FEXCore/JIT: Ignore local encoding limit checks

These are guaranteed not to hit encoding distance limits, so we can
ignore the returns.
This commit is contained in:
Ryan Houdek authored and Billy Laws committed 2025-10-28 23:53:15 +00:00
1 parent 65fff73959
commit f653c5e0c0
5 files changed
+117 -120

No files matched your search

+22 -22
View File
@@ -588,7 +588,7 @@ DEF_OP(ShiftFlags) {
and_(ARMEmitter::Size::i32Bit, TMP1, Src2, OpSize == IR::OpSize::i64Bit ? 0x3f : 0x1f);
ARMEmitter::ForwardLabel Done;
cbz(EmitSize, TMP1, &Done);
(void)cbz(EmitSize, TMP1, &Done);
{
// PF/SF/ZF/OF
if (OpSize >= IR::OpSize::i32Bit) {
@@ -652,7 +652,7 @@ DEF_OP(ShiftFlags) {
msr(ARMEmitter::SystemRegister::NZCV, TMP2);
}
}
Bind(&Done);
(void)Bind(&Done);
// TODO: Make RA less dumb so this can't happen (e.g. with late-kill).
if (PFOutput != PFTemp) {
@@ -669,7 +669,7 @@ DEF_OP(RotateFlags) {
// If shift=0, flags are unaffected. Wrap the whole implementation in a cbz.
ARMEmitter::ForwardLabel Done;
cbz(EmitSize, Shift, &Done);
(void)cbz(EmitSize, Shift, &Done);
{
// Extract the last bit shifted in to CF
const auto BitSize = IR::OpSizeToSize(Op->Size) * 8;
@@ -701,7 +701,7 @@ DEF_OP(RotateFlags) {
msr(ARMEmitter::SystemRegister::NZCV, TMP3);
}
}
Bind(&Done);
(void)Bind(&Done);
}
DEF_OP(Extr) {
@@ -767,14 +767,14 @@ DEF_OP(PDep) {
// Now, they're copied, so we can start setting Dest (even if it overlaps with
// one of them). Handle early exit case
mov(EmitSize, Dest, 0);
cbz(EmitSize, OrigMask, &Done);
(void)cbz(EmitSize, OrigMask, &Done);
// Setup for first iteration
neg(EmitSize, T0, Mask);
and_(EmitSize, T0, T0, Mask);
// Main loop
Bind(&NextBit);
(void)Bind(&NextBit);
sbfx(EmitSize, T1, Input, 0, 1);
eor(EmitSize, Mask, Mask, T0);
and_(EmitSize, T0, T1, T0);
@@ -782,10 +782,10 @@ DEF_OP(PDep) {
orr(EmitSize, Dest, Dest, T0);
lsr(EmitSize, Input, Input, 1);
and_(EmitSize, T0, Mask, T1);
cbnz(EmitSize, T0, &NextBit);
(void)cbnz(EmitSize, T0, &NextBit);
// All done with nothing to do.
Bind(&Done);
(void)Bind(&Done);
}
}
@@ -821,27 +821,27 @@ DEF_OP(PExt) {
ARMEmitter::BackwardLabel NextBit;
ARMEmitter::ForwardLabel Done;
cbz(EmitSize, Mask, &EarlyExit);
(void)cbz(EmitSize, Mask, &EarlyExit);
mov(EmitSize, MaskReg, Mask);
mov(EmitSize, ValueReg, Input);
mov(EmitSize, Dest, ARMEmitter::Reg::zr);
// Main loop
Bind(&NextBit);
cbz(EmitSize, MaskReg, &Done);
(void)Bind(&NextBit);
(void)cbz(EmitSize, MaskReg, &Done);
clz(EmitSize, BitReg, MaskReg);
lslv(EmitSize, ValueReg, ValueReg, BitReg);
lslv(EmitSize, MaskReg, MaskReg, BitReg);
extr(EmitSize, Dest, Dest, ValueReg, OpSizeBitsM1);
bfc(EmitSize, MaskReg, OpSizeBitsM1, 1);
b(&NextBit);
(void)b(&NextBit);
// Early exit
Bind(&EarlyExit);
(void)Bind(&EarlyExit);
mov(EmitSize, Dest, ARMEmitter::Reg::zr);
// All done with nothing to do.
