File: stack-overalign.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 (30 lines) | stat: -rw-r--r-- 1,152 bytes parent folder | download | duplicates (12)
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: opt < %s -passes=instcombine -S | grep "align 32" | count 2

; It's tempting to have an instcombine in which the src pointer of a
; memcpy is aligned up to the alignment of the destination, however
; there are pitfalls. If the src is an alloca, aligning it beyond what
; the target's stack pointer is aligned at will require dynamic
; stack realignment, which can require functions that don't otherwise
; need a frame pointer to need one.
;
; Abstaining from this transform is not the only way to approach this
; issue. Some late phase could be smart enough to reduce alloca
; alignments when they are greater than they need to be. Or, codegen
; could do dynamic alignment for just the one alloca, and leave the
; main stack pointer at its standard alignment.
;


@dst = global [1024 x i8] zeroinitializer, align 32

define void @foo() nounwind {
entry:
  %src = alloca [1024 x i8], align 64
  call void @llvm.memcpy.p0.p0.i32(ptr align 32 @dst, ptr align 32 %src, i32 1024, i1 false)
  call void @frob(ptr %src) nounwind
  ret void
}

declare void @frob(ptr)

declare void @llvm.memcpy.p0.p0.i32(ptr nocapture, ptr nocapture, i32, i1) nounwind