A64: Implement SQRDMULH (vector), vector variant
This commit is contained in:
parent
a9ffcf08b1
commit
12243692f5
3 changed files with 19 additions and 2 deletions
|
@ -3194,7 +3194,7 @@ void EmitX64::EmitVectorSignedSaturatedDoublingMultiply32(EmitContext& ctx, IR::
|
|||
const Xbyak::Xmm lower_result = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code.vpsllq(lower_result, even, 32);
|
||||
code.vblendps(lower_result, lower_result, even, 0b0101);
|
||||
code.vblendps(lower_result, lower_result, odds, 0b0101);
|
||||
|
||||
ctx.reg_alloc.DefineValue(lower_inst, lower_result);
|
||||
ctx.EraseInstruction(lower_inst);
|
||||
|
|
|
@ -763,7 +763,7 @@ INST(MLS_vec, "MLS (vector)", "0Q101
|
|||
INST(PMUL, "PMUL", "0Q101110zz1mmmmm100111nnnnnddddd")
|
||||
INST(UMAXP, "UMAXP", "0Q101110zz1mmmmm101001nnnnnddddd")
|
||||
INST(UMINP, "UMINP", "0Q101110zz1mmmmm101011nnnnnddddd")
|
||||
//INST(SQRDMULH_vec_2, "SQRDMULH (vector)", "0Q101110zz1mmmmm101101nnnnnddddd")
|
||||
INST(SQRDMULH_vec_2, "SQRDMULH (vector)", "0Q101110zz1mmmmm101101nnnnnddddd")
|
||||
INST(FMAXNMP_vec_2, "FMAXNMP (vector)", "0Q1011100z1mmmmm110001nnnnnddddd")
|
||||
//INST(FMLAL_vec_2, "FMLAL, FMLAL2 (vector)", "0Q1011100z1mmmmm110011nnnnnddddd")
|
||||
INST(FADDP_vec_2, "FADDP (vector)", "0Q1011100z1mmmmm110101nnnnnddddd")
|
||||
|
|
|
@ -438,6 +438,23 @@ bool TranslatorVisitor::SQDMULH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec V
|
|||
return true;
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::SQRDMULH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) {
|
||||
if (size == 0b00 || size == 0b11) {
|
||||
return ReservedValue();
|
||||
}
|
||||
|
||||
const size_t esize = 8 << size.ZeroExtend();
|
||||
const size_t datasize = Q ? 128 : 64;
|
||||
|
||||
const IR::U128 operand1 = V(datasize, Vn);
|
||||
const IR::U128 operand2 = V(datasize, Vm);
|
||||
const IR::UpperAndLower multiply = ir.VectorSignedSaturatedDoublingMultiply(esize, operand1, operand2);
|
||||
const IR::U128 result = ir.VectorAdd(esize, multiply.upper, ir.VectorLogicalShiftRight(esize, multiply.lower, static_cast<u8>(esize - 1)));
|
||||
|
||||
V(datasize, Vd, result);
|
||||
return true;
|
||||
}
|
||||
|
||||
bool TranslatorVisitor::ADD_vector(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) {
|
||||
if (size == 0b11 && !Q) return ReservedValue();
|
||||
const size_t esize = 8 << size.ZeroExtend<size_t>();
|
||||
|
|
Loading…
Reference in a new issue