llvm-project

1
; REQUIRES: x86
2
; RUN: llvm-as %s -o %t.o
3

4
;; Verify that LTO behavior can be tweaked using -mattr.
5

6
; RUN: %lld -mcpu haswell -mllvm -mattr=+fma %t.o -o %t.dylib -dylib
7
; RUN: llvm-objdump --no-print-imm-hex -d --section="__text" --no-leading-addr --no-show-raw-insn %t.dylib | FileCheck %s --check-prefix=FMA
8

9
; RUN: %lld -mcpu haswell -mllvm -mattr=-fma %t.o -o %t.dylib -dylib
10
; RUN: llvm-objdump --no-print-imm-hex -d --section="__text" --no-leading-addr --no-show-raw-insn %t.dylib | FileCheck %s --check-prefix=NO-FMA
11

12
; FMA:      <_foo>:
13
; FMA-NEXT: vrcpss       %xmm0, %xmm0, %xmm1
14
; FMA-NEXT: vfmsub213ss  [[#]](%rip), %xmm1, %xmm0 ## xmm0 = (xmm1 * xmm0) - mem
15
; FMA-NEXT:                                        ## 0x
16
; FMA-NEXT: vfnmadd132ss %xmm1, %xmm1, %xmm0
17
; FMA-NEXT: retq
18

19
; NO-FMA:      <_foo>:
20
; NO-FMA-NEXT: vrcpss %xmm0, %xmm0, %xmm1
21
; NO-FMA-NEXT: vmulss %xmm1, %xmm0, %xmm0
22
; NO-FMA-NEXT: vmovss [[#]](%rip), %xmm2  ## 0x
23
; NO-FMA-NEXT: vsubss %xmm0, %xmm2, %xmm0
24
; NO-FMA-NEXT: vmulss %xmm0, %xmm1, %xmm0
25
; NO-FMA-NEXT: vaddss %xmm0, %xmm1, %xmm0
26
; NO-FMA-NEXT: retq
27

28
target triple = "x86_64-apple-darwin"
29
target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
30

31
define float @foo(float %x) #0 {
32
  %div = fdiv fast float 1.0, %x
33
  ret float %div
34
}
35

36
attributes #0 = { "unsafe-fp-math"="true" "reciprocal-estimates"="divf,vec-divf" }
37