Page MenuHomePhabricator
Differential D106033 Diff 358964 llvm/lib/Transforms/IPO/OpenMPOpt.cpp

Changeset View

Standalone View

View Options

llvm/lib/Transforms/IPO/OpenMPOpt.cpp

Show First 20 Lines • Show All 1,757 Lines • ▼ Show 20 Lines bool runAttributor(bool IsModulePass) {
ChangeStatus Changed = A.run(); ChangeStatus Changed = A.run();
LLVM_DEBUG(dbgs() << "[Attributor] Done with " << SCC.size() LLVM_DEBUG(dbgs() << "[Attributor] Done with " << SCC.size()
<< " functions, result: " << Changed << ".\n"); << " functions, result: " << Changed << ".\n");
return Changed == ChangeStatus::CHANGED; return Changed == ChangeStatus::CHANGED;
} }
/// Populate the Attributor with abstract attribute opportunities in the /// Populate the Attributor with abstract attribute opportunities in the
/// function. /// function.
jdoerfertUnsubmitted
Done
ReplyInline Actions

copy and paste.

jdoerfert: copy and paste.
void registerFoldRuntimeCall(RuntimeFunction RF);
/// Populate the Attributor with abstract attribute opportunities in the
/// function.
void registerAAs(bool IsModulePass); void registerAAs(bool IsModulePass);
}; };
Kernel OpenMPOpt::getUniqueKernelFor(Function &F) { Kernel OpenMPOpt::getUniqueKernelFor(Function &F) {
if (!OMPInfoCache.ModuleSlice.count(&F)) if (!OMPInfoCache.ModuleSlice.count(&F))
return nullptr; return nullptr;
// Use a scope to keep the lifetime of the CachedKernel short. // Use a scope to keep the lifetime of the CachedKernel short.
▲ Show 20 Lines • Show All 1,577 Lines • ▼ Show 20 Lines void initialize(Attributor &A) override {
const unsigned int WrapperFunctionArgNo = 6; const unsigned int WrapperFunctionArgNo = 6;
RuntimeFunction RF = It->getSecond(); RuntimeFunction RF = It->getSecond();
switch (RF) { switch (RF) {
// All the functions we know are compatible with SPMD mode. // All the functions we know are compatible with SPMD mode.
case OMPRTL___kmpc_is_spmd_exec_mode: case OMPRTL___kmpc_is_spmd_exec_mode:
case OMPRTL___kmpc_for_static_fini: case OMPRTL___kmpc_for_static_fini:
case OMPRTL___kmpc_global_thread_num: case OMPRTL___kmpc_global_thread_num:
case OMPRTL___kmpc_get_hardware_num_threads_in_block:
case OMPRTL___kmpc_get_hardware_num_blocks:
case OMPRTL___kmpc_single: case OMPRTL___kmpc_single:
case OMPRTL___kmpc_end_single: case OMPRTL___kmpc_end_single:
case OMPRTL___kmpc_master: case OMPRTL___kmpc_master:
case OMPRTL___kmpc_end_master: case OMPRTL___kmpc_end_master:
case OMPRTL___kmpc_barrier: case OMPRTL___kmpc_barrier:
break; break;
case OMPRTL___kmpc_for_static_init_4: case OMPRTL___kmpc_for_static_init_4:
case OMPRTL___kmpc_for_static_init_4u: case OMPRTL___kmpc_for_static_init_4u:
▲ Show 20 Lines • Show All 137 Lines • ▼ Show 20 Lines A.registerSimplificationCallback(
A.recordDependence(*this, *AA, DepClassTy::OPTIONAL); A.recordDependence(*this, *AA, DepClassTy::OPTIONAL);
} }
return SimplifiedValue; return SimplifiedValue;
}); });
} }
ChangeStatus updateImpl(Attributor &A) override { ChangeStatus updateImpl(Attributor &A) override {
ChangeStatus Changed = ChangeStatus::UNCHANGED; ChangeStatus Changed = ChangeStatus::UNCHANGED;
switch (RFKind) { switch (RFKind) {
case OMPRTL___kmpc_is_spmd_exec_mode: case OMPRTL___kmpc_is_spmd_exec_mode:
Changed = Changed | foldIsSPMDExecMode(A); Changed = Changed | foldIsSPMDExecMode(A);
break; break;
case OMPRTL___kmpc_get_hardware_num_threads_in_block:
Changed = Changed | foldHardwareNumThreads(A);
break;
case OMPRTL___kmpc_get_hardware_num_blocks:
Changed = Changed | foldHardwareNumTeams(A);
break;
default: default:
llvm_unreachable("Unhandled OpenMP runtime function!"); llvm_unreachable("Unhandled OpenMP runtime function!");
} }
return Changed; return Changed;
} }
ChangeStatus manifest(Attributor &A) override { ChangeStatus manifest(Attributor &A) override {
Show All 14 Linesprivate:
ChangeStatus foldIsSPMDExecMode(Attributor &A) { ChangeStatus foldIsSPMDExecMode(Attributor &A) {
BooleanState StateBefore = getState(); BooleanState StateBefore = getState();
unsigned AssumedSPMDCount = 0, KnownSPMDCount = 0; unsigned AssumedSPMDCount = 0, KnownSPMDCount = 0;
unsigned AssumedNonSPMDCount = 0, KnownNonSPMDCount = 0; unsigned AssumedNonSPMDCount = 0, KnownNonSPMDCount = 0;
auto &CallerKernelInfoAA = A.getAAFor<AAKernelInfo>( auto &CallerKernelInfoAA = A.getAAFor<AAKernelInfo>(
*this, IRPosition::function(*getAnchorScope()), DepClassTy::REQUIRED); *this, IRPosition::function(*getAnchorScope()), DepClassTy::REQUIRED);
// What actual kernels reach this function
for (Kernel K : CallerKernelInfoAA.ReachingKernelEntries) { for (Kernel K : CallerKernelInfoAA.ReachingKernelEntries) {
auto &AA = A.getAAFor<AAKernelInfo>(*this, IRPosition::function(*K), auto &AA = A.getAAFor<AAKernelInfo>(*this, IRPosition::function(*K),
DepClassTy::REQUIRED); DepClassTy::REQUIRED);
if (!AA.isValidState()) { if (!AA.isValidState()) {
SimplifiedValue = nullptr; SimplifiedValue = nullptr;
return indicatePessimisticFixpoint(); return indicatePessimisticFixpoint();
} }
Show All 37 Lines if (KnownSPMDCount || AssumedSPMDCount) {
// calls to __kmpc_is_spmd_exec_mode to 0. // calls to __kmpc_is_spmd_exec_mode to 0.
SimplifiedValue = ConstantInt::get(Type::getInt8Ty(Ctx), false); SimplifiedValue = ConstantInt::get(Type::getInt8Ty(Ctx), false);
} }
return getState() == StateBefore ? ChangeStatus::UNCHANGED return getState() == StateBefore ? ChangeStatus::UNCHANGED
: ChangeStatus::CHANGED; : ChangeStatus::CHANGED;
} }
/// Fold __kmpc_get_hardware_num_blocks into a constant if possible.
JonChesterfieldUnsubmitted
Done
ReplyInline Actions

I haven't read through this part, but if we can only fold it to a constant sometimes, we shouldn't mark the calls that survive noinline, as that'll be expensive for the cases that this pass misses.

