luau-lang/luau
1
0
Fork 0
mirror of https://github.com/luau-lang/luau.git synced 2025-04-05 19:30:54 +01:00
Code Issues Projects Releases Packages Wiki Activity

CodeGen: Implement a naive version of A64 DOT_VEC

Browse source 
This is using existing instructions and scalar adds to have a baseline.
This is still faster than the original implementation of vector. ops.
This commit is contained in:
Arseny Kapoulkine 2024-11-08 10:40:12 -08:00
parent cd73807c09
commit 8fc458edbd
1 changed files with 13 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
CodeGen/src/IrLoweringA64.cpp
View file

@ -730,7 +730,19 @@ void IrLoweringA64::lowerInst(IrInst& inst, uint32_t index, const IrBlock& next)
} }
case IrCmd::DOT_VEC: case IrCmd::DOT_VEC:
{ {
CODEGEN_ASSERT(!"DOT_VEC is not implemented for A64"); inst.regA64 = regs.allocReg(KindA64::d, index);
RegisterA64 temp1 = regs.allocTemp(KindA64::q);
RegisterA64 temp2 = regs.allocTemp(KindA64::q);
RegisterA64 temp3 = regs.allocTemp(KindA64::q);
build.fmul(temp1, regOp(inst.a), regOp(inst.b));
build.dup_4s(temp2, temp1, 1);
build.dup_4s(temp3, temp1, 2);
build.fadd(castReg(KindA64::s, temp1), castReg(KindA64::s, temp1), castReg(KindA64::s, temp2));
build.fadd(castReg(KindA64::s, temp1), castReg(KindA64::s, temp1), castReg(KindA64::s, temp3));
build.fcvt(inst.regA64, castReg(KindA64::s, temp1));
break; break;
} }
case IrCmd::NOT_ANY: case IrCmd::NOT_ANY: