Lines Matching full:xmm3
1604 * xmm3 is tmp (operate on high bits of 16bit lanes)
1606 * xmm3 = xmm1
1609 * xmm3 = xmm3 .>> 8
1611 * xmm2 = xmm2 .* xmm3 // xmm2 now has high bits
1645 * - Get xmm2 and xmm3 as temp
1653 * movdqa %xmm3, %xmm0
1654 * psrlq %xmm3, $0x20
1655 * pmuludq %xmm3, %xmm2
1658 * paddq %xmm1, %xmm3
1664 * calculation only once. Thus we don't need the xmm3 temp above. That sequence becomes:
