[WebAssembly] Enable a limited amount of stackification for debug code #136510
+220
−143
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
SingleAccretion
force-pushed
the
Dbg-Stk
branch
from
8b39c4a to
3cc8acd
Compare
This change is a step towards fixing one long-standing problem with LLVM's debug WASM codegen: excessive use of locals. One local for each temporary value in IR (roughly speaking). This has a lot of problems: 1) It makes it easy to hit engine limitations of 50K locals with certain code patterns and large functions. 2) It makes for larger binaries that are slower to load and slower to compile to native code. 3) It makes certain compilation strategies (spill all WASM locals to stack, for example) for debug code excessively expensive and makes debug WASM code either run very slow, or be less debuggable. 4) It slows down LLVM itself. This change addresses these partially by running a limited version of the stackification pass for unoptimized code, one that gets rid of the most 'obviously' unnecessary locals. Care needs to be taken to not impact LLVM's ability to produce high quality debug variable locations with this pass. To that end: 1) We only allow stackification when it doesn't require moving any instructions. 2) We disable stackification of any locals that are used in DEBUG_VALUEs, or as a frame base. I have verified on a moderately large example that the baseline and the diff produce the same kinds (local/global/stack) of locations, and the only differences are due to the shifting of instruction offsets, with many local.[get|set]s not being present anymore. Even with this quite conservative approach, the results are pretty good: 1) 30% reduction in raw code size, up to 10x reduction in the number of locals for select large methods (~1000 => ~100). 2) ~10% reduction in instructions retired for an "llc -O0" run on a moderately sized input.
SingleAccretion
force-pushed
the
Dbg-Stk
branch
from
3cc8acd to
bb9e2ae
Compare
|
@llvm/pr-subscribers-backend-webassembly Author: None (SingleAccretion) ChangesThis change is a step towards fixing one long-standing problem with LLVM's debug WASM codegen: excessive use of locals. One local for each temporary value in IR (roughly speaking). This has a lot of problems:
This change addresses these partially by running a limited version of the stackification pass for unoptimized code, one that gets rid of the most 'obviously' unnecessary locals. Care needs to be taken to not impact LLVM's ability to produce high quality debug variable locations with this pass. To that end:
Even with this quite conservative approach, the results are pretty good:
Patch is 31.52 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/136510.diff 9 Files Affected:
diff --git a/llvm/lib/Target/WebAssembly/WebAssembly.h b/llvm/lib/Target/WebAssembly/WebAssembly.h
index 8f142fa0928ca..7c455c2168cde 100644
--- a/llvm/lib/Target/WebAssembly/WebAssembly.h
+++ b/llvm/lib/Target/WebAssembly/WebAssembly.h
@@ -44,7 +44,7 @@ FunctionPass *createWebAssemblyReplacePhysRegs();
FunctionPass *createWebAssemblyNullifyDebugValueLists();
FunctionPass *createWebAssemblyOptimizeLiveIntervals();
FunctionPass *createWebAssemblyMemIntrinsicResults();
-FunctionPass *createWebAssemblyRegStackify();
+FunctionPass *createWebAssemblyRegStackify(CodeGenOptLevel OptLevel);
FunctionPass *createWebAssemblyRegColoring();
FunctionPass *createWebAssemblyFixBrTableDefaults();
FunctionPass *createWebAssemblyFixIrreducibleControlFlow();
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
index 428d573668fb4..05c89bd0e09fa 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
@@ -41,14 +41,18 @@ using namespace llvm;
namespace {
class WebAssemblyRegStackify final : public MachineFunctionPass {
+ bool Optimize;
+
StringRef getPassName() const override {
return "WebAssembly Register Stackify";
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
- AU.addRequired<MachineDominatorTreeWrapperPass>();
- AU.addRequired<LiveIntervalsWrapperPass>();
+ if (Optimize) {
+ AU.addRequired<LiveIntervalsWrapperPass>();
+ AU.addRequired<MachineDominatorTreeWrapperPass>();
+ }
AU.addPreserved<MachineBlockFrequencyInfoWrapperPass>();
AU.addPreserved<SlotIndexesWrapperPass>();
AU.addPreserved<LiveIntervalsWrapperPass>();
@@ -61,7 +65,9 @@ class WebAssemblyRegStackify final : public MachineFunctionPass {
public:
static char ID; // Pass identification, replacement for typeid
- WebAssemblyRegStackify() : MachineFunctionPass(ID) {}
+ WebAssemblyRegStackify(CodeGenOptLevel OptLevel)
+ : MachineFunctionPass(ID), Optimize(OptLevel != CodeGenOptLevel::None) {}
+ WebAssemblyRegStackify() : WebAssemblyRegStackify(CodeGenOptLevel::Default) {}
};
} // end anonymous namespace
@@ -70,8 +76,8 @@ INITIALIZE_PASS(WebAssemblyRegStackify, DEBUG_TYPE,
"Reorder instructions to use the WebAssembly value stack",
false, false)
-FunctionPass *llvm::createWebAssemblyRegStackify() {
- return new WebAssemblyRegStackify();
+FunctionPass *llvm::createWebAssemblyRegStackify(CodeGenOptLevel OptLevel) {
+ return new WebAssemblyRegStackify(OptLevel);
}
// Decorate the given instruction with implicit operands that enforce the
@@ -96,8 +102,7 @@ static void imposeStackOrdering(MachineInstr *MI) {
static void convertImplicitDefToConstZero(MachineInstr *MI,
MachineRegisterInfo &MRI,
const TargetInstrInfo *TII,
- MachineFunction &MF,
- LiveIntervals &LIS) {
+ MachineFunction &MF) {
assert(MI->getOpcode() == TargetOpcode::IMPLICIT_DEF);
const auto *RegClass = MRI.getRegClass(MI->getOperand(0).getReg());
@@ -262,36 +267,52 @@ static bool shouldRematerialize(const MachineInstr &Def,
// LiveIntervals to handle complex cases.
static MachineInstr *getVRegDef(unsigned Reg, const MachineInstr *Insert,
const MachineRegisterInfo &MRI,
- const LiveIntervals &LIS) {
+ const LiveIntervals *LIS) {
// Most registers are in SSA form here so we try a quick MRI query first.
if (MachineInstr *Def = MRI.getUniqueVRegDef(Reg))
return Def;
// MRI doesn't know what the Def is. Try asking LIS.
- if (const VNInfo *ValNo = LIS.getInterval(Reg).getVNInfoBefore(
- LIS.getInstructionIndex(*Insert)))
- return LIS.getInstructionFromIndex(ValNo->def);
+ if (LIS != nullptr) {
+ SlotIndex InstIndex = LIS->getInstructionIndex(*Insert);
+ if (const VNInfo *ValNo = LIS->getInterval(Reg).getVNInfoBefore(InstIndex))
+ return LIS->getInstructionFromIndex(ValNo->def);
+ }
return nullptr;
}
// Test whether Reg, as defined at Def, has exactly one use. This is a
// generalization of MachineRegisterInfo::hasOneNonDBGUse that uses
-// LiveIntervals to handle complex cases.
-static bool hasOneNonDBGUse(unsigned Reg, MachineInstr *Def,
- MachineRegisterInfo &MRI, MachineDominatorTree &MDT,
- LiveIntervals &LIS) {
+// LiveIntervals to handle complex cases in optimized code.
+static bool hasSingleUse(unsigned Reg, MachineRegisterInfo &MRI,
+ WebAssemblyFunctionInfo &MFI, bool Optimize,
+ MachineInstr *Def, LiveIntervals *LIS) {
+ if (!Optimize) {
+ // We don't want to stackify DBG_VALUE operands since WASM stack locations
+ // are less useful and less widely supported than WASM local locations.
+ if (!MRI.hasOneUse(Reg))
+ return false;
+ // The frame base always has an implicit DBG use as DW_AT_frame_base.
