[llvm] [AMDGPU] Remove isKernelLDS, add isKernel(const Function &). NFC. (PR #167300)
Jay Foad via llvm-commits
llvm-commits at lists.llvm.org
Tue Nov 25 06:39:15 PST 2025
https://github.com/jayfoad updated https://github.com/llvm/llvm-project/pull/167300
>From 2da20d84413c385801c20d407fb149d28dfce9d8 Mon Sep 17 00:00:00 2001
From: Jay Foad <jay.foad at amd.com>
Date: Mon, 10 Nov 2025 11:06:08 +0000
Subject: [PATCH 1/2] [AMDGPU] Remove isKernelLDS, add isKernel(const Function
&). NFC.
Since #142598 isKernelLDS has been a pointless wrapper around isKernel.
---
.../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp | 2 +-
.../AMDGPU/AMDGPULowerModuleLDSPass.cpp | 28 +++++++++----------
llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.cpp | 16 ++++-------
llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.h | 2 --
llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp | 2 +-
llvm/lib/Target/AMDGPU/AMDGPUSwLowerLDS.cpp | 6 ++--
llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 2 +-
llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h | 2 ++
8 files changed, 28 insertions(+), 32 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 1a13b2226ecd6..8b4396cd63e9a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -7724,7 +7724,7 @@ bool AMDGPULegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
case Intrinsic::amdgcn_make_buffer_rsrc:
return legalizePointerAsRsrcIntrin(MI, MRI, B);
case Intrinsic::amdgcn_kernarg_segment_ptr:
- if (!AMDGPU::isKernel(B.getMF().getFunction().getCallingConv())) {
+ if (!AMDGPU::isKernel(B.getMF().getFunction())) {
// This only makes sense to call in a kernel, so just lower to null.
B.buildConstant(MI.getOperand(0).getReg(), 0);
MI.eraseFromParent();
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp b/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
index a4ef524c43466..56365284ffe32 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
@@ -441,7 +441,7 @@ class AMDGPULowerModuleLDS {
return KernelSet;
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || !isKernelLDS(&Func))
+ if (Func.isDeclaration() || !isKernel(Func))
continue;
for (GlobalVariable *GV : LDSUsesInfo.indirect_access[&Func]) {
if (VariableSet.contains(GV)) {
@@ -555,7 +555,7 @@ class AMDGPULowerModuleLDS {
for (Function &Func : M->functions()) {
if (Func.isDeclaration())
continue;
- if (!isKernelLDS(&Func))
+ if (!isKernel(Func))
continue;
if (KernelsThatAllocateTableLDS.contains(&Func) ||
@@ -703,7 +703,7 @@ class AMDGPULowerModuleLDS {
return false;
}
Function *F = I->getFunction();
- return !isKernelLDS(F);
+ return !isKernel(*F);
});
// Replace uses of module scope variable from kernel functions that
@@ -711,7 +711,7 @@ class AMDGPULowerModuleLDS {
// Record on each kernel whether the module scope global is used by it
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || !isKernelLDS(&Func))
+ if (Func.isDeclaration() || !isKernel(Func))
continue;
if (KernelsThatAllocateModuleLDS.contains(&Func)) {
@@ -743,7 +743,7 @@ class AMDGPULowerModuleLDS {
DenseMap<Function *, LDSVariableReplacement> KernelToReplacement;
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || !isKernelLDS(&Func))
+ if (Func.isDeclaration() || !isKernel(Func))
continue;
DenseSet<GlobalVariable *> KernelUsedVariables;
@@ -828,7 +828,7 @@ class AMDGPULowerModuleLDS {
// semantics. Setting the alignment here allows this IR pass to accurately
// predict the exact constant at which it will be allocated.
