This is an archive of the discontinued LLVM Phabricator instance.

Differential D45070

[NVPTX] Enable StructuredCFG for NVPTX
ClosedPublic

Authored by timshen on Mar 29 2018, 4:45 PM.

Details

Reviewers
jlebar
Commits
rG75f4129be13b: Merging r328885:
rG8f9f02696579: [NVPTX] Enable StructuredCFG for NVPTX
rL333475: Merging r328885:
rL328885: [NVPTX] Enable StructuredCFG for NVPTX
Summary

Make NVPTX require structured CFG. Added a temporary flag to
"roll back" the behavior for easy deployment.

Combined with D45008, this fixes several internal Nvidia GPU test
failures that we suspect to be ptxas miscompiles (PR27738).

Diff Detail

Repository
rL LLVM

Event Timeline

timshen created this revision.Mar 29 2018, 4:45 PM
Herald added subscribers: hiraditya, sanjoy, jholewinski. · View Herald TranscriptMar 29 2018, 4:45 PM
jlebar accepted this revision.Mar 29 2018, 4:46 PM
This revision is now accepted and ready to land.Mar 29 2018, 4:46 PM
vchuravy added a subscriber: vchuravy.Mar 29 2018, 10:44 PM
Closed by commit rL328885: [NVPTX] Enable StructuredCFG for NVPTX (authored by timshen). · Explain WhyMar 30 2018, 10:53 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
trunk/
lib/
Target/
NVPTX/
10 lines

Diff 140458

llvm/trunk/lib/Target/NVPTX/NVPTXTargetMachine.cpp

Show All 38 Lines
// LSV is still relatively new; this switch lets us turn it off in case we // LSV is still relatively new; this switch lets us turn it off in case we
// encounter (or suspect) a bug. // encounter (or suspect) a bug.
static cl::opt<bool> static cl::opt<bool>
DisableLoadStoreVectorizer("disable-nvptx-load-store-vectorizer", DisableLoadStoreVectorizer("disable-nvptx-load-store-vectorizer",
cl::desc("Disable load/store vectorizer"), cl::desc("Disable load/store vectorizer"),
cl::init(false), cl::Hidden); cl::init(false), cl::Hidden);
// TODO: Remove this flag when we are confident with no regressions.
static cl::opt<bool> DisableRequireStructuredCFG(
"disable-nvptx-require-structured-cfg",
cl::desc("Transitional flag to turn off NVPTX's requirement on preserving "
"structured CFG. The requirement should be disabled only when "
"unexpected regressions happen."),
cl::init(false), cl::Hidden);
namespace llvm { namespace llvm {
void initializeNVVMIntrRangePass(PassRegistry&); void initializeNVVMIntrRangePass(PassRegistry&);
void initializeNVVMReflectPass(PassRegistry&); void initializeNVVMReflectPass(PassRegistry&);
void initializeGenericToNVVMPass(PassRegistry&); void initializeGenericToNVVMPass(PassRegistry&);
void initializeNVPTXAllocaHoistingPass(PassRegistry &); void initializeNVPTXAllocaHoistingPass(PassRegistry &);
void initializeNVPTXAssignValidGlobalNamesPass(PassRegistry&); void initializeNVPTXAssignValidGlobalNamesPass(PassRegistry&);
void initializeNVPTXLowerAggrCopiesPass(PassRegistry &); void initializeNVPTXLowerAggrCopiesPass(PassRegistry &);
▲ Show 20 Lines • Show All 48 Lines • ▼ Show 20 LinesNVPTXTargetMachine::NVPTXTargetMachine(const Target &T, const Triple &TT,
: LLVMTargetMachine(T, computeDataLayout(is64bit), TT, CPU, FS, Options, : LLVMTargetMachine(T, computeDataLayout(is64bit), TT, CPU, FS, Options,
Reloc::PIC_, getEffectiveCodeModel(CM), OL), Reloc::PIC_, getEffectiveCodeModel(CM), OL),
is64bit(is64bit), TLOF(llvm::make_unique<NVPTXTargetObjectFile>()), is64bit(is64bit), TLOF(llvm::make_unique<NVPTXTargetObjectFile>()),
Subtarget(TT, CPU, FS, *this) { Subtarget(TT, CPU, FS, *this) {
if (TT.getOS() == Triple::NVCL) if (TT.getOS() == Triple::NVCL)
drvInterface = NVPTX::NVCL; drvInterface = NVPTX::NVCL;
else else
drvInterface = NVPTX::CUDA; drvInterface = NVPTX::CUDA;
if (!DisableRequireStructuredCFG)
setRequiresStructuredCFG(true);
initAsmInfo(); initAsmInfo();
} }
NVPTXTargetMachine::~NVPTXTargetMachine() = default; NVPTXTargetMachine::~NVPTXTargetMachine() = default;
void NVPTXTargetMachine32::anchor() {} void NVPTXTargetMachine32::anchor() {}
NVPTXTargetMachine32::NVPTXTargetMachine32(const Target &T, const Triple &TT, NVPTXTargetMachine32::NVPTXTargetMachine32(const Target &T, const Triple &TT,
▲ Show 20 Lines • Show All 251 Lines • Show Last 20 Lines