mirror/freebsd-ports
1
0
Fork 0
mirror of https://git.FreeBSD.org/ports.git synced 2025-01-20 08:27:15 +00:00
Code Issues Releases Activity

devel/llvm18: fix host dependent compiler output for i386

Browse Source 
Merge fixes for a couple cases where the compiler generated different
i386 code depending on the host.  In the base system this showed up as
very small differences in a couple object files in buildworld for i386
depending on the host architecture (i386 or amd64).

PR:		276961
This commit is contained in:
Brooks Davis 2024-07-25 23:00:32 +01:00
parent ecfb3034bf
commit a4e0cb2fd9
3 changed files with 172 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
devel/llvm18/Makefile
View File

@ -1,6 +1,6 @@
PORTNAME= llvm
DISTVERSION= 18.1.8
PORTREVISION= 0
PORTREVISION= 1
CATEGORIES= devel lang
MASTER_SITES= https://github.com/llvm/llvm-project/releases/download/llvmorg-${DISTVERSION:S/rc/-rc/}/
PKGNAMESUFFIX= ${LLVM_SUFFIX}

43
devel/llvm18/files/patch-backport-freebsd-397c2693fa6 Normal file
View File

@ -0,0 +1,43 @@
commit 397c2693fa66508cb5e6b173650a1f3bc6c4dd4f
Author: Dimitry Andric <dim@FreeBSD.org>
Date: Sun Jul 21 22:37:27 2024 +0200
Fix llvm register allocator for native/cross build differences
Work around an issue in LLVM's register allocator, which can cause
slightly different i386 object files, when produced by a native or cross
build of clang.
This adds another volatile qualifier to a float variable declaration in
the weightCalcHelper() function, which otherwise produces slightly
different float results on amd64 and i386 hosts. In turn, this can lead
to different (but equivalent) register choices, and thus non-identical
assembly code.
See https://github.com/llvm/llvm-project/issues/99396 for more details.
Note this is a temporary fix, meant to merge in time for 13.4. As soon
as upstream has a permanent solution we will import that.
PR: 276961
Reported by: cperciva
MFC after: 3 days
diff --git llvm/lib/CodeGen/CalcSpillWeights.cpp llvm/lib/CodeGen/CalcSpillWeights.cpp
index f3cb7fa5af61..afde8d001f88 100644
--- llvm/lib/CodeGen/CalcSpillWeights.cpp
+++ llvm/lib/CodeGen/CalcSpillWeights.cpp
@@ -256,7 +256,12 @@ float VirtRegAuxInfo::weightCalcHelper(LiveInterval &LI, SlotIndex *Start,
return -1.0f;
}
- float Weight = 1.0f;
+ // FreeBSD customization: similar to the HWeight declaration below, add a
+ // volatile qualifier to avoid slightly different weight results on amd64
+ // and i386 hosts, and possibly choosing different registers in the register
+ // allocator. See <https://github.com/llvm/llvm-project/issues/99396> for
+ // more details.
+ volatile float Weight = 1.0f;
if (IsSpillable) {
// Get loop info for mi.
if (MI->getParent() != MBB) {