+ if (MFI.isFrameBaseVirtual() && MFI.getFrameBaseVreg() == Reg)
+ return false;
+ return true;
+ }
+
// Most registers are in SSA form here so we try a quick MRI query first.
if (MRI.hasOneNonDBGUse(Reg))
return true;
+ if (LIS == nullptr)
+ return false;
+
bool HasOne = false;
- const LiveInterval &LI = LIS.getInterval(Reg);
+ const LiveInterval &LI = LIS->getInterval(Reg);
const VNInfo *DefVNI =
- LI.getVNInfoAt(LIS.getInstructionIndex(*Def).getRegSlot());
+ LI.getVNInfoAt(LIS->getInstructionIndex(*Def).getRegSlot());
assert(DefVNI);
for (auto &I : MRI.use_nodbg_operands(Reg)) {
- const auto &Result = LI.Query(LIS.getInstructionIndex(*I.getParent()));
+ const auto &Result = LI.Query(LIS->getInstructionIndex(*I.getParent()));
if (Result.valueIn() == DefVNI) {
if (!Result.isKill())
return false;
@@ -311,7 +332,7 @@ static bool hasOneNonDBGUse(unsigned Reg, MachineInstr *Def,
static bool isSafeToMove(const MachineOperand *Def, const MachineOperand *Use,
const MachineInstr *Insert,
const WebAssemblyFunctionInfo &MFI,
- const MachineRegisterInfo &MRI) {
+ const MachineRegisterInfo &MRI, bool Optimize) {
const MachineInstr *DefI = Def->getParent();
const MachineInstr *UseI = Use->getParent();
assert(DefI->getParent() == Insert->getParent());
@@ -357,6 +378,12 @@ static bool isSafeToMove(const MachineOperand *Def, const MachineOperand *Use,
if (NextI == Insert)
return true;
+ // When not optimizing, we only handle the trivial case above
+ // to guarantee no impact to debugging and to avoid spending
+ // compile time.
+ if (!Optimize)
+ return false;
+
// 'catch' and 'catch_all' should be the first instruction of a BB and cannot
// move.
if (WebAssembly::isCatch(DefI->getOpcode()))
@@ -520,14 +547,15 @@ static void shrinkToUses(LiveInterval &LI, LiveIntervals &LIS) {
/// dependencies; move the def down and nest it with the current instruction.
static MachineInstr *moveForSingleUse(unsigned Reg, MachineOperand &Op,
MachineInstr *Def, MachineBasicBlock &MBB,
- MachineInstr *Insert, LiveIntervals &LIS,
+ MachineInstr *Insert, LiveIntervals *LIS,
WebAssemblyFunctionInfo &MFI,
MachineRegisterInfo &MRI) {
LLVM_DEBUG(dbgs() << "Move for single use: "; Def->dump());
WebAssemblyDebugValueManager DefDIs(Def);
DefDIs.sink(Insert);
- LIS.handleMove(*Def);
+ if (LIS != nullptr)
+ LIS->handleMove(*Def);
if (MRI.hasOneDef(Reg) && MRI.hasOneNonDBGUse(Reg)) {
// No one else is using this register for anything so we can just stackify
@@ -540,17 +568,18 @@ static MachineInstr *moveForSingleUse(unsigned Reg, MachineOperand &Op,
Op.setReg(NewReg);
DefDIs.updateReg(NewReg);
- // Tell LiveIntervals about the new register.
- LIS.createAndComputeVirtRegInterval(NewReg);
+ if (LIS != nullptr) {
+ // Tell LiveIntervals about the new register.
+ LIS->createAndComputeVirtRegInterval(NewReg);
- // Tell LiveIntervals about the changes to the old register.
- LiveInterval &LI = LIS.getInterval(Reg);
- LI.removeSegment(LIS.getInstructionIndex(*Def).getRegSlot(),
- LIS.getInstructionIndex(*Op.getParent()).getRegSlot(),
- /*RemoveDeadValNo=*/true);
+ // Tell LiveIntervals about the changes to the old register.
+ LiveInterval &LI = LIS->getInterval(Reg);
+ LI.removeSegment(LIS->getInstructionIndex(*Def).getRegSlot(),
+ LIS->getInstructionIndex(*Op.getParent()).getRegSlot(),
+ /*RemoveDeadValNo=*/true);
+ }
MFI.stackifyVReg(MRI, NewReg);
-
LLVM_DEBUG(dbgs() << " - Replaced register: "; Def->dump());
}
@@ -567,11 +596,12 @@ static MachineInstr *getPrevNonDebugInst(MachineInstr *MI) {
/// A trivially cloneable instruction; clone it and nest the new copy with the
/// current instruction.
-static MachineInstr *rematerializeCheapDef(
- unsigned Reg, MachineOperand &Op, MachineInstr &Def, MachineBasicBlock &MBB,
- MachineBasicBlock::instr_iterator Insert, LiveIntervals &LIS,
- WebAssemblyFunctionInfo &MFI, MachineRegisterInfo &MRI,
- const WebAssemblyInstrInfo *TII, const WebAssemblyRegisterInfo *TRI) {
+static MachineInstr *
+rematerializeCheapDef(unsigned Reg, MachineOperand &Op, MachineInstr &Def,
+ MachineBasicBlock::instr_iterator Insert,
+ LiveIntervals &LIS, WebAssemblyFunctionInfo &MFI,
+ MachineRegisterInfo &MRI,
+ const WebAssemblyInstrInfo *TII) {
LLVM_DEBUG(dbgs() << "Rematerializing cheap def: "; Def.dump());
LLVM_DEBUG(dbgs() << " - for use in "; Op.getParent()->dump());
@@ -811,9 +841,12 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
MachineRegisterInfo &MRI = MF.getRegInfo();
WebAssemblyFunctionInfo &MFI = *MF.getInfo<WebAssemblyFunctionInfo>();
const auto *TII = MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo();
- const auto *TRI = MF.getSubtarget<WebAssemblySubtarget>().getRegisterInfo();
- auto &MDT = getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
- auto &LIS = getAnalysis<LiveIntervalsWrapperPass>().getLIS();
+ MachineDominatorTree *MDT = nullptr;
+ LiveIntervals *LIS = nullptr;
+ if (Optimize) {
+ MDT = &getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
+ LIS = &getAnalysis<LiveIntervalsWrapperPass>().getLIS();
+ }
// Walk the instructions from the bottom up. Currently we don't look past
// block boundaries, and the blocks aren't ordered so the block visitation
@@ -876,23 +909,28 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
// supports intra-block moves) and it's MachineSink's job to catch all
// the sinking opportunities anyway.
bool SameBlock = DefI->getParent() == &MBB;
- bool CanMove = SameBlock && isSafeToMove(Def, &Use, Insert, MFI, MRI) &&
+ bool CanMove = SameBlock &&
+ isSafeToMove(Def, &Use, Insert, MFI, MRI, Optimize) &&
!TreeWalker.isOnStack(Reg);
- if (CanMove && hasOneNonDBGUse(Reg, DefI, MRI, MDT, LIS)) {
+ if (CanMove && hasSingleUse(Reg, MRI, MFI, Optimize, DefI, LIS)) {
Insert = moveForSingleUse(Reg, Use, DefI, MBB, Insert, LIS, MFI, MRI);
// If we are removing the frame base reg completely, remove the debug
// info as well.
// TODO: Encode this properly as a stackified value.
- if (MFI.isFrameBaseVirtual() && MFI.getFrameBaseVreg() == Reg)
+ if (MFI.isFrameBaseVirtual() && MFI.getFrameBaseVreg() == Reg) {
+ assert(
+ Optimize &&
+ "Stackifying away frame base in unoptimized code not expected");
MFI.clearFrameBaseVreg();
- } else if (shouldRematerialize(*DefI, TII)) {
- Insert =
- rematerializeCheapDef(Reg, Use, *DefI, MBB, Insert->getIterator(),
- LIS, MFI, MRI, TII, TRI);
- } else if (CanMove && oneUseDominatesOtherUses(Reg, Use, MBB, MRI, MDT,
- LIS, MFI)) {
- Insert = moveAndTeeForMultiUse(Reg, Use, DefI, MBB, Insert, LIS, MFI,
+ }
+ } else if (Optimize && shouldRematerialize(*DefI, TII)) {
+ Insert = rematerializeCheapDef(Reg, Use, *DefI, Insert->getIterator(),
+ *LIS, MFI, MRI, TII);
+ } else if (Optimize && CanMove &&
+ oneUseDominatesOtherUses(Reg, Use, MBB, MRI, *MDT, *LIS,
+ MFI)) {
+ Insert = moveAndTeeForMultiUse(Reg, Use, DefI, MBB, Insert, *LIS, MFI,
MRI, TII);
} else {
// We failed to stackify the operand. If the problem was ordering
@@ -915,7 +953,8 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
Register DefReg = SubsequentDef->getReg();
Register UseReg = SubsequentUse->getReg();
// TODO: This single-use restriction could be relaxed by using tees
- if (DefReg != UseReg || !MRI.hasOneNonDBGUse(DefReg))
+ if (DefReg != UseReg ||
+ !hasSingleUse(DefReg, MRI, MFI, Optimize, nullptr, nullptr))
break;
MFI.stackifyVReg(MRI, DefReg);
++SubsequentDef;
@@ -926,7 +965,7 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
// to a constant 0 so that the def is explicit, and the push/pop
// correspondence is maintained.
if (Insert->getOpcode() == TargetOpcode::IMPLICIT_DEF)
- convertImplicitDefToConstZero(Insert, MRI, TII, MF, LIS);
+ convertImplicitDefToConstZero(Insert, MRI, TII, MF);
// We stackified an operand. Add the defining instruction's operands to
// the worklist stack now to continue to build an ever deeper tree.
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
index 8ec72d5c47833..6c637f11be0ec 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp
@@ -603,14 +603,16 @@ void WebAssemblyPassConfig::addPreEmitPass() {
// Prepare memory intrinsic calls for register stackifying.
addPass(createWebAssemblyMemIntrinsicResults());
+ }
- // Mark registers as representing wasm's value stack. This is a key
- // code-compression technique in WebAssembly. We run this pass (and
- // MemIntrinsicResults above) very late, so that it sees as much code as
- // possible, including code emitted by PEI and expanded by late tail
- // duplication.
- addPass(createWebAssemblyRegStackify());
+ // Mark registers as representing wasm's value stack. This is a key
+ // code-compression technique in WebAssembly. We run this pass (and
+ // MemIntrinsicResults above) very late, so that it sees as much code as
+ // possible, including code emitted by PEI and expanded by late tail
+ // duplication.
+ addPass(createWebAssemblyRegStackify(getOptLevel()));
+ if (getOptLevel() != CodeGenOptLevel::None) {
// Run the register coloring pass to reduce the total number of registers.
// This runs after stackification so that it doesn't consider registers
// that become stackified.
diff --git a/llvm/test/CodeGen/WebAssembly/PR40172.ll b/llvm/test/CodeGen/WebAssembly/PR40172.ll
index ed1630c4926d4..e70b115e67478 100644
--- a/llvm/test/CodeGen/WebAssembly/PR40172.ll
+++ b/llvm/test/CodeGen/WebAssembly/PR40172.ll
@@ -10,9 +10,9 @@ target triple = "wasm32-unknown-unknown"
; CHECK: i32.sub $[[BASE:[0-9]+]]=,
; CHECK: local.copy $[[ARG:[0-9]+]]=, $0{{$}}
-; CHECK: i32.const $[[A0:[0-9]+]]=, 1{{$}}
-; CHECK: i32.and $[[A1:[0-9]+]]=, $[[ARG]], $[[A0]]{{$}}
-; CHECK: i32.store8 8($[[BASE]]), $[[A1]]{{$}}
+; CHECK: i32.const $push[[A0:[0-9]+]]=, 1{{$}}
+; CHECK: i32.and $push[[A1:[0-9]+]]=, $[[ARG]], $pop[[A0]]{{$}}
+; CHECK: i32.store8 8($[[BASE]]), $pop[[A1]]{{$}}
define void @test(i8 %byte) {
%t = alloca { i8, i8 }, align 8
diff --git a/llvm/test/CodeGen/WebAssembly/PR41841.ll b/llvm/test/CodeGen/WebAssembly/PR41841.ll
index 4e3166adf7bb1..76ec83ecf8163 100644
--- a/llvm/test/CodeGen/WebAssembly/PR41841.ll
+++ b/llvm/test/CodeGen/WebAssembly/PR41841.ll
@@ -6,9 +6,9 @@ declare void @foo(i128)
; CHECK-LABEL: test_zext:
; CHECK-NEXT: .functype test_zext (i32) -> (){{$}}
-; CHECK-NEXT: i64.extend_i32_u $[[TMP3:[0-9]+]]=, $0{{$}}
-; CHECK-NEXT: i64.const $[[TMP4:[0-9]+]]=, 1{{$}}
-; CHECK-NEXT: i64.and $[[TMP1:[0-9]+]]=, $[[TMP3]], $[[TMP4]]{{$}}
+; CHECK-NEXT: i64.extend_i32_u $push[[TMP3:[0-9]+]]=, $0{{$}}
+; CHECK-NEXT: i64.const $push[[TMP4:[0-9]+]]=, 1{{$}}
+; CHECK-NEXT: i64.and $[[TMP1:[0-9]+]]=, $pop[[TMP3]], $pop[[TMP4]]{{$}}
; CHECK-NEXT: i64.const $[[TMP2:[0-9]+]]=, 0{{$}}
; CHECK-NEXT: call foo, $[[TMP1]], $[[TMP2]]{{$}}
; CHECK-NEXT: return{{$}}
@@ -23,11 +23,11 @@ next: ; preds = %start
; CHECK-LABEL: test_sext:
; CHECK-NEXT:.functype test_sext (i32) -> (){{$}}
-; CHECK-NEXT: i64.extend_i32_u $[[TMP3:[0-9]+]]=, $0{{$}}
-; CHECK-NEXT: i64.const $[[TMP4:[0-9]+]]=, 1{{$}}
-; CHECK-NEXT: i64.and $[[TMP5:[0-9]+]]=, $[[TMP3]], $[[TMP4]]{{$}}
-; CHECK-NEXT: i64.const $[[TMP6:[0-9]+]]=, 0{{$}}
-; CHECK-NEXT: i64.sub $[[TMP1:[0-9]+]]=, $[[TMP6]], $[[TMP5]]{{$}}
+; CHECK-NEXT: i64.extend_i32_u $push[[TMP3:[0-9]+]]=, $0{{$}}
+; CHECK-NEXT: i64.const $push[[TMP4:[0-9]+]]=, 1{{$}}
+; CHECK-NEXT: i64.and $[[TMP5:[0-9]+]]=, $pop[[TMP3]], $pop[[TMP4]]{{$}}
+; CHECK-NEXT: i64.const $push[[TMP6:[0-9]+]]=, 0{{$}}
+; CHECK-NEXT: i64.sub $[[TMP1:[0-9]+]]=, $pop[[TMP6]], $[[TMP5]]{{$}}
; CHECK-NEXT: local.copy $[[TMP2:[0-9]+]]=, $[[TMP1]]{{$}}
; CHECK-NEXT: call foo, $[[TMP1]], $[[TMP2]]{{$}}
; CHECK-NEXT: return{{$}}
diff --git a/llvm/test/CodeGen/WebAssembly/debug-code-stackification.ll b/llvm/test/CodeGen/WebAssembly/debug-code-stackification.ll
new file mode 100644
index 0000000000000..a2a0ae7267290
--- /dev/null
+++ b/llvm/test/CodeGen/WebAssembly/debug-code-stackification.ll
@@ -0,0 +1,89 @@
+; RUN: llc < %s -O0 --filetype=obj -o - | llvm-dwarfdump - | FileCheck %s --check-prefixes DBG_USE
+
+target triple = "wasm32-unknown-unknown"
+
+declare i32 @extern_func(i32, i32)
+
+; We want to produce local DW_OP_WASM_locations in debug code instead
+; of operand stack locations since local locations are more widely
+; supported and can cover the entirety of the method.
+; DBG_USE: DW_TAG_subprogram
+; DBG_USE: DW_AT_name ("single_non_dbg_use")
+; DBG_USE: DW_TAG_variable
+; DBG_USE: DW_AT_location
+; DBG_USE: DW_OP_WASM_location 0x0
+; DBG_USE: DW_AT_name ("call_value")
+; DBG_USE: DW_TAG_variable
+; DBG_USE: DW_AT_location
+; DBG_USE: DW_OP_WASM_location 0x0
+; DBG_USE: DW_AT_name ("sub_value")
+define i32 @single_non_dbg_use(i32 %0, i32 %1) !dbg !6 {
+ %call_value = call i32 @extern_func(i32 1, i32 2), !dbg !20
+ call void @llvm.dbg.value(metadata i32 %call_value, metadata !11, metadata !DIExpression()), !dbg !20
+ %div = udiv i32 %0, %1, !dbg !21
+ %sub = sub i32 %call_value, %div, !dbg !22
+ call void @llvm.dbg.value(metadata i32 %sub, metadata !12, metadata !DIExpression()), !dbg !22
+ ret i32 %sub, !dbg !23
+}
+
+!6 = distinct !DISubprogram(name: "single_non_dbg_use", scope: !1, file: !1, type: !7, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0)
+!7 = !DISubroutineType(types: !8)
+!8 = !{!9, !9, !9}
+!9 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!11 = !DILocalVariable(name: "call_value", scope: !6, type: !9)
+!12 = !DILocalVariabl...
[truncated]
|
|
I think this is an interesting idea. Thanks for keeping in mind those constraints for debug info, I think those are basically the right goals to have. Does the code you're testing come from C#/CLR or somewhere else? I guess another variation on this could be to do something like enabling register coloring instead, which would reduce the local count (and maybe code size a little, since it could reduce the LEB-encoded local numbers). It's probably easy to try, but my guess is that it wouldn't be as effective as doing this. /cc @aheejin @sunfishcode |
It is indeed C#/IL, mostly from the standard library, compiled with our LLVM-based compiler. It differs from typical C/C++ WASM in that it has a lot of I expect these wins to translate very well to C/C++ code as well (most any code, really). Compiling this little GC implementation, I see a 45% reduction in code size (600Kb -> 335Kb).
Indeed. In (The same is true for our compiler, so to test the location coverage, I forced a mode where it would emit |
|
<Ping> |
sunfishcode
reviewed
May 27, 2025
| @@ -915,7 +953,8 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) { | |||
| Register DefReg = SubsequentDef->getReg(); | |||
| Register UseReg = SubsequentUse->getReg(); | |||
| // TODO: This single-use restriction could be relaxed by using tees | |||
| if (DefReg != UseReg || !MRI.hasOneNonDBGUse(DefReg)) | |||
| if (DefReg != UseReg || | |||
| !hasSingleUse(DefReg, MRI, MFI, Optimize, nullptr, nullptr)) | |||
There was a problem hiding this comment.
This passes nullptr as the Def argument, which appears to be dereferenced unconditionally in hasSingleUse.
There was a problem hiding this comment.
That's ok because Def is only used for the LIS != nullptr case, and we're passing LIS: nullptr here as well. The LIS/Def pair of arguments is sort of an optional precision enhancement.
In fact this part of the change is mostly about defensive coding, since currently we don't collect debug values for the non-first defs.
Already measured - this is actually a big throughput win.
This is because we don't need to insert so many explicit |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This change is a step towards fixing one long-standing problem with LLVM's debug WASM codegen: excessive use of locals. One local for each temporary value in IR (roughly speaking).
This has a lot of problems:
This change addresses these partially by running a limited version of the stackification pass for unoptimized code, one that gets rid of the most 'obviously' unnecessary locals.
Care needs to be taken to not impact LLVM's ability to produce high quality debug variable locations with this pass. To that end:
I have verified on a moderately large example that the baseline and the diff produce the same kinds (local/global/stack) of locations, and the only differences are due to the shifting of instruction offsets, with many local.[get|set]s not being present anymore.
Even with this quite conservative approach, the results are pretty good: