[MC] [Win64EH] Try to generate packed unwind info where possible

In practice, this only gives modest savings (for a 6.5 MB DLL with
230 KB xdata, the xdata sections shrinks by around 2.5 KB); to
gain more, the frame lowering would need to be tweaked to more often
generate frame layouts that match the canonical layouts that can
be written in packed form.

Differential Revision: https://reviews.llvm.org/D87371
This commit is contained in:
Martin Storsjö 2020-09-09 10:34:15 +03:00
parent fe3c1195cf
commit f69e090d7d
3 changed files with 1195 additions and 6 deletions

View File

@ -45,6 +45,7 @@ struct FrameInfo {
const MCSymbol *PrologEnd = nullptr;
const MCSymbol *Symbol = nullptr;
MCSection *TextSection = nullptr;
uint32_t PackedInfo = 0;
bool HandlesUnwind = false;
bool HandlesExceptions = false;

View File

@ -648,9 +648,233 @@ static int checkPackedEpilog(MCStreamer &streamer, WinEH::FrameInfo *info,
return Offset;
}
static bool tryPackedUnwind(WinEH::FrameInfo *info, uint32_t FuncLength,
int PackedEpilogOffset) {
if (PackedEpilogOffset == 0) {
// Fully symmetric prolog and epilog, should be ok for packed format.
// For CR=3, the corresponding synthesized epilog actually lacks the
// SetFP opcode, but unwinding should work just fine despite that
// (if at the SetFP opcode, the unwinder considers it as part of the
// function body and just unwinds the full prolog instead).
} else if (PackedEpilogOffset == 1) {
// One single case of differences between prolog and epilog is allowed:
// The epilog can lack a single SetFP that is the last opcode in the
// prolog, for the CR=3 case.
if (info->Instructions.back().Operation != Win64EH::UOP_SetFP)
return false;
} else {
// Too much difference between prolog and epilog.
return false;
}
unsigned RegI = 0, RegF = 0;
int Predecrement = 0;
enum {
Start,
Start2,
IntRegs,
FloatRegs,
InputArgs,
StackAdjust,
FrameRecord,
End
} Location = Start;
bool StandaloneLR = false, FPLRPair = false;
int StackOffset = 0;
int Nops = 0;
// Iterate over the prolog and check that all opcodes exactly match
// the canonical order and form. A more lax check could verify that
// all saved registers are in the expected locations, but not enforce
// the order - that would work fine when unwinding from within
// functions, but not be exactly right if unwinding happens within
// prologs/epilogs.
for (const WinEH::Instruction &Inst : info->Instructions) {
switch (Inst.Operation) {
case Win64EH::UOP_End:
if (Location != Start)
return false;
Location = Start2;
break;
case Win64EH::UOP_SaveR19R20X:
if (Location != Start2)
return false;
Predecrement = Inst.Offset;
RegI = 2;
Location = IntRegs;
break;
case Win64EH::UOP_SaveRegX:
if (Location != Start2)
return false;
Predecrement = Inst.Offset;
if (Inst.Register == 19)
RegI += 1;
else if (Inst.Register == 30)
StandaloneLR = true;
else
return false;
// Odd register; can't be any further int registers.
Location = FloatRegs;
break;
case Win64EH::UOP_SaveRegPX:
// Can't have this in a canonical prologue. Either this has been
// canonicalized into SaveR19R20X or SaveFPLRX, or it's an unsupported
// register pair.
// It can't be canonicalized into SaveR19R20X if the offset is
// larger than 248 bytes, but even with the maximum case with
// RegI=10/RegF=8/CR=1/H=1, we end up with SavSZ = 216, which should
// fit into SaveR19R20X.
// The unwinding opcodes can't describe the otherwise seemingly valid
// case for RegI=1 CR=1, that would start with a
// "stp x19, lr, [sp, #-...]!" as that fits neither SaveRegPX nor
// SaveLRPair.
return false;
case Win64EH::UOP_SaveRegP:
if (Location != IntRegs || Inst.Offset != 8 * RegI ||
Inst.Register != 19 + RegI)
return false;
RegI += 2;
break;
case Win64EH::UOP_SaveReg:
if (Location != IntRegs || Inst.Offset != 8 * RegI)
return false;
if (Inst.Register == 19 + RegI)
RegI += 1;
else if (Inst.Register == 30)
StandaloneLR = true;
else
return false;
// Odd register; can't be any further int registers.
Location = FloatRegs;
break;
case Win64EH::UOP_SaveLRPair:
if (Location != IntRegs || Inst.Offset != 8 * RegI ||
Inst.Register != 19 + RegI)
return false;
RegI += 1;
StandaloneLR = true;
Location = FloatRegs;
break;
case Win64EH::UOP_SaveFRegX:
// Packed unwind can't handle prologs that only save one single
// float register.
return false;
case Win64EH::UOP_SaveFReg:
if (Location != FloatRegs || RegF == 0 || Inst.Register != 8 + RegF ||
Inst.Offset != 8 * (RegI + (StandaloneLR ? 1 : 0) + RegF))
return false;
RegF += 1;
Location = InputArgs;
break;
case Win64EH::UOP_SaveFRegPX:
if (Location != Start2 || Inst.Register != 8)
return false;
Predecrement = Inst.Offset;
RegF = 2;
Location = FloatRegs;
break;
case Win64EH::UOP_SaveFRegP:
if ((Location != IntRegs && Location != FloatRegs) ||
Inst.Register != 8 + RegF ||
Inst.Offset != 8 * (RegI + (StandaloneLR ? 1 : 0) + RegF))
return false;
RegF += 2;
Location = FloatRegs;
break;
case Win64EH::UOP_SaveNext:
if (Location == IntRegs)
RegI += 2;
else if (Location == FloatRegs)
RegF += 2;
else
return false;
break;
case Win64EH::UOP_Nop:
if (Location != IntRegs && Location != FloatRegs && Location != InputArgs)
return false;
Location = InputArgs;
Nops++;
break;
case Win64EH::UOP_AllocSmall:
case Win64EH::UOP_AllocMedium:
if (Location != Start2 && Location != IntRegs && Location != FloatRegs &&
Location != InputArgs && Location != StackAdjust)
return false;
// Can have either a single decrement, or a pair of decrements with
// 4080 and another decrement.
if (StackOffset == 0)
StackOffset = Inst.Offset;
else if (StackOffset != 4080)
return false;
else
StackOffset += Inst.Offset;
Location = StackAdjust;
break;
case Win64EH::UOP_SaveFPLRX:
// Not allowing FPLRX after StackAdjust; if a StackAdjust is used, it
// should be followed by a FPLR instead.
if (Location != Start2 && Location != IntRegs && Location != FloatRegs &&
Location != InputArgs)
return false;
StackOffset = Inst.Offset;
Location = FrameRecord;
FPLRPair = true;
break;
case Win64EH::UOP_SaveFPLR:
// This can only follow after a StackAdjust
if (Location != StackAdjust || Inst.Offset != 0)
return false;
Location = FrameRecord;
FPLRPair = true;
break;
case Win64EH::UOP_SetFP:
if (Location != FrameRecord)
return false;
Location = End;
break;
}
}
if (RegI > 10 || RegF > 8)
return false;
if (StandaloneLR && FPLRPair)
return false;
if (FPLRPair && Location != End)
return false;
if (Nops != 0 && Nops != 4)
return false;
int H = Nops == 4;
int IntSZ = 8 * RegI;
if (StandaloneLR)
IntSZ += 8;
int FpSZ = 8 * RegF; // RegF not yet decremented
int SavSZ = (IntSZ + FpSZ + 8 * 8 * H + 0xF) & ~0xF;
if (Predecrement != SavSZ)
return false;
if (FPLRPair && StackOffset < 16)
return false;
if (StackOffset % 16)
return false;
uint32_t FrameSize = (StackOffset + SavSZ) / 16;
if (FrameSize > 0x1FF)
return false;
assert(RegF != 1 && "One single float reg not allowed");
if (RegF > 0)
RegF--; // Convert from actual number of registers, to value stored
assert(FuncLength <= 0x7FF && "FuncLength should have been checked earlier");
int Flag = 0x01; // Function segments not supported yet
int CR = FPLRPair ? 3 : StandaloneLR ? 1 : 0;
info->PackedInfo |= Flag << 0;
info->PackedInfo |= (FuncLength & 0x7FF) << 2;
info->PackedInfo |= (RegF & 0x7) << 13;
info->PackedInfo |= (RegI & 0xF) << 16;
info->PackedInfo |= (H & 0x1) << 20;
info->PackedInfo |= (CR & 0x3) << 21;
info->PackedInfo |= (FrameSize & 0x1FF) << 23;
return true;
}
// Populate the .xdata section. The format of .xdata on ARM64 is documented at
// https://docs.microsoft.com/en-us/cpp/build/arm64-exception-handling
static void ARM64EmitUnwindInfo(MCStreamer &streamer, WinEH::FrameInfo *info) {
static void ARM64EmitUnwindInfo(MCStreamer &streamer, WinEH::FrameInfo *info,
bool TryPacked = true) {
// If this UNWIND_INFO already has a symbol, it's already been emitted.
if (info->Symbol)
return;
@ -728,6 +952,20 @@ static void ARM64EmitUnwindInfo(MCStreamer &streamer, WinEH::FrameInfo *info) {
int PackedEpilogOffset = checkPackedEpilog(streamer, info, PrologCodeBytes);
if (PackedEpilogOffset >= 0 && !info->HandlesExceptions &&
FuncLength <= 0x7ff && TryPacked) {
// Matching prolog/epilog and no exception handlers; check if the
// prolog matches the patterns that can be described by the packed
// format.
// info->Symbol was already set even if we didn't actually write any
// unwind info there. Keep using that as indicator that this unwind
// info has been generated already.
if (tryPackedUnwind(info, FuncLength, PackedEpilogOffset))
return;
}
// Process epilogs.
MapVector<MCSymbol *, uint32_t> EpilogInfo;
// Epilogs processed so far.
@ -835,10 +1073,13 @@ static void ARM64EmitRuntimeFunction(MCStreamer &streamer,
streamer.emitValueToAlignment(4);
EmitSymbolRefWithOfs(streamer, info->Function, info->Begin);
streamer.emitValue(MCSymbolRefExpr::create(info->Symbol,
MCSymbolRefExpr::VK_COFF_IMGREL32,
context),
4);
if (info->PackedInfo)
streamer.emitInt32(info->PackedInfo);
else
streamer.emitValue(
MCSymbolRefExpr::create(info->Symbol, MCSymbolRefExpr::VK_COFF_IMGREL32,
context),
4);
}
void llvm::Win64EH::ARM64UnwindEmitter::Emit(MCStreamer &Streamer) const {
@ -882,5 +1123,5 @@ void llvm::Win64EH::ARM64UnwindEmitter::EmitUnwindInfo(
// here and from Emit().
MCSection *XData = Streamer.getAssociatedXDataSection(info->TextSection);
Streamer.SwitchSection(XData);
ARM64EmitUnwindInfo(Streamer, info);
ARM64EmitUnwindInfo(Streamer, info, false);
}

View File

@ -0,0 +1,947 @@
// Check that we generate the packed unwind info format whe possible.
// For tests that don't generate packed unwind info, we still check that
// the epilog was packed (so that the testcase otherwise had all other
// preconditions for possibly making packed unwind info).
// REQUIRES: aarch64-registered-target
// RUN: llvm-mc -filetype=obj -triple aarch64-windows %s -o %t.o
// RUN: llvm-readobj --unwind %t.o | FileCheck %s
// CHECK: UnwindInformation [
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func1
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 88
// CHECK-NEXT: RegF: 7
// CHECK-NEXT: RegI: 10
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 160
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #16
// CHECK-NEXT: stp d14, d15, [sp, #128]
// CHECK-NEXT: stp d12, d13, [sp, #112]
// CHECK-NEXT: stp d10, d11, [sp, #96]
// CHECK-NEXT: stp d8, d9, [sp, #80]
// CHECK-NEXT: stp x27, x28, [sp, #64]
// CHECK-NEXT: stp x25, x26, [sp, #48]
// CHECK-NEXT: stp x23, x24, [sp, #32]
// CHECK-NEXT: stp x21, x22, [sp, #16]
// CHECK-NEXT: stp x19, x20, [sp, #-144]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func2
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 48
// CHECK-NEXT: RegF: 2
// CHECK-NEXT: RegI: 3
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 48
// CHECK-NEXT: Prologue [
// CHECK-NEXT: str d10, [sp, #40]
// CHECK-NEXT: stp d8, d9, [sp, #24]
// CHECK-NEXT: str x21, [sp, #16]
// CHECK-NEXT: stp x19, x20, [sp, #-48]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func3
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 32
// CHECK-NEXT: RegF: 3
// CHECK-NEXT: RegI: 1
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 48
// CHECK-NEXT: Prologue [
// CHECK-NEXT: stp d10, d11, [sp, #24]
// CHECK-NEXT: stp d8, d9, [sp, #8]
// CHECK-NEXT: str x19, [sp, #-48]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func4
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 24
// CHECK-NEXT: RegF: 1
// CHECK-NEXT: RegI: 0
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 48
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #32
// CHECK-NEXT: stp d8, d9, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func5
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 56
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 1
// CHECK-NEXT: HomedParameters: Yes
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 112
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #32
// CHECK-NEXT: stp x6, x7, [sp, #56]
// CHECK-NEXT: stp x4, x5, [sp, #40]
// CHECK-NEXT: stp x2, x3, [sp, #24]
// CHECK-NEXT: stp x0, x1, [sp, #8]
// CHECK-NEXT: str x19, [sp, #-80]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func6
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 24
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 0
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 1
// CHECK-NEXT: FrameSize: 32
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #16
// CHECK-NEXT: str lr, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func7
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 24
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 2
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 1
// CHECK-NEXT: FrameSize: 32
// CHECK-NEXT: Prologue [
// CHECK-NEXT: str lr, [sp, #16]
// CHECK-NEXT: stp x19, x20, [sp, #-32]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func8
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 32
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 3
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 1
// CHECK-NEXT: FrameSize: 48
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #16
// CHECK-NEXT: stp x21, lr, [sp, #16]
// CHECK-NEXT: stp x19, x20, [sp, #-32]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func9
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 32
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 2
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 3
// CHECK-NEXT: FrameSize: 48
// CHECK-NEXT: Prologue [
// CHECK-NEXT: mov x29, sp
// CHECK-NEXT: stp x29, lr, [sp, #-32]!
// CHECK-NEXT: stp x19, x20, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func10
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 24
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 0
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 3
// CHECK-NEXT: FrameSize: 32
// CHECK-NEXT: Prologue [
// CHECK-NEXT: mov x29, sp
// CHECK-NEXT: stp x29, lr, [sp, #-32]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func11
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 40
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 2
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 3
// CHECK-NEXT: FrameSize: 544
// CHECK-NEXT: Prologue [
// CHECK-NEXT: mov x29, sp
// CHECK-NEXT: stp x29, lr, [sp, #0]
// CHECK-NEXT: sub sp, sp, #528
// CHECK-NEXT: stp x19, x20, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func12
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 48
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 2
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 3
// CHECK-NEXT: FrameSize: 4112
// CHECK-NEXT: Prologue [
// CHECK-NEXT: mov x29, sp
// CHECK-NEXT: stp x29, lr, [sp, #0]
// CHECK-NEXT: sub sp, sp, #16
// CHECK-NEXT: sub sp, sp, #4080
// CHECK-NEXT: stp x19, x20, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func13
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 32
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 2
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 0
// CHECK-NEXT: FrameSize: 4112
// CHECK-NEXT: Prologue [
// CHECK-NEXT: sub sp, sp, #16
// CHECK-NEXT: sub sp, sp, #4080
// CHECK-NEXT: stp x19, x20, [sp, #-16]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func14
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 32
// CHECK-NEXT: RegF: 2
// CHECK-NEXT: RegI: 0
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 1
// CHECK-NEXT: FrameSize: 32
// CHECK-NEXT: Prologue [
// CHECK-NEXT: str d10, [sp, #24]
// CHECK-NEXT: stp d8, d9, [sp, #8]
// CHECK-NEXT: str lr, [sp, #-32]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK-NEXT: RuntimeFunction {
// CHECK-NEXT: Function: func15
// CHECK-NEXT: Fragment: No
// CHECK-NEXT: FunctionLength: 20
// CHECK-NEXT: RegF: 0
// CHECK-NEXT: RegI: 0
// CHECK-NEXT: HomedParameters: No
// CHECK-NEXT: CR: 3
// CHECK-NEXT: FrameSize: 32
// CHECK-NEXT: Prologue [
// CHECK-NEXT: mov x29, sp
// CHECK-NEXT: stp x29, lr, [sp, #-32]!
// CHECK-NEXT: end
// CHECK-NEXT: ]
// CHECK-NEXT: }
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked1
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked2
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked3
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked4
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked5
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked6
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked7
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked8
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked9
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked10
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked11
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked12
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
// CHECK: RuntimeFunction {
// CHECK-NEXT: Function: nonpacked13
// CHECK-NEXT: ExceptionRecord:
// CHECK-NEXT: ExceptionData {
// CHECK: EpiloguePacked: Yes
.text
func1:
.seh_proc func1
stp x19, x20, [sp, #-144]!
.seh_save_r19r20_x 144
stp x21, x22, [sp, #16]
.seh_save_regp x21, 16
stp x23, x24, [sp, #32]
.seh_save_next
stp x25, x26, [sp, #48]
.seh_save_next
stp x27, x28, [sp, #64]
.seh_save_next
stp d8, d9, [sp, #80]
.seh_save_fregp d8, 80
stp d10, d11, [sp, #96]
.seh_save_fregp d10, 96
stp d12, d13, [sp, #112]
.seh_save_fregp d12, 112
stp d14, d15, [sp, #128]
.seh_save_fregp d14, 128
sub sp, sp, #16
.seh_stackalloc 16
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #16
.seh_stackalloc 16
ldp d14, d15, [sp, #128]
.seh_save_fregp d14, 128
ldp d12, d13, [sp, #112]
.seh_save_fregp d12, 112
ldp d10, d11, [sp, #96]
.seh_save_fregp d10, 96
ldp d8, d9, [sp, #80]
.seh_save_fregp d8, 80
ldp x27, x28, [sp, #64]
.seh_save_next
ldp x25, x26, [sp, #48]
.seh_save_next
ldp x23, x24, [sp, #32]
.seh_save_next
ldp x21, x22, [sp, #16]
.seh_save_next
ldp x19, x20, [sp], #144
.seh_save_regp_x x19, 144
.seh_endepilogue
ret
.seh_endproc
func2:
.seh_proc func2
stp x19, x20, [sp, #-48]!
.seh_save_r19r20_x 48
str x21, [sp, #16]
.seh_save_reg x21, 16
stp d8, d9, [sp, #24]
.seh_save_fregp d8, 24
str d10, [sp, #40]
.seh_save_freg d10, 40
sub sp, sp, #0
.seh_stackalloc 0
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #0
.seh_stackalloc 0
ldr d10, [sp, #40]
.seh_save_freg d10, 40
ldp d8, d9, [sp, #24]
.seh_save_fregp d8, 24
ldr x21, [sp, #16]
.seh_save_reg x21, 16
ldp x19, x20, [sp], #48
.seh_save_r19r20_x 48
.seh_endepilogue
ret
.seh_endproc
func3:
.seh_proc func3
str x19, [sp, #-48]!
.seh_save_reg_x x19, 48
stp d8, d9, [sp, #8]
.seh_save_fregp d8, 8
stp d10, d11, [sp, #24]
.seh_save_fregp d10, 24
.seh_endprologue
nop
.seh_startepilogue
ldp d10, d11, [sp, #24]
.seh_save_fregp d10, 24
ldp d8, d9, [sp, #8]
.seh_save_fregp d8, 8
ldr x19, [sp], #48
.seh_save_reg_x x19, 48
.seh_endepilogue
ret
.seh_endproc
func4:
.seh_proc func4
stp d8, d9, [sp, #-16]!
.seh_save_fregp_x d8, 16
sub sp, sp, #32
.seh_stackalloc 32
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #32
.seh_stackalloc 32
ldp d8, d9, [sp], #16
.seh_save_fregp_x d8, 16
.seh_endepilogue
ret
.seh_endproc
func5:
.seh_proc func5
str x19, [sp, #-80]!
.seh_save_reg_x x19, 80
stp x0, x1, [sp, #8]
.seh_nop
stp x2, x3, [sp, #24]
.seh_nop
stp x4, x5, [sp, #40]
.seh_nop
stp x6, x7, [sp, #56]
.seh_nop
sub sp, sp, #32
.seh_stackalloc 32
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #32
.seh_stackalloc 32
nop
.seh_nop
nop
.seh_nop
nop
.seh_nop
nop
.seh_nop
ldr x19, [sp], #80
.seh_save_reg_x x19, 80
.seh_endepilogue
ret
.seh_endproc
func6:
.seh_proc func6
str lr, [sp, #-16]!
.seh_save_reg_x lr, 16
sub sp, sp, #16
.seh_stackalloc 16
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #16
.seh_stackalloc 16
ldr lr, [sp], #16
.seh_save_reg_x lr, 16
.seh_endepilogue
ret
.seh_endproc
func7:
.seh_proc func7
stp x19, x20, [sp, #-32]!
.seh_save_r19r20_x 32
str lr, [sp, #16]
.seh_save_reg lr, 16
.seh_endprologue
nop
.seh_startepilogue
ldr lr, [sp, #16]
.seh_save_reg lr, 16
ldp x19, x20, [sp], #32
.seh_save_r19r20_x 32
.seh_endepilogue
ret
.seh_endproc
func8:
.seh_proc func8
stp x19, x20, [sp, #-32]!
.seh_save_r19r20_x 32
stp x21, lr, [sp, #16]
.seh_save_lrpair x21, 16
sub sp, sp, #16
.seh_stackalloc 16
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #16
.seh_stackalloc 16
ldp x21, lr, [sp, #16]
.seh_save_lrpair x21, 16
ldp x19, x20, [sp], #32
.seh_save_r19r20_x 32
.seh_endepilogue
ret
.seh_endproc
func9:
.seh_proc func9
stp x19, x20, [sp, #-16]!
.seh_save_r19r20_x 16
stp x29, lr, [sp, #-32]!
.seh_save_fplr_x 32
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp], #32
.seh_save_fplr_x 32
ldp x19, x20, [sp], #16
.seh_save_r19r20_x 16
.seh_endepilogue
ret
.seh_endproc
func10:
.seh_proc func10
stp x29, lr, [sp, #-32]!
.seh_save_fplr_x 32
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp], #32
.seh_save_fplr_x 32
.seh_endepilogue
ret
.seh_endproc
func11:
.seh_proc func11
stp x19, x20, [sp, #-16]!
.seh_save_r19r20_x 16
sub sp, sp, #528
.seh_stackalloc 528
stp x29, lr, [sp, #0]
.seh_save_fplr 0
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp, #0]
.seh_save_fplr 0
add sp, sp, #528
.seh_stackalloc 528
ldp x19, x20, [sp], #16
.seh_save_r19r20_x 16
.seh_endepilogue
ret
.seh_endproc
func12:
.seh_proc func12
stp x19, x20, [sp, #-16]!
.seh_save_r19r20_x 16
sub sp, sp, #4080
.seh_stackalloc 4080
sub sp, sp, #16
.seh_stackalloc 16
stp x29, lr, [sp, #0]
.seh_save_fplr 0
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp, #0]
.seh_save_fplr 0
add sp, sp, #16
.seh_stackalloc 16
add sp, sp, #4080
.seh_stackalloc 4080
ldp x19, x20, [sp], #16
.seh_save_r19r20_x 16
.seh_endepilogue
ret
.seh_endproc
func13:
.seh_proc func13
stp x19, x20, [sp, #-16]!
.seh_save_r19r20_x 16
sub sp, sp, #4080
.seh_stackalloc 4080
sub sp, sp, #16
.seh_stackalloc 16
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #16
.seh_stackalloc 16
add sp, sp, #4080
.seh_stackalloc 4080
ldp x19, x20, [sp], #16
.seh_save_r19r20_x 16
.seh_endepilogue
ret
.seh_endproc
func14:
.seh_proc func14
str lr, [sp, #-32]!
.seh_save_reg_x lr, 32
stp d8, d9, [sp, #8]
.seh_save_fregp d8, 8
str d10, [sp, #24]
.seh_save_freg d10, 24
.seh_endprologue
nop
.seh_startepilogue
ldr d10, [sp, #24]
.seh_save_freg d10, 24
ldp d8, d9, [sp, #8]
.seh_save_fregp d8, 8
ldr lr, [sp], #32
.seh_save_reg_x lr, 32
.seh_endepilogue
ret
.seh_endproc
func15:
.seh_proc func15
stp x29, lr, [sp, #-32]!
.seh_save_fplr_x 32
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
// Epilogue missing the .seh_set_fp, but still generating packed info.
ldp x29, lr, [sp], #32
.seh_save_fplr_x 32
.seh_endepilogue
ret
.seh_endproc
nonpacked1:
.seh_proc nonpacked1
// Can't be packed; can't save integer registers after float registers.
stp d8, d9, [sp, #-32]!
.seh_save_fregp_x d8, 32
stp x19, x20, [sp, #16]!
.seh_save_regp x19, 16
.seh_endprologue
nop
.seh_startepilogue
ldp x19, x20, [sp, #16]
.seh_save_regp x19, 16
ldp d8, d9, [sp], #32
.seh_save_fregp_x d8, 32
.seh_endepilogue
ret
.seh_endproc
nonpacked2:
.seh_proc nonpacked2
// Can't be packed; x21/x22 aren't saved in the expected spot
stp x19, x20, [sp, #-48]!
.seh_save_r19r20_x 48
stp x21, x22, [sp, #32]
.seh_save_regp x21, 32
.seh_endprologue
nop
.seh_startepilogue
ldp x21, x22, [sp, #32]
.seh_save_regp x21, 32
ldp x19, x20, [sp], #48
.seh_save_r19r20_x 48
.seh_endepilogue
ret
.seh_endproc
nonpacked3:
.seh_proc nonpacked3
// Can't be packed; x29/x30 can't be treated as the other saved registers
stp x19, x20, [sp, #-96]!
.seh_save_r19r20_x 96
stp x21, x22, [sp, #16]
.seh_save_next
stp x23, x24, [sp, #32]
.seh_save_next
stp x25, x26, [sp, #48]
.seh_save_next
stp x27, x28, [sp, #64]
.seh_save_next
stp x29, x30, [sp, #80]
.seh_save_next
.seh_endprologue
nop
.seh_startepilogue
ldp x29, x30, [sp, #80]
.seh_save_next
ldp x27, x28, [sp, #64]
.seh_save_next
ldp x25, x26, [sp, #48]
.seh_save_next
ldp x23, x24, [sp, #32]
.seh_save_next
ldp x21, x22, [sp, #16]
.seh_save_next
ldp x19, x20, [sp], #96
.seh_save_r19r20_x 96
.seh_endepilogue
ret
.seh_endproc
nonpacked4:
.seh_proc nonpacked4
// Can't be packed; more predecrement for x19/x20 than used for
// corresponding RegI/RegF/LR saves
stp x19, x20, [sp, #-32]!
.seh_save_r19r20_x 32
.seh_endprologue
nop
.seh_startepilogue
ldp x19, x20, [sp], #32
.seh_save_r19r20_x 32
.seh_endepilogue
ret
.seh_endproc
nonpacked5:
.seh_proc nonpacked5
// Can't be packed; can't save LR twice
stp x19, x20, [sp, #-32]!
.seh_save_r19r20_x 32
str lr, [sp, #16]
.seh_save_reg lr, 16
str lr, [sp, #24]
.seh_save_reg lr, 24
.seh_endprologue
nop
.seh_startepilogue
ldr lr, [sp, #24]
.seh_save_reg lr, 24
ldr lr, [sp, #16]
.seh_save_reg lr, 16
ldp x19, x20, [sp], #32
.seh_save_r19r20_x 32
.seh_endepilogue
ret
.seh_endproc
nonpacked6:
.seh_proc nonpacked6
// Can't be packed; can't save LR both standalone (CR 1) and as FPLR (CR 3)
stp x19, x20, [sp, #-32]!
.seh_save_r19r20_x 32
str lr, [sp, #16]
.seh_save_reg lr, 16
stp x29, lr, [sp, #-16]!
.seh_save_fplr_x 16
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp], #32
.seh_save_fplr_x 16
ldr lr, [sp, #16]
.seh_save_reg lr, 16
ldp x19, x20, [sp], #32
.seh_save_r19r20_x 32
.seh_endepilogue
ret
.seh_endproc
nonpacked7:
.seh_proc nonpacked7
// Can't be packed; too many saved FP regs
stp d8, d9, [sp, #-80]!
.seh_save_fregp_x d8, 80
stp d10, d11, [sp, #16]
.seh_save_fregp d10, 16
stp d12, d13, [sp, #32]
.seh_save_fregp d12, 32
stp d14, d15, [sp, #48]
.seh_save_fregp d14, 48
stp d16, d17, [sp, #64]
.seh_save_next
.seh_endprologue
nop
.seh_startepilogue
ldp d16, d17, [sp, #64]
.seh_save_next
ldp d14, d15, [sp, #48]
.seh_save_fregp d14, 48
ldp d12, d13, [sp, #32]
.seh_save_fregp d12, 32
ldp d10, d11, [sp, #16]
.seh_save_fregp d10, 16
ldp d8, d9, [sp], #80
.seh_save_fregp_x d8, 80
.seh_endepilogue
ret
.seh_endproc
nonpacked8:
.seh_proc nonpacked8
// Can't be packed; Can't handle only a single FP reg
str d8, [sp, #-16]!
.seh_save_freg_x d8, 16
.seh_endprologue
nop
.seh_startepilogue
ldr d8, [sp], #16
.seh_save_freg_x d8, 16
.seh_endepilogue
ret
.seh_endproc
nonpacked9:
.seh_proc nonpacked9
// Can't be packed; can't have a separate stack adjustment with save_fplr_x
sub sp, sp, #32
.seh_stackalloc 32
stp x29, lr, [sp, #-16]!
.seh_save_fplr_x 16
mov x29, sp
.seh_set_fp
.seh_endprologue
nop
.seh_startepilogue
mov sp, x29
.seh_set_fp
ldp x29, lr, [sp], #32
.seh_save_fplr_x 16
add sp, sp, #32
.seh_stackalloc 32
.seh_endepilogue
ret
.seh_endproc
nonpacked10:
.seh_proc nonpacked10
// Can't be packed; wrong predecrement
stp x19, x20, [sp, #-16]!
.seh_save_r19r20_x 16
stp x21, x22, [sp, #16]
.seh_save_next
.seh_endprologue
nop
.seh_startepilogue
ldp x21, x22, [sp, #16]
.seh_save_next
ldp x19, x20, [sp], #16
.seh_save_r19r20_x 16
.seh_endepilogue
ret
.seh_endproc
nonpacked11:
.seh_proc nonpacked11
// Can't be packed; too big stack allocation
sub sp, sp, #4080
.seh_stackalloc 4080
sub sp, sp, #8192
.seh_stackalloc 8192
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #8192
.seh_stackalloc 8192
add sp, sp, #4080
.seh_stackalloc 4080
.seh_endepilogue
ret
.seh_endproc
nonpacked12:
.seh_proc nonpacked12
// Can't be packed; missing .seh_set_fp
stp x29, lr, [sp, #-32]!
.seh_save_fplr_x 32
.seh_endprologue
nop
.seh_startepilogue
ldp x29, lr, [sp], #32
.seh_save_fplr_x 32
.seh_endepilogue
ret
.seh_endproc
nonpacked13:
.seh_proc nonpacked13
// Can't be packed; not doing a packed info if .seh_handlerdata is used
sub sp, sp, #16
.seh_stackalloc 16
.seh_endprologue
nop
.seh_startepilogue
add sp, sp, #16
.seh_stackalloc 16
.seh_endepilogue
ret
.seh_endfunclet
.seh_handlerdata
.long 0
.text
.seh_endproc