JonChesterfield: I haven't read through this part, but if we can only fold it to a constant sometimes, we…
jdoerfertUnsubmitted
Done
ReplyInline Actions

We are about to remove noinline from known runtime functions such that we can keep them around until we get to OpenMP-Opt as calls. This will have the effect we want without any drawbacks. Thus, adding noinline in the runitme will be totally fine.

jdoerfert: We are about to remove noinline from known runtime functions such that we can keep them around…
/// The value is an attribute in the kernel
jdoerfertUnsubmitted
Done
ReplyInline Actions
/// Fold an attribute of a kerrnel into a constant if possible.
- /// The value is an attribute in the kernel
- ChangeStatus foldKernelFnAttribute(Attributor &A, llvm::StringRef Attr) {
+ /// The value \p Attr is an attribute of the kernel. ChangeStatus foldKernelFnAttribute(Attributor &A, llvm::StringRef Attr) {
jdoerfert:
ChangeStatus foldHardwareNumTeams(Attributor &A) {
// Specialize only if all the calls agree with the number of teams
int32_t CurrentNumTeams = -1;
BooleanState StateBefore = getState();
auto &CallerKernelInfoAA = A.getAAFor<AAKernelInfo>(
*this, IRPosition::function(*getAnchorScope()), DepClassTy::REQUIRED);
jdoerfertUnsubmitted
Done
ReplyInline Actions
if (!CallerKernelInfoAA.ReachingKernelEntries.isValid())
  return indicatePessimisticFixpoint();

also below

jdoerfert: ``` if (!CallerKernelInfoAA.ReachingKernelEntries.isValid()) return…
// What actual kernels reach this function
for (Kernel K : CallerKernelInfoAA.ReachingKernelEntries) {
if (K->hasFnAttribute("NumTeams")) {
int32_t NumT =
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-        int32_t NumT = 
+        int32_t NumT =
Lint: Pre-merge checks: clang-format: please reformat the code ``` - int32_t NumT = + int32_t NumT = ```
std::stoi(K->getFnAttribute("NumTeams").getValueAsString().str());
if (CurrentNumTeams != -1 && CurrentNumTeams != NumT) {
SimplifiedValue = nullptr;
return indicatePessimisticFixpoint();
} else {
Lint: Pre-merge checks
Inline Actions

clang-tidy: warning: do not use 'else' after 'return' [llvm-else-after-return]
not useful

Lint: Pre-merge checks: clang-tidy: warning: do not use 'else' after 'return' [llvm-else-after-return] [[https://github.
tianshilei1992Unsubmitted
Done
ReplyInline Actions

No need to have else here because the code above already returns.

tianshilei1992: No need to have `else` here because the code above already returns.
CurrentNumTeams = NumT;
}
jdoerfertUnsubmitted
Done
ReplyInline Actions

use early exit instead,
if (!...)

return ...
jdoerfert: use early exit instead, if (!...) return ...
} else {
tianshilei1992Unsubmitted
Done
ReplyInline Actions

no need for else as well

tianshilei1992: no need for `else` as well
// TODO: No attribute, then default?
tianshilei1992Unsubmitted
Done
ReplyInline Actions

Directly indicate pessimistic state because we don't know clearly what the number is.

tianshilei1992: Directly indicate pessimistic state because we don't know clearly what the number is.
}
jdoerfertUnsubmitted
Done
ReplyInline Actions

You should set CurrentAttrValue = NextAttrValue at the end of the loop.

jdoerfert: You should set CurrentAttrValue = NextAttrValue at the end of the loop.
}
if (CurrentNumTeams != -1) {
auto &Ctx = getAnchorValue().getContext();
SimplifiedValue =
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-      SimplifiedValue = 
-          ConstantInt::get(Type::getInt8Ty(Ctx), CurrentNumTeams);
+      SimplifiedValue = ConstantInt::get(Type::getInt8Ty(Ctx), CurrentNumTeams);
Lint: Pre-merge checks: clang-format: please reformat the code ``` - SimplifiedValue = - ConstantInt…
jdoerfertUnsubmitted
Done
ReplyInline Actions
SimplifiedValue =
- ConstantInt::get(Type::getInt8Ty(Ctx), CurrentAttrValue);
+ ConstantInt::get(Type::getInt32Ty(Ctx), CurrentAttrValue);
}
return SimplifiedValue == SimplifiedValueBefore ? ChangeStatus::UNCHANGED
jdoerfert:
ConstantInt::get(Type::getInt8Ty(Ctx), CurrentNumTeams);
jdoerfertUnsubmitted
Done
ReplyInline Actions

no reaching kernels is fine, keep it at none (which is the default) and just return UNCHANGED.

jdoerfert: no reaching kernels is fine, keep it at none (which is the default) and just return UNCHANGED.
}
jdoerfertUnsubmitted
Done
ReplyInline Actions

make this function take the string attribute such that we can have a single one and not two.

jdoerfert: make this function take the string attribute such that we can have a single one and not two.
return getState() == StateBefore ? ChangeStatus::UNCHANGED
tianshilei1992Unsubmitted
Done
ReplyInline Actions

This has to be updated as SimplifiedValue is not part of BooleanState. Refer to foldIsSPMDExecMode.

tianshilei1992: This has to be updated as `SimplifiedValue` is not part of `BooleanState`. Refer to…
: ChangeStatus::CHANGED;
}
/// Fold __kmpc_get_hardware_num_threads_in_block into a constant if possible.
/// The value is an attribute in the kernel
ChangeStatus foldHardwareNumThreads(Attributor &A) {
tianshilei1992Unsubmitted
Done
ReplyInline Actions

Update accordingly

tianshilei1992: Update accordingly
// Specialize only if all the calls agree with the number of threads
int32_t CurrentNumThreads = -1;
BooleanState StateBefore = getState();
auto &CallerKernelInfoAA = A.getAAFor<AAKernelInfo>(
*this, IRPosition::function(*getAnchorScope()), DepClassTy::REQUIRED);
// What actual kernels reach this function
for (Kernel K : CallerKernelInfoAA.ReachingKernelEntries) {
if (K->hasFnAttribute("NumThreads")) {
int32_t NumT =
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-        int32_t NumT = 
+        int32_t NumT =
Lint: Pre-merge checks: clang-format: please reformat the code ``` - int32_t NumT = + int32_t NumT = ```
std::stoi(K->getFnAttribute("NumThreads").getValueAsString().str());
if (CurrentNumThreads != -1 && CurrentNumThreads != NumT) {
SimplifiedValue = nullptr;
return indicatePessimisticFixpoint();
} else {
Lint: Pre-merge checks
Inline Actions

clang-tidy: warning: do not use 'else' after 'return' [llvm-else-after-return]
not useful

Lint: Pre-merge checks: clang-tidy: warning: do not use 'else' after 'return' [llvm-else-after-return] [[https://github.
CurrentNumThreads = NumT;
}
} else {
// TODO: No attribute, then default?
}
}
if (CurrentNumThreads != -1) {
auto &Ctx = getAnchorValue().getContext();
SimplifiedValue =
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-      SimplifiedValue = 
+      SimplifiedValue =
Lint: Pre-merge checks: clang-format: please reformat the code ``` - SimplifiedValue = + SimplifiedValue =…
ConstantInt::get(Type::getInt8Ty(Ctx), CurrentNumThreads);
}
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-    
+
Lint: Pre-merge checks: clang-format: please reformat the code ``` - + ```
return getState() == StateBefore ? ChangeStatus::UNCHANGED
: ChangeStatus::CHANGED;
}
/// An optional value the associated value is assumed to fold to. That is, we /// An optional value the associated value is assumed to fold to. That is, we
/// assume the associated value (which is a call) can be replaced by this /// assume the associated value (which is a call) can be replaced by this
/// simplified value. /// simplified value.
Optional<Value *> SimplifiedValue; Optional<Value *> SimplifiedValue;
/// The runtime function kind of the callee of the associated call site. /// The runtime function kind of the callee of the associated call site.
RuntimeFunction RFKind; RuntimeFunction RFKind;
}; };
} // namespace } // namespace
/// Register folding callsite
void OpenMPOpt::registerFoldRuntimeCall(RuntimeFunction RF) {
auto &RFI = OMPInfoCache.RFIs[RF];
RFI.foreachUse(SCC, [&](Use &U, Function &) {
CallInst *CI = OpenMPOpt::getCallIfRegularCall(U, &RFI);
if (!CI)
jdoerfertUnsubmitted
Done
ReplyInline Actions

leftover. (and you can also just pipe an IRPosition into errs())

jdoerfert: leftover. (and you can also just pipe an IRPosition into errs())
return false;
A.getOrCreateAAFor<AAFoldRuntimeCall>(
IRPosition::callsite_function(*CI), /* QueryingAA */ nullptr,
tianshilei1992Unsubmitted
Done
ReplyInline Actions

This part needs to be changed. Refer to the trunk for more details. Basically it should be IRPosition::callsite_returned(*CI).

tianshilei1992: This part needs to be changed. Refer to the trunk for more details. Basically it should be…
DepClassTy::NONE, /* ForceUpdate */ false,
/* UpdateAfterInit */ false);
return false;
});
}
void OpenMPOpt::registerAAs(bool IsModulePass) { void OpenMPOpt::registerAAs(bool IsModulePass) {
if (SCC.empty()) if (SCC.empty())
return; return;
if (IsModulePass) { if (IsModulePass) {
// Ensure we create the AAKernelInfo AAs first and without triggering an // Ensure we create the AAKernelInfo AAs first and without triggering an
// update. This will make sure we register all value simplification // update. This will make sure we register all value simplification
// callbacks before any other AA has the chance to create an AAValueSimplify // callbacks before any other AA has the chance to create an AAValueSimplify
// or similar. // or similar.
for (Function *Kernel : OMPInfoCache.Kernels) for (Function *Kernel : OMPInfoCache.Kernels) {
A.getOrCreateAAFor<AAKernelInfo>( A.getOrCreateAAFor<AAKernelInfo>(
IRPosition::function(*Kernel), /* QueryingAA */ nullptr, IRPosition::function(*Kernel), /* QueryingAA */ nullptr,
DepClassTy::NONE, /* ForceUpdate */ false, DepClassTy::NONE, /* ForceUpdate */ false,
/* UpdateAfterInit */ false); /* UpdateAfterInit */ false);
}
auto &IsSPMDRFI = OMPInfoCache.RFIs[OMPRTL___kmpc_is_spmd_exec_mode]; registerFoldRuntimeCall(OMPRTL___kmpc_is_spmd_exec_mode);
IsSPMDRFI.foreachUse(SCC, [&](Use &U, Function &) { registerFoldRuntimeCall(OMPRTL___kmpc_get_hardware_num_threads_in_block);
CallInst *CI = OpenMPOpt::getCallIfRegularCall(U, &IsSPMDRFI); registerFoldRuntimeCall(OMPRTL___kmpc_get_hardware_num_blocks);
if (!CI)
return false;
A.getOrCreateAAFor<AAFoldRuntimeCall>(
IRPosition::callsite_function(*CI), /* QueryingAA */ nullptr,
DepClassTy::NONE, /* ForceUpdate */ false,
/* UpdateAfterInit */ false);
return false;
});
} }
// Create CallSite AA for all Getters. // Create CallSite AA for all Getters.
for (int Idx = 0; Idx < OMPInfoCache.ICVs.size() - 1; ++Idx) { for (int Idx = 0; Idx < OMPInfoCache.ICVs.size() - 1; ++Idx) {
auto ICVInfo = OMPInfoCache.ICVs[static_cast<InternalControlVar>(Idx)]; auto ICVInfo = OMPInfoCache.ICVs[static_cast<InternalControlVar>(Idx)];
auto &GetterRFI = OMPInfoCache.RFIs[ICVInfo.Getter]; auto &GetterRFI = OMPInfoCache.RFIs[ICVInfo.Getter];
▲ Show 20 Lines • Show All 395 Lines • Show Last 20 Lines