReg() && MO.getReg() == Reg);
+unsigned SubReg = MO.getSubReg();
+if (SubReg == 0 && MO.isUse()) {
+ if (MO.isUndef())
+continue;
+ return false;
qcolombet wrote:
Ah that makes sense. Thanks for updating the comment of the function.
https://github.com/llvm/llvm-project/pull/120199
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] RegAllocGreedy: Fix subrange based instruction split logic (PR #120199)
https://github.com/qcolombet approved this pull request. https://github.com/llvm/llvm-project/pull/120199 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
https://github.com/TIFitis created
https://github.com/llvm/llvm-project/pull/120994
This patch adds the mapper field to the omp.map.info op.
>From 21178b07176869e000f7e027bc5c9593b715b5ad Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 19:05:30 +
Subject: [PATCH] Add mapper field to mapInfoOp.
---
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 2 ++
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 2 +-
mlir/test/Dialect/OpenMP/ops.mlir | 4 ++--
3 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 881f28d42e6312..7cc2c8fa8ce1e1 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1000,6 +1000,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
OptionalAttr:$members_index,
Variadic:$bounds, /* rank-0 to
rank-{n-1} */
OptionalAttr:$map_type,
+ OptionalAttr:$mapper_id,
OptionalAttr:$map_capture_type,
OptionalAttr:$name,
DefaultValuedAttr:$partial_map);
@@ -1064,6 +1065,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
`var_ptr` `(` $var_ptr `:` type($var_ptr) `,` $var_type `)`
oilist(
`var_ptr_ptr` `(` $var_ptr_ptr `:` type($var_ptr_ptr) `)`
+ | `mapper` `(` $mapper_id `)`
| `map_clauses` `(` custom($map_type) `)`
| `capture` `(` custom($map_capture_type) `)`
| `members` `(` $members `:` custom($members_index) `:`
type($members) `)`
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index b15d6ed15244ed..92233654ba1ddc 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -1592,7 +1592,7 @@ static LogicalResult verifyMapClause(Operation *op,
OperandRange mapVars) {
to ? updateToVars.insert(updateVar) : updateFromVars.insert(updateVar);
}
-} else {
+} else if (!isa(op)) {
emitError(op->getLoc(), "map argument is not a map entry operation");
}
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir
b/mlir/test/Dialect/OpenMP/ops.mlir
index a167c8bd1abbf9..97ef132f6dfa53 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -2523,13 +2523,13 @@ func.func @omp_targets_with_map_bounds(%arg0:
!llvm.ptr, %arg1: !llvm.ptr) -> ()
// CHECK: %[[C_12:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[C_13:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[BOUNDS1:.*]] = omp.map.bounds lower_bound(%[[C_11]] : i64)
upper_bound(%[[C_10]] : i64) stride(%[[C_12]] : i64) start_idx(%[[C_13]] : i64)
- // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) map_clauses(exit_release_or_enter_alloc)
capture(ByCopy) bounds(%[[BOUNDS1]]) -> !llvm.ptr {name = ""}
+ // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) mapper(@my_mapper)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%[[BOUNDS1]])
-> !llvm.ptr {name = ""}
%6 = llvm.mlir.constant(9 : index) : i64
%7 = llvm.mlir.constant(1 : index) : i64
%8 = llvm.mlir.constant(2 : index) : i64
%9 = llvm.mlir.constant(2 : index) : i64
%10 = omp.map.bounds lower_bound(%7 : i64) upper_bound(%6 : i64)
stride(%8 : i64) start_idx(%9 : i64)
-%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%10) ->
!llvm.ptr {name = ""}
+%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
mapper(@my_mapper) map_clauses(exit_release_or_enter_alloc) capture(ByCopy)
bounds(%10) -> !llvm.ptr {name = ""}
// CHECK: omp.target map_entries(%[[MAP0]] -> {{.*}}, %[[MAP1]] -> {{.*}}
: !llvm.ptr, !llvm.ptr)
omp.target map_entries(%mapv1 -> %arg2, %mapv2 -> %arg3 : !llvm.ptr,
!llvm.ptr) {
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
llvmbot wrote:
@llvm/pr-subscribers-flang-openmp
Author: Akash Banerjee (TIFitis)
Changes
This patch adds the mapper field to the omp.map.info op.
---
Full diff: https://github.com/llvm/llvm-project/pull/120994.diff
3 Files Affected:
- (modified) mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td (+2)
- (modified) mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp (+1-1)
- (modified) mlir/test/Dialect/OpenMP/ops.mlir (+2-2)
``diff
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 881f28d42e6312..7cc2c8fa8ce1e1 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1000,6 +1000,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
OptionalAttr:$members_index,
Variadic:$bounds, /* rank-0 to
rank-{n-1} */
OptionalAttr:$map_type,
+ OptionalAttr:$mapper_id,
OptionalAttr:$map_capture_type,
OptionalAttr:$name,
DefaultValuedAttr:$partial_map);
@@ -1064,6 +1065,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
`var_ptr` `(` $var_ptr `:` type($var_ptr) `,` $var_type `)`
oilist(
`var_ptr_ptr` `(` $var_ptr_ptr `:` type($var_ptr_ptr) `)`
+ | `mapper` `(` $mapper_id `)`
| `map_clauses` `(` custom($map_type) `)`
| `capture` `(` custom($map_capture_type) `)`
| `members` `(` $members `:` custom($members_index) `:`
type($members) `)`
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index b15d6ed15244ed..92233654ba1ddc 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -1592,7 +1592,7 @@ static LogicalResult verifyMapClause(Operation *op,
OperandRange mapVars) {
to ? updateToVars.insert(updateVar) : updateFromVars.insert(updateVar);
}
-} else {
+} else if (!isa(op)) {
emitError(op->getLoc(), "map argument is not a map entry operation");
}
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir
b/mlir/test/Dialect/OpenMP/ops.mlir
index a167c8bd1abbf9..97ef132f6dfa53 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -2523,13 +2523,13 @@ func.func @omp_targets_with_map_bounds(%arg0:
!llvm.ptr, %arg1: !llvm.ptr) -> ()
// CHECK: %[[C_12:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[C_13:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[BOUNDS1:.*]] = omp.map.bounds lower_bound(%[[C_11]] : i64)
upper_bound(%[[C_10]] : i64) stride(%[[C_12]] : i64) start_idx(%[[C_13]] : i64)
- // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) map_clauses(exit_release_or_enter_alloc)
capture(ByCopy) bounds(%[[BOUNDS1]]) -> !llvm.ptr {name = ""}
+ // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) mapper(@my_mapper)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%[[BOUNDS1]])
-> !llvm.ptr {name = ""}
%6 = llvm.mlir.constant(9 : index) : i64
%7 = llvm.mlir.constant(1 : index) : i64
%8 = llvm.mlir.constant(2 : index) : i64
%9 = llvm.mlir.constant(2 : index) : i64
%10 = omp.map.bounds lower_bound(%7 : i64) upper_bound(%6 : i64)
stride(%8 : i64) start_idx(%9 : i64)
-%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%10) ->
!llvm.ptr {name = ""}
+%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
mapper(@my_mapper) map_clauses(exit_release_or_enter_alloc) capture(ByCopy)
bounds(%10) -> !llvm.ptr {name = ""}
// CHECK: omp.target map_entries(%[[MAP0]] -> {{.*}}, %[[MAP1]] -> {{.*}}
: !llvm.ptr, !llvm.ptr)
omp.target map_entries(%mapv1 -> %arg2, %mapv2 -> %arg3 : !llvm.ptr,
!llvm.ptr) {
``
https://github.com/llvm/llvm-project/pull/120994
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
llvmbot wrote:
@llvm/pr-subscribers-mlir-openmp
Author: Akash Banerjee (TIFitis)
Changes
This patch adds the mapper field to the omp.map.info op.
---
Full diff: https://github.com/llvm/llvm-project/pull/120994.diff
3 Files Affected:
- (modified) mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td (+2)
- (modified) mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp (+1-1)
- (modified) mlir/test/Dialect/OpenMP/ops.mlir (+2-2)
``diff
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 881f28d42e6312..7cc2c8fa8ce1e1 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1000,6 +1000,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
OptionalAttr:$members_index,
Variadic:$bounds, /* rank-0 to
rank-{n-1} */
OptionalAttr:$map_type,
+ OptionalAttr:$mapper_id,
OptionalAttr:$map_capture_type,
OptionalAttr:$name,
DefaultValuedAttr:$partial_map);
@@ -1064,6 +1065,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
`var_ptr` `(` $var_ptr `:` type($var_ptr) `,` $var_type `)`
oilist(
`var_ptr_ptr` `(` $var_ptr_ptr `:` type($var_ptr_ptr) `)`
+ | `mapper` `(` $mapper_id `)`
| `map_clauses` `(` custom($map_type) `)`
| `capture` `(` custom($map_capture_type) `)`
| `members` `(` $members `:` custom($members_index) `:`
type($members) `)`
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index b15d6ed15244ed..92233654ba1ddc 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -1592,7 +1592,7 @@ static LogicalResult verifyMapClause(Operation *op,
OperandRange mapVars) {
to ? updateToVars.insert(updateVar) : updateFromVars.insert(updateVar);
}
-} else {
+} else if (!isa(op)) {
emitError(op->getLoc(), "map argument is not a map entry operation");
}
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir
b/mlir/test/Dialect/OpenMP/ops.mlir
index a167c8bd1abbf9..97ef132f6dfa53 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -2523,13 +2523,13 @@ func.func @omp_targets_with_map_bounds(%arg0:
!llvm.ptr, %arg1: !llvm.ptr) -> ()
// CHECK: %[[C_12:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[C_13:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[BOUNDS1:.*]] = omp.map.bounds lower_bound(%[[C_11]] : i64)
upper_bound(%[[C_10]] : i64) stride(%[[C_12]] : i64) start_idx(%[[C_13]] : i64)
- // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) map_clauses(exit_release_or_enter_alloc)
capture(ByCopy) bounds(%[[BOUNDS1]]) -> !llvm.ptr {name = ""}
+ // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) mapper(@my_mapper)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%[[BOUNDS1]])
-> !llvm.ptr {name = ""}
%6 = llvm.mlir.constant(9 : index) : i64
%7 = llvm.mlir.constant(1 : index) : i64
%8 = llvm.mlir.constant(2 : index) : i64
%9 = llvm.mlir.constant(2 : index) : i64
%10 = omp.map.bounds lower_bound(%7 : i64) upper_bound(%6 : i64)
stride(%8 : i64) start_idx(%9 : i64)
-%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%10) ->
!llvm.ptr {name = ""}
+%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
mapper(@my_mapper) map_clauses(exit_release_or_enter_alloc) capture(ByCopy)
bounds(%10) -> !llvm.ptr {name = ""}
// CHECK: omp.target map_entries(%[[MAP0]] -> {{.*}}, %[[MAP1]] -> {{.*}}
: !llvm.ptr, !llvm.ptr)
omp.target map_entries(%mapv1 -> %arg2, %mapv2 -> %arg3 : !llvm.ptr,
!llvm.ptr) {
``
https://github.com/llvm/llvm-project/pull/120994
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
llvmbot wrote:
@llvm/pr-subscribers-mlir
Author: Akash Banerjee (TIFitis)
Changes
This patch adds the mapper field to the omp.map.info op.
