File: consecutive-load-shuffle.ll

package info (click to toggle)
llvm-toolchain-17 1%3A17.0.6-22
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 1,799,624 kB
  • sloc: cpp: 6,428,607; ansic: 1,383,196; asm: 793,408; python: 223,504; objc: 75,364; f90: 60,502; lisp: 33,869; pascal: 15,282; sh: 9,684; perl: 7,453; ml: 4,937; awk: 3,523; makefile: 2,889; javascript: 2,149; xml: 888; fortran: 619; cs: 573
file content (25 lines) | stat: -rw-r--r-- 1,186 bytes parent folder | download | duplicates (13)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-pc-windows | FileCheck %s

; We should be able to prodcue a single 128-bit load for these two 64-bit loads.
; But we previously weren't because we weren't consistently looking through
; WrapperRIP.

@f = local_unnamed_addr global [4 x float] zeroinitializer, align 16
@ms = common local_unnamed_addr global <4 x float> zeroinitializer, align 16

define void @foo2() {
; CHECK-LABEL: foo2:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    movaps f(%rip), %xmm0
; CHECK-NEXT:    movaps %xmm0, ms(%rip)
; CHECK-NEXT:    retq
entry:
  %0 = load <2 x float>, ptr getelementptr inbounds ([4 x float], ptr @f, i64 0, i64 2), align 8
  %shuffle.i10 = shufflevector <2 x float> %0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
  %1 = load <2 x float>, ptr @f, align 16
  %shuffle.i7 = shufflevector <2 x float> %1, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
  %shuffle.i = shufflevector <4 x float> %shuffle.i7, <4 x float> %shuffle.i10, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
  store <4 x float> %shuffle.i, ptr @ms, align 16
  ret void
}