ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge latest ArmEmitter changes from ppsspp while we're at it.

This commit is contained in:
Ryan Houdek 2013-02-27 11:53:24 -06:00
parent 8c779bb242
commit f1d727cf13
2 changed files with 123 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

131
Source/Core/Common/Src/ArmEmitter.cpp
View File

@ -22,11 +22,16 @@
#include <assert.h> #include <assert.h>
#include <stdarg.h> #include <stdarg.h>
// For cache flushing on Symbian/Blackberry // For cache flushing on Symbian/iOS/Blackberry
#ifdef __SYMBIAN32__ #ifdef __SYMBIAN32__
#include <e32std.h> #include <e32std.h>
#endif #endif
#ifdef IOS
#include <libkern/OSCacheControl.h>
#include <sys/mman.h>
#endif
#ifdef BLACKBERRY #ifdef BLACKBERRY
#include <sys/mman.h> #include <sys/mman.h>
#endif #endif
@ -185,18 +190,16 @@ void ARMXEmitter::FlushIcache()
void ARMXEmitter::FlushIcacheSection(u8 *start, u8 *end) void ARMXEmitter::FlushIcacheSection(u8 *start, u8 *end)
{ {
#ifdef __SYMBIAN32__ #ifdef __SYMBIAN32__
User::IMB_Range( start, end); User::IMB_Range(start, end);
#elif defined(BLACKBERRY) #elif defined(BLACKBERRY)
msync(start, end - start, MS_SYNC | MS_INVALIDATE_ICACHE); msync(start, end - start, MS_SYNC | MS_INVALIDATE_ICACHE);
#else #elif defined(IOS)
#ifndef _WIN32 sys_cache_control(kCacheFunctionPrepareForExecution, start, end - start);
#ifdef ANDROID #elif !defined(_WIN32)
__builtin___clear_cache (start, end); #ifndef ANDROID
#else start = startcode;
// If on Linux, we HAVE to clear from start addr or else everything gets /really/ unstable
__builtin___clear_cache (startcode, end);
#endif
#endif #endif
__builtin___clear_cache(start, end);
#endif #endif
} }
@ -493,6 +496,7 @@ void ARMXEmitter::SMULL(ARMReg destLo, ARMReg destHi, ARMReg rm, ARMReg rn)
{ {
Write4OpMultiply(0xC, destLo, destHi, rn, rm); Write4OpMultiply(0xC, destLo, destHi, rn, rm);
} }
void ARMXEmitter::SXTB (ARMReg dest, ARMReg op2) void ARMXEmitter::SXTB (ARMReg dest, ARMReg op2)
{ {
Write32(condition | (0x6AF << 16) | (dest << 12) | (7 << 4) | op2); Write32(condition | (0x6AF << 16) | (dest << 12) | (7 << 4) | op2);
@ -528,9 +532,10 @@ void ARMXEmitter::MRS (ARMReg dest)
{ {
Write32(condition | (16 << 20) | (15 << 16) | (dest << 12)); Write32(condition | (16 << 20) | (15 << 16) | (dest << 12));
} }
void ARMXEmitter::WriteStoreOp(u32 op, ARMReg dest, ARMReg src, Operand2 op2) void ARMXEmitter::WriteStoreOp(u32 op, ARMReg dest, ARMReg src, Operand2 op2)
{ {
if (op2.GetData() == 0) // Don't index if (op2.GetData() == 0) // set the preindex bit, but not the W bit!
