File: stack-overalign.ll

package info (click to toggle)
llvm-toolchain-16 1%3A16.0.6-15~deb12u1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,634,792 kB
  • sloc: cpp: 6,179,261; ansic: 1,216,205; asm: 741,319; python: 196,614; objc: 75,325; f90: 49,640; lisp: 32,396; pascal: 12,286; sh: 9,394; perl: 7,442; ml: 5,494; awk: 3,523; makefile: 2,723; javascript: 1,206; xml: 886; fortran: 581; 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