Files
ppsspp/Core/MIPS/ARM/ArmIRRegCache.cpp
T
Henrik RydgårdandClaude Opus 5.5 bc7afa8845 ARM32 IR JIT: Call a NEON dot routine for the accurate dot products
After clang's ARMv7 output for vfpu_dot_neon(), in Q8-Q15, which the register
cache leaves alone. The end is done in 32-bit integers instead of 64-bit ones,
checked bit-exact against vfpu_dot_finish(). The call site gathers the lanes into
Q8 and Q9 and flushes whatever guest state is in R2, R3, R12 and LR, usually
nothing. Inf and NaN go through a routine that calls vfpu_dot_reference().
Without NEON, Vec4VfpuDot still goes through the IR interpreter.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 22:30:37 +02:00

239 lines
7.2 KiB
C++

// Copyright (c) 2012- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "ppsspp_config.h"
#if PPSSPP_ARCH(ARM)
#ifndef offsetof
#include <cstddef>
#endif
#include "Common/LogReporting.h"
#include "Core/MemMap.h"
#include "Core/MIPS/IR/IRInst.h"
#include "Core/MIPS/IR/IRAnalysis.h"
#include "Core/MIPS/ARM/ArmIRRegCache.h"
#include "Core/MIPS/JitCommon/JitState.h"
using namespace ArmGen;
using namespace ArmIRJitConstants;
ArmIRRegCache::ArmIRRegCache(MIPSComp::JitOptions *jo)
: IRNativeRegCacheBase(jo) {
config_.totalNativeRegs = NUM_GPRS + NUM_FPRS;
config_.mapFPUSIMD = false;
config_.mapUseVRegs = false;
}
void ArmIRRegCache::Init(ARMXEmitter *emitter) {
emit_ = emitter;
}
const int *ArmIRRegCache::GetAllocationOrder(MIPSLoc type, MIPSMap flags, int &count, int &base) const {
// ARMReg values are used directly as native reg numbers.
base = 0;
if (type == MIPSLoc::REG) {
// R0/R1 are scratch, R9-R11 are fixed (see ArmIRJitConstants), R13/R15 are SP/PC.
// The callee-saved ones come first, since they survive calls.
static const int allocationOrder[] = {
R4, R5, R6, R7, R8, R2, R3, R12, R14,
};
count = ARRAY_SIZE(allocationOrder);
return allocationOrder;
} else if (type == MIPSLoc::FREG) {
// S16-S31 (D8-D15) are callee-saved. S0-S3 are scratch.
static const int allocationOrder[] = {
S16, S17, S18, S19, S20, S21, S22, S23,
S24, S25, S26, S27, S28, S29, S30, S31,
S4, S5, S6, S7, S8, S9, S10, S11,
S12, S13, S14, S15,
};
count = ARRAY_SIZE(allocationOrder);
return allocationOrder;
} else {
_assert_msg_(false, "Allocation order not yet implemented");
count = 0;
return nullptr;
}
}
void ArmIRRegCache::FlushHostRegs(std::initializer_list<ARMReg> regs) {
for (ARMReg r : regs)
FlushNativeReg((IRNativeReg)r);
}
void ArmIRRegCache::FlushBeforeCall() {
// These are not preserved by calls.
for (int r : { R2, R3, R12, R14 })
FlushNativeReg((IRNativeReg)r);
for (int r = S4; r <= S15; ++r)
FlushNativeReg((IRNativeReg)r);
}
ARMReg ArmIRRegCache::TryMapTempImm(IRReg r) {
_dbg_assert_(IsValidGPR(r));
// If already mapped, no need for a temporary.
if (IsGPRMapped(r)) {
return R(r);
}
if (mr[r].loc == MIPSLoc::IMM) {
// Try our luck - check for an exact match in another reg.
for (int i = 1; i < TOTAL_MAPPABLE_IRREGS; ++i) {
if (mr[i].loc == MIPSLoc::REG_IMM && mr[i].imm == mr[r].imm) {
return (ARMReg)mr[i].nReg;
}
}
}
return INVALID_REG;
}
ARMReg ArmIRRegCache::GetAndLockTempGPR() {
IRNativeReg reg = AllocateReg(MIPSLoc::REG, MIPSMap::INIT);
if (reg != -1) {
nr[reg].tempLockIRIndex = irIndex_;
}
return (ARMReg)reg;
}
ARMReg ArmIRRegCache::GetAndLockTempFPR() {
IRNativeReg reg = AllocateReg(MIPSLoc::FREG, MIPSMap::INIT);
if (reg != -1) {
nr[reg].tempLockIRIndex = irIndex_;
}
return (ARMReg)reg;
}
ARMReg ArmIRRegCache::MapWithFPRTemp(const IRInst &inst) {
return (ARMReg)MapWithTemp(inst, MIPSLoc::FREG);
}
ARMReg ArmIRRegCache::MapGPR(IRReg mipsReg, MIPSMap mapFlags) {
_dbg_assert_(IsValidGPR(mipsReg));
IRNativeReg nreg = MapNativeReg(MIPSLoc::REG, mipsReg, 1, mapFlags);
return (ARMReg)nreg;
}
ARMReg ArmIRRegCache::MapFPR(IRReg mipsReg, MIPSMap mapFlags) {
_dbg_assert_(IsValidFPR(mipsReg));
_dbg_assert_(mr[mipsReg + 32].loc == MIPSLoc::MEM || mr[mipsReg + 32].loc == MIPSLoc::FREG);
IRNativeReg nreg = MapNativeReg(MIPSLoc::FREG, mipsReg + 32, 1, mapFlags);
if (nreg != -1)
return (ARMReg)nreg;
return INVALID_REG;
}
void ArmIRRegCache::AdjustNativeRegAsPtr(IRNativeReg nreg, bool state) {
_assert_(nreg >= 0 && nreg < NUM_GPRS);
ARMReg r = (ARMReg)nreg;
// 32-bit builds always mask PSP addresses, which destroys the value.
if (state) {
_dbg_assert_(!nr[nreg].isDirty);
emit_->BIC(r, r, Operand2(0xC0, 4));
emit_->ADD(r, r, MEMBASEREG);
} else {
_dbg_assert_(!nr[nreg].isDirty);
emit_->SUB(r, r, MEMBASEREG);
}
}
bool ArmIRRegCache::IsNativeRegCompatible(IRNativeReg nreg, MIPSLoc type, MIPSMap flags, int lanes) {
// Every reg holds one lane.
return lanes == 1 && IRNativeRegCacheBase::IsNativeRegCompatible(nreg, type, flags, lanes);
}
void ArmIRRegCache::LoadNativeReg(IRNativeReg nreg, IRReg first, int lanes) {
_dbg_assert_(first != MIPS_REG_ZERO);
_assert_msg_(lanes == 1, "Can't load %d lanes", lanes);
ARMReg r = (ARMReg)nreg;
if (nreg < NUM_GPRS) {
emit_->LDR(r, CTXREG, GetMipsRegOffset(first));
} else {
_assert_msg_(mr[first].loc == MIPSLoc::FREG, "Cannot load this type: %d", (int)mr[first].loc);
emit_->VLDR(r, CTXREG, GetMipsRegOffset(first));
}
}
void ArmIRRegCache::StoreNativeReg(IRNativeReg nreg, IRReg first, int lanes) {
_dbg_assert_(first != MIPS_REG_ZERO);
_assert_msg_(lanes == 1, "Can't store %d lanes", lanes);
ARMReg r = (ARMReg)nreg;
if (nreg < NUM_GPRS) {
_assert_(mr[first].loc == MIPSLoc::REG || mr[first].loc == MIPSLoc::REG_IMM);
emit_->STR(r, CTXREG, GetMipsRegOffset(first));
} else {
_assert_msg_(mr[first].loc == MIPSLoc::FREG, "Cannot store this type: %d", (int)mr[first].loc);
emit_->VSTR(r, CTXREG, GetMipsRegOffset(first));
}
}
void ArmIRRegCache::SetNativeRegValue(IRNativeReg nreg, uint32_t imm) {
_dbg_assert_(nreg >= 0 && nreg < NUM_GPRS);
emit_->MOVI2R((ARMReg)nreg, imm);
}
void ArmIRRegCache::StoreRegValue(IRReg mreg, uint32_t imm) {
_assert_(IsValidGPRNoZero(mreg));
// Could we get lucky? Check for an exact match in another reg.
ARMReg storeReg = INVALID_REG;
for (int i = 1; i < TOTAL_MAPPABLE_IRREGS; ++i) {
if (mr[i].loc == MIPSLoc::REG_IMM && mr[i].imm == imm) {
storeReg = (ARMReg)mr[i].nReg;
break;
}
}
if (storeReg == INVALID_REG) {
emit_->MOVI2R(SCRATCH1, imm);
storeReg = SCRATCH1;
}
emit_->STR(storeReg, CTXREG, GetMipsRegOffset(mreg));
}
void ArmIRRegCache::FlushAll(bool gprs, bool fprs) {
// Note: make sure not to change the registers when flushing:
// Branching code may expect the armreg to retain its value.
IRNativeRegCacheBase::FlushAll(gprs, fprs);
}
ARMReg ArmIRRegCache::R(IRReg mipsReg) {
_dbg_assert_(IsValidGPR(mipsReg));
_dbg_assert_(mr[mipsReg].loc == MIPSLoc::REG || mr[mipsReg].loc == MIPSLoc::REG_IMM);
if (mr[mipsReg].loc == MIPSLoc::REG || mr[mipsReg].loc == MIPSLoc::REG_IMM) {
return (ARMReg)mr[mipsReg].nReg;
} else {
ERROR_LOG_REPORT(Log::JIT, "Reg %i not in arm reg", mipsReg);
return INVALID_REG; // BAAAD
}
}
ARMReg ArmIRRegCache::F(IRReg mipsReg) {
_dbg_assert_(IsValidFPR(mipsReg));
_dbg_assert_(mr[mipsReg + 32].loc == MIPSLoc::FREG);
if (mr[mipsReg + 32].loc == MIPSLoc::FREG) {
return (ARMReg)mr[mipsReg + 32].nReg;
} else {
ERROR_LOG_REPORT(Log::JIT, "Reg %i not in arm reg", mipsReg);
return INVALID_REG; // BAAAD
}
}
#endif