steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-06 14:00:16 +02:00
Code Issues Packages Projects Releases 66 Wiki Activity

Merge pull request #3172 from CallumDev/x87f64-opts

Browse Source 
X87F64: Use Bfe for rounding mode, FCHS use float instruction
This commit is contained in:
Ryan Houdek authored and GitHub committed 2023-09-30 18:41:29 -07:00
2 parents ab4642af38 d4a623a3fb
commit ccfd770d9d
2 files changed
+9 -23
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp
+3 -13
View File
@@ -72,11 +72,7 @@ void OpDispatchBuilder::X87LDENVF64(OpcodeArgs) {
auto NewFCW = _LoadMem(GPRClass, 2, Mem, 2);
//ignore the rounding precision, we're always 64-bit in F64.
//extract rounding mode
OrderedNode *roundingMode = NewFCW;
auto roundShift = _Constant(10);
auto roundMask = _Constant(3);
roundingMode = _Lshr(OpSize::i32Bit, roundingMode, roundShift);
roundingMode = _And(OpSize::i32Bit, roundingMode, roundMask);
OrderedNode *roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
_SetRoundingMode(roundingMode);
_StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
@@ -96,11 +92,7 @@ void OpDispatchBuilder::X87FLDCWF64(OpcodeArgs) {
OrderedNode *NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, -1);
//ignore the rounding precision, we're always 64-bit in F64.
//extract rounding mode
OrderedNode *roundingMode = NewFCW;
auto shift = _Constant(10);
auto mask = _Constant(3);
roundingMode = _Lshr(OpSize::i32Bit, roundingMode, shift);
roundingMode = _And(OpSize::i32Bit, roundingMode, mask);
OrderedNode *roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
_SetRoundingMode(roundingMode);
_StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
}
@@ -589,9 +581,7 @@ void OpDispatchBuilder::FSUBF64<32, true, true, OpDispatchBuilder::OpResult::RES
void OpDispatchBuilder::FCHSF64(OpcodeArgs) {
auto top = GetX87Top();
auto a = _LoadContextIndexed(top, 8, MMBaseOffset(), 16, FPRClass);
auto b = _VCastFromGPR(8, 8, _Constant(0x8000000000000000));
auto result = _VXor(8, 8, a, b);
auto result = _VFNeg(8, 8, a);
// Write to ST[TOP]
_StoreContextIndexed(result, top, 8, MMBaseOffset(), 16, FPRClass);
}
unittests/InstructionCountCI/x87_f64.json
+6 -10
View File
@@ -1746,15 +1746,14 @@
]
},
"fldenv [rax]": {
"ExpectedInstructionCount": 55,
"ExpectedInstructionCount": 54,
"Optimal": "No",
"Comment": [
"0xd9 !11b /4"
],
"ExpectedArm64ASM": [
"ldrh w20, [x4]",
"lsr w21, w20, #10",
"and w21, w21, #0x3",
"ubfx w21, w20, #10, #3",
"rbit w1, w21",
"lsr w1, w1, #30",
"mrs x0, fpcr",
@@ -1810,15 +1809,14 @@
]
},
"fldcw [rax]": {
"ExpectedInstructionCount": 11,
"ExpectedInstructionCount": 10,
"Optimal": "No",
"Comment": [
"0xd9 !11b /5"
],
"ExpectedArm64ASM": [
"ldrh w20, [x4]",
"lsr w21, w20, #10",
"and w21, w21, #0x3",
"ubfx w21, w20, #10, #3",
"rbit w1, w21",
"lsr w1, w1, #30",
"mrs x0, fpcr",
@@ -2272,7 +2270,7 @@
"ExpectedArm64ASM": []
},
"fchs": {
"ExpectedInstructionCount": 8,
"ExpectedInstructionCount": 6,
"Optimal": "No",
"Comment": [
"0xd9 11b 0xe0 /4"
@@ -2281,9 +2279,7 @@
"ldrb w20, [x28, #747]",
"add x0, x28, x20, lsl #4",
"ldr d2, [x0, #752]",
"mov x21, #0x8000000000000000",
"fmov d3, x21",
"eor v2.16b, v2.16b, v3.16b",
"fneg v2.2d, v2.2d",
"add x0, x28, x20, lsl #4",
"str d2, [x0, #752]"
]