- assert(isKernelLDS(func));
+ assert(isKernel(*func));
LLVMContext &Ctx = M.getContext();
const DataLayout &DL = M.getDataLayout();
@@ -878,7 +878,7 @@ class AMDGPULowerModuleLDS {
for (auto &func : OrderedKernels) {
if (KernelsThatIndirectlyAllocateDynamicLDS.contains(func)) {
- assert(isKernelLDS(func));
+ assert(isKernel(*func));
if (!func->hasName()) {
reportFatalUsageError("anonymous kernels cannot use LDS variables");
}
@@ -912,7 +912,7 @@ class AMDGPULowerModuleLDS {
auto *I = dyn_cast<Instruction>(U.getUser());
if (!I)
continue;
- if (isKernelLDS(I->getFunction()))
+ if (isKernel(*I->getFunction()))
continue;
replaceUseWithTableLookup(M, Builder, table, GV, U, nullptr);
@@ -928,7 +928,7 @@ class AMDGPULowerModuleLDS {
for (Use &U : GV->uses()) {
if (auto *I = dyn_cast<Instruction>(U.getUser())) {
Function *F = I->getFunction();
- if (isKernelLDS(F) && F != KF) {
+ if (isKernel(*F) && F != KF) {
NeedsReplacement = true;
break;
}
@@ -945,7 +945,7 @@ class AMDGPULowerModuleLDS {
for (Use &U : make_early_inc_range(GV->uses())) {
if (auto *I = dyn_cast<Instruction>(U.getUser())) {
Function *F = I->getFunction();
- if (!isKernelLDS(F) || F == KF) {
+ if (!isKernel(*F) || F == KF) {
U.getUser()->replaceUsesOfWith(GV, NewGV);
}
}
@@ -997,7 +997,7 @@ class AMDGPULowerModuleLDS {
std::vector<Function *> OrderedKernels;
for (auto &K : LDSUsesInfo.direct_access) {
Function *F = K.first;
- assert(isKernelLDS(F));
+ assert(isKernel(*F));
OrderedKernels.push_back(F);
}
OrderedKernels = sortByName(std::move(OrderedKernels));
@@ -1033,7 +1033,7 @@ class AMDGPULowerModuleLDS {
}
// Also erase those special LDS variables from indirect_access.
for (auto &K : LDSUsesInfo.indirect_access) {
- assert(isKernelLDS(K.first));
+ assert(isKernel(*K.first));
for (GlobalVariable *GV : K.second) {
if (isNamedBarrier(*GV))
K.second.erase(GV);
@@ -1058,7 +1058,7 @@ class AMDGPULowerModuleLDS {
VariableFunctionMap LDSToKernelsThatNeedToAccessItIndirectly;
for (auto &K : LDSUsesInfo.indirect_access) {
Function *F = K.first;
- assert(isKernelLDS(F));
+ assert(isKernel(*F));
for (GlobalVariable *GV : K.second) {
LDSToKernelsThatNeedToAccessItIndirectly[GV].insert(F);
}
@@ -1157,7 +1157,7 @@ class AMDGPULowerModuleLDS {
const DataLayout &DL = M.getDataLayout();
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || !isKernelLDS(&Func))
+ if (Func.isDeclaration() || !isKernel(Func))
continue;
// All three of these are optional. The first variable is allocated at
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.cpp b/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.cpp
index e17c2113ca398..f274471ffe6a4 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.cpp
@@ -126,7 +126,7 @@ void getUsesOfLDSByFunction(const CallGraph &CG, Module &M,
for (User *V : GV.users()) {
if (auto *I = dyn_cast<Instruction>(V)) {
Function *F = I->getFunction();
- if (isKernelLDS(F))
+ if (isKernel(*F))
kernels[F].insert(&GV);
else
Functions[F].insert(&GV);
@@ -135,10 +135,6 @@ void getUsesOfLDSByFunction(const CallGraph &CG, Module &M,
}
}
-bool isKernelLDS(const Function *F) {
- return AMDGPU::isKernel(F->getCallingConv());
-}
-
LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
FunctionVariableMap DirectMapKernel;
@@ -148,7 +144,7 @@ LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
// Collect functions whose address has escaped
DenseSet<Function *> AddressTakenFuncs;
for (Function &F : M.functions()) {
- if (!isKernelLDS(&F))
+ if (!isKernel(F))
if (F.hasAddressTaken(nullptr,
/* IgnoreCallbackUses */ false,
/* IgnoreAssumeLikeCalls */ false,
@@ -180,7 +176,7 @@ LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
// access all variables accessed by functions whose address escaped
for (Function &F : M.functions()) {
if (!F.isDeclaration() && FunctionMakesUnknownCall(&F)) {
- if (!isKernelLDS(&F)) {
+ if (!isKernel(F)) {
set_union(TransitiveMapFunction[&F],
VariablesReachableThroughFunctionPointer);
}
@@ -190,7 +186,7 @@ LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
// Direct implementation of collecting all variables reachable from each
// function
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || isKernelLDS(&Func))
+ if (Func.isDeclaration() || isKernel(Func))
continue;
DenseSet<Function *> seen; // catches cycles
@@ -227,7 +223,7 @@ LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
FunctionVariableMap IndirectMapKernel;
for (Function &Func : M.functions()) {
- if (Func.isDeclaration() || !isKernelLDS(&Func))
+ if (Func.isDeclaration() || !isKernel(Func))
continue;
for (const CallGraphNode::CallRecord &R : *CG[&Func]) {
@@ -335,7 +331,7 @@ void removeFnAttrFromReachable(CallGraph &CG, Function *KernelRoot,
Function *PotentialCallee =
ExternalCallRecord.second->getFunction();
assert(PotentialCallee);
- if (!isKernelLDS(PotentialCallee)) {
+ if (!isKernel(*PotentialCallee)) {
for (StringRef Attr : FnAttrs)
PotentialCallee->removeFnAttr(Attr);
}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.h b/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.h
index 058e74452573c..8868b93440768 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMemoryUtils.h
@@ -53,8 +53,6 @@ void getUsesOfLDSByFunction(const CallGraph &CG, Module &M,
FunctionVariableMap &kernels,
FunctionVariableMap &functions);
-bool isKernelLDS(const Function *F);
-
LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M);
/// Strip FnAttr attribute from any functions where we may have
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
index 26e0b3dfc2e8a..38e7e17b85491 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -350,7 +350,7 @@ bool AMDGPUSubtarget::makeLIDRangeMetadata(Instruction *I) const {
}
unsigned AMDGPUSubtarget::getImplicitArgNumBytes(const Function &F) const {
- assert(AMDGPU::isKernel(F.getCallingConv()));
+ assert(AMDGPU::isKernel(F));
// We don't allocate the segment if we know the implicit arguments weren't
// used, even if the ABI implies we need them.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSwLowerLDS.cpp b/llvm/lib/Target/AMDGPU/AMDGPUSwLowerLDS.cpp
index 4a9437b37aa39..722b71b38141b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSwLowerLDS.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSwLowerLDS.cpp
@@ -271,7 +271,7 @@ void AMDGPUSwLowerLDS::getNonKernelsWithLDSArguments(const CallGraph &CG) {
Function *CalledFunc = CallerCGN->getFunction();
if (!CalledFunc || CalledFunc->isDeclaration())
continue;
- if (AMDGPU::isKernelLDS(CalledFunc))
+ if (AMDGPU::isKernel(*CalledFunc))
continue;
for (auto AI = CalledFunc->arg_begin(), E = CalledFunc->arg_end();
AI != E; ++AI) {
@@ -297,7 +297,7 @@ void AMDGPUSwLowerLDS::getUsesOfLDSByNonKernels() {
for (User *V : GV->users()) {
if (auto *I = dyn_cast<Instruction>(V)) {
Function *F = I->getFunction();
- if (!isKernelLDS(F) && !F->isDeclaration())
+ if (!isKernel(*F) && !F->isDeclaration())
FuncLDSAccessInfo.NonKernelToLDSAccessMap[F].insert(GV);
}
}
@@ -1169,7 +1169,7 @@ bool AMDGPUSwLowerLDS::run() {
if (!F || K.second.empty())
continue;
- assert(isKernelLDS(F));
+ assert(isKernel(*F));
// Only inserts if key isn't already in the map.
