project64/Source/Project64-rsp/Recompiler Sections.cpp

1275 lines
36 KiB
C++

#include <stdio.h>
#include <windows.h>
#include "CPU.h"
#include "RSP Command.h"
#include "RSP Registers.h"
#include "Recompiler CPU.h"
#include "Rsp.h"
#include "cpu/RspTypes.h"
#include "dma.h"
#include "log.h"
#include "memory.h"
#include "x86.h"
#include "cpu/RSPInstruction.h"
#pragma warning(disable : 4152) // Non-standard extension, function/data pointer conversion in expression
void RSP_Sections_VMUDH(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMUDH - affects the upper 32-bits
if (AccumStyle == Low16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMUDH
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3, &RSP_Vect[RspOp.rt].s16(4), Reg);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM3);
}
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM2);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM2);
}
}
else
{
RSP_MultiElement2Mmx(x86_MM2, x86_MM3);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM3);
}
}
}
void RSP_Sections_VMADH(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMADH - affects the upper 32-bits
if (AccumStyle == Low16BitAccum)
{
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMUDH
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2 + 2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3 + 2, &RSP_Vect[RspOp.rt].s16(4), Reg);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2 + 2);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
}
else
{
RSP_MultiElement2Mmx(x86_MM2 + 2, x86_MM3 + 2);
if (AccumStyle == Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
MmxPaddswRegToReg(x86_MM0, x86_MM0 + 2);
MmxPaddswRegToReg(x86_MM1, x86_MM1 + 2);
}
void RSP_Sections_VMUDL(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMUDL - affects the lower 16-bits
if (AccumStyle != Low16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMUDL
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2, x86_MM3);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
}
void RSP_Sections_VMADL(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMADL - affects the lower 16-bits
if (AccumStyle != Low16BitAccum)
{
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMADL
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2 + 2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3 + 2, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2 + 2, x86_MM3 + 2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
MmxPaddswRegToReg(x86_MM0, x86_MM0 + 2);
MmxPaddswRegToReg(x86_MM1, x86_MM1 + 2);
}
void RSP_Sections_VMUDM(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMUDM - affects the middle 32-bits, s16*u16
if (AccumStyle == High16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMUDM
if (AccumStyle != Middle16BitAccum)
{
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2, x86_MM3);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
}
else
{
if ((RSPOpC.rs & 0xF) < 2)
{
sprintf(Reg, "RSP_Vect[%i].UHW[0]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM4, &RSP_Vect[RSPOpC.vt].u16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].UHW[4]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM5, &RSP_Vect[RSPOpC.vt].u16(4), Reg);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2, x86_MM0);
