This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU: Fix crash with invariant markers
ClosedPublic

Authored by arsenm on Jan 21 2016, 10:42 AM.

Download Raw Diff

Details

Reviewers

Summary

The promote alloca pass didn't handle these intrinsics and crashed.
These intrinsics should accept any address space, but for now just
erase them to avoid breaking.

Diff Detail

Event Timeline

arsenm updated this revision to Diff 45562.Jan 21 2016, 10:42 AM

arsenm retitled this revision from to AMDGPU: Fix crash with invariant markers.

arsenm updated this object.

arsenm added a reviewer: • tstellarAMD.

arsenm added a subscriber: llvm-commits.

Herald added a subscriber: arsenm. · View Herald TranscriptJan 21 2016, 10:42 AM

LGTM.

This revision is now accepted and ready to land.Jan 22 2016, 7:36 AM

r258537

Revision Contents

Path

Size

lib/

Target/

AMDGPU/

AMDGPUPromoteAlloca.cpp

8 lines

test/

CodeGen/

AMDGPU/

promote-alloca-invariant-markers.ll

25 lines

Diff 45562

lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

Show First 20 Lines • Show All 406 Lines • ▼ Show 20 Lines	for (std::vector<Value*>::iterator i = WorkList.begin(),
case Intrinsic::memset: {		case Intrinsic::memset: {
MemSetInst *MemSet = cast<MemSetInst>(Intr);		MemSetInst *MemSet = cast<MemSetInst>(Intr);
Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),		Builder.CreateMemSet(MemSet->getRawDest(), MemSet->getValue(),
MemSet->getLength(), MemSet->getAlignment(),		MemSet->getLength(), MemSet->getAlignment(),
MemSet->isVolatile());		MemSet->isVolatile());
Intr->eraseFromParent();		Intr->eraseFromParent();
continue;		continue;
}		}
		case Intrinsic::invariant_start:
		case Intrinsic::invariant_end:
		case Intrinsic::invariant_group_barrier:
		Intr->eraseFromParent();
		// FIXME: I think the invariant marker should still theoretically apply,
		// but the intrinsics need to be changed to accept pointers with any
		// address space.
		continue;
default:		default:
Intr->dump();		Intr->dump();
llvm_unreachable("Don't know how to promote alloca intrinsic use.");		llvm_unreachable("Don't know how to promote alloca intrinsic use.");
}		}
}		}
}		}

FunctionPass *llvm::createAMDGPUPromoteAlloca(const AMDGPUSubtarget &ST) {		FunctionPass *llvm::createAMDGPUPromoteAlloca(const AMDGPUSubtarget &ST) {
return new AMDGPUPromoteAlloca(ST);		return new AMDGPUPromoteAlloca(ST);
}		}

test/CodeGen/AMDGPU/promote-alloca-invariant-markers.ll

This file was added.

				; RUN: llc -march=amdgcn -mattr=+promote-alloca -verify-machineinstrs < %s \| FileCheck -check-prefix=GCN %s

				declare {}* @llvm.invariant.start(i64, i8* nocapture) #0
				declare void @llvm.invariant.end({}, i64, i8 nocapture) #0
				declare i8* @llvm.invariant.group.barrier(i8*) #1

				; GCN-LABEL: {{^}}use_invariant_promotable_lds:
				; GCN: buffer_load_dword
				; GCN: ds_write_b32
				define void @use_invariant_promotable_lds(i32 addrspace(1)* %arg) #2 {
				bb:
				%tmp = alloca i32, align 4
				%tmp1 = bitcast i32* %tmp to i8*
				%tmp2 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 1
				%tmp3 = load i32, i32 addrspace(1)* %tmp2
				store i32 %tmp3, i32* %tmp
				%tmp4 = call {}* @llvm.invariant.start(i64 4, i8* %tmp1) #0
				call void @llvm.invariant.end({}* %tmp4, i64 4, i8* %tmp1) #0
				%tmp5 = call i8* @llvm.invariant.group.barrier(i8* %tmp1) #1
				ret void
				}

				attributes #0 = { argmemonly nounwind }
				attributes #1 = { nounwind readnone }
				attributes #2 = { nounwind }