128
devel/llvm18/files/patch-backport-freebsd-55a2a91c5e1 Normal file
View File

@ -0,0 +1,128 @@
commit 55a2a91c5e1bb39dd625ba56597608883fbcb318
Author: Dimitry Andric <dim@FreeBSD.org>
Date: Thu Jul 25 13:13:45 2024 +0200
Merge commit 28a2b85602a5 from llvm-project (by Kazu Hirata):
[DeadStoreElimination] Use SmallSetVector (NFC) (#79410)
The use of SmallSetVector saves 0.58% of heap allocations during the
compilation of a large preprocessed file, namely X86ISelLowering.cpp,
for the X86 target. During the experiment, the final size of ToCheck
was 8 or less 88% of the time.
Merge commit 9e95c4947d31 from llvm-project (by Nikita Popov):
[DSE] Fix non-determinism due to address reuse (#84943)
The malloc->calloc fold creates a new MemoryAccess, which may end of at
the same address as a previously deleted access inside SkipStores.
To the most part, this is not a problem, because SkipStores is normally
only used together with MemDefs. Neither the old malloc access nor the
new calloc access will be part of MemDefs, so there is no problem here.
However, SkipStores is also used in one more place: In the main DSE
loop, ToCheck entries are checked against it. Fix this by not using
SkipStores here, and instead using a separate set to track deletions
inside this loop. This way it is not affected by the calloc optimization
that happens outside it.
This is all pretty ugly, but I haven't found another good way to fix it.
Suggestions welcome.
No test case as I don't have a reliable DSE-only test-case for this.
Fixes https://github.com/llvm/llvm-project/issues/84458.
This fixes another possible difference in output when building i386
object files with a native or cross build of clang. (Specifically, the
file sbin/ipf/ipmon/ipmon.o.)
PR: 276961
Reported by: cperciva
MFC after: 3 days
diff --git llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp
index 380d65836553..f0f0f5f28025 100644
--- llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp
+++ llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp
@@ -1697,7 +1697,9 @@ struct DSEState {
/// Delete dead memory defs and recursively add their operands to ToRemove if
/// they became dead.
- void deleteDeadInstruction(Instruction *SI) {
+ void
+ deleteDeadInstruction(Instruction *SI,
+ SmallPtrSetImpl<MemoryAccess *> *Deleted = nullptr) {
MemorySSAUpdater Updater(&MSSA);
SmallVector<Instruction *, 32> NowDeadInsts;
NowDeadInsts.push_back(SI);
@@ -1718,6 +1720,8 @@ struct DSEState {
if (IsMemDef) {
auto *MD = cast<MemoryDef>(MA);
SkipStores.insert(MD);
+ if (Deleted)
+ Deleted->insert(MD);
if (auto *SI = dyn_cast<StoreInst>(MD->getMemoryInst())) {
if (SI->getValueOperand()->getType()->isPointerTy()) {
const Value *UO = getUnderlyingObject(SI->getValueOperand());
@@ -2111,7 +2115,12 @@ static bool eliminateDeadStores(Function &F, AliasAnalysis &AA, MemorySSA &MSSA,
unsigned WalkerStepLimit = MemorySSAUpwardsStepLimit;
unsigned PartialLimit = MemorySSAPartialStoreLimit;
// Worklist of MemoryAccesses that may be killed by KillingDef.
- SetVector<MemoryAccess *> ToCheck;
+ SmallSetVector<MemoryAccess *, 8> ToCheck;
+ // Track MemoryAccesses that have been deleted in the loop below, so we can
+ // skip them. Don't use SkipStores for this, which may contain reused
+ // MemoryAccess addresses.
+ SmallPtrSet<MemoryAccess *, 8> Deleted;
+ [[maybe_unused]] unsigned OrigNumSkipStores = State.SkipStores.size();
ToCheck.insert(KillingDef->getDefiningAccess());
bool Shortend = false;
@@ -2119,7 +2128,7 @@ static bool eliminateDeadStores(Function &F, AliasAnalysis &AA, MemorySSA &MSSA,
// Check if MemoryAccesses in the worklist are killed by KillingDef.
for (unsigned I = 0; I < ToCheck.size(); I++) {
MemoryAccess *Current = ToCheck[I];
- if (State.SkipStores.count(Current))
+ if (Deleted.contains(Current))
continue;
std::optional<MemoryAccess *> MaybeDeadAccess = State.getDomMemoryDef(
@@ -2166,7 +2175,7 @@ static bool eliminateDeadStores(Function &F, AliasAnalysis &AA, MemorySSA &MSSA,
continue;
LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *DeadI
<< "\n KILLER: " << *KillingI << '\n');
- State.deleteDeadInstruction(DeadI);
+ State.deleteDeadInstruction(DeadI, &Deleted);
++NumFastStores;
MadeChange = true;
} else {
@@ -2203,7 +2212,7 @@ static bool eliminateDeadStores(Function &F, AliasAnalysis &AA, MemorySSA &MSSA,
Shortend = true;
// Remove killing store and remove any outstanding overlap
// intervals for the updated store.
- State.deleteDeadInstruction(KillingSI);
+ State.deleteDeadInstruction(KillingSI, &Deleted);
auto I = State.IOLs.find(DeadSI->getParent());
if (I != State.IOLs.end())
I->second.erase(DeadSI);
@@ -2215,13 +2224,16 @@ static bool eliminateDeadStores(Function &F, AliasAnalysis &AA, MemorySSA &MSSA,
if (OR == OW_Complete) {
LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *DeadI
<< "\n KILLER: " << *KillingI << '\n');
- State.deleteDeadInstruction(DeadI);
+ State.deleteDeadInstruction(DeadI, &Deleted);
++NumFastStores;
MadeChange = true;
}
}
}
+ assert(State.SkipStores.size() - OrigNumSkipStores == Deleted.size() &&
+ "SkipStores and Deleted out of sync?");
+
// Check if the store is a no-op.
if (!Shortend && State.storeIsNoop(KillingDef, KillingUndObj)) {
LLVM_DEBUG(dbgs() << "DSE: Remove No-Op Store:\n DEAD: " << *KillingI