================
@@ -2745,6 +2745,72 @@ static bool interp__builtin_ia32_addsub(InterpState &S,
CodePtr OpPC,
return true;
}
+static bool interp__builtin_ia32_pclmulqdq(InterpState &S, CodePtr OpPC,
+ const CallExpr *Call) {
+ // PCLMULQDQ: carry-less multiplication of selected 64-bit halves
+ // imm8 bit 0: selects lower (0) or upper (1) 64 bits of first operand
+ // imm8 bit 4: selects lower (0) or upper (1) 64 bits of second operand
+ assert(Call->getArg(0)->getType()->isVectorType() &&
+ Call->getArg(1)->getType()->isVectorType());
+
+ // Extract imm8 argument
+ APSInt Imm8 = popToAPSInt(S, Call->getArg(2));
+ bool SelectUpperA = (Imm8 & 0x01) != 0;
+ bool SelectUpperB = (Imm8 & 0x10) != 0;
+
+ const Pointer &RHS = S.Stk.pop<Pointer>();
+ const Pointer &LHS = S.Stk.pop<Pointer>();
+ const Pointer &Dst = S.Stk.peek<Pointer>();
+
+ const auto *VT = Call->getArg(0)->getType()->castAs<VectorType>();
+ PrimType ElemT = *S.getContext().classify(VT->getElementType());
+ unsigned NumElems = VT->getNumElements();
+ const auto *DestVT = Call->getType()->castAs<VectorType>();
+ PrimType DestElemT = *S.getContext().classify(DestVT->getElementType());
+ bool DestUnsigned = Call->getType()->isUnsignedIntegerOrEnumerationType();
+
+ // Process each 128-bit lane (2 elements at a time)
+ for (unsigned Lane = 0; Lane < NumElems; Lane += 2) {
+ APSInt A0, A1, B0, B1;
+ INT_TYPE_SWITCH_NO_BOOL(ElemT, {
+ A0 = LHS.elem<T>(Lane + 0).toAPSInt();
+ A1 = LHS.elem<T>(Lane + 1).toAPSInt();
+ B0 = RHS.elem<T>(Lane + 0).toAPSInt();
+ B1 = RHS.elem<T>(Lane + 1).toAPSInt();
+ });
+
+ // Select the appropriate 64-bit values based on imm8
+ APInt A = SelectUpperA ? A1 : A0;
+ APInt B = SelectUpperB ? B1 : B0;
+
+ // Perform carry-less multiplication (polynomial multiplication in
GF(2^64))
+ // This multiplies two 64-bit values to produce a 128-bit result
+ APInt AVal = A.zextOrTrunc(64);
+ APInt BVal = B.zextOrTrunc(64);
+ APInt Result(128, 0);
+
+ // For each bit in A, if set, XOR B shifted left by that bit position
+ for (unsigned i = 0; i < 64; ++i) {
+ if (AVal[i]) {
+ APInt ShiftedB = BVal.zext(128) << i;
+ Result ^= ShiftedB;
+ }
+ }
----------------
ahmednoursphinx wrote:
Good idea, updated it and extended to use equal bit width
https://github.com/llvm/llvm-project/pull/169214
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