Bind(&Done);
(void)Bind(&Done);
}
}
@@ -909,7 +909,7 @@ DEF_OP(Div) {
eor(EmitSize, TMP1, TMP1, Upper);
// If the sign bit matches then the result is zero
cbz(EmitSize, TMP1, &Only64Bit);
(void)cbz(EmitSize, TMP1, &Only64Bit);
// Long divide
{
@@ -928,17 +928,17 @@ DEF_OP(Div) {
mov(EmitSize, Remainder, TMP2);
// Skip 64-bit path
b(&LongDIVRet);
(void)b(&LongDIVRet);
}
Bind(&Only64Bit);
(void)Bind(&Only64Bit);
// 64-Bit only
{
sdiv(EmitSize, Quotient, Lower, Divisor);
msub(EmitSize, Remainder, Quotient, Divisor, Lower);
}
Bind(&LongDIVRet);
(void)Bind(&LongDIVRet);
break;
}
default: LOGMAN_MSG_A_FMT("Unknown DIV Size: {}", OpSize); break;
@@ -992,7 +992,7 @@ DEF_OP(UDiv) {
// Check the upper bits for zero
// If the upper bits are zero then we can do a 64-bit divide
cbz(EmitSize, Upper, &Only64Bit);
(void)cbz(EmitSize, Upper, &Only64Bit);
// Long divide
{
@@ -1011,17 +1011,17 @@ DEF_OP(UDiv) {
mov(EmitSize, Remainder, TMP2);
// Skip 64-bit path
b(&LongDIVRet);
(void)b(&LongDIVRet);
}
Bind(&Only64Bit);
(void)Bind(&Only64Bit);
// 64-Bit only
{
udiv(EmitSize, Quotient, Lower, Divisor);
msub(EmitSize, Remainder, Quotient, Divisor, Lower);
}
Bind(&LongDIVRet);
(void)Bind(&LongDIVRet);
break;
}
default: LOGMAN_MSG_A_FMT("Unknown LUDIV Size: {}", OpSize); break;
+34 -34
View File
@@ -62,27 +62,27 @@ DEF_OP(CASPair) {
ARMEmitter::BackwardLabel LoopTop;
ARMEmitter::ForwardLabel LoopNotExpected;
ARMEmitter::ForwardLabel LoopExpected;
Bind(&LoopTop);
(void)Bind(&LoopTop);
// This instruction sequence must be synced with HandleCASPAL_Armv8.
ldaxp(EmitSize, TMP2, TMP3, MemSrc);
cmp(EmitSize, TMP2, Expected0);
ccmp(EmitSize, TMP3, Expected1, ARMEmitter::StatusFlags::None, ARMEmitter::Condition::CC_EQ);
b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
(void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
stlxp(EmitSize, TMP2, Desired0, Desired1, MemSrc);
cbnz(EmitSize, TMP2, &LoopTop);
(void)cbnz(EmitSize, TMP2, &LoopTop);
mov(EmitSize, Dst0, Expected0);
mov(EmitSize, Dst1, Expected1);
b(&LoopExpected);
(void)b(&LoopExpected);
Bind(&LoopNotExpected);
(void)Bind(&LoopNotExpected);
mov(EmitSize, Dst0, TMP2.R());
mov(EmitSize, Dst1, TMP3.R());
// exclusive monitor needs to be cleared here
// Might have hit the case where ldaxr was hit but stlxr wasn't
clrex();
Bind(&LoopExpected);
(void)Bind(&LoopExpected);
// Restore
msr(ARMEmitter::SystemRegister::NZCV, TMP1);
@@ -114,7 +114,7 @@ DEF_OP(CAS) {
ARMEmitter::BackwardLabel LoopTop;
ARMEmitter::ForwardLabel LoopNotExpected;
ARMEmitter::ForwardLabel LoopExpected;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
if (IROp->Size == IR::OpSize::i8Bit) {
cmp(EmitSize, TMP2, Expected, ARMEmitter::ExtendedType::UXTB, 0);
@@ -123,18 +123,18 @@ DEF_OP(CAS) {
} else {
cmp(EmitSize, TMP2, Expected);
}
b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
(void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
stlxr(SubEmitSize, TMP3, Desired, MemSrc);
cbnz(EmitSize, TMP3, &LoopTop);
(void)cbnz(EmitSize, TMP3, &LoopTop);
mov(EmitSize, Dst, Expected);
b(&LoopExpected);
(void)b(&LoopExpected);
Bind(&LoopNotExpected);
(void)Bind(&LoopNotExpected);
mov(EmitSize, Dst, TMP2.R());
// exclusive monitor needs to be cleared here
// Might have hit the case where ldaxr was hit but stlxr wasn't
clrex();
Bind(&LoopExpected);