MmxMoveRegToReg(x86_MM3, x86_MM1);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0, x86_MM4);
MmxPmulhuwRegToReg(x86_MM1, x86_MM5);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2, 15);
MmxPsrawImmed(x86_MM3, 15);
MmxPmullwRegToReg(x86_MM2, x86_MM4);
MmxPmullwRegToReg(x86_MM3, x86_MM5);
}
else if ((RSPOpC.rs & 0xF) >= 8)
{
RSP_Element2Mmx(x86_MM4);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2, x86_MM0);
MmxMoveRegToReg(x86_MM3, x86_MM1);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0, x86_MM4);
MmxPmulhuwRegToReg(x86_MM1, x86_MM4);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2, 15);
MmxPsrawImmed(x86_MM3, 15);
MmxPmullwRegToReg(x86_MM2, x86_MM4);
MmxPmullwRegToReg(x86_MM3, x86_MM4);
}
else
{
RSP_MultiElement2Mmx(x86_MM4, x86_MM5);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2, x86_MM0);
MmxMoveRegToReg(x86_MM3, x86_MM1);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0, x86_MM4);
MmxPmulhuwRegToReg(x86_MM1, x86_MM5);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2, 15);
MmxPsrawImmed(x86_MM3, 15);
MmxPmullwRegToReg(x86_MM2, x86_MM4);
MmxPmullwRegToReg(x86_MM3, x86_MM5);
}
// Add them up
MmxPaddwRegToReg(x86_MM0, x86_MM2);
MmxPaddwRegToReg(x86_MM1, x86_MM3);
}
}
void RSP_Sections_VMADM(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMADM - affects the middle 32-bits, s16*u16
if (AccumStyle == High16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMADM
if (AccumStyle != Middle16BitAccum)
{
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2 + 2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3 + 2, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2 + 2, x86_MM3 + 2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
else
{
if ((RSPOpC.rs & 0xF) < 2)
{
sprintf(Reg, "RSP_Vect[%i].UHW[0]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM4 + 2, &RSP_Vect[RSPOpC.vt].u16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].UHW[4]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM5 + 2, &RSP_Vect[RSPOpC.vt].u16(4), Reg);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2 + 2, x86_MM0 + 2);
MmxMoveRegToReg(x86_MM3 + 2, x86_MM1 + 2);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0 + 2, x86_MM4 + 2);
MmxPmulhuwRegToReg(x86_MM1 + 2, x86_MM5 + 2);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2 + 2, 15);
MmxPsrawImmed(x86_MM3 + 2, 15);
MmxPmullwRegToReg(x86_MM2 + 2, x86_MM4 + 2);
MmxPmullwRegToReg(x86_MM3 + 2, x86_MM5 + 2);
}
else if ((RSPOpC.rs & 0xF) >= 8)
{
RSP_Element2Mmx(x86_MM4 + 2);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2 + 2, x86_MM0 + 2);
MmxMoveRegToReg(x86_MM3 + 2, x86_MM1 + 2);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0 + 2, x86_MM4 + 2);
MmxPmulhuwRegToReg(x86_MM1 + 2, x86_MM4 + 2);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2 + 2, 15);
MmxPsrawImmed(x86_MM3 + 2, 15);
MmxPmullwRegToReg(x86_MM2 + 2, x86_MM4 + 2);
MmxPmullwRegToReg(x86_MM3 + 2, x86_MM4 + 2);
}
else
{
RSP_MultiElement2Mmx(x86_MM4 + 2, x86_MM5 + 2);
// Copy the signed portion
MmxMoveRegToReg(x86_MM2 + 2, x86_MM0 + 2);
