aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/X86
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/X86')
-rw-r--r--llvm/lib/Target/X86/X86.h14
-rw-r--r--llvm/lib/Target/X86/X86AvoidTrailingCall.cpp33
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp11
-rw-r--r--llvm/lib/Target/X86/X86PassRegistry.def2
-rw-r--r--llvm/lib/Target/X86/X86TargetMachine.cpp4
5 files changed, 51 insertions, 13 deletions
diff --git a/llvm/lib/Target/X86/X86.h b/llvm/lib/Target/X86/X86.h
index fa23656..2b83d57 100644
--- a/llvm/lib/Target/X86/X86.h
+++ b/llvm/lib/Target/X86/X86.h
@@ -14,6 +14,7 @@
#ifndef LLVM_LIB_TARGET_X86_X86_H
#define LLVM_LIB_TARGET_X86_X86_H
+#include "llvm/CodeGen/MachineFunctionAnalysisManager.h"
#include "llvm/IR/Analysis.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Support/CodeGen.h"
@@ -104,7 +105,16 @@ FunctionPass *createX86LowerTileCopyPass();
/// CALL instruction. The pass does the same for each funclet as well. This
/// ensures that the open interval of function start and end PCs contains all
/// return addresses for the benefit of the Windows x64 unwinder.
-FunctionPass *createX86AvoidTrailingCallPass();
+class X86AvoidTrailingCallPass
+ : public PassInfoMixin<X86AvoidTrailingCallPass> {
+public:
+ X86AvoidTrailingCallPass() = default;
+ PreservedAnalyses run(MachineFunction &MF,
+ MachineFunctionAnalysisManager &MFAM);
+ static bool isRequired() { return true; }
+};
+
+FunctionPass *createX86AvoidTrailingCallLegacyPass();
/// Return a pass that optimizes the code-size of x86 call sequences. This is
/// done by replacing esp-relative movs with pushes.
@@ -222,7 +232,7 @@ void initializeX86FixupInstTuningPassPass(PassRegistry &);
void initializeX86FixupVectorConstantsPassPass(PassRegistry &);
void initializeWinEHStatePassPass(PassRegistry &);
void initializeX86AvoidSFBPassPass(PassRegistry &);
-void initializeX86AvoidTrailingCallPassPass(PassRegistry &);
+void initializeX86AvoidTrailingCallLegacyPassPass(PassRegistry &);
void initializeX86CallFrameOptimizationPass(PassRegistry &);
void initializeX86CmovConverterPassPass(PassRegistry &);
void initializeX86DAGToDAGISelLegacyPass(PassRegistry &);
diff --git a/llvm/lib/Target/X86/X86AvoidTrailingCall.cpp b/llvm/lib/Target/X86/X86AvoidTrailingCall.cpp
index 2ecf493..ebd4284 100644
--- a/llvm/lib/Target/X86/X86AvoidTrailingCall.cpp
+++ b/llvm/lib/Target/X86/X86AvoidTrailingCall.cpp
@@ -37,6 +37,8 @@
#include "X86Subtarget.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/IR/Analysis.h"
+#include "llvm/IR/PassManager.h"
#define AVOIDCALL_DESC "X86 avoid trailing call pass"
#define AVOIDCALL_NAME "x86-avoid-trailing-call"
@@ -46,9 +48,9 @@
using namespace llvm;
namespace {
-class X86AvoidTrailingCallPass : public MachineFunctionPass {
+class X86AvoidTrailingCallLegacyPass : public MachineFunctionPass {
public:
- X86AvoidTrailingCallPass() : MachineFunctionPass(ID) {}
+ X86AvoidTrailingCallLegacyPass() : MachineFunctionPass(ID) {}
bool runOnMachineFunction(MachineFunction &MF) override;
@@ -59,13 +61,14 @@ private:
};
} // end anonymous namespace
-char X86AvoidTrailingCallPass::ID = 0;
+char X86AvoidTrailingCallLegacyPass::ID = 0;
-FunctionPass *llvm::createX86AvoidTrailingCallPass() {
- return new X86AvoidTrailingCallPass();
+FunctionPass *llvm::createX86AvoidTrailingCallLegacyPass() {
+ return new X86AvoidTrailingCallLegacyPass();
}
-INITIALIZE_PASS(X86AvoidTrailingCallPass, AVOIDCALL_NAME, AVOIDCALL_DESC, false, false)
+INITIALIZE_PASS(X86AvoidTrailingCallLegacyPass, AVOIDCALL_NAME, AVOIDCALL_DESC,
+ false, false)
// A real instruction is a non-meta, non-pseudo instruction. Some pseudos
// expand to nothing, and some expand to code. This logic conservatively assumes
@@ -79,7 +82,7 @@ static bool isCallInstruction(const MachineInstr &MI) {
return MI.isCall() && !MI.isReturn();
}
-bool X86AvoidTrailingCallPass::runOnMachineFunction(MachineFunction &MF) {
+bool UpdatedOnX86AvoidTrailingCallPass(MachineFunction &MF) {
const X86Subtarget &STI = MF.getSubtarget<X86Subtarget>();
const X86InstrInfo &TII = *STI.getInstrInfo();
assert(STI.isTargetWin64() && "pass only runs on Win64");
@@ -134,3 +137,19 @@ bool X86AvoidTrailingCallPass::runOnMachineFunction(MachineFunction &MF) {
return Changed;
}
+
+bool X86AvoidTrailingCallLegacyPass::runOnMachineFunction(MachineFunction &MF) {
+ return UpdatedOnX86AvoidTrailingCallPass(MF);
+}
+
+PreservedAnalyses
+X86AvoidTrailingCallPass::run(MachineFunction &MF,
+ MachineFunctionAnalysisManager &MFAM) {
+ bool Changed = UpdatedOnX86AvoidTrailingCallPass(MF);
+ if (!Changed)
+ return PreservedAnalyses::all();
+
+ PreservedAnalyses PA = PreservedAnalyses::none();
+ PA.preserveSet<CFGAnalyses>();
+ return PA;
+}
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 168e041..090f649 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -3454,6 +3454,12 @@ bool X86TargetLowering::isLoadBitCastBeneficial(EVT LoadVT, EVT BitcastVT,
isTypeLegal(LoadVT) && isTypeLegal(BitcastVT))
return true;
+ // If we have a large vector type (even if illegal), don't bitcast to large
+ // (illegal) scalar types. Better to load fewer vectors and extract.
+ if (LoadVT.isVector() && !BitcastVT.isVector() && LoadVT.isInteger() &&
+ BitcastVT.isInteger() && (LoadVT.getSizeInBits() % 128) == 0)
+ return false;
+
return TargetLowering::isLoadBitCastBeneficial(LoadVT, BitcastVT, DAG, MMO);
}
@@ -53354,6 +53360,7 @@ static SDValue combineMaskedStore(SDNode *N, SelectionDAG &DAG,
// i32 sub value.
static SDValue narrowBitOpRMW(StoreSDNode *St, const SDLoc &DL,
SelectionDAG &DAG,
+ TargetLowering::DAGCombinerInfo &DCI,
const X86Subtarget &Subtarget) {
using namespace SDPatternMatch;
SDValue StoredVal = St->getValue();
@@ -53451,6 +53458,8 @@ static SDValue narrowBitOpRMW(StoreSDNode *St, const SDLoc &DL,
if (!StoredVal.hasOneUse()) {
SDValue NewLoad =
DAG.getLoad(VT, DL, NewStore, Ld->getBasePtr(), Ld->getMemOperand());
+ for (SDNode *User : StoredVal->users())
+ DCI.AddToWorklist(User);
DAG.ReplaceAllUsesWith(StoredVal, NewLoad);
}
return NewStore;
@@ -53682,7 +53691,7 @@ static SDValue combineStore(SDNode *N, SelectionDAG &DAG,
}
}
- if (SDValue R = narrowBitOpRMW(St, dl, DAG, Subtarget))
+ if (SDValue R = narrowBitOpRMW(St, dl, DAG, DCI, Subtarget))
return R;
// Convert store(cmov(load(p), x, CC), p) to cstore(x, p, CC)
diff --git a/llvm/lib/Target/X86/X86PassRegistry.def b/llvm/lib/Target/X86/X86PassRegistry.def
index db25594..5246362 100644
--- a/llvm/lib/Target/X86/X86PassRegistry.def
+++ b/llvm/lib/Target/X86/X86PassRegistry.def
@@ -29,6 +29,7 @@ DUMMY_FUNCTION_PASS("x86-winehstate", WinEHStatePass())
#ifndef MACHINE_FUNCTION_PASS
#define MACHINE_FUNCTION_PASS(NAME, CREATE_PASS)
#endif
+MACHINE_FUNCTION_PASS("x86-avoid-trailing-call", X86AvoidTrailingCallPass())
MACHINE_FUNCTION_PASS("x86-isel", X86ISelDAGToDAGPass(*this))
#undef MACHINE_FUNCTION_PASS
@@ -36,7 +37,6 @@ MACHINE_FUNCTION_PASS("x86-isel", X86ISelDAGToDAGPass(*this))
#define DUMMY_MACHINE_FUNCTION_PASS(NAME, PASS_NAME)
#endif
DUMMY_MACHINE_FUNCTION_PASS("x86-avoid-SFB", X86AvoidSFBPass())
-DUMMY_MACHINE_FUNCTION_PASS("x86-avoid-trailing-call", X86AvoidTrailingCallPass())
DUMMY_MACHINE_FUNCTION_PASS("x86-cf-opt", X86CallFrameOptimization())
DUMMY_MACHINE_FUNCTION_PASS("x86-cmov-conversion", X86CmovConverterPass())
DUMMY_MACHINE_FUNCTION_PASS("x86-codege", FPS())
diff --git a/llvm/lib/Target/X86/X86TargetMachine.cpp b/llvm/lib/Target/X86/X86TargetMachine.cpp
index 5f0bcab..0c2bd7c 100644
--- a/llvm/lib/Target/X86/X86TargetMachine.cpp
+++ b/llvm/lib/Target/X86/X86TargetMachine.cpp
@@ -90,7 +90,7 @@ extern "C" LLVM_C_ABI void LLVMInitializeX86Target() {
initializeX86ExecutionDomainFixPass(PR);
initializeX86DomainReassignmentPass(PR);
initializeX86AvoidSFBPassPass(PR);
- initializeX86AvoidTrailingCallPassPass(PR);
+ initializeX86AvoidTrailingCallLegacyPassPass(PR);
initializeX86SpeculativeLoadHardeningPassPass(PR);
initializeX86SpeculativeExecutionSideEffectSuppressionPass(PR);
initializeX86FlagsCopyLoweringPassPass(PR);
@@ -589,7 +589,7 @@ void X86PassConfig::addPreEmitPass2() {
// Insert extra int3 instructions after trailing call instructions to avoid
// issues in the unwinder.
if (TT.isOSWindows() && TT.isX86_64())
- addPass(createX86AvoidTrailingCallPass());
+ addPass(createX86AvoidTrailingCallLegacyPass());
// Verify basic block incoming and outgoing cfa offset and register values and
// correct CFA calculation rule where needed by inserting appropriate CFI
generated by cgit v1.1 at 2025-11-07 17:27:42 +0000