1; RUN: llc < %s -mcpu=nehalem | FileCheck %s 2target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" 3target triple = "x86_64-apple-macosx10.7" 4 5; CHECK: f 6; 7; This function contains load / store / and operations that all can execute in 8; any domain. The only domain-specific operation is the %add = shl... operation 9; which is <4 x i32>. 10; 11; The paddd instruction can only influence the other operations through the loop 12; back-edge. Check that everything is still moved into the integer domain. 13 14define void @f(<4 x i32>* nocapture %p, i32 %n) nounwind uwtable ssp { 15entry: 16 br label %while.body 17 18; Materialize a zeroinitializer and a constant-pool load in the integer domain. 19; The order is not important. 20; CHECK: pxor 21; CHECK: movdqa 22 23; The instructions in the loop must all be integer domain as well. 24; CHECK: while.body 25; CHECK: pand 26; CHECK: movdqa 27; CHECK: movdqa 28; Finally, the controlling integer-only instruction. 29; CHECK: paddd 30while.body: 31 %p.addr.04 = phi <4 x i32>* [ %incdec.ptr, %while.body ], [ %p, %entry ] 32 %n.addr.03 = phi i32 [ %dec, %while.body ], [ %n, %entry ] 33 %x.02 = phi <4 x i32> [ %add, %while.body ], [ zeroinitializer, %entry ] 34 %dec = add nsw i32 %n.addr.03, -1 35 %and = and <4 x i32> %x.02, <i32 127, i32 127, i32 127, i32 127> 36 %incdec.ptr = getelementptr inbounds <4 x i32>, <4 x i32>* %p.addr.04, i64 1 37 store <4 x i32> %and, <4 x i32>* %p.addr.04, align 16 38 %0 = load <4 x i32>, <4 x i32>* %incdec.ptr, align 16 39 %add = shl <4 x i32> %0, <i32 1, i32 1, i32 1, i32 1> 40 %tobool = icmp eq i32 %dec, 0 41 br i1 %tobool, label %while.end, label %while.body 42 43while.end: 44 ret void 45} 46