Write32(condition | 0x01800000 | (op << 20) | (dest << 16) | (src << 12) | op2.Imm12()); Write32(condition | 0x01800000 | (op << 20) | (dest << 16) | (src << 12) | op2.Imm12());
else else
Write32(condition | (op << 20) | (3 << 23) | (dest << 16) | (src << 12) | op2.Imm12()); Write32(condition | (op << 20) | (3 << 23) | (dest << 16) | (src << 12) | op2.Imm12());
@ -620,6 +625,7 @@ ARMReg ARMXEmitter::SubBase(ARMReg Reg)
} }
return Reg; return Reg;
} }
// NEON Specific // NEON Specific
void ARMXEmitter::VADD(IntegerSize Size, ARMReg Vd, ARMReg Vn, ARMReg Vm) void ARMXEmitter::VADD(IntegerSize Size, ARMReg Vd, ARMReg Vn, ARMReg Vm)
{ {
@ -651,16 +657,20 @@ void ARMXEmitter::VSUB(IntegerSize Size, ARMReg Vd, ARMReg Vn, ARMReg Vm)
Write32((0xF3 << 24) | ((Vd & 0x10) << 18) | (Size << 20) | ((Vn & 0xF) << 16) \ Write32((0xF3 << 24) | ((Vd & 0x10) << 18) | (Size << 20) | ((Vn & 0xF) << 16) \
| ((Vd & 0xF) << 12) | (0x8 << 8) | ((Vn & 0x10) << 3) | (1 << 6) \ | ((Vd & 0xF) << 12) | (0x8 << 8) | ((Vn & 0x10) << 3) | (1 << 6) \
| ((Vm & 0x10) << 2) | (Vm & 0xF)); | ((Vm & 0x10) << 2) | (Vm & 0xF));
} }
// VFP Specific // VFP Specific
void ARMXEmitter::VLDR(ARMReg Dest, ARMReg Base, u16 op) void ARMXEmitter::VLDR(ARMReg Dest, ARMReg Base, u16 offset)
{ {
_assert_msg_(DYNA_REC, Dest >= S0 && Dest <= D31, "Passed Invalid dest register to VLDR"); _assert_msg_(DYNA_REC, Dest >= S0 && Dest <= D31, "Passed Invalid dest register to VLDR");
_assert_msg_(DYNA_REC, Base <= R15, "Passed invalid Base register to VLDR"); _assert_msg_(DYNA_REC, Base <= R15, "Passed invalid Base register to VLDR");
_assert_msg_(DYNA_REC, !(op & 3), "Offset needs to be word aligned"); _assert_msg_(DYNA_REC, (offset & 0xC03) == 0, "VLDR: Offset needs to be word aligned and small enough");
if (offset & 0xC03) {
ERROR_LOG(DYNA_REC, "VLDR: Bad offset %08x", offset);
}
bool single_reg = Dest < D0; bool single_reg = Dest < D0;
Dest = SubBase(Dest); Dest = SubBase(Dest);
@ -668,20 +678,25 @@ void ARMXEmitter::VLDR(ARMReg Dest, ARMReg Base, u16 op)
if (single_reg) if (single_reg)
{ {
Write32(NO_COND | (0x1B << 23) | ((Dest & 0x1) << 22) | (1 << 20) | (Base << 16) \ Write32(NO_COND | (0x1B << 23) | ((Dest & 0x1) << 22) | (1 << 20) | (Base << 16) \
| ((Dest & 0x1E) << 11) | (10 << 8) | (op >> 2)); | ((Dest & 0x1E) << 11) | (10 << 8) | (offset >> 2));
} }
else else
{ {
Write32(NO_COND | (0x1B << 23) | ((Dest & 0x10) << 18) | (1 << 20) | (Base << 16) \ Write32(NO_COND | (0x1B << 23) | ((Dest & 0x10) << 18) | (1 << 20) | (Base << 16) \
| ((Dest & 0xF) << 12) | (11 << 8) | (op >> 2)); | ((Dest & 0xF) << 12) | (11 << 8) | (offset >> 2));
} }
} }
void ARMXEmitter::VSTR(ARMReg Src, ARMReg Base, u16 op) void ARMXEmitter::VSTR(ARMReg Src, ARMReg Base, u16 offset)
{ {
_assert_msg_(DYNA_REC, Src >= S0 && Src <= D31, "Passed invalid src register to VSTR"); _assert_msg_(DYNA_REC, Src >= S0 && Src <= D31, "Passed invalid src register to VSTR");
_assert_msg_(DYNA_REC, Base <= R15, "Passed invalid base register to VSTR"); _assert_msg_(DYNA_REC, Base <= R15, "Passed invalid base register to VSTR");
_assert_msg_(DYNA_REC, !(op & 3), "Offset needs to be word aligned"); _assert_msg_(DYNA_REC, (offset & 0xC03) == 0, "VSTR: Offset needs to be word aligned");
if (offset & 0xC03) {
ERROR_LOG(DYNA_REC, "VSTR: Bad offset %08x", offset);
}
bool single_reg = Src < D0; bool single_reg = Src < D0;
Src = SubBase(Src); Src = SubBase(Src);
@ -689,13 +704,13 @@ void ARMXEmitter::VSTR(ARMReg Src, ARMReg Base, u16 op)
if (single_reg) if (single_reg)
{ {
Write32(NO_COND | (0x1B << 23) | ((Src & 0x1) << 22) | (Base << 16) \ Write32(NO_COND | (0x1B << 23) | ((Src & 0x1) << 22) | (Base << 16) \
| ((Src & 0x1E) << 11) | (10 << 8) | (op >> 2)); | ((Src & 0x1E) << 11) | (10 << 8) | (offset >> 2));
} }
else else
{ {
Write32(NO_COND | (0x1B << 23) | ((Src & 0x10) << 18) | (Base << 16) \ Write32(NO_COND | (0x1B << 23) | ((Src & 0x10) << 18) | (Base << 16) \
| ((Src & 0xF) << 12) | (11 << 8) | (op >> 2)); | ((Src & 0xF) << 12) | (11 << 8) | (offset >> 2));
} }
} }
void ARMXEmitter::VCMP(ARMReg Vd, ARMReg Vm) void ARMXEmitter::VCMP(ARMReg Vd, ARMReg Vm)
@ -779,12 +794,14 @@ void ARMXEmitter::VSQRT(ARMReg Vd, ARMReg Vm)
| ((Vd & 0xF) << 12) | (0x2F << 6) | ((Vm & 0x10) << 2) | (Vm & 0xF)); | ((Vd & 0xF) << 12) | (0x2F << 6) | ((Vm & 0x10) << 2) | (Vm & 0xF));
} }
} }
// VFP and ASIMD // VFP and ASIMD
void ARMXEmitter::VABS(ARMReg Vd, ARMReg Vm) void ARMXEmitter::VABS(ARMReg Vd, ARMReg Vm)
{ {
_assert_msg_(DYNA_REC, Vd < Q0, "VABS doesn't currently support Quad reg"); _assert_msg_(DYNA_REC, Vd < Q0, "VABS doesn't currently support Quad reg");
_assert_msg_(DYNA_REC, Vd >= S0, "VABS doesn't support ARM Regs"); _assert_msg_(DYNA_REC, Vd >= S0, "VABS doesn't support ARM Regs");
bool single_reg = Vd < D0; bool single_reg = Vd < D0;
bool double_reg = Vd < Q0;
Vd = SubBase(Vd); Vd = SubBase(Vd);
Vm = SubBase(Vm); Vm = SubBase(Vm);
@ -795,11 +812,20 @@ void ARMXEmitter::VABS(ARMReg Vd, ARMReg Vm)
| (0xAC << 4) | ((Vm & 0x1) << 5) | (Vm >> 1)); | (0xAC << 4) | ((Vm & 0x1) << 5) | (Vm >> 1));
} }
else else
{
if (double_reg)
{ {
Write32(NO_COND | (0xEB << 20) | ((Vd & 0x10) << 18) | ((Vd & 0xF) << 12) \ Write32(NO_COND | (0xEB << 20) | ((Vd & 0x10) << 18) | ((Vd & 0xF) << 12) \
| (0xBC << 4) | ((Vm & 0x10) << 1) | (Vm & 0xF)); | (0xBC << 4) | ((Vm & 0x10) << 1) | (Vm & 0xF));
} }
else
{
_assert_msg_(DYNA_REC, cpu_info.bNEON, "Trying to use VADD with Quad Reg without support!");
// XXX: TODO
}
}
} }
void ARMXEmitter::VADD(ARMReg Vd, ARMReg Vn, ARMReg Vm) void ARMXEmitter::VADD(ARMReg Vd, ARMReg Vn, ARMReg Vm)
{ {
_assert_msg_(DYNA_REC, Vd >= S0, "Passed invalid dest register to VADD"); _assert_msg_(DYNA_REC, Vd >= S0, "Passed invalid dest register to VADD");
@ -870,6 +896,70 @@ void ARMXEmitter::VSUB(ARMReg Vd, ARMReg Vn, ARMReg Vm)
} }
} }
} }
void ARMXEmitter::VMUL(ARMReg Vd, ARMReg Vn, ARMReg Vm)
{
_assert_msg_(DYNA_REC, Vd >= S0, "Passed invalid dest register to VADD");
_assert_msg_(DYNA_REC, Vn >= S0, "Passed invalid Vn to VADD");
_assert_msg_(DYNA_REC, Vm >= S0, "Passed invalid Vm to VADD");
bool single_reg = Vd < D0;
bool double_reg = Vd < Q0;
Vd = SubBase(Vd);
Vn = SubBase(Vn);
Vm = SubBase(Vm);
if (single_reg)
{
Write32(NO_COND | (0x1C << 23) | ((Vd & 0x1) << 22) | (0x2 << 20) \
| ((Vn & 0x1E) << 15) | ((Vd & 0x1E) << 11) | (0x5 << 9) \
| ((Vn & 0x1) << 7) | ((Vm & 0x1) << 5) | (Vm >> 1));
}
else
{
if (double_reg)
{
Write32(NO_COND | (0x1C << 23) | ((Vd & 0x10) << 18) | (0x2 << 20) \
| ((Vn & 0xF) << 16) | ((Vd & 0xF) << 12) | (0xB << 8) \
| ((Vn & 0x10) << 3) | ((Vm & 0x10) << 2) | (Vm & 0xF));
}
else
{
_assert_msg_(DYNA_REC, cpu_info.bNEON, "Trying to use VMUL with Quad Reg without support!");
// XXX: TODO
}
}
}
void ARMXEmitter::VNEG(ARMReg Vd, ARMReg Vm)
{
_assert_msg_(DYNA_REC, Vd < Q0, "VNEG doesn't currently support Quad reg");
_assert_msg_(DYNA_REC, Vd >= S0, "VNEG doesn't support ARM Regs");
bool single_reg = Vd < D0;
bool double_reg = Vd < Q0;
Vd = SubBase(Vd);
Vm = SubBase(Vm);
if (single_reg)
{
Write32(NO_COND | (0x1D << 23) | ((Vd & 0x1) << 22) | (0x31 << 16) \
| ((Vd & 0x1E) << 11) | (0x29 << 6) | ((Vm & 0x1) << 5) | (Vm >> 1));
}
else
{
if (double_reg)
{
Write32(NO_COND | (0x1D << 23) | ((Vd & 0x10) << 18) | (0x31 << 16) \
| ((Vd & 0xF) << 12) | (0x2D << 6) | ((Vm & 0x10) << 2) | (Vm & 0xF));
}
else
{
_assert_msg_(DYNA_REC, cpu_info.bNEON, "Trying to use VNEG with Quad Reg without support!");
// XXX: TODO
}
}
}
void ARMXEmitter::VMOV(ARMReg Dest, ARMReg Src, bool high) void ARMXEmitter::VMOV(ARMReg Dest, ARMReg Src, bool high)
{ {
@ -881,6 +971,7 @@ void ARMXEmitter::VMOV(ARMReg Dest, ARMReg Src, bool high)
Write32(NO_COND | (0xE << 24) | (high << 21) | ((Dest & 0xF) << 16) | (Src << 12) \ Write32(NO_COND | (0xE << 24) | (high << 21) | ((Dest & 0xF) << 16) | (Src << 12) \
| (11 << 8) | ((Dest & 0x10) << 3) | (1 << 4)); | (11 << 8) | ((Dest & 0x10) << 3) | (1 << 4));
} }
void ARMXEmitter::VMOV(ARMReg Dest, ARMReg Src) void ARMXEmitter::VMOV(ARMReg Dest, ARMReg Src)
{ {
if (Dest > R15) if (Dest > R15)

12
Source/Core/Common/Src/ArmEmitter.h
View File

@ -227,7 +227,7 @@ public:
} }
const u32 IMMSR() // IMM shifted register const u32 IMMSR() // IMM shifted register
{ {
_assert_msg_(DYNA_REC, Type = TYPE_IMMSREG, "IMMSR must be imm shifted register"); _assert_msg_(DYNA_REC, Type == TYPE_IMMSREG, "IMMSR must be imm shifted register");
return ((IndexOrShift & 0x1f) << 7 | (Shift << 5) | Value); return ((IndexOrShift & 0x1f) << 7 | (Shift << 5) | Value);
} }
const u32 RSR() // Register shifted register const u32 RSR() // Register shifted register
@ -343,6 +343,7 @@ private:
void WriteShiftedDataOp(u32 op, bool SetFlags, ARMReg dest, ARMReg src, Operand2 op2); void WriteShiftedDataOp(u32 op, bool SetFlags, ARMReg dest, ARMReg src, Operand2 op2);
void WriteSignedMultiply(u32 Op, u32 Op2, u32 Op3, ARMReg dest, ARMReg r1, ARMReg r2); void WriteSignedMultiply(u32 Op, u32 Op2, u32 Op3, ARMReg dest, ARMReg r1, ARMReg r2);
void Write4OpMultiply(u32 op, ARMReg destLo, ARMReg destHi, ARMReg rn, ARMReg rm); void Write4OpMultiply(u32 op, ARMReg destLo, ARMReg destHi, ARMReg rn, ARMReg rm);
// New Ops // New Ops
@ -498,8 +499,8 @@ public:
void VSUB(IntegerSize Size, ARMReg Vd, ARMReg Vn, ARMReg Vm); void VSUB(IntegerSize Size, ARMReg Vd, ARMReg Vn, ARMReg Vm);
// VFP Only // VFP Only
void VLDR(ARMReg Dest, ARMReg Base, u16 op); void VLDR(ARMReg Dest, ARMReg Base, u16 offset);
void VSTR(ARMReg Src, ARMReg Base, u16 op); void VSTR(ARMReg Src, ARMReg Base, u16 offset);
void VCMP(ARMReg Vd, ARMReg Vm); void VCMP(ARMReg Vd, ARMReg Vm);
// Compares against zero // Compares against zero
void VCMP(ARMReg Vd); void VCMP(ARMReg Vd);
@ -507,10 +508,11 @@ public:
void VSQRT(ARMReg Vd, ARMReg Vm); void VSQRT(ARMReg Vd, ARMReg Vm);
// NEON and VFP // NEON and VFP
void VABS(ARMReg Vd, ARMReg Vm);
void VADD(ARMReg Vd, ARMReg Vn, ARMReg Vm); void VADD(ARMReg Vd, ARMReg Vn, ARMReg Vm);
void VSUB(ARMReg Vd, ARMReg Vn, ARMReg Vm); void VSUB(ARMReg Vd, ARMReg Vn, ARMReg Vm);
void VABS(ARMReg Vd, ARMReg Vm);
void VNEG(ARMReg Vd, ARMReg Vm);
void VMUL(ARMReg Vd, ARMReg Vn, ARMReg Vm);
void VMOV(ARMReg Dest, ARMReg Src, bool high); void VMOV(ARMReg Dest, ARMReg Src, bool high);
void VMOV(ARMReg Dest, ARMReg Src); void VMOV(ARMReg Dest, ARMReg Src);