---
Full diff: https://github.com/llvm/llvm-project/pull/120994.diff
3 Files Affected:
- (modified) mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td (+2)
- (modified) mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp (+1-1)
- (modified) mlir/test/Dialect/OpenMP/ops.mlir (+2-2)
``diff
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 881f28d42e6312..7cc2c8fa8ce1e1 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1000,6 +1000,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
OptionalAttr:$members_index,
Variadic:$bounds, /* rank-0 to
rank-{n-1} */
OptionalAttr:$map_type,
+ OptionalAttr:$mapper_id,
OptionalAttr:$map_capture_type,
OptionalAttr:$name,
DefaultValuedAttr:$partial_map);
@@ -1064,6 +1065,7 @@ def MapInfoOp : OpenMP_Op<"map.info",
[AttrSizedOperandSegments]> {
`var_ptr` `(` $var_ptr `:` type($var_ptr) `,` $var_type `)`
oilist(
`var_ptr_ptr` `(` $var_ptr_ptr `:` type($var_ptr_ptr) `)`
+ | `mapper` `(` $mapper_id `)`
| `map_clauses` `(` custom($map_type) `)`
| `capture` `(` custom($map_capture_type) `)`
| `members` `(` $members `:` custom($members_index) `:`
type($members) `)`
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index b15d6ed15244ed..92233654ba1ddc 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -1592,7 +1592,7 @@ static LogicalResult verifyMapClause(Operation *op,
OperandRange mapVars) {
to ? updateToVars.insert(updateVar) : updateFromVars.insert(updateVar);
}
-} else {
+} else if (!isa(op)) {
emitError(op->getLoc(), "map argument is not a map entry operation");
}
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir
b/mlir/test/Dialect/OpenMP/ops.mlir
index a167c8bd1abbf9..97ef132f6dfa53 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -2523,13 +2523,13 @@ func.func @omp_targets_with_map_bounds(%arg0:
!llvm.ptr, %arg1: !llvm.ptr) -> ()
// CHECK: %[[C_12:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[C_13:.*]] = llvm.mlir.constant(2 : index) : i64
// CHECK: %[[BOUNDS1:.*]] = omp.map.bounds lower_bound(%[[C_11]] : i64)
upper_bound(%[[C_10]] : i64) stride(%[[C_12]] : i64) start_idx(%[[C_13]] : i64)
- // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) map_clauses(exit_release_or_enter_alloc)
capture(ByCopy) bounds(%[[BOUNDS1]]) -> !llvm.ptr {name = ""}
+ // CHECK: %[[MAP1:.*]] = omp.map.info var_ptr(%[[ARG1]] : !llvm.ptr,
!llvm.array<10 x i32>) mapper(@my_mapper)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%[[BOUNDS1]])
-> !llvm.ptr {name = ""}
%6 = llvm.mlir.constant(9 : index) : i64
%7 = llvm.mlir.constant(1 : index) : i64
%8 = llvm.mlir.constant(2 : index) : i64
%9 = llvm.mlir.constant(2 : index) : i64
%10 = omp.map.bounds lower_bound(%7 : i64) upper_bound(%6 : i64)
stride(%8 : i64) start_idx(%9 : i64)
-%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
map_clauses(exit_release_or_enter_alloc) capture(ByCopy) bounds(%10) ->
!llvm.ptr {name = ""}
+%mapv2 = omp.map.info var_ptr(%arg1 : !llvm.ptr, !llvm.array<10 x i32>)
mapper(@my_mapper) map_clauses(exit_release_or_enter_alloc) capture(ByCopy)
bounds(%10) -> !llvm.ptr {name = ""}
// CHECK: omp.target map_entries(%[[MAP0]] -> {{.*}}, %[[MAP1]] -> {{.*}}
: !llvm.ptr, !llvm.ptr)
omp.target map_entries(%mapv1 -> %arg2, %mapv2 -> %arg3 : !llvm.ptr,
!llvm.ptr) {
``
https://github.com/llvm/llvm-project/pull/120994
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [flang] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
https://github.com/TIFitis updated
https://github.com/llvm/llvm-project/pull/120994
>From 0b54cafc829119b931d04d1d9ec3e790e50f7b08 Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 19:05:30 +
Subject: [PATCH] Add mapper field to mapInfoOp.
---
flang/lib/Lower/OpenMP/Utils.cpp | 3 ++-
flang/lib/Lower/OpenMP/Utils.h| 3 ++-
.../lib/Optimizer/OpenMP/MapInfoFinalization.cpp | 5 -
.../Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp | 1 +
flang/test/Lower/OpenMP/map-mapper.f90| 15 +++
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 2 ++
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 2 +-
mlir/test/Dialect/OpenMP/ops.mlir | 4 ++--
8 files changed, 29 insertions(+), 6 deletions(-)
create mode 100644 flang/test/Lower/OpenMP/map-mapper.f90
diff --git a/flang/lib/Lower/OpenMP/Utils.cpp b/flang/lib/Lower/OpenMP/Utils.cpp
index 9971dc8e0b0014..af3a3df825fb88 100644
--- a/flang/lib/Lower/OpenMP/Utils.cpp
+++ b/flang/lib/Lower/OpenMP/Utils.cpp
@@ -125,7 +125,7 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
llvm::ArrayRef members,
mlir::ArrayAttr membersIndex, uint64_t mapType,
mlir::omp::VariableCaptureKind mapCaptureType, mlir::Type
retTy,
-bool partialMap) {
+bool partialMap, mlir::FlatSymbolRefAttr mapperId) {
if (auto boxTy = llvm::dyn_cast(baseAddr.getType())) {
baseAddr = builder.create(loc, baseAddr);
retTy = baseAddr.getType();
@@ -144,6 +144,7 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
mlir::omp::MapInfoOp op = builder.create(
loc, retTy, baseAddr, varType, varPtrPtr, members, membersIndex, bounds,
builder.getIntegerAttr(builder.getIntegerType(64, false), mapType),
+ mapperId,
builder.getAttr(mapCaptureType),
builder.getStringAttr(name), builder.getBoolAttr(partialMap));
return op;
diff --git a/flang/lib/Lower/OpenMP/Utils.h b/flang/lib/Lower/OpenMP/Utils.h
index f2e378443e5f29..3943eb633b04e3 100644
--- a/flang/lib/Lower/OpenMP/Utils.h
+++ b/flang/lib/Lower/OpenMP/Utils.h
@@ -116,7 +116,8 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
llvm::ArrayRef members,
mlir::ArrayAttr membersIndex, uint64_t mapType,
mlir::omp::VariableCaptureKind mapCaptureType, mlir::Type
retTy,
-bool partialMap = false);
+bool partialMap = false,
+mlir::FlatSymbolRefAttr mapperId = mlir::FlatSymbolRefAttr());
void insertChildMapInfoIntoParent(
Fortran::lower::AbstractConverter &converter,
diff --git a/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
b/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
index b08ccf3b88eb46..247ea6383cacda 100644
--- a/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
+++ b/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
@@ -171,6 +171,7 @@ class MapInfoFinalizationPass
baseAddrAddr, /*members=*/mlir::SmallVector{},
/*membersIndex=*/mlir::ArrayAttr{}, bounds,
builder.getIntegerAttr(builder.getIntegerType(64, false), mapType),
+/*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
mlir::omp::VariableCaptureKind::ByRef),
/*name=*/builder.getStringAttr(""),
@@ -316,7 +317,8 @@ class MapInfoFinalizationPass
builder.getIntegerAttr(
builder.getIntegerType(64, false),
getDescriptorMapType(op.getMapType().value_or(0), target)),
-op.getMapCaptureTypeAttr(), op.getNameAttr(),
+/*mapperId*/ mlir::FlatSymbolRefAttr(), op.getMapCaptureTypeAttr(),
+op.getNameAttr(),
/*partial_map=*/builder.getBoolAttr(false));
op.replaceAllUsesWith(newDescParentMapOp.getResult());
op->erase();
@@ -611,6 +613,7 @@ class MapInfoFinalizationPass
/*members=*/mlir::ValueRange{},
/*members_index=*/mlir::ArrayAttr{},
/*bounds=*/bounds, op.getMapTypeAttr(),
+ /*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
mlir::omp::VariableCaptureKind::ByRef),
builder.getStringAttr(op.getNameAttr().strref() + "." +
diff --git a/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
b/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
index c990bebcabde42..9f3b4dcdd898f4 100644
--- a/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
+++ b/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
@@ -89,6 +89,7 @@ class MapsForPrivatizedSymbolsPass
/*bounds=*/ValueRange{},
builder.getIntegerAttr(builder.getIntegerType(64, /*isSigned=*/false),
mapTypeTo),
+/*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
omp
[llvm-branch-commits] [flang] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
https://github.com/TIFitis updated
https://github.com/llvm/llvm-project/pull/120994
>From 002d7153f4029c02722b6a4f45bbde8fad83ba95 Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 20:53:47 +
Subject: [PATCH] Add mapper field to mapInfoOp.
---
flang/lib/Lower/OpenMP/Utils.cpp| 3 ++-
flang/lib/Lower/OpenMP/Utils.h | 3 ++-
flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp | 5 -
flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp | 1 +
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 2 ++
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp| 2 +-
mlir/test/Dialect/OpenMP/ops.mlir | 4 ++--
7 files changed, 14 insertions(+), 6 deletions(-)
diff --git a/flang/lib/Lower/OpenMP/Utils.cpp b/flang/lib/Lower/OpenMP/Utils.cpp
index 9971dc8e0b0014..af3a3df825fb88 100644
--- a/flang/lib/Lower/OpenMP/Utils.cpp
+++ b/flang/lib/Lower/OpenMP/Utils.cpp
@@ -125,7 +125,7 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
llvm::ArrayRef members,
mlir::ArrayAttr membersIndex, uint64_t mapType,
mlir::omp::VariableCaptureKind mapCaptureType, mlir::Type
retTy,
-bool partialMap) {
+bool partialMap, mlir::FlatSymbolRefAttr mapperId) {
if (auto boxTy = llvm::dyn_cast(baseAddr.getType())) {
baseAddr = builder.create(loc, baseAddr);
retTy = baseAddr.getType();
@@ -144,6 +144,7 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
mlir::omp::MapInfoOp op = builder.create(
loc, retTy, baseAddr, varType, varPtrPtr, members, membersIndex, bounds,
builder.getIntegerAttr(builder.getIntegerType(64, false), mapType),
+ mapperId,
builder.getAttr(mapCaptureType),
builder.getStringAttr(name), builder.getBoolAttr(partialMap));
return op;
diff --git a/flang/lib/Lower/OpenMP/Utils.h b/flang/lib/Lower/OpenMP/Utils.h
index f2e378443e5f29..3943eb633b04e3 100644
--- a/flang/lib/Lower/OpenMP/Utils.h
+++ b/flang/lib/Lower/OpenMP/Utils.h
@@ -116,7 +116,8 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location
loc,
llvm::ArrayRef members,
mlir::ArrayAttr membersIndex, uint64_t mapType,
mlir::omp::VariableCaptureKind mapCaptureType, mlir::Type
retTy,
-bool partialMap = false);
+bool partialMap = false,
+mlir::FlatSymbolRefAttr mapperId = mlir::FlatSymbolRefAttr());
void insertChildMapInfoIntoParent(
Fortran::lower::AbstractConverter &converter,
diff --git a/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
b/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
index b08ccf3b88eb46..247ea6383cacda 100644
--- a/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
+++ b/flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp
@@ -171,6 +171,7 @@ class MapInfoFinalizationPass
baseAddrAddr, /*members=*/mlir::SmallVector{},
/*membersIndex=*/mlir::ArrayAttr{}, bounds,
builder.getIntegerAttr(builder.getIntegerType(64, false), mapType),
+/*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
mlir::omp::VariableCaptureKind::ByRef),
/*name=*/builder.getStringAttr(""),
@@ -316,7 +317,8 @@ class MapInfoFinalizationPass
builder.getIntegerAttr(
builder.getIntegerType(64, false),
getDescriptorMapType(op.getMapType().value_or(0), target)),
-op.getMapCaptureTypeAttr(), op.getNameAttr(),
+/*mapperId*/ mlir::FlatSymbolRefAttr(), op.getMapCaptureTypeAttr(),
+op.getNameAttr(),
/*partial_map=*/builder.getBoolAttr(false));
op.replaceAllUsesWith(newDescParentMapOp.getResult());
op->erase();
@@ -611,6 +613,7 @@ class MapInfoFinalizationPass
/*members=*/mlir::ValueRange{},
/*members_index=*/mlir::ArrayAttr{},
/*bounds=*/bounds, op.getMapTypeAttr(),
+ /*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
mlir::omp::VariableCaptureKind::ByRef),
builder.getStringAttr(op.getNameAttr().strref() + "." +
diff --git a/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
b/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
index c990bebcabde42..9f3b4dcdd898f4 100644
--- a/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
+++ b/flang/lib/Optimizer/OpenMP/MapsForPrivatizedSymbols.cpp
@@ -89,6 +89,7 @@ class MapsForPrivatizedSymbolsPass
/*bounds=*/ValueRange{},
builder.getIntegerAttr(builder.getIntegerType(64, /*isSigned=*/false),
mapTypeTo),
+/*mapperId*/ mlir::FlatSymbolRefAttr(),
builder.getAttr(
omp::VariableCaptureKind::ByRef),
StringAttr(), builder.getBoolAttr(false));
diff --git a/
[llvm-branch-commits] [clang] [llvm] [Coverage] Make additional counters available for BranchRegion. NFC. (PR #120930)
https://github.com/Fznamznon edited https://github.com/llvm/llvm-project/pull/120930 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [flang] [MLIR][OpenMP] Add Lowering support for OpenMP custom mappers in map clause (PR #121001)
https://github.com/TIFitis created
https://github.com/llvm/llvm-project/pull/121001
Add Lowering support for OpenMP mapper field in mapInfoOp.
Depends on #120994.
>From 30611f81421028ccc8fceed006e06f8afc6848b9 Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 19:05:30 +
Subject: [PATCH 1/2] Add mapper field to mapInfoOp.
---
flang/test/Lower/OpenMP/map-mapper.f90 | 15 +++
1 file changed, 15 insertions(+)
create mode 100644 flang/test/Lower/OpenMP/map-mapper.f90
diff --git a/flang/test/Lower/OpenMP/map-mapper.f90
b/flang/test/Lower/OpenMP/map-mapper.f90
new file mode 100644
index 00..e7fded3034406e
--- /dev/null
+++ b/flang/test/Lower/OpenMP/map-mapper.f90
@@ -0,0 +1,15 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 %s -o - | FileCheck
%s
+program p
+ integer, parameter :: n = 256
+ real(8) :: a(256)
+ type t1
+ integer :: x
+ end type t1
+ !$omp declare mapper(xx : t1 :: nn) map(nn, nn%x)
+ !$omp target map(mapper(xx), from:a)
+!CHECK: not yet implemented: Support for mapper modifiers is not implemented
yet
+ do i = 1, n
+ a(i) = 4.2
+ end do
+ !$omp end target
+end program p
>From ea048f78cdec2ea749562f17ae6d3614c2ef1f46 Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 21:13:42 +
Subject: [PATCH 2/2] Add flang lowering changes for mapper field in map
clause.
---
flang/lib/Lower/OpenMP/ClauseProcessor.cpp | 32 +
flang/lib/Lower/OpenMP/ClauseProcessor.h| 3 +-
flang/test/Lower/OpenMP/Todo/map-mapper.f90 | 16 ---
flang/test/Lower/OpenMP/map-mapper.f90 | 20 +
4 files changed, 43 insertions(+), 28 deletions(-)
delete mode 100644 flang/test/Lower/OpenMP/Todo/map-mapper.f90
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 3c9831120351ee..0bc9f4919330e4 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -936,8 +936,10 @@ void ClauseProcessor::processMapObjects(
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
std::map &parentMemberIndices,
llvm::SmallVectorImpl &mapVars,
-llvm::SmallVectorImpl &mapSyms) const {
+llvm::SmallVectorImpl &mapSyms,
+std::string mapperIdName) const {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::FlatSymbolRefAttr mapperId;
for (const omp::Object &object : objects) {
llvm::SmallVector bounds;
@@ -970,6 +972,20 @@ void ClauseProcessor::processMapObjects(
}
}
+if (!mapperIdName.empty()) {
+ if (mapperIdName == "default") {
+auto &typeSpec = object.sym()->owner().IsDerivedType()
+ ? *object.sym()->owner().derivedTypeSpec()
+ : object.sym()->GetType()->derivedTypeSpec();
+mapperIdName = typeSpec.name().ToString() + ".default";
+mapperIdName = converter.mangleName(mapperIdName,
*typeSpec.GetScope());
+ }
+ assert(converter.getMLIRSymbolTable()->lookup(mapperIdName) &&
+ "mapper not found");
+ mapperId = mlir::FlatSymbolRefAttr::get(&converter.getMLIRContext(),
+ mapperIdName);
+ mapperIdName.clear();
+}
// Explicit map captures are captured ByRef by default,
// optimisation passes may alter this to ByCopy or other capture
// types to optimise
@@ -983,7 +999,8 @@ void ClauseProcessor::processMapObjects(
static_cast<
std::underlying_type_t>(
mapTypeBits),
-mlir::omp::VariableCaptureKind::ByRef, baseOp.getType());
+mlir::omp::VariableCaptureKind::ByRef, baseOp.getType(), false,
+mapperId);
if (parentObj.has_value()) {
parentMemberIndices[parentObj.value()].addChildIndexAndMapToParent(
@@ -1014,6 +1031,7 @@ bool ClauseProcessor::processMap(
const auto &[mapType, typeMods, mappers, iterator, objects] = clause.t;
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits =
llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_NONE;
+std::string mapperIdName;
// If the map type is specified, then process it else Tofrom is the
// default.
Map::MapType type = mapType.value_or(Map::MapType::Tofrom);
@@ -1057,13 +1075,17 @@ bool ClauseProcessor::processMap(
"Support for iterator modifiers is not implemented yet");
}
if (mappers) {
- TODO(currentLocation,
- "Support for mapper modifiers is not implemented yet");
+ assert(mappers->size() == 1 && "more than one mapper");
+ mapperIdName = mappers->front().v.id().symbol->name().ToString();
+ if (mapperIdName != "default")
+mapperIdName = converter.mangleName(
+mapperIdName, mappers->front().v.id().symbol->owner());
}
processMapObjects(stmtCtx, clauseLocation,
std::get(clause.t), mapTypeBits,
-
[llvm-branch-commits] [flang] [mlir] [MLIR][OpenMP] Add OMP Mapper field to MapInfoOp (PR #120994)
https://github.com/TIFitis edited https://github.com/llvm/llvm-project/pull/120994 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [flang] [MLIR][OpenMP] Add Lowering support for OpenMP custom mappers in map clause (PR #121001)
llvmbot wrote:
@llvm/pr-subscribers-flang-fir-hlfir
Author: Akash Banerjee (TIFitis)
Changes
Add Lowering support for OpenMP mapper field in mapInfoOp.
Depends on #120994.
---
Full diff: https://github.com/llvm/llvm-project/pull/121001.diff
4 Files Affected:
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.cpp (+27-5)
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.h (+2-1)
- (removed) flang/test/Lower/OpenMP/Todo/map-mapper.f90 (-16)
- (added) flang/test/Lower/OpenMP/map-mapper.f90 (+23)
``diff
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 3c9831120351ee..0bc9f4919330e4 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -936,8 +936,10 @@ void ClauseProcessor::processMapObjects(
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
std::map &parentMemberIndices,
llvm::SmallVectorImpl &mapVars,
-llvm::SmallVectorImpl &mapSyms) const {
+llvm::SmallVectorImpl &mapSyms,
+std::string mapperIdName) const {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::FlatSymbolRefAttr mapperId;
for (const omp::Object &object : objects) {
llvm::SmallVector bounds;
@@ -970,6 +972,20 @@ void ClauseProcessor::processMapObjects(
}
}
+if (!mapperIdName.empty()) {
+ if (mapperIdName == "default") {
+auto &typeSpec = object.sym()->owner().IsDerivedType()
+ ? *object.sym()->owner().derivedTypeSpec()
+ : object.sym()->GetType()->derivedTypeSpec();
+mapperIdName = typeSpec.name().ToString() + ".default";
+mapperIdName = converter.mangleName(mapperIdName,
*typeSpec.GetScope());
+ }
+ assert(converter.getMLIRSymbolTable()->lookup(mapperIdName) &&
+ "mapper not found");
+ mapperId = mlir::FlatSymbolRefAttr::get(&converter.getMLIRContext(),
+ mapperIdName);
+ mapperIdName.clear();
+}
// Explicit map captures are captured ByRef by default,
// optimisation passes may alter this to ByCopy or other capture
// types to optimise
@@ -983,7 +999,8 @@ void ClauseProcessor::processMapObjects(
static_cast<
std::underlying_type_t>(
mapTypeBits),
-mlir::omp::VariableCaptureKind::ByRef, baseOp.getType());
+mlir::omp::VariableCaptureKind::ByRef, baseOp.getType(), false,
+mapperId);
if (parentObj.has_value()) {
parentMemberIndices[parentObj.value()].addChildIndexAndMapToParent(
@@ -1014,6 +1031,7 @@ bool ClauseProcessor::processMap(
const auto &[mapType, typeMods, mappers, iterator, objects] = clause.t;
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits =
llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_NONE;
+std::string mapperIdName;
// If the map type is specified, then process it else Tofrom is the
// default.
Map::MapType type = mapType.value_or(Map::MapType::Tofrom);
@@ -1057,13 +1075,17 @@ bool ClauseProcessor::processMap(
"Support for iterator modifiers is not implemented yet");
}
if (mappers) {
- TODO(currentLocation,
- "Support for mapper modifiers is not implemented yet");
+ assert(mappers->size() == 1 && "more than one mapper");
+ mapperIdName = mappers->front().v.id().symbol->name().ToString();
+ if (mapperIdName != "default")
+mapperIdName = converter.mangleName(
+mapperIdName, mappers->front().v.id().symbol->owner());
}
processMapObjects(stmtCtx, clauseLocation,
std::get(clause.t), mapTypeBits,
- parentMemberIndices, result.mapVars, *ptrMapSyms);
+ parentMemberIndices, result.mapVars, *ptrMapSyms,
+ mapperIdName);
};
bool clauseFound = findRepeatableClause(process);
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h
b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 7b047d4a7567ad..34eebf16ca17b0 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -171,7 +171,8 @@ class ClauseProcessor {
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
std::map &parentMemberIndices,
llvm::SmallVectorImpl &mapVars,
- llvm::SmallVectorImpl &mapSyms) const;
+ llvm::SmallVectorImpl &mapSyms,
+ std::string mapperIdName = "") const;
lower::AbstractConverter &converter;
semantics::SemanticsContext &semaCtx;
diff --git a/flang/test/Lower/OpenMP/Todo/map-mapper.f90
b/flang/test/Lower/OpenMP/Todo/map-mapper.f90
deleted file mode 100644
index 9554ffd5fda7bd..00
--- a/flang/test/Lower/OpenMP/Todo/map-mapper.f90
+++ /dev/null
@@ -1,16 +0,0 @@
-! RUN: not %flang_fc1 -emit-fir -fopenmp -fopenmp-version=50 %s 2>&1 |
FileCheck %s
-program p
- integer, parameter :: n = 256
- rea
[llvm-branch-commits] [flang] [MLIR][OpenMP] Add Lowering support for OpenMP custom mappers in map clause (PR #121001)
llvmbot wrote:
@llvm/pr-subscribers-flang-openmp
Author: Akash Banerjee (TIFitis)
Changes
Add Lowering support for OpenMP mapper field in mapInfoOp.
Depends on #120994.
---
Full diff: https://github.com/llvm/llvm-project/pull/121001.diff
4 Files Affected:
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.cpp (+27-5)
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.h (+2-1)
- (removed) flang/test/Lower/OpenMP/Todo/map-mapper.f90 (-16)
- (added) flang/test/Lower/OpenMP/map-mapper.f90 (+23)
``diff
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 3c9831120351ee..0bc9f4919330e4 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -936,8 +936,10 @@ void ClauseProcessor::processMapObjects(
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
std::map &parentMemberIndices,
llvm::SmallVectorImpl &mapVars,
-llvm::SmallVectorImpl &mapSyms) const {
+llvm::SmallVectorImpl &mapSyms,
+std::string mapperIdName) const {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::FlatSymbolRefAttr mapperId;
for (const omp::Object &object : objects) {
llvm::SmallVector bounds;
@@ -970,6 +972,20 @@ void ClauseProcessor::processMapObjects(
}
}
+if (!mapperIdName.empty()) {
+ if (mapperIdName == "default") {
+auto &typeSpec = object.sym()->owner().IsDerivedType()
+ ? *object.sym()->owner().derivedTypeSpec()
+ : object.sym()->GetType()->derivedTypeSpec();
+mapperIdName = typeSpec.name().ToString() + ".default";
+mapperIdName = converter.mangleName(mapperIdName,
*typeSpec.GetScope());
+ }
+ assert(converter.getMLIRSymbolTable()->lookup(mapperIdName) &&
+ "mapper not found");
+ mapperId = mlir::FlatSymbolRefAttr::get(&converter.getMLIRContext(),
+ mapperIdName);
+ mapperIdName.clear();
+}
// Explicit map captures are captured ByRef by default,
// optimisation passes may alter this to ByCopy or other capture
// types to optimise
@@ -983,7 +999,8 @@ void ClauseProcessor::processMapObjects(
static_cast<
std::underlying_type_t>(
mapTypeBits),
-mlir::omp::VariableCaptureKind::ByRef, baseOp.getType());
+mlir::omp::VariableCaptureKind::ByRef, baseOp.getType(), false,
+mapperId);
if (parentObj.has_value()) {
parentMemberIndices[parentObj.value()].addChildIndexAndMapToParent(
@@ -1014,6 +1031,7 @@ bool ClauseProcessor::processMap(
const auto &[mapType, typeMods, mappers, iterator, objects] = clause.t;
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits =
llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_NONE;
+std::string mapperIdName;
// If the map type is specified, then process it else Tofrom is the
// default.
