The-Homebrew-Cloud/dolphin
2
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-01-25 07:21:14 +01:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Stop violating the AAPCS64 so much.

Browse Source
This commit is contained in:
Ryan Houdek 2015-03-02 11:21:15 -06:00
parent fad46729b0
commit f1a9db9bdc
3 changed files with 44 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
Source/Core/Common/Arm64Emitter.cpp
View File

@ -2874,7 +2874,7 @@ void ARM64FloatEmitter::FMUL(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, u8
EmitVectorxElement(0, 2 | (size >> 6), L, 0b1001, H, Rd, Rn, Rm); EmitVectorxElement(0, 2 | (size >> 6), L, 0b1001, H, Rd, Rn, Rm);
} }
void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers) void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers, ARM64Reg tmp)
{ {
bool bundled_loadstore = false; bool bundled_loadstore = false;
int num_regs = registers.Count(); int num_regs = registers.Count();
@ -2904,15 +2904,10 @@ void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers)
} }
} }
if (!bundled_loadstore) if (bundled_loadstore && tmp != INVALID_REG)
{ {
for (auto it : registers)
STR(128, INDEX_PRE, (ARM64Reg)(Q0 + it), SP, -16);
}
else
{
// Violating the AAPCS64 never felt so right.
m_emit->SUB(SP, SP, num_regs * 16); m_emit->SUB(SP, SP, num_regs * 16);
m_emit->ADD(tmp, SP, 0);
for (int i = 0; i < 32; ++i) for (int i = 0; i < 32; ++i)
{ {
if (!registers[i]) if (!registers[i])
@ -2927,14 +2922,18 @@ void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers)
// 4 < 4 && registers[i + 4] false! // 4 < 4 && registers[i + 4] false!
while (++count < 4 && (i + count) < 32 && registers[i + count]) {} while (++count < 4 && (i + count) < 32 && registers[i + count]) {}
ST1(64, count, INDEX_POST, (ARM64Reg)(Q0 + i), SP); ST1(64, count, INDEX_POST, (ARM64Reg)(Q0 + i), tmp);
i += count - 1; i += count - 1;
} }
m_emit->SUB(SP, SP, num_regs * 16); }
else
{
for (auto it : registers)
STR(128, INDEX_PRE, (ARM64Reg)(Q0 + it), SP, -16);
} }
} }
void ARM64FloatEmitter::ABI_PopRegisters(BitSet32 registers) void ARM64FloatEmitter::ABI_PopRegisters(BitSet32 registers, ARM64Reg tmp)
{ {
bool bundled_loadstore = false; bool bundled_loadstore = false;
int num_regs = registers.Count(); int num_regs = registers.Count();
@ -2964,18 +2963,9 @@ void ARM64FloatEmitter::ABI_PopRegisters(BitSet32 registers)
} }
} }
if (!bundled_loadstore) if (bundled_loadstore && tmp != INVALID_REG)
{
for (int i = 31; i >= 0; --i)
{
if (!registers[i])
continue;
LDR(128, INDEX_POST, (ARM64Reg)(Q0 + i), SP, 16);
}
}
else
{ {
// The temporary register is only used to indicate that we can use this code path
for (int i = 0; i < 32; ++i) for (int i = 0; i < 32; ++i)
{ {
if (!registers[i]) if (!registers[i])
@ -2989,6 +2979,16 @@ void ARM64FloatEmitter::ABI_PopRegisters(BitSet32 registers)
i += count - 1; i += count - 1;
} }
} }
else
{
for (int i = 31; i >= 0; --i)
{
if (!registers[i])
continue;
LDR(128, INDEX_POST, (ARM64Reg)(Q0 + i), SP, 16);
}
}
} }
} }

4
Source/Core/Common/Arm64Emitter.h
View File

@ -755,8 +755,8 @@ public:
void FMUL(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, u8 index); void FMUL(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, u8 index);
// ABI related // ABI related
void ABI_PushRegisters(BitSet32 registers); void ABI_PushRegisters(BitSet32 registers, ARM64Reg tmp = INVALID_REG);
void ABI_PopRegisters(BitSet32 registers); void ABI_PopRegisters(BitSet32 registers, ARM64Reg tmp = INVALID_REG);
private: private:
ARM64XEmitter* m_emit; ARM64XEmitter* m_emit;

40
Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
View File

@ -279,12 +279,12 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(64, X0, Q0, 0); float_emit.UMOV(64, X0, Q0, 0);
ORR(X0, SP, X0, ArithOption(X0, ST_ROR, 32)); ORR(X0, SP, X0, ArithOption(X0, ST_ROR, 32));
MOVI2R(X30, (u64)PowerPC::Write_U64); MOVI2R(X30, (u64)PowerPC::Write_U64);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -310,12 +310,12 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(16, W0, Q0, 0); float_emit.UMOV(16, W0, Q0, 0);
REV16(W0, W0); REV16(W0, W0);
MOVI2R(X30, (u64)PowerPC::Write_U16); MOVI2R(X30, (u64)PowerPC::Write_U16);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -341,12 +341,12 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(16, W0, Q0, 0); float_emit.UMOV(16, W0, Q0, 0);
REV16(W0, W0); REV16(W0, W0);
MOVI2R(X30, (u64)PowerPC::Write_U16); MOVI2R(X30, (u64)PowerPC::Write_U16);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -372,12 +372,12 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.REV32(8, D0, D0); float_emit.REV32(8, D0, D0);
float_emit.UMOV(32, W0, Q0, 0); float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)PowerPC::Write_U32); MOVI2R(X30, (u64)PowerPC::Write_U32);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -402,12 +402,12 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.REV32(8, D0, D0); float_emit.REV32(8, D0, D0);
float_emit.UMOV(32, W0, Q0, 0); float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)PowerPC::Write_U32); MOVI2R(X30, (u64)PowerPC::Write_U32);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -428,11 +428,11 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(32, W0, Q0, 0); float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&PowerPC::Write_U32); MOVI2R(X30, (u64)&PowerPC::Write_U32);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -457,11 +457,11 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(32, W0, Q0, 0); float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&PowerPC::Write_U8); MOVI2R(X30, (u64)&PowerPC::Write_U8);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -486,11 +486,11 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.SMOV(32, W0, Q0, 0); float_emit.SMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&PowerPC::Write_U8); MOVI2R(X30, (u64)&PowerPC::Write_U8);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -515,11 +515,11 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.UMOV(32, W0, Q0, 0); float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&PowerPC::Write_U16); MOVI2R(X30, (u64)&PowerPC::Write_U16);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }
@ -544,11 +544,11 @@ void JitArm64AsmRoutineManager::GenerateCommon()
SetJumpTarget(argh); SetJumpTarget(argh);
ABI_PushRegisters(gprs); ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs); float_emit.ABI_PushRegisters(fprs, X3);
float_emit.SMOV(32, W0, Q0, 0); float_emit.SMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&PowerPC::Write_U16); MOVI2R(X30, (u64)&PowerPC::Write_U16);
BLR(X30); BLR(X30);
float_emit.ABI_PopRegisters(fprs); float_emit.ABI_PopRegisters(fprs, X3);
ABI_PopRegisters(gprs); ABI_PopRegisters(gprs);
RET(X30); RET(X30);
} }