mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

A64: Implement FMULX (by element), single and double precision variants

Browse source
This commit is contained in:
MerryMage 2018-11-17 21:51:39 +00:00
parent c106d8cedf
commit e3898e628e
2 changed files with 10 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/frontend/A64/decoder/a64.inc
View file

@ -899,7 +899,7 @@ INST(UMULL_elt, "UMULL, UMULL2 (by element)", "0Q101
INST(UDOT_elt, "UDOT (by element)", "0Q101111zzLMmmmm1110H0nnnnnddddd")
//INST(SQRDMLSH_elt_2, "SQRDMLSH (by element)", "0Q101111zzLMmmmm1111H0nnnnnddddd")
//INST(FMULX_elt_3, "FMULX (by element)", "0Q10111100LMmmmm1001H0nnnnnddddd")
//INST(FMULX_elt_4, "FMULX (by element)", "0Q1011111zLMmmmm1001H0nnnnnddddd")
INST(FMULX_elt_4, "FMULX (by element)", "0Q1011111zLMmmmm1001H0nnnnnddddd")
//INST(FCMLA_elt, "FCMLA (by element)", "0Q101111zzLMmmmm0rr1H0nnnnnddddd")
// Data Processing - FP and SIMD - Cryptographic three register

11
src/frontend/A64/translate/impl/simd_vector_x_indexed_element.cpp
View file

@ -20,6 +20,7 @@ std::pair<size_t, Vec> Combine(Imm<2> size, Imm<1> H, Imm<1> L, Imm<1> M, Imm<4>
enum class ExtraBehavior {
None,
Extended,
Accumulate,
Subtract,
};
@ -34,11 +35,11 @@ bool MultiplyByElement(TranslatorVisitor& v, bool Q, Imm<2> size, Imm<1> L, Imm<
const size_t idxdsize = H == 1 ? 128 : 64;
const size_t esize = 8 << size.ZeroExtend();
const size_t datasize = Q ? 128 : 64;
const IR::U128 operand1 = v.V(datasize, Vn);
const IR::U128 operand2 = v.ir.VectorBroadcast(esize, v.ir.VectorGetElement(esize, v.V(idxdsize, Vm), index));
const IR::U128 operand3 = v.V(datasize, Vd);
IR::U128 result = v.ir.VectorMultiply(esize, operand1, operand2);
if (extra_behavior == ExtraBehavior::Accumulate) {
result = v.ir.VectorAdd(esize, operand3, result);
@ -74,6 +75,8 @@ bool FPMultiplyByElement(TranslatorVisitor& v, bool Q, bool sz, Imm<1> L, Imm<1>
switch (extra_behavior) {
case ExtraBehavior::None:
return v.ir.FPVectorMul(esize, operand1, operand2);
case ExtraBehavior::Extended:
return v.ir.FPVectorMulX(esize, operand1, operand2);
case ExtraBehavior::Accumulate:
return v.ir.FPVectorMulAdd(esize, operand3, operand1, operand2);
case ExtraBehavior::Subtract:
@ -202,6 +205,10 @@ bool TranslatorVisitor::FMUL_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> V
return FPMultiplyByElement(*this, Q, sz, L, M, Vmlo, H, Vn, Vd, ExtraBehavior::None);
}
bool TranslatorVisitor::FMULX_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) {
return FPMultiplyByElement(*this, Q, sz, L, M, Vmlo, H, Vn, Vd, ExtraBehavior::Extended);
}
bool TranslatorVisitor::SMLAL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) {
return MultiplyLong(*this, Q, size, L, M, Vmlo, H, Vn, Vd, ExtraBehavior::Accumulate, Signedness::Signed);
}