MmxMoveRegToReg(x86_MM3 + 2, x86_MM1 + 2);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0 + 2, x86_MM4 + 2);
MmxPmulhuwRegToReg(x86_MM1 + 2, x86_MM5 + 2);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2 + 2, 15);
MmxPsrawImmed(x86_MM3 + 2, 15);
MmxPmullwRegToReg(x86_MM2 + 2, x86_MM4 + 2);
MmxPmullwRegToReg(x86_MM3 + 2, x86_MM5 + 2);
}
// Add them up
MmxPaddwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPaddwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
MmxPaddswRegToReg(x86_MM0, x86_MM0 + 2);
MmxPaddswRegToReg(x86_MM1, x86_MM1 + 2);
}
void RSP_Sections_VMUDN(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMUDN - affects the middle 32-bits, u16*s16
if (AccumStyle == High16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// VMUDN
if (AccumStyle != Middle16BitAccum)
{
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RspOp.rd].s16(4), Reg);
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2, x86_MM3);
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
}
else
{
// NOTE: for code clarity, this is the same as VMUDM,
// just the MMX registers are swapped, this is easier
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM4, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM5, &RSP_Vect[RspOp.rd].s16(4), Reg);
if ((RSPOpC.rs & 0xF) < 2)
{
sprintf(Reg, "RSP_Vect[%i].UHW[0]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RSPOpC.vt].u16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].UHW[4]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RSPOpC.vt].u16(4), Reg);
}
else if ((RSPOpC.rs & 0xF) >= 8)
{
RSP_Element2Mmx(x86_MM0);
MmxMoveRegToReg(x86_MM1, x86_MM0);
}
else
{
RSP_MultiElement2Mmx(x86_MM0, x86_MM1);
}
// Copy the signed portion
MmxMoveRegToReg(x86_MM2, x86_MM0);
MmxMoveRegToReg(x86_MM3, x86_MM1);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0, x86_MM4);
MmxPmulhuwRegToReg(x86_MM1, x86_MM5);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2, 15);
MmxPsrawImmed(x86_MM3, 15);
MmxPmullwRegToReg(x86_MM2, x86_MM4);
MmxPmullwRegToReg(x86_MM3, x86_MM5);
// Add them up
MmxPaddwRegToReg(x86_MM0, x86_MM2);
MmxPaddwRegToReg(x86_MM1, x86_MM3);
}
}
void RSP_Sections_VMADN(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMADN - affects the middle 32-bits, u16*s16
if (AccumStyle == High16BitAccum)
{
return;
}
RSPOpC = RspOp;
// VMADN
if (AccumStyle != Middle16BitAccum)
{
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2 + 2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3 + 2, &RSP_Vect[RspOp.rt].s16(4), Reg);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
else
{
RSP_MultiElement2Mmx(x86_MM2 + 2, x86_MM3 + 2);
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
else
{
// NOTE: for code clarity, this is the same as VMADM,
// just the MMX registers are swapped, this is easier
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM4 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM5 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
if ((RSPOpC.rs & 0xF) < 2)
{
sprintf(Reg, "RSP_Vect[%i].UHW[0]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RSPOpC.vt].u16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].UHW[4]", RSPOpC.rt);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RSPOpC.vt].u16(4), Reg);
}
else if ((RSPOpC.rs & 0xF) >= 8)
{
RSP_Element2Mmx(x86_MM0 + 2);
MmxMoveRegToReg(x86_MM1 + 2, x86_MM0 + 2);
}
else
{
RSP_MultiElement2Mmx(x86_MM0 + 2, x86_MM1 + 2);
}
// Copy the signed portion
MmxMoveRegToReg(x86_MM2 + 2, x86_MM0 + 2);
MmxMoveRegToReg(x86_MM3 + 2, x86_MM1 + 2);
// high((u16)a * b)
MmxPmulhuwRegToReg(x86_MM0 + 2, x86_MM4 + 2);
MmxPmulhuwRegToReg(x86_MM1 + 2, x86_MM5 + 2);
// low((a >> 15) * b)
MmxPsrawImmed(x86_MM2 + 2, 15);
MmxPsrawImmed(x86_MM3 + 2, 15);
MmxPmullwRegToReg(x86_MM2 + 2, x86_MM4 + 2);
MmxPmullwRegToReg(x86_MM3 + 2, x86_MM5 + 2);
// Add them up
MmxPaddwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPaddwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
// Only thing is when we are responsible for clamping
// So we adopt unsigned here?
MmxPaddswRegToReg(x86_MM0, x86_MM0 + 2);
MmxPaddswRegToReg(x86_MM1, x86_MM1 + 2);
}
void RSP_Sections_VMULF(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMULF - affects the middle 32-bits, s16*s16*2
if (AccumStyle == High16BitAccum)
{
MmxXorRegToReg(x86_MM0, x86_MM0);
MmxXorRegToReg(x86_MM1, x86_MM1);
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMULF
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3, &RSP_Vect[RspOp.rt].s16(4), Reg);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM3);
}
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM2);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM2);
}
}
else
{
RSP_MultiElement2Mmx(x86_MM2, x86_MM3);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0, x86_MM2);
MmxPmullwRegToReg(x86_MM1, x86_MM3);
}
else
{
MmxPmulhwRegToReg(x86_MM0, x86_MM2);
MmxPmulhwRegToReg(x86_MM1, x86_MM3);
}
}
MmxPsllwImmed(x86_MM0, 1);
MmxPsllwImmed(x86_MM1, 1);
}
void RSP_Sections_VMACF(RSPOpcode RspOp, DWORD AccumStyle)
{
char Reg[256];
// VMACF - affects the upper 32-bits, s16*s16*2
if (AccumStyle == High16BitAccum)
{
return;
}
RSPOpC = RspOp;
// Load source registers
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM0 + 2, &RSP_Vect[RspOp.rd].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rd);
MmxMoveQwordVariableToReg(x86_MM1 + 2, &RSP_Vect[RspOp.rd].s16(4), Reg);
// VMACF
if ((RspOp.rs & 0x0f) < 2)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM2 + 2, &RSP_Vect[RspOp.rt].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RspOp.rt);
MmxMoveQwordVariableToReg(x86_MM3 + 2, &RSP_Vect[RspOp.rt].s16(4), Reg);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
else if ((RspOp.rs & 0x0f) >= 8)
{
RSP_Element2Mmx(x86_MM2 + 2);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM2 + 2);
}
}
else
{
RSP_MultiElement2Mmx(x86_MM2 + 2, x86_MM3 + 2);
if (AccumStyle != Middle16BitAccum)
{
MmxPmullwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmullwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
else
{
MmxPmulhwRegToReg(x86_MM0 + 2, x86_MM2 + 2);
MmxPmulhwRegToReg(x86_MM1 + 2, x86_MM3 + 2);
}
}
MmxPsllwImmed(x86_MM0 + 2, 1);
MmxPsllwImmed(x86_MM1 + 2, 1);
MmxPaddswRegToReg(x86_MM0, x86_MM0 + 2);
MmxPaddswRegToReg(x86_MM1, x86_MM1 + 2);
}
// Microcode sections
static DWORD Section_000_VMADN; // Yeah I know, but leave it
