================
@@ -3019,6 +3019,54 @@ static bool interp__builtin_ia32_dbpsadbw(InterpState
&S, CodePtr OpPC,
return true;
}
+static bool interp__builtin_ia32_mpsadbw(InterpState &S, CodePtr OpPC,
+ const CallExpr *Call) {
+ assert(Call->getNumArgs() == 3);
+ uint64_t Imm;
+ if (!popToUInt64(S, Call->getArg(2), Imm))
+ return false;
+
+ const Pointer &Src2 = S.Stk.pop<Pointer>();
+ const Pointer &Src1 = S.Stk.pop<Pointer>();
+ const Pointer &Dst = S.Stk.peek<Pointer>();
+
+ const auto *SrcVT = Call->getArg(0)->getType()->castAs<VectorType>();
+ PrimType SrcElemT = *S.getContext().classify(SrcVT->getElementType());
+ unsigned SourceLen = SrcVT->getNumElements(); // 16 or 32
+
+ const auto *DestVT = Call->getType()->castAs<VectorType>();
+ PrimType DestElemT = *S.getContext().classify(DestVT->getElementType());
+ bool DestUnsigned = Call->getType()->isUnsignedIntegerOrEnumerationType();
+
+ constexpr unsigned LaneSize = 16; // 128-bit lane = 16 bytes
+ unsigned NumLanes = SourceLen / LaneSize;
+
+ for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
+ unsigned Ctrl = (Imm >> (3 * Lane)) & 0x7;
+ unsigned AOff = ((Ctrl >> 2) & 1) * 4;
+ unsigned BOff = (Ctrl & 3) * 4;
+ for (unsigned J = 0; J < 8; ++J) {
+ uint16_t Sad = 0;
+ for (unsigned K = 0; K < 4; ++K) {
----------------
tbaederr wrote:
```suggestion
for (unsigned K = 0; K != 4; ++K) {
```
https://github.com/llvm/llvm-project/pull/202257
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