FuncLDSAccessInfo.KernelToLDSParametersMap.insert(
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 8bb28084159e8..528ea73184a14 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -9737,7 +9737,7 @@ SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
AMDGPUFunctionArgInfo::IMPLICIT_ARG_PTR);
}
case Intrinsic::amdgcn_kernarg_segment_ptr: {
- if (!AMDGPU::isKernel(MF.getFunction().getCallingConv())) {
+ if (!AMDGPU::isKernel(MF.getFunction())) {
// This only makes sense to call in a kernel, so just lower to null.
return DAG.getConstant(0, DL, VT);
}
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 5e3195b36fe4c..86b3cf767e565 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -1513,6 +1513,8 @@ constexpr inline bool isKernel(CallingConv::ID CC) {
}
}
+inline bool isKernel(const Function &F) { return isKernel(F.getCallingConv()); }
+
LLVM_READNONE
constexpr bool canGuaranteeTCO(CallingConv::ID CC) {
return CC == CallingConv::Fast;
>From e971df367767bc872b92c4546b97989690e1a36f Mon Sep 17 00:00:00 2001
From: Jay Foad <jay.foad at amd.com>
Date: Tue, 25 Nov 2025 14:38:57 +0000
Subject: [PATCH 2/2] Update AMDGPULowerExecSync
---
llvm/lib/Target/AMDGPU/AMDGPULowerExecSync.cpp | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULowerExecSync.cpp b/llvm/lib/Target/AMDGPU/AMDGPULowerExecSync.cpp
index 89f6b38df9d56..38b01dc1a551c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULowerExecSync.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULowerExecSync.cpp
@@ -44,7 +44,7 @@ static GlobalVariable *uniquifyGVPerKernel(Module &M, GlobalVariable *GV,
for (Use &U : GV->uses()) {
if (auto *I = dyn_cast<Instruction>(U.getUser())) {
Function *F = I->getFunction();
- if (isKernelLDS(F) && F != KF) {
+ if (isKernel(*F) && F != KF) {
NeedsReplacement = true;
break;
}
@@ -61,7 +61,7 @@ static GlobalVariable *uniquifyGVPerKernel(Module &M, GlobalVariable *GV,
for (Use &U : make_early_inc_range(GV->uses())) {
if (auto *I = dyn_cast<Instruction>(U.getUser())) {
Function *F = I->getFunction();
- if (!isKernelLDS(F) || F == KF) {
+ if (!isKernel(*F) || F == KF) {
U.getUser()->replaceUsesOfWith(GV, NewGV);
}
}
@@ -133,7 +133,7 @@ static bool lowerExecSyncGlobalVariables(
SmallVector<Function *> OrderedKernels;
for (auto &K : LDSUsesInfo.direct_access) {
Function *F = K.first;
- assert(isKernelLDS(F));
+ assert(isKernel(*F));
OrderedKernels.push_back(F);
}
OrderedKernels = sortByName(std::move(OrderedKernels));
@@ -169,7 +169,7 @@ static bool lowerExecSyncGlobalVariables(
}
// Also erase those special LDS variables from indirect_access.
for (auto &K : LDSUsesInfo.indirect_access) {
- assert(isKernelLDS(K.first));
+ assert(isKernel(*K.first));
for (GlobalVariable *GV : K.second) {
if (isNamedBarrier(*GV))
K.second.erase(GV);
@@ -191,7 +191,7 @@ static bool runLowerExecSyncGlobals(Module &M) {
VariableFunctionMap LDSToKernelsThatNeedToAccessItIndirectly;
for (auto &K : LDSUsesInfo.indirect_access) {
Function *F = K.first;
- assert(isKernelLDS(F));
+ assert(isKernel(*F));
for (GlobalVariable *GV : K.second) {
LDSToKernelsThatNeedToAccessItIndirectly[GV].insert(F);
}
More information about the llvm-commits
mailing list