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

VECTOR_COMPARE_UGT/UGE (probably).

This commit is contained in:
Ben Vanik 2014-06-09 21:29:35 -07:00
parent 289075d052
commit ff8c03046a
3 changed files with 116 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/alloy/backend/x64/x64_emitter.cc
View File

@ -505,6 +505,10 @@ Address X64Emitter::GetXmmConstPtr(XmmConst id) {
/* XMMShiftByteMask */ vec128i(0x000000FFu, 0x000000FFu, 0x000000FFu, 0x000000FFu), /* XMMShiftByteMask */ vec128i(0x000000FFu, 0x000000FFu, 0x000000FFu, 0x000000FFu),
/* XMMUnsignedDwordMax */ vec128i(0xFFFFFFFFu, 0x00000000u, 0xFFFFFFFFu, 0x00000000u), /* XMMUnsignedDwordMax */ vec128i(0xFFFFFFFFu, 0x00000000u, 0xFFFFFFFFu, 0x00000000u),
/* XMM255 */ vec128f(255.0f, 255.0f, 255.0f, 255.0f), /* XMM255 */ vec128f(255.0f, 255.0f, 255.0f, 255.0f),
/* XMMSignMaskI8 */ vec128i(0x80808080u, 0x80808080u, 0x80808080u, 0x80808080u),
/* XMMSignMaskI16 */ vec128i(0x80008000u, 0x80008000u, 0x80008000u, 0x80008000u),
/* XMMSignMaskI32 */ vec128i(0x80000000u, 0x80000000u, 0x80000000u, 0x80000000u),
/* XMMSignMaskF32 */ vec128i(0x80000000u, 0x80000000u, 0x80000000u, 0x80000000u),
}; };
// TODO(benvanik): cache base pointer somewhere? stack? It'd be nice to // TODO(benvanik): cache base pointer somewhere? stack? It'd be nice to
// prevent this move. // prevent this move.

4
src/alloy/backend/x64/x64_emitter.h
View File

@ -57,6 +57,10 @@ enum XmmConst {
XMMShiftByteMask, XMMShiftByteMask,
XMMUnsignedDwordMax, XMMUnsignedDwordMax,
XMM255, XMM255,
XMMSignMaskI8,
XMMSignMaskI16,
XMMSignMaskI32,
XMMSignMaskF32,
}; };
// Unfortunately due to the design of xbyak we have to pass this to the ctor. // Unfortunately due to the design of xbyak we have to pass this to the ctor.

124
src/alloy/backend/x64/x64_sequences.cc
View File

@ -2337,25 +2337,117 @@ EMITTER_OPCODE_TABLE(
// ============================================================================ // ============================================================================
// OPCODE_VECTOR_COMPARE_UGT // OPCODE_VECTOR_COMPARE_UGT
// ============================================================================ // ============================================================================
//EMITTER(VECTOR_COMPARE_UGT_V128, MATCH(I<OPCODE_VECTOR_COMPARE_UGT, V128<>, V128<>, V128<>>)) { EMITTER(VECTOR_COMPARE_UGT_V128, MATCH(I<OPCODE_VECTOR_COMPARE_UGT, V128<>, V128<>, V128<>>)) {
// static void Emit(X64Emitter& e, const EmitArgType& i) { static void Emit(X64Emitter& e, const EmitArgType& i) {
// } Xbyak::Address sign_addr = e.ptr[e.rax]; // dummy
//}; switch (i.instr->flags) {
//EMITTER_OPCODE_TABLE( case INT8_TYPE:
// OPCODE_VECTOR_COMPARE_UGT, sign_addr = e.GetXmmConstPtr(XMMSignMaskI8);
// VECTOR_COMPARE_UGT_V128); break;
case INT16_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskI16);
break;
case INT32_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskI32);
break;
case FLOAT32_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskF32);
break;
}
if (i.src1.is_constant) {
// TODO(benvanik): make this constant.
e.LoadConstantXmm(e.xmm0, i.src1.constant());
e.vpxor(e.xmm0, sign_addr);
} else {
e.vpxor(e.xmm0, i.src1, sign_addr);
}
if (i.src2.is_constant) {
// TODO(benvanik): make this constant.
e.LoadConstantXmm(e.xmm1, i.src1.constant());
e.vpxor(e.xmm1, sign_addr);
} else {
e.vpxor(e.xmm1, i.src2, sign_addr);
}
switch (i.instr->flags) {
case INT8_TYPE:
e.vpcmpgtb(i.dest, e.xmm0, e.xmm1);
break;
case INT16_TYPE:
e.vpcmpgtw(i.dest, e.xmm0, e.xmm1);
break;
case INT32_TYPE:
e.vpcmpgtd(i.dest, e.xmm0, e.xmm1);
break;
case FLOAT32_TYPE:
e.vcmpgtps(i.dest, e.xmm0, e.xmm1);
break;
}
}
};
EMITTER_OPCODE_TABLE(
OPCODE_VECTOR_COMPARE_UGT,
VECTOR_COMPARE_UGT_V128);
// ============================================================================ // ============================================================================
// OPCODE_VECTOR_COMPARE_UGE // OPCODE_VECTOR_COMPARE_UGE
// ============================================================================ // ============================================================================
//EMITTER(VECTOR_COMPARE_UGE_V128, MATCH(I<OPCODE_VECTOR_COMPARE_UGE, V128<>, V128<>, V128<>>)) { EMITTER(VECTOR_COMPARE_UGE_V128, MATCH(I<OPCODE_VECTOR_COMPARE_UGE, V128<>, V128<>, V128<>>)) {
// static void Emit(X64Emitter& e, const EmitArgType& i) { static void Emit(X64Emitter& e, const EmitArgType& i) {
// } Xbyak::Address sign_addr = e.ptr[e.rax]; // dummy
//}; switch (i.instr->flags) {
//EMITTER_OPCODE_TABLE( case INT8_TYPE:
// OPCODE_VECTOR_COMPARE_UGE, sign_addr = e.GetXmmConstPtr(XMMSignMaskI8);
// VECTOR_COMPARE_UGE_V128); break;
case INT16_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskI16);
break;
case INT32_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskI32);
break;
case FLOAT32_TYPE:
sign_addr = e.GetXmmConstPtr(XMMSignMaskF32);
break;
}
if (i.src1.is_constant) {
// TODO(benvanik): make this constant.
e.LoadConstantXmm(e.xmm0, i.src1.constant());
e.vpxor(e.xmm0, sign_addr);
} else {
e.vpxor(e.xmm0, i.src1, sign_addr);
}
if (i.src2.is_constant) {
// TODO(benvanik): make this constant.
e.LoadConstantXmm(e.xmm1, i.src1.constant());
e.vpxor(e.xmm1, sign_addr);
} else {
e.vpxor(e.xmm1, i.src2, sign_addr);
}
switch (i.instr->flags) {
case INT8_TYPE:
e.vpcmpeqb(e.xmm2, e.xmm0, e.xmm1);
e.vpcmpgtb(i.dest, e.xmm0, e.xmm1);
e.vpor(i.dest, e.xmm2);
break;
case INT16_TYPE:
e.vpcmpeqw(e.xmm2, e.xmm0, e.xmm1);
e.vpcmpgtw(i.dest, e.xmm0, e.xmm1);
e.vpor(i.dest, e.xmm2);
break;
case INT32_TYPE:
e.vpcmpeqd(e.xmm2, e.xmm0, e.xmm1);
e.vpcmpgtd(i.dest, e.xmm0, e.xmm1);
e.vpor(i.dest, e.xmm2);
break;
case FLOAT32_TYPE:
e.vcmpgeps(i.dest, e.xmm0, e.xmm1);
break;
}
}
};
EMITTER_OPCODE_TABLE(
OPCODE_VECTOR_COMPARE_UGE,
VECTOR_COMPARE_UGE_V128);
// ============================================================================ // ============================================================================
@ -4968,8 +5060,8 @@ void alloy::backend::x64::RegisterSequences() {
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_EQ); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_EQ);
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_SGT); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_SGT);
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_SGE); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_SGE);
//REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_UGT); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_UGT);
//REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_UGE); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_COMPARE_UGE);
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_ADD); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_ADD);
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_ADD_CARRY); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_ADD_CARRY);
REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_ADD); REGISTER_EMITTER_OPCODE_TABLE(OPCODE_VECTOR_ADD);