1 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=prescott | FileCheck %s --check-prefix=PRESCOTT
2 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mcpu=nehalem | FileCheck %s --check-prefix=NEHALEM
4 ;;; TODO: The last run line chooses cpu=nehalem to reveal possible bugs in the "foo" test case.
6 ;;; Nehalem has a 'fast unaligned memory' attribute, so (1) some of the loads and stores
7 ;;; are certainly unaligned and (2) the first load and first store overlap with the second
8 ;;; load and second store respectively.
10 ;;; Is either of these sequences ideal?
11 ;;; Is the ideal code being generated for all CPU models?
13 define float @foo(i8* nocapture %buf, float %a, float %b) nounwind uwtable {
14 ; PRESCOTT-LABEL: foo:
15 ; PRESCOTT: # BB#0: # %entry
16 ; PRESCOTT-NEXT: movw .Ltmp0+20(%rip), %ax
17 ; PRESCOTT-NEXT: movw %ax, 20(%rdi)
18 ; PRESCOTT-NEXT: movl .Ltmp0+16(%rip), %eax
19 ; PRESCOTT-NEXT: movl %eax, 16(%rdi)
20 ; PRESCOTT-NEXT: movq .Ltmp0+8(%rip), %rax
21 ; PRESCOTT-NEXT: movq %rax, 8(%rdi)
22 ; PRESCOTT-NEXT: movq .Ltmp0(%rip), %rax
23 ; PRESCOTT-NEXT: movq %rax, (%rdi)
26 ; NEHALEM: # BB#0: # %entry
27 ; NEHALEM-NEXT: movq .Ltmp0+14(%rip), %rax
28 ; NEHALEM-NEXT: movq %rax, 14(%rdi)
29 ; NEHALEM-NEXT: movups .Ltmp0(%rip), %xmm2
30 ; NEHALEM-NEXT: movups %xmm2, (%rdi)
33 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %buf, i8* blockaddress(@foo, %out), i64 22, i32 1, i1 false)
37 %add = fadd float %a, %b
41 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind