steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-06 13:00:15 +02:00
Code Issues Packages Projects Releases 66 Wiki Activity

Vector: Trim one instruction off insertq

Browse Source 
We can fold a bitwise not and and pair into a bic
This commit is contained in:
LC committed 2026-07-09 14:58:04 -04:00
1 parent e508b6df0d
commit 9b7c9f0fb6
2 files changed
+10 -19
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

FEXCore/Source/Interface/Core/OpcodeDispatcher/Vector.cpp
+2 -7
View File
@@ -5686,10 +5686,7 @@ void OpDispatchBuilder::Insertq_imm(OpcodeArgs) {
MaskVector = _VShlI(OpSize::i128Bit, OpSize::i64Bit, MaskVector, Shift);
}
// Negate the mask.
MaskVector = _VNot(OpSize::i64Bit, MaskVector);
Dest = _VAnd(OpSize::i64Bit, Dest, MaskVector);
Dest = _VAndn(OpSize::i64Bit, Dest, MaskVector);
const Ref Result = _VOr(OpSize::i64Bit, Dest, Src);
StoreResult_WithAVXInsert(VectorOpType::SSE, RegClass::FPR, Op, Result);
@@ -5745,9 +5742,7 @@ void OpDispatchBuilder::Insertq(OpcodeArgs) {
SrcData = _VUShl(OpSize::i128Bit, OpSize::i64Bit, SrcData, ShiftBits, false);
// Generate a destination mask
const Ref DstMask = _VNot(OpSize::i64Bit, _VUShl(OpSize::i128Bit, OpSize::i64Bit, SrcMask, ShiftBits, false));
Ref Result = _VAnd(OpSize::i64Bit, Dest, DstMask);
Ref Result = _VAndn(OpSize::i64Bit, Dest, _VUShl(OpSize::i128Bit, OpSize::i64Bit, SrcMask, ShiftBits, false));
Result = _VOr(OpSize::i64Bit, Result, SrcData);
StoreResult_WithAVXInsert(VectorOpType::SSE, RegClass::FPR, Op, Result);
}
unittests/InstructionCountCI/Secondary_REPNE.json
+8 -12
View File
@@ -320,7 +320,7 @@
]
},
"insertq xmm0, xmm1, 0, 0": {
"ExpectedInstructionCount": 6,
"ExpectedInstructionCount": 5,
"Comment": [
"SSE4a",
"0xf2 0x0f 0x78"
@@ -329,13 +329,12 @@
"mov x20, #0xffffffffffffffff",
"fmov d2, x20",
"and v3.8b, v17.8b, v2.8b",
"mvn v2.16b, v2.16b",
"and v2.8b, v16.8b, v2.8b",
"bic v2.8b, v16.8b, v2.8b",
"orr v16.8b, v2.8b, v3.8b"
]
},
"insertq xmm0, xmm1, 64, 0": {
"ExpectedInstructionCount": 6,
"ExpectedInstructionCount": 5,
"Comment": [
"SSE4a",
"0xf2 0x0f 0x78"
@@ -344,13 +343,12 @@
"mov x20, #0xffffffffffffffff",
"fmov d2, x20",
"and v3.8b, v17.8b, v2.8b",
"mvn v2.16b, v2.16b",
"and v2.8b, v16.8b, v2.8b",
"bic v2.8b, v16.8b, v2.8b",
"orr v16.8b, v2.8b, v3.8b"
]
},
"insertq xmm0, xmm1, 32, 32": {
"ExpectedInstructionCount": 8,
"ExpectedInstructionCount": 7,
"Comment": [
"SSE4a",
"0xf2 0x0f 0x78"
@@ -361,13 +359,12 @@
"and v3.8b, v17.8b, v2.8b",
"shl v3.2d, v3.2d, #32",
"shl v2.2d, v2.2d, #32",
"mvn v2.16b, v2.16b",
"and v2.8b, v16.8b, v2.8b",
"bic v2.8b, v16.8b, v2.8b",
"orr v16.8b, v2.8b, v3.8b"
]
},
"insertq xmm0, xmm1": {
"ExpectedInstructionCount": 20,
"ExpectedInstructionCount": 19,
"Comment": [
"SSE4a",
"0xf2 0x0f 0x79"
@@ -389,8 +386,7 @@
"and v4.16b, v17.16b, v3.16b",
"ushl v4.2d, v4.2d, v2.2d",
"ushl v3.2d, v3.2d, v2.2d",
"mvn v2.16b, v3.16b",
"and v2.8b, v16.8b, v2.8b",
"bic v2.8b, v16.8b, v3.8b",
"orr v16.8b, v2.8b, v4.8b",
"msr nzcv, x21"
]