Linuxydable/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Merge pull request #4086 from MerryMage/abi

Browse source 
xbyak_abi: Cleanup
This commit is contained in:
bunnei 2020-06-17 11:20:52 -04:00 committed by GitHub
commit a8ac99b619
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 36 additions and 73 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

97
src/common/x64/xbyak_abi.h
View file

@ -11,7 +11,7 @@
namespace Common::X64 { namespace Common::X64 {
inline int RegToIndex(const Xbyak::Reg& reg) { inline std::size_t RegToIndex(const Xbyak::Reg& reg) {
using Kind = Xbyak::Reg::Kind; using Kind = Xbyak::Reg::Kind;
ASSERT_MSG((reg.getKind() & (Kind::REG | Kind::XMM)) != 0, ASSERT_MSG((reg.getKind() & (Kind::REG | Kind::XMM)) != 0,
"RegSet only support GPRs and XMM registers."); "RegSet only support GPRs and XMM registers.");
@ -19,17 +19,17 @@ inline int RegToIndex(const Xbyak::Reg& reg) {
return reg.getIdx() + (reg.getKind() == Kind::REG ? 0 : 16); return reg.getIdx() + (reg.getKind() == Kind::REG ? 0 : 16);
} }
inline Xbyak::Reg64 IndexToReg64(int reg_index) { inline Xbyak::Reg64 IndexToReg64(std::size_t reg_index) {
ASSERT(reg_index < 16); ASSERT(reg_index < 16);
return Xbyak::Reg64(reg_index); return Xbyak::Reg64(static_cast<int>(reg_index));
} }
inline Xbyak::Xmm IndexToXmm(int reg_index) { inline Xbyak::Xmm IndexToXmm(std::size_t reg_index) {
ASSERT(reg_index >= 16 && reg_index < 32); ASSERT(reg_index >= 16 && reg_index < 32);
return Xbyak::Xmm(reg_index - 16); return Xbyak::Xmm(static_cast<int>(reg_index - 16));
} }
inline Xbyak::Reg IndexToReg(int reg_index) { inline Xbyak::Reg IndexToReg(std::size_t reg_index) {
if (reg_index < 16) { if (reg_index < 16) {
return IndexToReg64(reg_index); return IndexToReg64(reg_index);
} else { } else {
@ -151,9 +151,13 @@ constexpr size_t ABI_SHADOW_SPACE = 0;
#endif #endif
inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment, struct ABIFrameInfo {
size_t needed_frame_size, s32* out_subtraction, s32 subtraction;
s32* out_xmm_offset) { s32 xmm_offset;
};
inline ABIFrameInfo ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
size_t needed_frame_size) {
const auto count = (regs & ABI_ALL_GPRS).count(); const auto count = (regs & ABI_ALL_GPRS).count();
rsp_alignment -= count * 8; rsp_alignment -= count * 8;
size_t subtraction = 0; size_t subtraction = 0;
@ -170,33 +174,28 @@ inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
rsp_alignment -= subtraction; rsp_alignment -= subtraction;
subtraction += rsp_alignment & 0xF; subtraction += rsp_alignment & 0xF;
*out_subtraction = (s32)subtraction; return ABIFrameInfo{static_cast<s32>(subtraction),
*out_xmm_offset = (s32)(subtraction - xmm_base_subtraction); static_cast<s32>(subtraction - xmm_base_subtraction)};
} }
inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs, inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
size_t rsp_alignment, size_t needed_frame_size = 0) { size_t rsp_alignment, size_t needed_frame_size = 0) {
s32 subtraction, xmm_offset; auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
for (std::size_t i = 0; i < regs.size(); ++i) { for (std::size_t i = 0; i < regs.size(); ++i) {
if (regs[i] && ABI_ALL_GPRS[i]) { if (regs[i] && ABI_ALL_GPRS[i]) {
code.push(IndexToReg64(static_cast<int>(i)));
}
}
if (subtraction != 0) {
code.sub(code.rsp, subtraction);
}
for (int i = 0; i < regs.count(); i++) {
if (regs.test(i) & ABI_ALL_GPRS.test(i)) {
code.push(IndexToReg64(i)); code.push(IndexToReg64(i));
} }
} }
if (frame_info.subtraction != 0) {
code.sub(code.rsp, frame_info.subtraction);
}
for (std::size_t i = 0; i < regs.size(); ++i) { for (std::size_t i = 0; i < regs.size(); ++i) {
if (regs[i] && ABI_ALL_XMMS[i]) { if (regs[i] && ABI_ALL_XMMS[i]) {
code.movaps(code.xword[code.rsp + xmm_offset], IndexToXmm(static_cast<int>(i))); code.movaps(code.xword[code.rsp + frame_info.xmm_offset], IndexToXmm(i));
xmm_offset += 0x10; frame_info.xmm_offset += 0x10;
} }
} }
@ -205,59 +204,23 @@ inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::b
inline void ABI_PopRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs, inline void ABI_PopRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
size_t rsp_alignment, size_t needed_frame_size = 0) { size_t rsp_alignment, size_t needed_frame_size = 0) {
s32 subtraction, xmm_offset; auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
for (std::size_t i = 0; i < regs.size(); ++i) { for (std::size_t i = 0; i < regs.size(); ++i) {
if (regs[i] && ABI_ALL_XMMS[i]) { if (regs[i] && ABI_ALL_XMMS[i]) {
code.movaps(IndexToXmm(static_cast<int>(i)), code.xword[code.rsp + xmm_offset]); code.movaps(IndexToXmm(i), code.xword[code.rsp + frame_info.xmm_offset]);
xmm_offset += 0x10; frame_info.xmm_offset += 0x10;
} }
} }
if (subtraction != 0) { if (frame_info.subtraction != 0) {
code.add(code.rsp, subtraction); code.add(code.rsp, frame_info.subtraction);
} }
// GPRs need to be popped in reverse order // GPRs need to be popped in reverse order
for (int i = 15; i >= 0; i--) { for (std::size_t j = 0; j < regs.size(); ++j) {
if (regs[i]) { const std::size_t i = regs.size() - j - 1;
code.pop(IndexToReg64(i));
}
}
}
inline size_t ABI_PushRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
size_t rsp_alignment,
size_t needed_frame_size = 0) {
s32 subtraction, xmm_offset;
ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
for (std::size_t i = 0; i < regs.size(); ++i) {
if (regs[i] && ABI_ALL_GPRS[i]) { if (regs[i] && ABI_ALL_GPRS[i]) {
code.push(IndexToReg64(static_cast<int>(i)));
}
}
if (subtraction != 0) {
code.sub(code.rsp, subtraction);
}
return ABI_SHADOW_SPACE;
}
inline void ABI_PopRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
size_t rsp_alignment, size_t needed_frame_size = 0) {
s32 subtraction, xmm_offset;
ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
if (subtraction != 0) {
code.add(code.rsp, subtraction);
}
// GPRs need to be popped in reverse order
for (int i = 15; i >= 0; i--) {
if (regs[i]) {
code.pop(IndexToReg64(i)); code.pop(IndexToReg64(i));
} }
} }

12
src/video_core/macro/macro_jit_x64.cpp
View file

@ -295,22 +295,22 @@ void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
sub(result, opcode.immediate * -1); sub(result, opcode.immediate * -1);
} }
} }
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(Common::X64::ABI_PARAM1, qword[STATE]); mov(Common::X64::ABI_PARAM1, qword[STATE]);
mov(Common::X64::ABI_PARAM2, RESULT); mov(Common::X64::ABI_PARAM2, RESULT);
Common::X64::CallFarFunction(*this, &Read); Common::X64::CallFarFunction(*this, &Read);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(RESULT, Common::X64::ABI_RETURN.cvt32()); mov(RESULT, Common::X64::ABI_RETURN.cvt32());
Compile_ProcessResult(opcode.result_operation, opcode.dst); Compile_ProcessResult(opcode.result_operation, opcode.dst);
} }
void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) { void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(Common::X64::ABI_PARAM1, qword[STATE]); mov(Common::X64::ABI_PARAM1, qword[STATE]);
mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS); mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS);
mov(Common::X64::ABI_PARAM3, value); mov(Common::X64::ABI_PARAM3, value);
Common::X64::CallFarFunction(*this, &Send); Common::X64::CallFarFunction(*this, &Send);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0); Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
Xbyak::Label dont_process{}; Xbyak::Label dont_process{};
// Get increment // Get increment
@ -414,7 +414,7 @@ void MacroJITx64Impl::Compile() {
bool keep_executing = true; bool keep_executing = true;
labels.fill(Xbyak::Label()); labels.fill(Xbyak::Label());
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8); Common::X64::ABI_PushRegistersAndAdjustStack(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
// JIT state // JIT state
mov(STATE, Common::X64::ABI_PARAM1); mov(STATE, Common::X64::ABI_PARAM1);
mov(PARAMETERS, Common::X64::ABI_PARAM2); mov(PARAMETERS, Common::X64::ABI_PARAM2);
@ -452,7 +452,7 @@ void MacroJITx64Impl::Compile() {
L(end_of_code); L(end_of_code);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8); Common::X64::ABI_PopRegistersAndAdjustStack(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
ret(); ret();
ready(); ready();
program = getCode<ProgramType>(); program = getCode<ProgramType>();