Map::MapType type = mapType.value_or(Map::MapType::Tofrom);
@@ -1057,13 +1075,17 @@ bool ClauseProcessor::processMap(
"Support for iterator modifiers is not implemented yet");
}
if (mappers) {
- TODO(currentLocation,
- "Support for mapper modifiers is not implemented yet");
+ assert(mappers->size() == 1 && "more than one mapper");
+ mapperIdName = mappers->front().v.id().symbol->name().ToString();
+ if (mapperIdName != "default")
+mapperIdName = converter.mangleName(
+mapperIdName, mappers->front().v.id().symbol->owner());
}
processMapObjects(stmtCtx, clauseLocation,
std::get(clause.t), mapTypeBits,
- parentMemberIndices, result.mapVars, *ptrMapSyms);
+ parentMemberIndices, result.mapVars, *ptrMapSyms,
+ mapperIdName);
};
bool clauseFound = findRepeatableClause(process);
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h
b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 7b047d4a7567ad..34eebf16ca17b0 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -171,7 +171,8 @@ class ClauseProcessor {
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
std::map &parentMemberIndices,
llvm::SmallVectorImpl &mapVars,
- llvm::SmallVectorImpl &mapSyms) const;
+ llvm::SmallVectorImpl &mapSyms,
+ std::string mapperIdName = "") const;
lower::AbstractConverter &converter;
semantics::SemanticsContext &semaCtx;
diff --git a/flang/test/Lower/OpenMP/Todo/map-mapper.f90
b/flang/test/Lower/OpenMP/Todo/map-mapper.f90
deleted file mode 100644
index 9554ffd5fda7bd..00
--- a/flang/test/Lower/OpenMP/Todo/map-mapper.f90
+++ /dev/null
@@ -1,16 +0,0 @@
-! RUN: not %flang_fc1 -emit-fir -fopenmp -fopenmp-version=50 %s 2>&1 |
FileCheck %s
-program p
- integer, parameter :: n = 256
- real(8
[llvm-branch-commits] [flang] [mlir] [MLIR][OpenMP] Add conversion support from FIR to LLVM Dialect for OMP DeclareMapper (PR #121005)
https://github.com/TIFitis created
https://github.com/llvm/llvm-project/pull/121005
Add conversion support from FIR to LLVM Dialect for OMP DeclareMapper.
>From 8a90d208ace9b8fe34f427f6268968e8b4b08976 Mon Sep 17 00:00:00 2001
From: Akash Banerjee
Date: Mon, 23 Dec 2024 21:50:03 +
Subject: [PATCH] Add OpenMP to LLVM dialect conversion support for
DeclareMapperOp.
---
.../Fir/convert-to-llvm-openmp-and-fir.fir| 27 --
.../Conversion/OpenMPToLLVM/OpenMPToLLVM.cpp | 50 +++
.../OpenMPToLLVM/convert-to-llvmir.mlir | 13 +
3 files changed, 75 insertions(+), 15 deletions(-)
diff --git a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
index 8e4e1fe824d9f5..82f2aea3ad983c 100644
--- a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
+++ b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
@@ -936,9 +936,9 @@ func.func @omp_map_info_descriptor_type_conversion(%arg0 :
!fir.ref>, i32)
map_clauses(tofrom) capture(ByRef) -> !fir.llvm_ptr> {name = ""}
// CHECK: %[[DESC_MAP:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<(ptr, i64, i32, i8, i8, i8, i8)>) map_clauses(always, delete)
capture(ByRef) members(%[[MEMBER_MAP]] : [0] : !llvm.ptr) -> !llvm.ptr {name =
""}
%2 = omp.map.info var_ptr(%arg0 : !fir.ref>>,
!fir.box>) map_clauses(always, delete) capture(ByRef) members(%1
: [0] : !fir.llvm_ptr>) -> !fir.ref>>
{name = ""}
- // CHECK: omp.target_exit_data map_entries(%[[DESC_MAP]] : !llvm.ptr)
+ // CHECK: omp.target_exit_data map_entries(%[[DESC_MAP]] : !llvm.ptr)
omp.target_exit_data map_entries(%2 : !fir.ref>>)
- return
+ return
}
// -
@@ -956,8 +956,8 @@ func.func
@omp_map_info_derived_type_explicit_member_conversion(%arg0 : !fir.ref
%3 = fir.field_index real,
!fir.type<_QFderived_type{real:f32,array:!fir.array<10xi32>,int:i32}>
%4 = fir.coordinate_of %arg0, %3 :
(!fir.ref,int:i32}>>,
!fir.field) -> !fir.ref
// CHECK: %[[MAP_MEMBER_2:.*]] = omp.map.info var_ptr(%[[GEP_2]] :
!llvm.ptr, f32) map_clauses(tofrom) capture(ByRef) -> !llvm.ptr {name =
"dtype%real"}
- %5 = omp.map.info var_ptr(%4 : !fir.ref, f32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "dtype%real"}
- // CHECK: %[[MAP_PARENT:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<"_QFderived_type", (f32, array<10 x i32>, i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[MAP_MEMBER_1]], %[[MAP_MEMBER_2]]
: [2], [0] : !llvm.ptr, !llvm.ptr) -> !llvm.ptr {name = "dtype", partial_map =
true}
+ %5 = omp.map.info var_ptr(%4 : !fir.ref, f32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "dtype%real"}
+ // CHECK: %[[MAP_PARENT:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<"_QFderived_type", (f32, array<10 x i32>, i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[MAP_MEMBER_1]], %[[MAP_MEMBER_2]]
: [2], [0] : !llvm.ptr, !llvm.ptr) -> !llvm.ptr {name = "dtype", partial_map =
true}
%6 = omp.map.info var_ptr(%arg0 :
!fir.ref,int:i32}>>,
!fir.type<_QFderived_type{real:f32,array:!fir.array<10xi32>,int:i32}>)
map_clauses(tofrom) capture(ByRef) members(%2, %5 : [2], [0] : !fir.ref,
!fir.ref) ->
!fir.ref,int:i32}>>
{name = "dtype", partial_map = true}
// CHECK: omp.target map_entries(%[[MAP_MEMBER_1]] -> %[[ARG_1:.*]],
%[[MAP_MEMBER_2]] -> %[[ARG_2:.*]], %[[MAP_PARENT]] -> %[[ARG_3:.*]] :
!llvm.ptr, !llvm.ptr, !llvm.ptr) {
omp.target map_entries(%2 -> %arg1, %5 -> %arg2, %6 -> %arg3 :
!fir.ref, !fir.ref,
!fir.ref,int:i32}>>)
{
@@ -1275,3 +1275,22 @@ func.func @map_nested_dtype_alloca_mem2(%arg0 :
!fir.ref {
+omp.declare_mapper @my_mapper : !fir.type<_QFdeclare_mapperTmy_type{data:i32}>
{
+// CHECK: ^bb0(%[[VAL_0:.*]]: !llvm.ptr):
+^bb0(%0: !fir.ref>):
+// CHECK: %[[VAL_1:.*]] = llvm.mlir.constant(0 : i32) : i32
+ %1 = fir.field_index data, !fir.type<_QFdeclare_mapperTmy_type{data:i32}>
+// CHECK: %[[VAL_2:.*]] = llvm.getelementptr %[[VAL_0]][0, 0] :
(!llvm.ptr) -> !llvm.ptr, !llvm.struct<"_QFdeclare_mapperTmy_type", (i32)>
+ %2 = fir.coordinate_of %0, %1 :
(!fir.ref>, !fir.field) ->
!fir.ref
+// CHECK: %[[VAL_3:.*]] = omp.map.info var_ptr(%[[VAL_2]] :
!llvm.ptr, i32) map_clauses(tofrom) capture(ByRef) -> !llvm.ptr {name =
"var%[[VAL_4:.*]]"}
+ %3 = omp.map.info var_ptr(%2 : !fir.ref, i32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "var%data"}
+// CHECK: %[[VAL_5:.*]] = omp.map.info var_ptr(%[[VAL_0]] :
!llvm.ptr, !llvm.struct<"_QFdeclare_mapperTmy_type", (i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[VAL_3]] : [0] : !llvm.ptr) ->
!llvm.ptr {name = "var", partial_map = true}
+ %4 = omp.map.info var_ptr(%0 :
!fir.ref>,
!fir.type<_QFdeclare_mapperTmy_type{data:i32}>) map_clauses(tofrom)
capture(ByRef) members(%3 : [0] : !fir.ref) ->
!fir.ref> {name = "var",
partial_map = true}
+// CHECK: omp.declare_mapper_info map_entries
[llvm-branch-commits] [flang] [mlir] [MLIR][OpenMP] Add conversion support from FIR to LLVM Dialect for OMP DeclareMapper (PR #121005)
llvmbot wrote:
@llvm/pr-subscribers-mlir
@llvm/pr-subscribers-mlir-openmp
Author: Akash Banerjee (TIFitis)
Changes
Add conversion support from FIR to LLVM Dialect for OMP DeclareMapper.
---
Full diff: https://github.com/llvm/llvm-project/pull/121005.diff
3 Files Affected:
- (modified) flang/test/Fir/convert-to-llvm-openmp-and-fir.fir (+23-4)
- (modified) mlir/lib/Conversion/OpenMPToLLVM/OpenMPToLLVM.cpp (+39-11)
- (modified) mlir/test/Conversion/OpenMPToLLVM/convert-to-llvmir.mlir (+13)
``diff
diff --git a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
index 8e4e1fe824d9f5..82f2aea3ad983c 100644
--- a/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
+++ b/flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
@@ -936,9 +936,9 @@ func.func @omp_map_info_descriptor_type_conversion(%arg0 :
!fir.ref>, i32)
map_clauses(tofrom) capture(ByRef) -> !fir.llvm_ptr> {name = ""}
// CHECK: %[[DESC_MAP:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<(ptr, i64, i32, i8, i8, i8, i8)>) map_clauses(always, delete)
capture(ByRef) members(%[[MEMBER_MAP]] : [0] : !llvm.ptr) -> !llvm.ptr {name =
""}
%2 = omp.map.info var_ptr(%arg0 : !fir.ref>>,
!fir.box>) map_clauses(always, delete) capture(ByRef) members(%1
: [0] : !fir.llvm_ptr>) -> !fir.ref>>
{name = ""}
- // CHECK: omp.target_exit_data map_entries(%[[DESC_MAP]] : !llvm.ptr)
+ // CHECK: omp.target_exit_data map_entries(%[[DESC_MAP]] : !llvm.ptr)
omp.target_exit_data map_entries(%2 : !fir.ref>>)
- return
+ return
}
// -
@@ -956,8 +956,8 @@ func.func
@omp_map_info_derived_type_explicit_member_conversion(%arg0 : !fir.ref
%3 = fir.field_index real,
!fir.type<_QFderived_type{real:f32,array:!fir.array<10xi32>,int:i32}>
%4 = fir.coordinate_of %arg0, %3 :
(!fir.ref,int:i32}>>,
!fir.field) -> !fir.ref
// CHECK: %[[MAP_MEMBER_2:.*]] = omp.map.info var_ptr(%[[GEP_2]] :
!llvm.ptr, f32) map_clauses(tofrom) capture(ByRef) -> !llvm.ptr {name =
"dtype%real"}
- %5 = omp.map.info var_ptr(%4 : !fir.ref, f32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "dtype%real"}
- // CHECK: %[[MAP_PARENT:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<"_QFderived_type", (f32, array<10 x i32>, i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[MAP_MEMBER_1]], %[[MAP_MEMBER_2]]
: [2], [0] : !llvm.ptr, !llvm.ptr) -> !llvm.ptr {name = "dtype", partial_map =
true}
+ %5 = omp.map.info var_ptr(%4 : !fir.ref, f32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "dtype%real"}
+ // CHECK: %[[MAP_PARENT:.*]] = omp.map.info var_ptr(%[[ARG_0]] : !llvm.ptr,
!llvm.struct<"_QFderived_type", (f32, array<10 x i32>, i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[MAP_MEMBER_1]], %[[MAP_MEMBER_2]]
: [2], [0] : !llvm.ptr, !llvm.ptr) -> !llvm.ptr {name = "dtype", partial_map =
true}
%6 = omp.map.info var_ptr(%arg0 :
!fir.ref,int:i32}>>,
!fir.type<_QFderived_type{real:f32,array:!fir.array<10xi32>,int:i32}>)
map_clauses(tofrom) capture(ByRef) members(%2, %5 : [2], [0] : !fir.ref,
!fir.ref) ->
!fir.ref,int:i32}>>
{name = "dtype", partial_map = true}
// CHECK: omp.target map_entries(%[[MAP_MEMBER_1]] -> %[[ARG_1:.*]],
%[[MAP_MEMBER_2]] -> %[[ARG_2:.*]], %[[MAP_PARENT]] -> %[[ARG_3:.*]] :
!llvm.ptr, !llvm.ptr, !llvm.ptr) {
omp.target map_entries(%2 -> %arg1, %5 -> %arg2, %6 -> %arg3 :
!fir.ref, !fir.ref,
!fir.ref,int:i32}>>)
{
@@ -1275,3 +1275,22 @@ func.func @map_nested_dtype_alloca_mem2(%arg0 :
!fir.ref {
+omp.declare_mapper @my_mapper : !fir.type<_QFdeclare_mapperTmy_type{data:i32}>
{
+// CHECK: ^bb0(%[[VAL_0:.*]]: !llvm.ptr):
+^bb0(%0: !fir.ref>):
+// CHECK: %[[VAL_1:.*]] = llvm.mlir.constant(0 : i32) : i32
+ %1 = fir.field_index data, !fir.type<_QFdeclare_mapperTmy_type{data:i32}>
+// CHECK: %[[VAL_2:.*]] = llvm.getelementptr %[[VAL_0]][0, 0] :
(!llvm.ptr) -> !llvm.ptr, !llvm.struct<"_QFdeclare_mapperTmy_type", (i32)>
+ %2 = fir.coordinate_of %0, %1 :
(!fir.ref>, !fir.field) ->
!fir.ref
+// CHECK: %[[VAL_3:.*]] = omp.map.info var_ptr(%[[VAL_2]] :
!llvm.ptr, i32) map_clauses(tofrom) capture(ByRef) -> !llvm.ptr {name =
"var%[[VAL_4:.*]]"}
+ %3 = omp.map.info var_ptr(%2 : !fir.ref, i32) map_clauses(tofrom)
capture(ByRef) -> !fir.ref {name = "var%data"}
+// CHECK: %[[VAL_5:.*]] = omp.map.info var_ptr(%[[VAL_0]] :
!llvm.ptr, !llvm.struct<"_QFdeclare_mapperTmy_type", (i32)>)
map_clauses(tofrom) capture(ByRef) members(%[[VAL_3]] : [0] : !llvm.ptr) ->
!llvm.ptr {name = "var", partial_map = true}
+ %4 = omp.map.info var_ptr(%0 :
!fir.ref>,
!fir.type<_QFdeclare_mapperTmy_type{data:i32}>) map_clauses(tofrom)
capture(ByRef) members(%3 : [0] : !fir.ref) ->
!fir.ref> {name = "var",
partial_map = true}
+// CHECK: omp.declare_mapper_info map_entries(%[[VAL_5]], %[[VAL_3]]
: !llvm.ptr, !llvm.ptr)
+ omp.declare_mapper_info map_entries(%4, %3 :
!fir
[llvm-branch-commits] [llvm] 5337f1c - Revert "Reapply "[llvm]Add a simple Telemetry framework" (#120769) (#121003)"
Author: Vy Nguyen
Date: 2024-12-23T19:29:03-05:00
New Revision: 5337f1cc44dfa6b39c02eaf2a2dc3dec567018bc
URL:
https://github.com/llvm/llvm-project/commit/5337f1cc44dfa6b39c02eaf2a2dc3dec567018bc
DIFF:
https://github.com/llvm/llvm-project/commit/5337f1cc44dfa6b39c02eaf2a2dc3dec567018bc.diff
LOG: Revert "Reapply "[llvm]Add a simple Telemetry framework" (#120769)
(#121003)"
This reverts commit dbae7176a6ecf558dc5e92016cdda387c9d74d66.
Added:
Modified:
llvm/docs/UserGuides.rst
llvm/lib/CMakeLists.txt
llvm/unittests/CMakeLists.txt
Removed:
llvm/docs/Telemetry.rst
llvm/docs/llvm_telemetry_design.png
llvm/include/llvm/Telemetry/Telemetry.h
llvm/lib/Telemetry/CMakeLists.txt
llvm/lib/Telemetry/Telemetry.cpp
llvm/unittests/Telemetry/CMakeLists.txt
llvm/unittests/Telemetry/TelemetryTest.cpp
diff --git a/llvm/docs/Telemetry.rst b/llvm/docs/Telemetry.rst
deleted file mode 100644
index 4f30ae82b5628f..00
--- a/llvm/docs/Telemetry.rst
+++ /dev/null
@@ -1,257 +0,0 @@
-===
-Telemetry framework in LLVM
-===
-
-.. contents::
- :local:
-
-.. toctree::
- :hidden:
-
-Objective
-=
-
-Provides a common framework in LLVM for collecting various usage and
performance
-metrics.
-It is located at ``llvm/Telemetry/Telemetry.h``.
-
-Characteristics
-* Configurable and extensible by:
-
- * Tools: any tool that wants to use Telemetry can extend and customize it.
- * Vendors: Toolchain vendors can also provide custom implementation of the
-library, which could either override or extend the given tool's upstream
-implementation, to best fit their organization's usage and privacy models.
- * End users of such tool can also configure Telemetry (as allowed by their
-vendor).
-
-Important notes
-
-* There is no concrete implementation of a Telemetry library in upstream LLVM.
- We only provide the abstract API here. Any tool that wants telemetry will
- implement one.
-
- The rationale for this is that all the tools in LLVM are very
diff erent in
- what they care about (what/where/when to instrument data). Hence, it might
not
- be practical to have a single implementation.
- However, in the future, if we see enough common pattern, we can extract them
- into a shared place. This is TBD - contributions are welcome.
-
-* No implementation of Telemetry in upstream LLVM shall store any of the
- collected data due to privacy and security reasons:
-
- * Different organizations have
diff erent privacy models:
-
-* Which data is sensitive, which is not?
-* Whether it is acceptable for instrumented data to be stored anywhere?
- (to a local file, what not?)
-
- * Data ownership and data collection consents are hard to accommodate from
-LLVM developers' point of view:
-
-* E.g., data collected by Telemetry is not necessarily owned by the user
- of an LLVM tool with Telemetry enabled, hence the user's consent to data
- collection is not meaningful. On the other hand, LLVM developers have no
- reasonable ways to request consent from the "real" owners.
-
-
-High-level design
-=
-
-Key components
---
-
-The framework consists of four important classes:
-
-* ``llvm::telemetry::Manager``: The class responsible for collecting and
- transmitting telemetry data. This is the main point of interaction between
the
- framework and any tool that wants to enable telemetry.
-* ``llvm::telemetry::TelemetryInfo``: Data courier
-* ``llvm::telemetry::Destination``: Data sink to which the Telemetry framework
- sends data.
- Its implementation is transparent to the framework.
- It is up to the vendor to decide which pieces of data to forward and where
- to forward them to for their final storage.
-* ``llvm::telemetry::Config``: Configurations for the ``Manager``.
-
-.. image:: llvm_telemetry_design.png
-
-How to implement and interact with the API
---
-
-To use Telemetry in your tool, you need to provide a concrete implementation
of the ``Manager`` class and ``Destination``.
-
-1) Define a custom ``Serializer``, ``Manager``, ``Destination`` and optionally
a subclass of ``TelemetryInfo``
-
-.. code-block:: c++
-
- class JsonSerializer : public Serializer {
- public:
-json::Object *getOutputObject() { return Out.get(); }
-
-Error init() override {
- if (Started)
-return createStringError("Serializer already in use");
- started = true;
- Out = std::make_unique();
- return Error::success();
-}
-
-// Serialize the given value.
-void write(StringRef KeyName, bool Value) override {
- writeHelper(KeyName, Value);
-}
-
-void write(StringRef KeyName, int Value) override {
- writeHelper(KeyName, Value);
-}
-
-void wri
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/121012 This introduces `@llvm.dx.resource.load.rawbuffer` and generalizes the buffer load docs under DirectX/DXILResources. This resolves the "load" parts of #106188 >From 53bba63c6a55464ac1872dc60f39510eff426d64 Mon Sep 17 00:00:00 2001 From: Justin Bogner Date: Thu, 19 Dec 2024 16:33:48 -0700 Subject: [PATCH] [DirectX] Implement the resource.load.rawbuffer intrinsic This introduces `@llvm.dx.resource.load.rawbuffer` and generalizes the buffer load docs under DirectX/DXILResources. This resolves the "load" parts of #106188 --- llvm/docs/DirectX/DXILResources.rst | 151 ++-- llvm/include/llvm/IR/IntrinsicsDirectX.td | 4 + llvm/lib/Target/DirectX/DXIL.td | 19 ++ llvm/lib/Target/DirectX/DXILOpBuilder.cpp | 4 + llvm/lib/Target/DirectX/DXILOpLowering.cpp| 45 llvm/test/CodeGen/DirectX/BufferLoad-sm61.ll | 60 + .../CodeGen/DirectX/RawBufferLoad-error64.ll | 24 ++ llvm/test/CodeGen/DirectX/RawBufferLoad.ll| 232 ++ llvm/utils/TableGen/DXILEmitter.cpp | 4 +- 9 files changed, 516 insertions(+), 27 deletions(-) create mode 100644 llvm/test/CodeGen/DirectX/BufferLoad-sm61.ll create mode 100644 llvm/test/CodeGen/DirectX/RawBufferLoad-error64.ll create mode 100644 llvm/test/CodeGen/DirectX/RawBufferLoad.ll diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst index fafcb247f49c94..bcce1b52d9a485 100644 --- a/llvm/docs/DirectX/DXILResources.rst +++ b/llvm/docs/DirectX/DXILResources.rst @@ -321,36 +321,40 @@ Examples: 16-byte Loads, Samples, and Gathers --- -*relevant types: TypedBuffer, CBuffer, and Textures* - -TypedBuffer, CBuffer, and Texture loads, as well as samples and gathers, can -return 1 to 4 elements from the given resource, to a maximum of 16 bytes of -data. DXIL's modeling of this is influenced by DirectX and DXBC's history and -it generally treats these operations as returning 4 32-bit values. For 16-bit -elements the values are 16-bit values, and for 64-bit values the operations -return 4 32-bit integers and emit further code to construct the double. - -In DXIL, these operations return `ResRet`_ and `CBufRet`_ values, are structs -containing 4 elements of the same type, and in the case of `ResRet` a 5th -element that is used by the `CheckAccessFullyMapped`_ operation. - -In LLVM IR the intrinsics will return the contained type of the resource -instead. That is, ``llvm.dx.resource.load.typedbuffer`` from a -``Buffer`` would return a single float, from ``Buffer`` a vector -of 4 floats, and from ``Buffer`` a vector of two doubles, etc. The -operations are then expanded out to match DXIL's format during lowering. - -In order to support ``CheckAccessFullyMapped``, we need these intrinsics to -return an anonymous struct with element-0 being the contained type, and -element-1 being the ``i1`` result of a ``CheckAccessFullyMapped`` call. We -don't have a separate call to ``CheckAccessFullyMapped`` at all, since that's -the only operation that can possibly be done on this value. In practice this -may mean we insert a DXIL operation for the check when this was missing in the -HLSL source, but this actually matches DXC's behaviour in practice. +*relevant types: Buffers, CBuffers, and Textures* + +All load, sample, and gather operations in DXIL return a `ResRet`_ type, and +CBuffer loads return a similar `CBufRet`_ type. These types are structs +containing 4 elements of some basic type, and in the case of `ResRet` a 5th +element that is used by the `CheckAccessFullyMapped`_ operation. Some of these +operations, like `RawBufferLoad`_ include a mask and/or alignment that tell us +some information about how to interpret those four values. + +In the LLVM IR representations of these operations we instead return scalars or +vectors, but we keep the requirement that we only return up to 4 elements of a +basic type. This avoids some unnecessary casting and structure manipulation in +the intermediate format while also keeping lowering to DXIL straightforward. + +LLVM intrinsics that map to operations returning `ResRet` return an anonymous +struct with element-0 being the scalar or vector type, and element-1 being the +``i1`` result of a ``CheckAccessFullyMapped`` call. We don't have a separate +call to ``CheckAccessFullyMapped`` at all, since that's the only operation that +can possibly be done on this value. In practice this may mean we insert a DXIL +operation for the check when this was missing in the HLSL source, but this +actually matches DXC's behaviour in practice. + +For TypedBuffer and Texture, we map directly from the contained type of the +resource to the return value of the intrinsic. Since these resources are +constrained to contain only scalars and vectors of up to 4 elements, the +lowering to DXIL ops is generally straightforward. The one exception we ha
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
llvmbot wrote: @llvm/pr-subscribers-tablegen Author: Justin Bogner (bogner) Changes This introduces `@llvm.dx.resource.load.rawbuffer` and generalizes the buffer load docs under DirectX/DXILResources. This resolves the "load" parts of #106188 --- Patch is 28.22 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121012.diff 9 Files Affected: - (modified) llvm/docs/DirectX/DXILResources.rst (+125-26) - (modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+4) - (modified) llvm/lib/Target/DirectX/DXIL.td (+19) - (modified) llvm/lib/Target/DirectX/DXILOpBuilder.cpp (+4) - (modified) llvm/lib/Target/DirectX/DXILOpLowering.cpp (+45) - (added) llvm/test/CodeGen/DirectX/BufferLoad-sm61.ll (+60) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad-error64.ll (+24) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad.ll (+232) - (modified) llvm/utils/TableGen/DXILEmitter.cpp (+3-1) ``diff diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst index fafcb247f49c94..bcce1b52d9a485 100644 --- a/llvm/docs/DirectX/DXILResources.rst +++ b/llvm/docs/DirectX/DXILResources.rst @@ -321,36 +321,40 @@ Examples: 16-byte Loads, Samples, and Gathers --- -*relevant types: TypedBuffer, CBuffer, and Textures* - -TypedBuffer, CBuffer, and Texture loads, as well as samples and gathers, can -return 1 to 4 elements from the given resource, to a maximum of 16 bytes of -data. DXIL's modeling of this is influenced by DirectX and DXBC's history and -it generally treats these operations as returning 4 32-bit values. For 16-bit -elements the values are 16-bit values, and for 64-bit values the operations -return 4 32-bit integers and emit further code to construct the double. - -In DXIL, these operations return `ResRet`_ and `CBufRet`_ values, are structs -containing 4 elements of the same type, and in the case of `ResRet` a 5th -element that is used by the `CheckAccessFullyMapped`_ operation. - -In LLVM IR the intrinsics will return the contained type of the resource -instead. That is, ``llvm.dx.resource.load.typedbuffer`` from a -``Buffer`` would return a single float, from ``Buffer`` a vector -of 4 floats, and from ``Buffer`` a vector of two doubles, etc. The -operations are then expanded out to match DXIL's format during lowering. - -In order to support ``CheckAccessFullyMapped``, we need these intrinsics to -return an anonymous struct with element-0 being the contained type, and -element-1 being the ``i1`` result of a ``CheckAccessFullyMapped`` call. We -don't have a separate call to ``CheckAccessFullyMapped`` at all, since that's -the only operation that can possibly be done on this value. In practice this -may mean we insert a DXIL operation for the check when this was missing in the -HLSL source, but this actually matches DXC's behaviour in practice. +*relevant types: Buffers, CBuffers, and Textures* + +All load, sample, and gather operations in DXIL return a `ResRet`_ type, and +CBuffer loads return a similar `CBufRet`_ type. These types are structs +containing 4 elements of some basic type, and in the case of `ResRet` a 5th +element that is used by the `CheckAccessFullyMapped`_ operation. Some of these +operations, like `RawBufferLoad`_ include a mask and/or alignment that tell us +some information about how to interpret those four values. + +In the LLVM IR representations of these operations we instead return scalars or +vectors, but we keep the requirement that we only return up to 4 elements of a +basic type. This avoids some unnecessary casting and structure manipulation in +the intermediate format while also keeping lowering to DXIL straightforward. + +LLVM intrinsics that map to operations returning `ResRet` return an anonymous +struct with element-0 being the scalar or vector type, and element-1 being the +``i1`` result of a ``CheckAccessFullyMapped`` call. We don't have a separate +call to ``CheckAccessFullyMapped`` at all, since that's the only operation that +can possibly be done on this value. In practice this may mean we insert a DXIL +operation for the check when this was missing in the HLSL source, but this +actually matches DXC's behaviour in practice. + +For TypedBuffer and Texture, we map directly from the contained type of the +resource to the return value of the intrinsic. Since these resources are +constrained to contain only scalars and vectors of up to 4 elements, the +lowering to DXIL ops is generally straightforward. The one exception we have +here is that `double` types in the elements are special - these are allowed in +the LLVM intrinsics, but are lowered to pairs of `i32` followed by +``MakeDouble`` operations for DXIL. .. _ResRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#resource-operation-return-types .. _CBufRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#cbufferloadlegacy .. _CheckAccessFullyMap
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
llvmbot wrote: @llvm/pr-subscribers-llvm-ir Author: Justin Bogner (bogner) Changes This introduces `@llvm.dx.resource.load.rawbuffer` and generalizes the buffer load docs under DirectX/DXILResources. This resolves the "load" parts of #106188 --- Patch is 28.22 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121012.diff 9 Files Affected: - (modified) llvm/docs/DirectX/DXILResources.rst (+125-26) - (modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+4) - (modified) llvm/lib/Target/DirectX/DXIL.td (+19) - (modified) llvm/lib/Target/DirectX/DXILOpBuilder.cpp (+4) - (modified) llvm/lib/Target/DirectX/DXILOpLowering.cpp (+45) - (added) llvm/test/CodeGen/DirectX/BufferLoad-sm61.ll (+60) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad-error64.ll (+24) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad.ll (+232) - (modified) llvm/utils/TableGen/DXILEmitter.cpp (+3-1) ``diff diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst index fafcb247f49c94..bcce1b52d9a485 100644 --- a/llvm/docs/DirectX/DXILResources.rst +++ b/llvm/docs/DirectX/DXILResources.rst @@ -321,36 +321,40 @@ Examples: 16-byte Loads, Samples, and Gathers --- -*relevant types: TypedBuffer, CBuffer, and Textures* - -TypedBuffer, CBuffer, and Texture loads, as well as samples and gathers, can -return 1 to 4 elements from the given resource, to a maximum of 16 bytes of -data. DXIL's modeling of this is influenced by DirectX and DXBC's history and -it generally treats these operations as returning 4 32-bit values. For 16-bit -elements the values are 16-bit values, and for 64-bit values the operations -return 4 32-bit integers and emit further code to construct the double. - -In DXIL, these operations return `ResRet`_ and `CBufRet`_ values, are structs -containing 4 elements of the same type, and in the case of `ResRet` a 5th -element that is used by the `CheckAccessFullyMapped`_ operation. - -In LLVM IR the intrinsics will return the contained type of the resource -instead. That is, ``llvm.dx.resource.load.typedbuffer`` from a -``Buffer`` would return a single float, from ``Buffer`` a vector -of 4 floats, and from ``Buffer`` a vector of two doubles, etc. The -operations are then expanded out to match DXIL's format during lowering. - -In order to support ``CheckAccessFullyMapped``, we need these intrinsics to -return an anonymous struct with element-0 being the contained type, and -element-1 being the ``i1`` result of a ``CheckAccessFullyMapped`` call. We -don't have a separate call to ``CheckAccessFullyMapped`` at all, since that's -the only operation that can possibly be done on this value. In practice this -may mean we insert a DXIL operation for the check when this was missing in the -HLSL source, but this actually matches DXC's behaviour in practice. +*relevant types: Buffers, CBuffers, and Textures* + +All load, sample, and gather operations in DXIL return a `ResRet`_ type, and +CBuffer loads return a similar `CBufRet`_ type. These types are structs +containing 4 elements of some basic type, and in the case of `ResRet` a 5th +element that is used by the `CheckAccessFullyMapped`_ operation. Some of these +operations, like `RawBufferLoad`_ include a mask and/or alignment that tell us +some information about how to interpret those four values. + +In the LLVM IR representations of these operations we instead return scalars or +vectors, but we keep the requirement that we only return up to 4 elements of a +basic type. This avoids some unnecessary casting and structure manipulation in +the intermediate format while also keeping lowering to DXIL straightforward. + +LLVM intrinsics that map to operations returning `ResRet` return an anonymous +struct with element-0 being the scalar or vector type, and element-1 being the +``i1`` result of a ``CheckAccessFullyMapped`` call. We don't have a separate +call to ``CheckAccessFullyMapped`` at all, since that's the only operation that +can possibly be done on this value. In practice this may mean we insert a DXIL +operation for the check when this was missing in the HLSL source, but this +actually matches DXC's behaviour in practice. + +For TypedBuffer and Texture, we map directly from the contained type of the +resource to the return value of the intrinsic. Since these resources are +constrained to contain only scalars and vectors of up to 4 elements, the +lowering to DXIL ops is generally straightforward. The one exception we have +here is that `double` types in the elements are special - these are allowed in +the LLVM intrinsics, but are lowered to pairs of `i32` followed by +``MakeDouble`` operations for DXIL. .. _ResRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#resource-operation-return-types .. _CBufRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#cbufferloadlegacy .. _CheckAccessFullyMapp
[llvm-branch-commits] [llvm] [DirectX] Implement the resource.load.rawbuffer intrinsic (PR #121012)
llvmbot wrote: @llvm/pr-subscribers-backend-directx Author: Justin Bogner (bogner) Changes This introduces `@llvm.dx.resource.load.rawbuffer` and generalizes the buffer load docs under DirectX/DXILResources. This resolves the "load" parts of #106188 --- Patch is 28.22 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121012.diff 9 Files Affected: - (modified) llvm/docs/DirectX/DXILResources.rst (+125-26) - (modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+4) - (modified) llvm/lib/Target/DirectX/DXIL.td (+19) - (modified) llvm/lib/Target/DirectX/DXILOpBuilder.cpp (+4) - (modified) llvm/lib/Target/DirectX/DXILOpLowering.cpp (+45) - (added) llvm/test/CodeGen/DirectX/BufferLoad-sm61.ll (+60) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad-error64.ll (+24) - (added) llvm/test/CodeGen/DirectX/RawBufferLoad.ll (+232) - (modified) llvm/utils/TableGen/DXILEmitter.cpp (+3-1) ``diff diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst index fafcb247f49c94..bcce1b52d9a485 100644 --- a/llvm/docs/DirectX/DXILResources.rst +++ b/llvm/docs/DirectX/DXILResources.rst @@ -321,36 +321,40 @@ Examples: 16-byte Loads, Samples, and Gathers --- -*relevant types: TypedBuffer, CBuffer, and Textures* - -TypedBuffer, CBuffer, and Texture loads, as well as samples and gathers, can -return 1 to 4 elements from the given resource, to a maximum of 16 bytes of -data. DXIL's modeling of this is influenced by DirectX and DXBC's history and -it generally treats these operations as returning 4 32-bit values. For 16-bit -elements the values are 16-bit values, and for 64-bit values the operations -return 4 32-bit integers and emit further code to construct the double. - -In DXIL, these operations return `ResRet`_ and `CBufRet`_ values, are structs -containing 4 elements of the same type, and in the case of `ResRet` a 5th -element that is used by the `CheckAccessFullyMapped`_ operation. - -In LLVM IR the intrinsics will return the contained type of the resource -instead. That is, ``llvm.dx.resource.load.typedbuffer`` from a -``Buffer`` would return a single float, from ``Buffer`` a vector -of 4 floats, and from ``Buffer`` a vector of two doubles, etc. The -operations are then expanded out to match DXIL's format during lowering. - -In order to support ``CheckAccessFullyMapped``, we need these intrinsics to -return an anonymous struct with element-0 being the contained type, and -element-1 being the ``i1`` result of a ``CheckAccessFullyMapped`` call. We -don't have a separate call to ``CheckAccessFullyMapped`` at all, since that's -the only operation that can possibly be done on this value. In practice this -may mean we insert a DXIL operation for the check when this was missing in the -HLSL source, but this actually matches DXC's behaviour in practice. +*relevant types: Buffers, CBuffers, and Textures* + +All load, sample, and gather operations in DXIL return a `ResRet`_ type, and +CBuffer loads return a similar `CBufRet`_ type. These types are structs +containing 4 elements of some basic type, and in the case of `ResRet` a 5th +element that is used by the `CheckAccessFullyMapped`_ operation. Some of these +operations, like `RawBufferLoad`_ include a mask and/or alignment that tell us +some information about how to interpret those four values. + +In the LLVM IR representations of these operations we instead return scalars or +vectors, but we keep the requirement that we only return up to 4 elements of a +basic type. This avoids some unnecessary casting and structure manipulation in +the intermediate format while also keeping lowering to DXIL straightforward. + +LLVM intrinsics that map to operations returning `ResRet` return an anonymous +struct with element-0 being the scalar or vector type, and element-1 being the +``i1`` result of a ``CheckAccessFullyMapped`` call. We don't have a separate +call to ``CheckAccessFullyMapped`` at all, since that's the only operation that +can possibly be done on this value. In practice this may mean we insert a DXIL +operation for the check when this was missing in the HLSL source, but this +actually matches DXC's behaviour in practice. + +For TypedBuffer and Texture, we map directly from the contained type of the +resource to the return value of the intrinsic. Since these resources are +constrained to contain only scalars and vectors of up to 4 elements, the +lowering to DXIL ops is generally straightforward. The one exception we have +here is that `double` types in the elements are special - these are allowed in +the LLVM intrinsics, but are lowered to pairs of `i32` followed by +``MakeDouble`` operations for DXIL. .. _ResRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#resource-operation-return-types .. _CBufRet: https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#cbufferloadlegacy .. _CheckAccessF
