1; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=nocona | FileCheck %s --check-prefix=PRESCOTT 2; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=nehalem | FileCheck %s --check-prefix=NEHALEM 3 4;;; TODO: (1) Some of the loads and stores are certainly unaligned and (2) the first load and first 5;;; store overlap with the second load and second store respectively. 6;;; 7;;; Is either of these sequences ideal? 8 9define float @foo(i8* nocapture %buf, float %a, float %b) nounwind uwtable { 10; PRESCOTT-LABEL: foo: 11; PRESCOTT: # %bb.0: # %entry 12; PRESCOTT-NEXT: movq .Ltmp0+14(%rip), %rax 13; PRESCOTT-NEXT: movq %rax, 14(%rdi) 14; PRESCOTT-NEXT: movq .Ltmp0+8(%rip), %rax 15; PRESCOTT-NEXT: movq %rax, 8(%rdi) 16; PRESCOTT-NEXT: movq .Ltmp0(%rip), %rax 17; PRESCOTT-NEXT: movq %rax, (%rdi) 18; 19; NEHALEM-LABEL: foo: 20; NEHALEM: # %bb.0: # %entry 21; NEHALEM-NEXT: movq .Ltmp0+14(%rip), %rax 22; NEHALEM-NEXT: movq %rax, 14(%rdi) 23; NEHALEM-NEXT: movups .Ltmp0(%rip), %xmm2 24; NEHALEM-NEXT: movups %xmm2, (%rdi) 25 26entry: 27 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %buf, i8* blockaddress(@foo, %out), i64 22, i1 false) 28 br label %out 29 30out: ; preds = %entry 31 %add = fadd float %a, %b 32 ret float %add 33} 34 35declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind 36