[LiveDebugValues] Use getNumSupportedRegs to reduce compile time, NFCI

[phoebewang]

3rd try to fix compile regression by #113532

[llvmbot]

@llvm/pr-subscribers-debuginfo

Author: Phoebe Wang (phoebewang)

Changes

3rd try to fix compile regression by #113532

---

Full diff: https://github.com/llvm/llvm-project/pull/114944.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.cpp (+12-8)
• (modified) llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.h (+2)

diff --git a/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.cpp b/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.cpp
index a9d28a39c4418b..d99c8039aa3497 100644
--- a/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.cpp
+++ b/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.cpp
@@ -1022,7 +1022,7 @@ MLocTracker::MLocTracker(MachineFunction &MF, const TargetInstrInfo &TII,
                          const TargetLowering &TLI)
     : MF(MF), TII(TII), TRI(TRI), TLI(TLI),
       LocIdxToIDNum(ValueIDNum::EmptyValue), LocIdxToLocID(0) {
-  NumRegs = TRI.getNumRegs();
+  NumRegs = TRI.getNumSupportedRegs(MF);
   reset();
   LocIDToLocIdx.resize(NumRegs, LocIdx::MakeIllegalLoc());
   assert(NumRegs < (1u << NUM_LOC_BITS)); // Detect bit packing failure
@@ -1878,7 +1878,8 @@ void InstrRefBasedLDV::transferRegisterDef(MachineInstr &MI) {
     if (MO.isReg() && MO.isDef() && MO.getReg() && MO.getReg().isPhysical() &&
         !IgnoreSPAlias(MO.getReg())) {
       // Remove ranges of all aliased registers.
-      for (MCRegAliasIterator RAI(MO.getReg(), TRI, true); RAI.isValid(); ++RAI)
+      for (MCRegAliasIterator RAI(MO.getReg(), TRI, true);
+           RAI.isValid() && *RAI < NumRegs; ++RAI)
         // FIXME: Can we break out of this loop early if no insertion occurs?
         DeadRegs.insert(*RAI);
     } else if (MO.isRegMask()) {
@@ -1952,7 +1953,8 @@ void InstrRefBasedLDV::transferRegisterDef(MachineInstr &MI) {
 
 void InstrRefBasedLDV::performCopy(Register SrcRegNum, Register DstRegNum) {
   // In all circumstances, re-def all aliases. It's definitely a new value now.
-  for (MCRegAliasIterator RAI(DstRegNum, TRI, true); RAI.isValid(); ++RAI)
+  for (MCRegAliasIterator RAI(DstRegNum, TRI, true);
+       RAI.isValid() && *RAI < NumRegs; ++RAI)
     MTracker->defReg(*RAI, CurBB, CurInst);
 
   ValueIDNum SrcValue = MTracker->readReg(SrcRegNum);
@@ -2117,7 +2119,8 @@ bool InstrRefBasedLDV::transferSpillOrRestoreInst(MachineInstr &MI) {
     // stack slot.
 
     // Def all registers that alias the destination.
-    for (MCRegAliasIterator RAI(Reg, TRI, true); RAI.isValid(); ++RAI)
+    for (MCRegAliasIterator RAI(Reg, TRI, true);
+         RAI.isValid() && *RAI < NumRegs; ++RAI)
       MTracker->defReg(*RAI, CurBB, CurInst);
 
     // Now find subregisters within the destination register, and load values
@@ -2302,11 +2305,12 @@ void InstrRefBasedLDV::produceMLocTransferFunction(
   // appropriate clobbers.
   SmallVector<BitVector, 32> BlockMasks;
   BlockMasks.resize(MaxNumBlocks);
+  NumRegs = TRI->getNumSupportedRegs(MF);
 
   // Reserve one bit per register for the masks described above.
-  unsigned BVWords = MachineOperand::getRegMaskSize(TRI->getNumRegs());
+  unsigned BVWords = MachineOperand::getRegMaskSize(NumRegs);
   for (auto &BV : BlockMasks)
-    BV.resize(TRI->getNumRegs(), true);
+    BV.resize(NumRegs, true);
 
   // Step through all instructions and inhale the transfer function.
   for (auto &MBB : MF) {
@@ -2370,11 +2374,11 @@ void InstrRefBasedLDV::produceMLocTransferFunction(
   }
 
   // Compute a bitvector of all the registers that are tracked in this block.
-  BitVector UsedRegs(TRI->getNumRegs());
+  BitVector UsedRegs(NumRegs);
   for (auto Location : MTracker->locations()) {
     unsigned ID = MTracker->LocIdxToLocID[Location.Idx];
     // Ignore stack slots, and aliases of the stack pointer.
-    if (ID >= TRI->getNumRegs() || MTracker->SPAliases.count(ID))
+    if (ID >= NumRegs || MTracker->SPAliases.count(ID))
       continue;
     UsedRegs.set(ID);
   }
diff --git a/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.h b/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.h
index 68db65ace9a427..cdb9da236a84e4 100644
--- a/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.h
+++ b/llvm/lib/CodeGen/LiveDebugValues/InstrRefBasedImpl.h
@@ -1253,6 +1253,8 @@ class InstrRefBasedLDV : public LDVImpl {
   /// pointer.
   StringRef StackProbeSymbolName;
 
+  unsigned NumRegs = 0;
+
   /// Tests whether this instruction is a spill to a stack slot.
   std::optional<SpillLocationNo> isSpillInstruction(const MachineInstr &MI,
                                                     MachineFunction *MF);

[phoebewang]

Verified with valgrind with below command
vg-in-place --tool=callgrind clang -DNDEBUG -O3 -fomit-frame-pointer -DNDEBUG -g -w -Werror=date-time -DSTDC_HEADERS=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_SYS_STAT_H=1 -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_MEMORY_H=1 -DHAVE_STRINGS_H=1 -DHAVE_INTTYPES_H=1 -DHAVE_STDINT_H=1 -DHAVE_UNISTD_H=1 -DSQLITE_OMIT_LOAD_EXTENSION=1 -DSQLITE_THREADSAFE=0 -I. -MD -MT shell.c.o -MF shell.c.o.d -o shell.c.o -c MultiSource/Applications/sqlite3/shell.c

Patches	Instructions	Ratio
Without #113532	1790650933	-
With #113532	1793289982	+0.15%
With both	1790138180	-0.03%

[phoebewang]

Need to investigate lit failures.

[jmorse]

(While this is marked draft, reducing the known set of register that might need tracking is a solid plan, so this is a good direction to take).

[nikic]

I wasn't able to confirm the compile-time impact because the LTO build crashed: https://llvm-compile-time-tracker.com/show_error.php?commit=428bd77c9a8807b3780d39ad5b5a757ac65dfaef (Presumably related to the lit test failures.)

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[phoebewang]

The crash issue should be solved, but there's still a lit failure: llvm/test/CodeGen/X86/debuginfo-locations-dce.ll
With this patch, we lack the information of DW_AT_location in

0x00000043:     DW_TAG_variable
                  DW_AT_name   ("MyVar")
                  DW_AT_decl_file      ("./test.cpp")
                  DW_AT_decl_line      (4)
                  DW_AT_type   (0x00000056 "float")

I know little about the pass, so I have difficulty to figure out what's wrong here. I wouldn't expect there's any functional change with this patch.

[nikic]

Compile-time: https://llvm-compile-time-tracker.com/compare.php?from=a256e89fd1b9f42d3f29df42d4f21ea823476ff3&to=a3ffd3d22936650dc23828b92afdb9e1b4d24f1e&stat=instructions:u