Skip to content

Fix reverse NVVM barrier calls for LLVM 21+ CUDA intrinsics #2785

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Copilot wants to merge 6 commits into
base: main
Choose a base branch
from
+122 −5
Open

Fix reverse NVVM barrier calls for LLVM 21+ CUDA intrinsics #2785

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
cbe6bef
Initial plan
Copilot Apr 9, 2026
54c66ce
Fix NVVM reverse barrier arguments
Copilot Apr 9, 2026
078c6be
Fix EnzymeLogic NVVM barrier args
Copilot Apr 10, 2026
01e0b55
Fix AdjointGenerator formatting
Copilot Apr 11, 2026
c4263b0
update
minansys Apr 29, 2026
2c0028f
update
minansys Apr 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
88 changes: 63 additions & 25 deletions enzyme/Enzyme/AdjointGenerator.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3996,43 +3996,81 @@ class AdjointGenerator : public llvm::InstVisitor<AdjointGenerator> {
(void)vdiff;

switch (ID) {
#if LLVM_VERSION_MAJOR < 22
#if LLVM_VERSION_MAJOR < 21
case Intrinsic::nvvm_barrier0_popc:
case Intrinsic::nvvm_barrier0_and:
Comment thread
minansys marked this conversation as resolved.
Outdated
case Intrinsic::nvvm_barrier0_or:
case Intrinsic::nvvm_barrier0_or: {
SmallVector<Value *, 1> args = {};
auto *Fn = getIntrinsicDeclaration(M, Intrinsic::nvvm_barrier0);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
#elif LLVM_VERSION_MAJOR < 22
case Intrinsic::nvvm_barrier0_popc:
case Intrinsic::nvvm_barrier0_and:
case Intrinsic::nvvm_barrier0_or: {
SmallVector<Value *, 1> args = {
ConstantInt::get(Type::getInt32Ty(M->getContext()), 0)};
auto *Fn = getIntrinsicDeclaration(
M, Intrinsic::nvvm_barrier_cta_sync_aligned_all);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
#else
case Intrinsic::nvvm_barrier_cta_red_and_aligned_all:
case Intrinsic::nvvm_barrier_cta_red_and_aligned_count:
case Intrinsic::nvvm_barrier_cta_red_or_aligned_all:
case Intrinsic::nvvm_barrier_cta_red_popc_aligned_all: {
SmallVector<Value *, 1> args = {I.getOperand(0)};
auto *Fn = getIntrinsicDeclaration(
M, Intrinsic::nvvm_barrier_cta_sync_aligned_all);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
case Intrinsic::nvvm_barrier_cta_red_and_aligned_count:
case Intrinsic::nvvm_barrier_cta_red_or_aligned_count:
case Intrinsic::nvvm_barrier_cta_red_popc_aligned_all:
case Intrinsic::nvvm_barrier_cta_red_popc_aligned_count:
case Intrinsic::nvvm_barrier_cta_red_popc_aligned_count: {
SmallVector<Value *, 2> args = {I.getOperand(0), I.getOperand(1)};
auto *Fn = getIntrinsicDeclaration(
M, Intrinsic::nvvm_barrier_cta_sync_aligned_count);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
#endif
{

#if LLVM_VERSION_MAJOR < 21
case Intrinsic::nvvm_barrier0: {
SmallVector<Value *, 1> args = {};
#if LLVM_VERSION_MAJOR > 20
auto cal = cast<CallInst>(Builder2.CreateCall(
getIntrinsicDeclaration(
M, Intrinsic::nvvm_barrier_cta_sync_aligned_all),
args));
cal->setCallingConv(getIntrinsicDeclaration(
M, Intrinsic::nvvm_barrier_cta_sync_aligned_all)
->getCallingConv());
#else
auto cal = cast<CallInst>(Builder2.CreateCall(
getIntrinsicDeclaration(M, Intrinsic::nvvm_barrier0), args));
cal->setCallingConv(getIntrinsicDeclaration(M, Intrinsic::nvvm_barrier0)
->getCallingConv());
#endif
auto *Fn = getIntrinsicDeclaration(M, ID);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}

#if LLVM_VERSION_MAJOR <= 20
case Intrinsic::nvvm_barrier0:
#else
case Intrinsic::nvvm_barrier_cta_sync_aligned_all:
case Intrinsic::nvvm_barrier_cta_sync_aligned_count:
case Intrinsic::nvvm_barrier_cta_sync_aligned_all: {
SmallVector<Value *, 1> args = {I.getOperand(0)};
auto *Fn = getIntrinsicDeclaration(M, ID);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
case Intrinsic::nvvm_barrier_cta_sync_aligned_count: {
SmallVector<Value *, 2> args = {I.getOperand(0), I.getOperand(1)};
auto *Fn = getIntrinsicDeclaration(M, ID);
auto cal = cast<CallInst>(Builder2.CreateCall(Fn, args));
cal->setCallingConv(Fn->getCallingConv());
cal->setDebugLoc(gutils->getNewFromOriginal(I.getDebugLoc()));
return false;
}
#endif
case Intrinsic::amdgcn_s_barrier:
case Intrinsic::nvvm_membar_cta:
Expand Down
9 changes: 8 additions & 1 deletion enzyme/Enzyme/EnzymeLogic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4605,10 +4605,17 @@ Function *EnzymeLogic::CreatePrimalAndGradient(
auto BarrierInst = Arch == Triple::amdgcn
? (llvm::Intrinsic::ID)Intrinsic::amdgcn_s_barrier
: (llvm::Intrinsic::ID)Intrinsic::nvvm_barrier0;
#endif
SmallVector<Value *, 1> BarrierArgs;
#if LLVM_VERSION_MAJOR > 20
if (Arch != Triple::amdgcn) {
BarrierArgs.push_back(ConstantInt::get(
Type::getInt32Ty(gutils->newFunc->getContext()), 0));
}
#endif
instbuilder.CreateCall(
getIntrinsicDeclaration(gutils->newFunc->getParent(), BarrierInst),
{});
BarrierArgs);
OldEntryInsts->moveAfter(entry);
sharedBlock->moveAfter(entry);
IRBuilder<> sbuilder(sharedBlock);
Expand Down
70 changes: 70 additions & 0 deletions enzyme/test/Enzyme/ReverseMode/cuda-barrier-sync.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,70 @@
; RUN: split-file %s %t
; RUN: if [ %llvmver -le 20 ]; then %opt < %t/llvm20.ll %newLoadEnzyme -enzyme-preopt=false -enzyme-detect-readthrow=0 -passes="enzyme" -S | FileCheck %t/llvm20.ll; fi
; RUN: if [ %llvmver -gt 20 ]; then %opt < %t/llvm21plus.ll %newLoadEnzyme -enzyme-preopt=false -enzyme-detect-readthrow=0 -passes="enzyme" -S | FileCheck %t/llvm21plus.ll; fi

;--- llvm20.ll
target triple = "nvptx64-nvidia-cuda"

declare void @llvm.nvvm.barrier0()
declare float @__enzyme_autodiff(float (float)*, ...)

define float @f_sync(float %x) {
entry:
call void @llvm.nvvm.barrier0()
%res = fadd float %x, 1.000000e+00
ret float %res
}

define float @test(float %x) {
entry:
%r = call float (float (float)*, ...) @__enzyme_autodiff(float (float)* @f_sync, float %x)
ret float %r
}

; CHECK: define internal { float } @diffef_sync(float %x, float %differeturn)
; CHECK: call void @llvm.nvvm.barrier0()
; CHECK: call void @llvm.nvvm.barrier0()
; CHECK: ret { float }

;--- llvm21plus.ll
target triple = "nvptx64-nvidia-cuda"

declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32)
declare void @llvm.nvvm.barrier.cta.sync.aligned.count(i32, i32)
declare float @__enzyme_autodiff(float (float)*, ...)

define float @f_sync_all(float %x) {
entry:
call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 7)
%res = fadd float %x, 1.000000e+00
ret float %res
}

define float @test_all(float %x) {
entry:
%r = call float (float (float)*, ...) @__enzyme_autodiff(float (float)* @f_sync_all, float %x)
ret float %r
}

define float @f_sync_count(float %x) {
entry:
call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 7, i32 16)
%res = fadd float %x, 1.000000e+00
ret float %res
}

define float @test_count(float %x) {
entry:
%r = call float (float (float)*, ...) @__enzyme_autodiff(float (float)* @f_sync_count, float %x)
ret float %r
}

; CHECK: define internal { float } @diffef_sync_all(float %x, float %differeturn)
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 7)
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 7)
; CHECK: ret { float }

; CHECK: define internal { float } @diffef_sync_count(float %x, float %differeturn)
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 7, i32 16)
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 7, i32 16)
; CHECK: ret { float }
Loading