zdoom/gzdoom-gles
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom-gles.git synced 2025-02-01 21:00:44 +00:00
Code Issues Projects Releases Packages Wiki Activity

- update JIT PARAM handling to match the VM instruction change

Browse source
This commit is contained in:
Magnus Norddahl 2018-10-14 23:13:30 +02:00
parent 70f18f87b9
commit 1f0add9067
2 changed files with 35 additions and 38 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

71
src/scripting/vm/jit_call.cpp
View file

@ -11,90 +11,90 @@ void JitCompiler::EmitPARAM()
X86Gp stackPtr, tmp; X86Gp stackPtr, tmp;
X86Xmm tmp2; X86Xmm tmp2;
switch (B) switch (A)
{ {
case REGT_NIL: case REGT_NIL:
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), (int64_t)0); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), (int64_t)0);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_NIL); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_NIL);
break; break;
case REGT_INT: case REGT_INT:
cc.mov(x86::dword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, i)), regD[C]); cc.mov(x86::dword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, i)), regD[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_INT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_INT);
break; break;
case REGT_INT | REGT_ADDROF: case REGT_INT | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetD + (int)(C * sizeof(int32_t)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetD + (int)(BC * sizeof(int32_t))));
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_INT | REGT_KONST: case REGT_INT | REGT_KONST:
cc.mov(x86::dword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, i)), konstd[C]); cc.mov(x86::dword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, i)), konstd[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_INT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_INT);
break; break;
case REGT_STRING: case REGT_STRING:
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, sp)), regS[C]); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, sp)), regS[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_STRING); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_STRING);
break; break;
case REGT_STRING | REGT_ADDROF: case REGT_STRING | REGT_ADDROF:
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), regS[C]); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), regS[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_STRING | REGT_KONST: case REGT_STRING | REGT_KONST:
tmp = newTempIntPtr(); tmp = newTempIntPtr();
cc.mov(tmp, asmjit::imm_ptr(&konsts[C])); cc.mov(tmp, asmjit::imm_ptr(&konsts[BC]));
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, sp)), tmp); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, sp)), tmp);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_STRING); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_STRING);
break; break;
case REGT_POINTER: case REGT_POINTER:
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), regA[C]); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), regA[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_POINTER | REGT_ADDROF: case REGT_POINTER | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetA + (int)(C * sizeof(void*)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetA + (int)(BC * sizeof(void*))));
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_POINTER | REGT_KONST: case REGT_POINTER | REGT_KONST:
tmp = newTempIntPtr(); tmp = newTempIntPtr();
cc.mov(tmp, asmjit::imm_ptr(konsta[C].v)); cc.mov(tmp, asmjit::imm_ptr(konsta[BC].v));
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), tmp); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), tmp);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_FLOAT: case REGT_FLOAT:
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
break; break;
case REGT_FLOAT | REGT_MULTIREG2: case REGT_FLOAT | REGT_MULTIREG2:
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
index = NumParam++; index = NumParam++;
ParamOpcodes.Push(pc); ParamOpcodes.Push(pc);
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C + 1]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC + 1]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
break; break;
case REGT_FLOAT | REGT_MULTIREG3: case REGT_FLOAT | REGT_MULTIREG3:
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
index = NumParam++; index = NumParam++;
ParamOpcodes.Push(pc); ParamOpcodes.Push(pc);
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C + 1]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC + 1]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
index = NumParam++; index = NumParam++;
ParamOpcodes.Push(pc); ParamOpcodes.Push(pc);
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[C + 2]); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), regF[BC + 2]);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
break; break;
case REGT_FLOAT | REGT_ADDROF: case REGT_FLOAT | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetF + (int)(C * sizeof(double)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetF + (int)(BC * sizeof(double))));
cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr); cc.mov(x86::ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, a)), stackPtr);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_POINTER);
break; break;
case REGT_FLOAT | REGT_KONST: case REGT_FLOAT | REGT_KONST:
tmp = newTempIntPtr(); tmp = newTempIntPtr();
tmp2 = newTempXmmSd(); tmp2 = newTempXmmSd();
cc.mov(tmp, asmjit::imm_ptr(konstf + C)); cc.mov(tmp, asmjit::imm_ptr(konstf + BC));
cc.movsd(tmp2, asmjit::x86::qword_ptr(tmp)); cc.movsd(tmp2, asmjit::x86::qword_ptr(tmp));
cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), tmp2); cc.movsd(x86::qword_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, f)), tmp2);
cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT); cc.mov(x86::byte_ptr(vmframe, offsetParams + index * sizeof(VMValue) + offsetof(VMValue, Type)), (int)REGT_FLOAT);
@ -236,37 +236,37 @@ void JitCompiler::StoreInOuts(int b)
for (unsigned int i = ParamOpcodes.Size() - b; i < ParamOpcodes.Size(); i++) for (unsigned int i = ParamOpcodes.Size() - b; i < ParamOpcodes.Size(); i++)
{ {
asmjit::X86Gp stackPtr; asmjit::X86Gp stackPtr;
auto c = ParamOpcodes[i]->c; auto bc = ParamOpcodes[i]->i16u;
switch (ParamOpcodes[i]->b) switch (ParamOpcodes[i]->a)
{ {
case REGT_INT | REGT_ADDROF: case REGT_INT | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetD + (int)(c * sizeof(int32_t)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetD + (int)(bc * sizeof(int32_t))));
cc.mov(x86::dword_ptr(stackPtr), regD[c]); cc.mov(x86::dword_ptr(stackPtr), regD[bc]);
break; break;
case REGT_STRING | REGT_ADDROF: case REGT_STRING | REGT_ADDROF:
// We don't have to do anything in this case. String values are never moved to virtual registers. // We don't have to do anything in this case. String values are never moved to virtual registers.
break; break;
case REGT_POINTER | REGT_ADDROF: case REGT_POINTER | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetA + (int)(c * sizeof(void*)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetA + (int)(bc * sizeof(void*))));
cc.mov(x86::ptr(stackPtr), regA[c]); cc.mov(x86::ptr(stackPtr), regA[bc]);
break; break;
case REGT_FLOAT | REGT_ADDROF: case REGT_FLOAT | REGT_ADDROF:
stackPtr = newTempIntPtr(); stackPtr = newTempIntPtr();
cc.lea(stackPtr, x86::ptr(vmframe, offsetF + (int)(c * sizeof(double)))); cc.lea(stackPtr, x86::ptr(vmframe, offsetF + (int)(bc * sizeof(double))));
cc.movsd(x86::qword_ptr(stackPtr), regF[c]); cc.movsd(x86::qword_ptr(stackPtr), regF[bc]);
// When passing the address to a float we don't know if the receiving function will treat it as float, vec2 or vec3. // When passing the address to a float we don't know if the receiving function will treat it as float, vec2 or vec3.
if ((unsigned int)c + 1 < regF.Size()) if ((unsigned int)bc + 1 < regF.Size())
{ {
cc.add(stackPtr, (int)sizeof(double)); cc.add(stackPtr, (int)sizeof(double));
cc.movsd(x86::qword_ptr(stackPtr), regF[c + 1]); cc.movsd(x86::qword_ptr(stackPtr), regF[bc + 1]);
} }
if ((unsigned int)c + 2 < regF.Size()) if ((unsigned int)bc + 2 < regF.Size())
{ {
cc.add(stackPtr, (int)sizeof(double)); cc.add(stackPtr, (int)sizeof(double));
cc.movsd(x86::qword_ptr(stackPtr), regF[c + 2]); cc.movsd(x86::qword_ptr(stackPtr), regF[bc + 2]);
} }
break; break;
default: default:
@ -280,9 +280,9 @@ void JitCompiler::LoadInOuts(int b)
for (unsigned int i = ParamOpcodes.Size() - b; i < ParamOpcodes.Size(); i++) for (unsigned int i = ParamOpcodes.Size() - b; i < ParamOpcodes.Size(); i++)
{ {
const VMOP &param = *ParamOpcodes[i]; const VMOP &param = *ParamOpcodes[i];
if (param.op == OP_PARAM && (param.b & REGT_ADDROF)) if (param.op == OP_PARAM && (param.a & REGT_ADDROF))
{ {
LoadCallResult(param, true); LoadCallResult(param.a, param.i16u, true);
} }
} }
} }
@ -294,15 +294,12 @@ void JitCompiler::LoadReturns(const VMOP *retval, int numret)
if (retval[i].op != OP_RESULT) if (retval[i].op != OP_RESULT)
I_FatalError("Expected OP_RESULT to follow OP_CALL\n"); I_FatalError("Expected OP_RESULT to follow OP_CALL\n");
LoadCallResult(retval[i], false); LoadCallResult(retval[i].b, retval[i].c, false);
} }
} }
void JitCompiler::LoadCallResult(const VMOP &opdata, bool addrof) void JitCompiler::LoadCallResult(int type, int regnum, bool addrof)
{ {
int type = opdata.b;
int regnum = opdata.c;
switch (type & REGT_TYPE) switch (type & REGT_TYPE)
{ {
case REGT_INT: case REGT_INT:

2
src/scripting/vm/jitintern.h
View file

@ -53,7 +53,7 @@ private:
void LoadInOuts(int b); void LoadInOuts(int b);
void LoadReturns(const VMOP *retval, int numret); void LoadReturns(const VMOP *retval, int numret);
void FillReturns(const VMOP *retval, int numret); void FillReturns(const VMOP *retval, int numret);
void LoadCallResult(const VMOP &opdata, bool addrof); void LoadCallResult(int type, int regnum, bool addrof);
template <typename Func> template <typename Func>
void EmitComparisonOpcode(Func jmpFunc) void EmitComparisonOpcode(Func jmpFunc)