(void)Bind(&LoopExpected);
}
}
@@ -150,11 +150,11 @@ DEF_OP(AtomicXor) {
steorl(SubEmitSize, Src, MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
eor(EmitSize, TMP2, TMP2, Src);
stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
cbnz(EmitSize, TMP2, &LoopTop);
(void)cbnz(EmitSize, TMP2, &LoopTop);
}
}
@@ -179,10 +179,10 @@ DEF_OP(AtomicSwap) {
ldswpal(SubEmitSize, Src, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
stlxr(SubEmitSize, TMP4, Src, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
ubfm(EmitSize, GetReg(Node), TMP2, 0, IR::OpSizeAsBits(OpSize) - 1);
}
}
@@ -199,11 +199,11 @@ DEF_OP(AtomicFetchAdd) {
ldaddal(SubEmitSize, Src, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
add(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -221,11 +221,11 @@ DEF_OP(AtomicFetchSub) {
ldaddal(SubEmitSize, TMP2, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
sub(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -243,11 +243,11 @@ DEF_OP(AtomicFetchAnd) {
ldclral(SubEmitSize, TMP2, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
and_(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -264,11 +264,11 @@ DEF_OP(AtomicFetchCLR) {
ldclral(SubEmitSize, Src, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
bic(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -285,11 +285,11 @@ DEF_OP(AtomicFetchOr) {
ldsetal(SubEmitSize, Src, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
orr(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -306,11 +306,11 @@ DEF_OP(AtomicFetchXor) {
ldeoral(SubEmitSize, Src, GetReg(Node), MemSrc);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
eor(EmitSize, TMP3, TMP2, Src);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -326,20 +326,20 @@ DEF_OP(AtomicFetchNeg) {
// Use a CAS loop to avoid needing to emulate unaligned LLSC atomics
ldr(SubEmitSize, TMP2, MemSrc);
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
mov(EmitSize, TMP4, TMP2);
neg(EmitSize, TMP3, TMP2);
casal(SubEmitSize, TMP2, TMP3, MemSrc);
sub(EmitSize, TMP3, TMP2, TMP4);
cbnz(EmitSize, TMP3, &LoopTop);
(void)cbnz(EmitSize, TMP3, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(SubEmitSize, TMP2, MemSrc);
neg(EmitSize, TMP3, TMP2);
stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
cbnz(EmitSize, TMP4, &LoopTop);
(void)cbnz(EmitSize, TMP4, &LoopTop);
mov(EmitSize, GetReg(Node), TMP2.R());
}
}
@@ -359,11 +359,11 @@ DEF_OP(TelemetrySetValue) {
stsetl(ARMEmitter::SubRegSize::i64Bit, TMP1, TMP2);
} else {
ARMEmitter::BackwardLabel LoopTop;
Bind(&LoopTop);
(void)Bind(&LoopTop);
ldaxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP2);
orr(ARMEmitter::Size::i32Bit, TMP3, TMP3, Src);
stlxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP3, TMP2);
cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
(void)cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
}
#endif
}
@@ -141,7 +141,7 @@ DEF_OP(ExitFunction) {
if (!Op->CallReturnBlock.IsInvalid()) {
auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
adr(TMP1, &l_CallReturn);
(void)adr(TMP1, &l_CallReturn);
stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
} else {
stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
@@ -149,16 +149,16 @@ DEF_OP(ExitFunction) {
} else if (Op->Hint == IR::BranchHint::CheckTF) {
ARMEmitter::ForwardLabel TFUnset;
ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
(void)cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
LoadConstant(ARMEmitter::Size::i64Bit, TMP1, NewRIP);
str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
blr(TMP2);
Bind(&TFUnset);
(void)Bind(&TFUnset);
}
EmitLinkedBranch(NewRIP, Op->Hint == IR::BranchHint::Call);
Bind(&l_CallReturn);
(void)Bind(&l_CallReturn);
#ifdef _M_ARM_64EC
}
#endif
@@ -170,7 +170,7 @@ DEF_OP(ExitFunction) {
// First try to pop from the call-ret stack, otherwise follow the normal path (but ending in a ret)
ldp<ARMEmitter::IndexType::POST>(TMP1, TMP2, REG_CALLRET_SP, 0x10);
sub(TMP1, TMP1, RipReg.X());
cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
}
// L1 Cache
@@ -185,23 +185,23 @@ DEF_OP(ExitFunction) {
// Note: sub+cbnz used over cmp+br to preserve flags.
sub(TMP1, TMP1, RipReg.X());
cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
Bind(&SkipFullLookup);
(void)Bind(&SkipFullLookup);
if (Op->Hint == IR::BranchHint::Call) {
ARMEmitter::ForwardLabel l_CallReturn;
if (!Op->CallReturnBlock.IsInvalid()) {
auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
adr(TMP1, &l_CallReturn);
(void)adr(TMP1, &l_CallReturn);
stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
} else {
stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
}
blr(TMP2);
Bind(&l_CallReturn);
(void)Bind(&l_CallReturn);
} else if (Op->Hint == IR::BranchHint::Return) {
ret(TMP2);
} else {
+5 -8
View File
@@ -11,8 +11,6 @@ desc: Main glue logic of the arm64 splatter backend
$end_info$
*/
#include "Common/SoftFloat.h"
#include "Interface/Context/Context.h"
#include "Interface/Core/LookupCache.h"
#include "Interface/Core/Dispatcher/Dispatcher.h"
@@ -37,7 +35,6 @@ $end_info$
#include <cstdio>
#include <cstring>
#include <limits>
#include <unistd.h>
namespace {
@@ -742,11 +739,11 @@ void Arm64JITCore::EmitTFCheck() {
// Note that this needs to be before the below suspend checks, as X86 checks this flag immediately after executing an instruction.
ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
(void)cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
// X86 semantically checks TF after executing each instruction, so e.g. setting a context with TF set will execute a single instruction
// and then raise an exception. However on the FEX side this is simpler to implement by checking at the start of each instruction, handle this by having bit 1 being unset in the flag state indicate that TF is blocked for a single instruction.
tbz(TMP1, 1, &l_TFBlocked);
(void)tbz(TMP1, 1, &l_TFBlocked);
// Block TF for a single instruction when the frontend jumps to a new context by unsetting bit 1.
ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
@@ -769,11 +766,11 @@ void Arm64JITCore::EmitTFCheck() {
ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
br(TMP1);
Bind(&l_TFBlocked);
(void)Bind(&l_TFBlocked);
// If TF was blocked for this instruction, unblock it for the next.
LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0b11);
strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
Bind(&l_TFUnset);
(void)Bind(&l_TFUnset);
}
void Arm64JITCore::EmitSuspendInterruptCheck() {
@@ -844,7 +841,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
// Put the code header at the start of the data block.
ARMEmitter::BackwardLabel JITCodeHeaderLabel {};
Bind(&JITCodeHeaderLabel);
(void)Bind(&JITCodeHeaderLabel);
JITCodeHeader* CodeHeader = GetCursorAddress<JITCodeHeader*>();
CursorIncrement(sizeof(JITCodeHeader));
+47 -47
View File
@@ -912,7 +912,7 @@ DEF_OP(VLoadVectorMasked) {
// If the sign bit is zero then skip the load
ARMEmitter::ForwardLabel Skip {};
tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
(void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
// Do the gather load for this element into the destination
switch (IROp->ElementSize) {
case IR::OpSize::i8Bit: ld1<ARMEmitter::SubRegSize::i8Bit>(TempDst.Q(), i, TempMemReg); break;
@@ -923,7 +923,7 @@ DEF_OP(VLoadVectorMasked) {
default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
}
Bind(&Skip);
(void)Bind(&Skip);
if ((i + 1) != NumElements) {
// Handle register rename to save a move.
@@ -1013,7 +1013,7 @@ DEF_OP(VStoreVectorMasked) {
// If the sign bit is zero then skip the load
ARMEmitter::ForwardLabel Skip {};
tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
(void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
// Do the gather load for this element into the destination
switch (IROp->ElementSize) {
case IR::OpSize::i8Bit: st1<ARMEmitter::SubRegSize::i8Bit>(RegData.Q(), i, TempMemReg); break;
@@ -1024,7 +1024,7 @@ DEF_OP(VStoreVectorMasked) {
default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
}
Bind(&Skip);
(void)Bind(&Skip);
if ((i + 1) != NumElements) {
// Handle register rename to save a move.
@@ -1102,7 +1102,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
PerformMove(ElementSize, WorkingReg, MaskReg, i);
// Skip if the mask's sign bit isn't set
tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
(void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
// Extract Index Element
if ((IndexElement * IR::OpSizeToSize(VectorIndexSize)) >= 16) {
@@ -1140,7 +1140,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, ElementSize); FEX_UNREACHABLE;
}
Bind(&Skip);
(void)Bind(&Skip);
}
if (NeedsDestTmp) {
@@ -1874,7 +1874,7 @@ DEF_OP(MemSet) {
if (!DirectionIsInline) {
// Backward or forwards implementation depends on flag
tbnz(DirectionReg, 1, &BackwardImpl);
(void)tbnz(DirectionReg, 1, &BackwardImpl);
}
auto MemStore = [this](auto Value, uint32_t OpSize, int32_t Size) {
@@ -1922,7 +1922,7 @@ DEF_OP(MemSet) {
ARMEmitter::ForwardLabel DoneInternal {};
// Early exit if zero count.
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
if (!IsAtomic) {
ARMEmitter::ForwardLabel AgainInternal256Exit {};
@@ -1939,50 +1939,50 @@ DEF_OP(MemSet) {
// Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
// single copy loop if size < 64.
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal128Exit);
(void)tbnz(TMP1, 63, &AgainInternal128Exit);
// Fill VTMP2 with the set pattern
dup(SubRegSize, VTMP2.Q(), Value);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal256Exit);
(void)tbnz(TMP1, 63, &AgainInternal256Exit);
Bind(&AgainInternal256);
(void)Bind(&AgainInternal256);
stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
tbz(TMP1, 63, &AgainInternal256);
(void)tbz(TMP1, 63, &AgainInternal256);
Bind(&AgainInternal256Exit);
(void)Bind(&AgainInternal256Exit);
add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal128Exit);
Bind(&AgainInternal128);
(void)tbnz(TMP1, 63, &AgainInternal128Exit);
(void)Bind(&AgainInternal128);
stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbz(TMP1, 63, &AgainInternal128);
(void)tbz(TMP1, 63, &AgainInternal128);
Bind(&AgainInternal128Exit);
(void)Bind(&AgainInternal128Exit);
add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
if (Direction == -1) {
add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
}
}
Bind(&AgainInternal);
(void)Bind(&AgainInternal);
if (IsAtomic) {
MemStoreTSO(Value, OpSize, SizeDirection);
} else {
MemStore(Value, OpSize, SizeDirection);
}
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
(void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
Bind(&DoneInternal);
(void)Bind(&DoneInternal);
if (SizeDirection >= 0) {
switch (OpSize) {
@@ -2012,12 +2012,12 @@ DEF_OP(MemSet) {
EmitMemset(Direction);
if (Direction == 1) {
b(&Done);
Bind(&BackwardImpl);
(void)b(&Done);
(void)Bind(&BackwardImpl);
}
}
Bind(&Done);
(void)Bind(&Done);
// Destination already set to the final pointer.
}
}
@@ -2067,7 +2067,7 @@ DEF_OP(MemCpy) {
if (!DirectionIsInline) {
// Backward or forwards implementation depends on flag
tbnz(DirectionReg, 1, &BackwardImpl);
(void)tbnz(DirectionReg, 1, &BackwardImpl);
}
auto MemCpy = [this](uint32_t OpSize, int32_t Size) {
@@ -2164,7 +2164,7 @@ DEF_OP(MemCpy) {
ARMEmitter::ForwardLabel DoneInternal {};
// Early exit if zero count.
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
if (!IsAtomic) {
ARMEmitter::ForwardLabel AbsPos {};
@@ -2174,11 +2174,11 @@ DEF_OP(MemCpy) {
ARMEmitter::BackwardLabel AgainInternal256 {};
sub(ARMEmitter::Size::i64Bit, TMP4, TMP2, TMP3);
tbz(TMP4, 63, &AbsPos);
(void)tbz(TMP4, 63, &AbsPos);
neg(ARMEmitter::Size::i64Bit, TMP4, TMP4);
Bind(&AbsPos);
(void)Bind(&AbsPos);
sub(ARMEmitter::Size::i64Bit, TMP4, TMP4, 32);
tbnz(TMP4, 63, &AgainInternal);
(void)tbnz(TMP4, 63, &AgainInternal);
if (Direction == -1) {
sub(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2190,30 +2190,30 @@ DEF_OP(MemCpy) {
// Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
// single copy loop if size < 64.
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal128Exit);
(void)tbnz(TMP1, 63, &AgainInternal128Exit);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal256Exit);
(void)tbnz(TMP1, 63, &AgainInternal256Exit);
Bind(&AgainInternal256);
(void)Bind(&AgainInternal256);
MemCpy(32, 32 * Direction);
MemCpy(32, 32 * Direction);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
tbz(TMP1, 63, &AgainInternal256);
(void)tbz(TMP1, 63, &AgainInternal256);
Bind(&AgainInternal256Exit);
(void)Bind(&AgainInternal256Exit);
add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbnz(TMP1, 63, &AgainInternal128Exit);
Bind(&AgainInternal128);
(void)tbnz(TMP1, 63, &AgainInternal128Exit);
(void)Bind(&AgainInternal128);
MemCpy(32, 32 * Direction);
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
tbz(TMP1, 63, &AgainInternal128);
(void)tbz(TMP1, 63, &AgainInternal128);
Bind(&AgainInternal128Exit);
(void)Bind(&AgainInternal128Exit);
add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
(void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
if (Direction == -1) {
add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2221,16 +2221,16 @@ DEF_OP(MemCpy) {
}
}
Bind(&AgainInternal);
(void)Bind(&AgainInternal);
if (IsAtomic) {
MemCpyTSO(OpSize, SizeDirection);
} else {
MemCpy(OpSize, SizeDirection);
}
sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
(void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
Bind(&DoneInternal);
(void)Bind(&DoneInternal);
// Needs to use temporaries just in case of overwrite
mov(TMP1, MemRegDest.X());
@@ -2288,11 +2288,11 @@ DEF_OP(MemCpy) {
for (int32_t Direction : {1, -1}) {
EmitMemcpy(Direction);
if (Direction == 1) {
b(&Done);
Bind(&BackwardImpl);
(void)b(&Done);
(void)Bind(&BackwardImpl);
}
}
Bind(&Done);
(void)Bind(&Done);
// Destination already set to the final pointer.
}
}