File: swift-vldm.ll

package info (click to toggle)
llvm-toolchain-9 1%3A9.0.1-16
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 882,436 kB
  • sloc: cpp: 4,167,636; ansic: 714,256; asm: 457,610; python: 155,927; objc: 65,094; sh: 42,856; lisp: 26,908; perl: 7,786; pascal: 7,722; makefile: 6,881; ml: 5,581; awk: 3,648; cs: 2,027; xml: 888; javascript: 381; ruby: 156
file content (30 lines) | stat: -rw-r--r-- 1,326 bytes parent folder | download | duplicates (18)
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
26
27
28
29
30
; RUN: llc < %s -mcpu=swift -mtriple=armv7s-apple-ios | FileCheck %s
; RUN: llc < %s -arm-assume-misaligned-load-store -mcpu=swift -mtriple=armv7s-apple-ios | FileCheck %s

; Check that we avoid producing vldm instructions using d registers that
; begin in the most-significant half of a q register. These require more
; micro-ops on swift and so aren't worth combining.

; CHECK-LABEL: test_vldm
; CHECK: vldmia r{{[0-9]+}}, {d2, d3, d4}
; CHECK-NOT: vldmia r{{[0-9]+}}, {d1, d2, d3, d4}

declare fastcc void @force_register(double %d0, double %d1, double %d2, double %d3, double %d4) 

define void @test_vldm(double* %x, double * %y) {
entry:
  %addr1 = getelementptr double, double * %x, i32 1
  %addr2 = getelementptr double, double * %x, i32 2
  %addr3 = getelementptr double, double * %x, i32 3
  %d0 = load double , double * %y
  %d1 = load double , double * %x
  %d2 = load double , double * %addr1
  %d3 = load double , double * %addr2
  %d4 = load double , double * %addr3
  ; We are trying to force x[0-3] in registers d1 to d4 so that we can test we
  ; don't form a "vldmia rX, {d1, d2, d3, d4}".
  ; We are relying on the calling convention and that register allocation
  ; properly coalesces registers.
  call fastcc void @force_register(double %d0, double %d1, double %d2, double %d3, double %d4)
  ret void
}