luau/CodeGen/src/CodeGenX64.cpp
2023-08-25 18:25:09 +03:00

261 lines
8.3 KiB
C++

// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
#include "CodeGenX64.h"
#include "Luau/AssemblyBuilderX64.h"
#include "Luau/UnwindBuilder.h"
#include "NativeState.h"
#include "EmitCommonX64.h"
#include "lstate.h"
/* An overview of native environment stack setup that we are making in the entry function:
* Each line is 8 bytes, stack grows downwards.
*
* | ... previous frames ...
* | rdx home space | (unused)
* | rcx home space | (unused)
* | return address |
* | ... saved non-volatile registers ... <-- rsp + kStackSizeFull
* | alignment |
* | xmm9 non-vol |
* | xmm9 cont. |
* | xmm8 non-vol |
* | xmm8 cont. |
* | xmm7 non-vol |
* | xmm7 cont. |
* | xmm6 non-vol |
* | xmm6 cont. |
* | spill slot 5 |
* | spill slot 4 |
* | spill slot 3 |
* | spill slot 2 |
* | spill slot 1 | <-- rsp + kStackOffsetToSpillSlots
* | sTemporarySlot |
* | sCode |
* | sClosure | <-- rsp + kStackOffsetToLocals
* | argument 6 | <-- rsp + 40
* | argument 5 | <-- rsp + 32
* | r9 home space |
* | r8 home space |
* | rdx home space |
* | rcx home space | <-- rsp points here
*
* Arguments to our entry function are saved to home space only on Windows.
* Space for arguments to function we call is always reserved, but used only on Windows.
*
* Right now we use a frame pointer, but because of a fixed layout we can omit it in the future
*/
namespace Luau
{
namespace CodeGen
{
namespace X64
{
struct EntryLocations
{
Label start;
Label prologueEnd;
Label epilogueStart;
};
static EntryLocations buildEntryFunction(AssemblyBuilderX64& build, UnwindBuilder& unwind)
{
EntryLocations locations;
build.align(kFunctionAlignment, X64::AlignmentDataX64::Ud2);
locations.start = build.setLabel();
unwind.startFunction();
RegisterX64 rArg1 = (build.abi == ABIX64::Windows) ? rcx : rdi;
RegisterX64 rArg2 = (build.abi == ABIX64::Windows) ? rdx : rsi;
RegisterX64 rArg3 = (build.abi == ABIX64::Windows) ? r8 : rdx;
RegisterX64 rArg4 = (build.abi == ABIX64::Windows) ? r9 : rcx;
// Save common non-volatile registers
if (build.abi == ABIX64::SystemV)
{
// We need to use a standard rbp-based frame setup for debuggers to work with JIT code
build.push(rbp);
build.mov(rbp, rsp);
}
build.push(rbx);
build.push(r12);
build.push(r13);
build.push(r14);
build.push(r15);
if (build.abi == ABIX64::Windows)
{
// Save non-volatile registers that are specific to Windows x64 ABI
build.push(rdi);
build.push(rsi);
// On Windows, rbp is available as a general-purpose non-volatile register and this might be freed up
build.push(rbp);
}
// Allocate stack space
uint8_t usableXmmRegCount = getXmmRegisterCount(build.abi);
unsigned xmmStorageSize = getNonVolXmmStorageSize(build.abi, usableXmmRegCount);
unsigned fullStackSize = getFullStackSize(build.abi, usableXmmRegCount);
build.sub(rsp, fullStackSize);
OperandX64 xmmStorageOffset = rsp + (fullStackSize - (kStackAlign + xmmStorageSize));
// On Windows, we have to save non-volatile xmm registers
std::vector<RegisterX64> savedXmmRegs;
if (build.abi == ABIX64::Windows)
{
if (usableXmmRegCount > kWindowsFirstNonVolXmmReg)
savedXmmRegs.reserve(usableXmmRegCount - kWindowsFirstNonVolXmmReg);
for (uint8_t i = kWindowsFirstNonVolXmmReg, offset = 0; i < usableXmmRegCount; i++, offset += 16)
{
RegisterX64 xmmReg = RegisterX64{SizeX64::xmmword, i};
build.vmovaps(xmmword[xmmStorageOffset + offset], xmmReg);
savedXmmRegs.push_back(xmmReg);
}
}
locations.prologueEnd = build.setLabel();
uint32_t prologueSize = build.getLabelOffset(locations.prologueEnd) - build.getLabelOffset(locations.start);
if (build.abi == ABIX64::SystemV)
unwind.prologueX64(prologueSize, fullStackSize, /* setupFrame= */ true, {rbx, r12, r13, r14, r15}, {});
else if (build.abi == ABIX64::Windows)
unwind.prologueX64(prologueSize, fullStackSize, /* setupFrame= */ false, {rbx, r12, r13, r14, r15, rdi, rsi, rbp}, savedXmmRegs);
// Setup native execution environment
build.mov(rState, rArg1);
build.mov(rNativeContext, rArg4);
build.mov(rBase, qword[rState + offsetof(lua_State, base)]); // L->base
build.mov(rax, qword[rState + offsetof(lua_State, ci)]); // L->ci
build.mov(rax, qword[rax + offsetof(CallInfo, func)]); // L->ci->func
build.mov(rax, qword[rax + offsetof(TValue, value.gc)]); // L->ci->func->value.gc aka cl
build.mov(sClosure, rax);
build.mov(rConstants, qword[rArg2 + offsetof(Proto, k)]); // proto->k
build.mov(rax, qword[rArg2 + offsetof(Proto, code)]); // proto->code
build.mov(sCode, rax);
// Jump to the specified instruction; further control flow will be handled with custom ABI with register setup from EmitCommonX64.h
build.jmp(rArg3);
// Even though we jumped away, we will return here in the end
locations.epilogueStart = build.setLabel();
// Epilogue and exit
if (build.abi == ABIX64::Windows)
{
// xmm registers are restored before the official epilogue that has to start with 'add rsp/lea rsp'
for (uint8_t i = kWindowsFirstNonVolXmmReg, offset = 0; i < usableXmmRegCount; i++, offset += 16)
build.vmovaps(RegisterX64{SizeX64::xmmword, i}, xmmword[xmmStorageOffset + offset]);
}
build.add(rsp, fullStackSize);
if (build.abi == ABIX64::Windows)
{
build.pop(rbp);
build.pop(rsi);
build.pop(rdi);
}
build.pop(r15);
build.pop(r14);
build.pop(r13);
build.pop(r12);
build.pop(rbx);
if (build.abi == ABIX64::SystemV)
build.pop(rbp);
build.ret();
// Our entry function is special, it spans the whole remaining code area
unwind.finishFunction(build.getLabelOffset(locations.start), kFullBlockFuncton);
return locations;
}
bool initHeaderFunctions(NativeState& data)
{
AssemblyBuilderX64 build(/* logText= */ false);
UnwindBuilder& unwind = *data.unwindBuilder.get();
unwind.startInfo(UnwindBuilder::X64);
EntryLocations entryLocations = buildEntryFunction(build, unwind);
build.finalize();
unwind.finishInfo();
LUAU_ASSERT(build.data.empty());
uint8_t* codeStart = nullptr;
if (!data.codeAllocator.allocate(
build.data.data(), int(build.data.size()), build.code.data(), int(build.code.size()), data.gateData, data.gateDataSize, codeStart))
{
LUAU_ASSERT(!"Failed to create entry function");
return false;
}
// Set the offset at the begining so that functions in new blocks will not overlay the locations
// specified by the unwind information of the entry function
unwind.setBeginOffset(build.getLabelOffset(entryLocations.prologueEnd));
data.context.gateEntry = codeStart + build.getLabelOffset(entryLocations.start);
data.context.gateExit = codeStart + build.getLabelOffset(entryLocations.epilogueStart);
return true;
}
void assembleHelpers(X64::AssemblyBuilderX64& build, ModuleHelpers& helpers)
{
if (build.logText)
build.logAppend("; updatePcAndContinueInVm\n");
build.setLabel(helpers.updatePcAndContinueInVm);
emitUpdatePcForExit(build);
if (build.logText)
build.logAppend("; exitContinueVmClearNativeFlag\n");
build.setLabel(helpers.exitContinueVmClearNativeFlag);
emitClearNativeFlag(build);
if (build.logText)
build.logAppend("; exitContinueVm\n");
build.setLabel(helpers.exitContinueVm);
emitExit(build, /* continueInVm */ true);
if (build.logText)
build.logAppend("; exitNoContinueVm\n");
build.setLabel(helpers.exitNoContinueVm);
emitExit(build, /* continueInVm */ false);
if (build.logText)
build.logAppend("; continueCallInVm\n");
build.setLabel(helpers.continueCallInVm);
emitContinueCallInVm(build);
if (build.logText)
build.logAppend("; interrupt\n");
build.setLabel(helpers.interrupt);
emitInterrupt(build);
if (build.logText)
build.logAppend("; return\n");
build.setLabel(helpers.return_);
emitReturn(build, helpers);
}
} // namespace X64
} // namespace CodeGen
} // namespace Luau