bool Check_Section_000(void)
{
DWORD i;
RSPOpcode op0, op1;
RSP_LW_IMEM(CompilePC + 0x00, &op0.Value);
// Example: (Mario audio microcode)
// 0x574 VMUDN $v30, $v3, $v23
// 0x578 VMADN $v30, $v4, $v23
if (!(op0.op == RSP_CP2 && (op0.rs & 0x10) != 0 && op0.funct == RSP_VECTOR_VMUDN))
{
return false;
}
Section_000_VMADN = 0;
for (i = 0; i < 0x20; i++)
{
RSP_LW_IMEM(CompilePC + 0x04 + (i * 4), &op1.Value);
if (!(op1.op == RSP_CP2 && (op1.rs & 0x10) != 0 && op1.funct == RSP_VECTOR_VMADN))
{
break;
}
else
{
Section_000_VMADN++;
}
if ((op1.rs & 0xF) >= 2 && (op1.rs & 0xF) <= 7 && IsMmx2Enabled == false)
{
return false;
}
}
// We need at least 1 VMADN
if (Section_000_VMADN == 0)
{
return false;
}
// TODO: check destination and flushes
if (true == WriteToAccum(7, CompilePC + 0x4 + (Section_000_VMADN * 4) - 0x4))
{
return false;
}
if (!IsMmxEnabled)
{
return false;
}
return true;
}
void Compile_Section_000(void)
{
char Reg[256];
RSPOpcode vmudn, vmadn = {0};
DWORD i;
RSP_LW_IMEM(CompilePC + 0x00, &vmudn.Value);
CPU_Message("Compiling: %X to ..., RSP optimization $000", CompilePC);
CPU_Message(" %X %s", CompilePC + 0x00, RSPInstruction(CompilePC + 0x00, vmudn.Value).NameAndParam().c_str());
if (LogRDP)
{
char str[40];
sprintf(str, "%X", CompilePC);
PushImm32(str, CompilePC);
Call_Direct(RDP_LogLoc, "RDP_LogLoc");
AddConstToX86Reg(x86_ESP, 4);
}
for (i = 0; i < Section_000_VMADN; i++)
{
RSP_LW_IMEM(CompilePC + 0x04 + (i * 4), &vmadn.Value);
CPU_Message(" %X %s", CompilePC + 0x04 + (i * 4), RSPInstruction(CompilePC + 0x04 + (i * 4), vmadn.Value).NameAndParam().c_str());
if (LogRDP)
{
char str[40];
sprintf(str, "%X", CompilePC + 0x04 + (i * 4));
PushImm32(str, CompilePC + 0x04 + (i * 4));
Call_Direct(RDP_LogLoc, "RDP_LogLoc");
AddConstToX86Reg(x86_ESP, 4);
}
}
RSP_Sections_VMUDN(vmudn, Low16BitAccum);
CompilePC += 4;
for (i = 0; i < Section_000_VMADN; i++)
{
RSP_LW_IMEM(CompilePC, &vmadn.Value);
CompilePC += 4;
RSP_Sections_VMADN(vmadn, Low16BitAccum);
if (WriteToVectorDest(vmadn.sa, CompilePC - 4) == true)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", vmadn.sa);
MmxMoveQwordRegToVariable(x86_MM0, &RSP_Vect[vmadn.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", vmadn.sa);
MmxMoveQwordRegToVariable(x86_MM1, &RSP_Vect[vmadn.sa].s16(4), Reg);
}
}
sprintf(Reg, "RSP_Vect[%i].HW[0]", vmadn.sa);
MmxMoveQwordRegToVariable(x86_MM0, &RSP_Vect[vmadn.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", vmadn.sa);
MmxMoveQwordRegToVariable(x86_MM1, &RSP_Vect[vmadn.sa].s16(4), Reg);
MmxEmptyMultimediaState();
}
static DWORD Section_001_VMACF;
bool Check_Section_001(void)
{
DWORD i;
RSPOpcode op0, op1;
RSP_LW_IMEM(CompilePC + 0x00, &op0.Value);
// Example: (Mario audio microcode)
// 0xCC0 VMULF $v28, $v28, $v10 [6]
// 0xCC4 VMACF $v28, $v17, $v16
if (!(op0.op == RSP_CP2 && (op0.rs & 0x10) != 0 && op0.funct == RSP_VECTOR_VMULF))
{
return false;
}
Section_001_VMACF = 0;
for (i = 0; i < 0x20; i++)
{
RSP_LW_IMEM(CompilePC + 0x04 + (i * 4), &op1.Value);
if (!(op1.op == RSP_CP2 && (op1.rs & 0x10) != 0 && op1.funct == RSP_VECTOR_VMACF))
{
break;
}
else
{
Section_001_VMACF++;
}
if ((op1.rs & 0xF) >= 2 && (op1.rs & 0xF) <= 7 && IsMmx2Enabled == false)
{
return false;
}
}
// We need at least 1 VMACF
if (Section_001_VMACF == 0)
{
return false;
}
if (!IsMmxEnabled)
{
return false;
}
// Destinations are checked elsewhere, this is fine
if (true == WriteToAccum(7, CompilePC + 0x4 + (Section_001_VMACF * 4) - 0x4))
{
return false;
}
return true;
}
void Compile_Section_001(void)
{
DWORD i;
char Reg[256];
RSPOpcode vmulf, vmacf;
RSP_LW_IMEM(CompilePC + 0x00, &vmulf.Value);
CPU_Message("Compiling: %X to ..., RSP optimization $001", CompilePC);
CPU_Message(" %X %s", CompilePC + 0x00, RSPInstruction(CompilePC + 0x00, vmulf.Value).NameAndParam().c_str());
for (i = 0; i < Section_001_VMACF; i++)
{
RSP_LW_IMEM(CompilePC + 0x04 + (i * 4), &vmacf.Value);
CPU_Message(" %X %s", CompilePC + 0x04 + (i * 4), RSPInstruction(CompilePC + 0x04 + (i * 4), vmacf.Value).NameAndParam().c_str());
}
RSP_Sections_VMULF(vmulf, Middle16BitAccum);
if (WriteToVectorDest(vmulf.sa, CompilePC) == true)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", vmulf.sa);
MmxMoveQwordRegToVariable(x86_MM0, &RSP_Vect[vmulf.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", vmulf.sa);
MmxMoveQwordRegToVariable(x86_MM1, &RSP_Vect[vmulf.sa].s16(4), Reg);
}
CompilePC += 4;
for (i = 0; i < Section_001_VMACF; i++)
{
RSP_LW_IMEM(CompilePC, &vmacf.Value);
CompilePC += 4;
RSP_Sections_VMACF(vmacf, Middle16BitAccum);
if (WriteToVectorDest(vmacf.sa, CompilePC - 4) == true)
{
sprintf(Reg, "RSP_Vect[%i].HW[0]", vmacf.sa);
MmxMoveQwordRegToVariable(x86_MM0, &RSP_Vect[vmacf.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", vmacf.sa);
MmxMoveQwordRegToVariable(x86_MM1, &RSP_Vect[vmacf.sa].s16(4), Reg);
}
}
MmxEmptyMultimediaState();
}
bool Check_Section_002(void)
{
DWORD Count;
RSPOpcode op[0x0C];
for (Count = 0; Count < 0x0C; Count++)
{
RSP_LW_IMEM(CompilePC + (Count * 0x04), &op[Count].Value);
}
/*
** Example: (Mario audio microcode)
** 5F4 VMUDH $v2, $v21, $v27 [6]
** 5F8 VMADH $v2, $v20, $v27 [7]
** 5FC VMADH $v2, $v19, $v30 [0]
** 600 VMADH $v2, $v18, $v30 [1]
** 604 VMADH $v2, $v17, $v30 [2]
** 608 VMADH $v2, $v16, $v30 [3]
** 60C VMADH $v28, $v15, $v30 [4]
** 610 VMADH $v2, $v14, $v30 [5]
** 614 VMADH $v2, $v13, $v30 [6]
** 618 VMADH $v2, $v30, $v31 [5]
** 61C VSAW $v26 [9], $v7, $v28
** 620 VSAW $v28 [8], $v7, $v28
*/
if (!IsMmxEnabled)
{
return false;
}
if (!(op[0].op == RSP_CP2 && (op[0].rs & 0x10) != 0 && op[0].funct == RSP_VECTOR_VMUDH))
{
return false;
}
if ((op[0].rs & 0xF) < 8)
{
return false;
}
for (Count = 1; Count < 10; Count++)
{
if (!(op[Count].op == RSP_CP2 && (op[Count].rs & 0x10) != 0 && op[Count].funct == RSP_VECTOR_VMADH))
{
return false;
}
if ((op[Count].rs & 0xF) < 8)
{
return false;
}
}
if (!(op[10].op == RSP_CP2 && (op[10].rs & 0x10) != 0 && op[10].funct == RSP_VECTOR_VSAW)) return false;
if (!(op[11].op == RSP_CP2 && (op[11].rs & 0x10) != 0 && op[11].funct == RSP_VECTOR_VSAW)) return false;
if ((op[10].rs & 0xF) != 9)
{
return false;
}
if ((op[11].rs & 0xF) != 8)
{
return false;
}
if (true == WriteToAccum(7, CompilePC + 0x2C))
return false;
return true;
}
void Compile_Section_002(void)
{
char Reg[256];
DWORD Count;
RSPOpcode op[0x0C];
RSPOpcode vmudh, vsaw;
CPU_Message("Compiling: %X to ..., RSP optimization $002", CompilePC);
for (Count = 0; Count < 0xC; Count++)
{
RSP_LW_IMEM(CompilePC + (Count * 0x04), &op[Count].Value);
CPU_Message(" %X %s", CompilePC + (Count * 0x04), RSPInstruction(CompilePC + (Count * 0x04), op[Count].Value).NameAndParam().c_str());
if (LogRDP)
{
char str[40];
sprintf(str, "%X", CompilePC + (Count * 0x04));
PushImm32(str, CompilePC + (Count * 0x04));
Call_Direct(RDP_LogLoc, "RDP_LogLoc");
AddConstToX86Reg(x86_ESP, 4);
}
}
vmudh = op[0];
RSP_Sections_VMUDH(vmudh, High16BitAccum);
// VMADHs
for (Count = 1; Count < 10; Count++)
{
RSP_Sections_VMADH(op[Count], High16BitAccum);
}
// VSAWs
vsaw = op[10];
MmxXorRegToReg(x86_MM4, x86_MM4);
sprintf(Reg, "RSP_Vect[%i].HW[0]", RSPOpC.sa);
MmxMoveQwordRegToVariable(x86_MM4, &RSP_Vect[vsaw.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RSPOpC.sa);
MmxMoveQwordRegToVariable(x86_MM4, &RSP_Vect[vsaw.sa].s16(4), Reg);
vsaw = op[11];
sprintf(Reg, "RSP_Vect[%i].HW[0]", RSPOpC.sa);
MmxMoveQwordRegToVariable(x86_MM0, &RSP_Vect[vsaw.sa].s16(0), Reg);
sprintf(Reg, "RSP_Vect[%i].HW[4]", RSPOpC.sa);
MmxMoveQwordRegToVariable(x86_MM1, &RSP_Vect[vsaw.sa].s16(4), Reg);
MmxEmptyMultimediaState();
CompilePC += 12 * sizeof(RSPOpcode);
}
bool Check_Section_003(void)
{
DWORD Count;
RSPOpcode op[4];
for (Count = 0; Count < 4; Count++)
{
RSP_LW_IMEM(CompilePC + (Count * 0x04), &op[Count].Value);
}
// Example: (Zelda audio microcode)
// VMUDM $v23, $v31, $v23 [7]
// VMADH $v23, $v31, $v22 [7]
// VMADM $v22, $v25, $v18 [4]
// VMADN $v23, $v31, $v30 [0]
if (op[0].Value == 0x4BF7FDC5 && op[1].Value == 0x4BF6FDCF && op[2].Value == 0x4B92CD8D && op[3].Value == 0x4B1EFDCE)
{
if (true == WriteToAccum(7, CompilePC + 0xc))
return false;
return true;
}
return false;
}
static void resampler_hle()
{
UDWORD accum, initial;
DWORD const2 = (DWORD)RSP_Vect[18].u16(4 ^ 7);
__int64 const3 = (__int64)((int)RSP_Vect[30].s16(0 ^ 7)) << 16;
// VMUDM $v23, $v31, $v23 [7]
initial.DW = (__int64)((DWORD)RSP_Vect[23].u16(7 ^ 7)) << 16;
// VMADH $v23, $v31, $v22 [7]
initial.W[1] += (int)RSP_Vect[22].s16(7 ^ 7);
for (uint8_t i = 0; i < 8; i++)
{
accum.DW = initial.DW;
// VMADM $v22, $v25, $v18 [4]
accum.DW += (__int64)((int)RSP_Vect[25].s16(i) * const2) << 16;
if (accum.W[1] > 0x7FFF)
{
RSP_Vect[22].s16(i) = 0x7FFF;
}
else if (accum.W[1] < -0x8000)
{
RSP_Vect[22].s16(i) = -0x8000;
}
else
{
RSP_Vect[22].s16(i) = accum.HW[2];
}
// VMADN $v23, $v31, $v30 [0]
accum.DW += const3;
if (accum.W[1] > 0x7FFF)
{
RSP_Vect[23].u16(i) = 0xFFFF;
}
else if (accum.W[1] < -0x8000)
{
RSP_Vect[23].s16(i) = 0;
}
else
{
RSP_Vect[23].s16(i) = accum.HW[1];
}
}
}
void Compile_Section_003(void)
{
CPU_Message("Compiling: %X to ..., RSP optimization $003", CompilePC);
Call_Direct(resampler_hle, "Resampler_HLE");
CompilePC += 4 * sizeof(RSPOpcode);
}
bool RSP_DoSections(void)
{
if (true == Check_Section_000())
{
Compile_Section_000();
return true;
}
if (true == Check_Section_001())
{
Compile_Section_001();
return true;
}
if (true == Check_Section_002())
{
Compile_Section_002();
return true;
}
if (true == Check_Section_003())
{
Compile_Section_003();
return true;
}
return false;
}