Lines Matching refs:compare
2178 // If we got a 512-bit vector then we'll definitely have a vXi1 compare.
2184 // compare for vXi32/vXi64 for sure. If we have BWI we will also support
5071 // a floating-point compare and we have blendv or conditional move, then it is
5073 // load that depends on the compare result.
9642 // SSE41 can compare v2i64 - select between indices 0 and 1.
20841 // Only promote the compare up to I32 if it is a 16 bit operation
21251 "Cannot set masked compare for this operation");
21388 // TODO: We could use following steps to handle a quiet compare with
21392 // 3. Get the compare results of masked A, B
21411 // In the two cases not handled by SSE compare predicates (SETUEQ/SETONE),
21497 // But there is no compare instruction for i8 and i16 elements in KNL.
21505 // Translate compare code to XOP PCOM compare mode.
21564 // Break 256-bit integer vector compare into smaller ones.
21581 // If both operands are known non-negative, then an unsigned compare is the
21582 // same as a signed compare and there's no need to flip signbits.
21678 // compare is always unsigned.
21721 // Do the compare.
23474 /// Return (and \p Op, \p Mask) for compare instructions or
25866 // Else, expand to a compare/select.
36286 // sign-extend to a 256-bit operation to match the compare.
36289 // the compare.
36301 // truncating the result of the compare to 128-bits.
37715 // of the condition can usually be folded into a compare predicate, but even
38168 // the need for an extra compare
38881 // compare. If so change it to the RHS of the compare.
39895 // register. In most cases we actually compare or select YMM-sized registers
46363 IA->getConstraintString().compare(0, 5, "=r,0,") == 0 &&
46376 IA->getConstraintString().compare(0, 5, "=r,0,") == 0 &&