https://llvm.org/bugs/show_bug.cgi?id=24588
Bug ID: 24588
Summary: aarch64 backend uses x18 on ios
Product: new-bugs
Version: trunk
Hardware: Macintosh
OS: other
Status: NEW
Severity: enhancement
Priority: P
Component: new bugs
Assignee: unassignedb...@nondot.org
Reporter: andrew.b.ad...@gmail.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
Created attachment 14780
--> https://llvm.org/bugs/attachment.cgi?id=14780&action=edit
A chunk of ll large enough to demonstrate the issue
This is a minor complaint, but it produces a very hard-to-debug error.
If you compile some .ll with a target triple aarch64-ios, then the aarch64
backend will happily use x18, even though that's reserved by the system on this
platform.
This causes crashes due to register-stomping. Clang avoids this by setting the
additional flag -mattr=+reserve-x18, but having IOS in the target triple seems
like it should be enough.
To reproduce, compile the attached .ll like so:
llc -mtriple aarch64-ios initmod.opengl_64.ll -filetype=asm -o - | grep x18
orr w0, wzr, #0x18
orr w0, wzr, #0x18
ldrsw x18, [x22, #40]
lsl x18, x18, #1
add x5, x5, x18
ldrsw x18, [x22, #40]
lsl x18, x18, #2
add x5, x5, x18
ldrsw x18, [x21, #40]
lsl x18, x18, #1
add x3, x3, x18
ldrsw x18, [x21, #40]
lsl x18, x18, #2
add x3, x3, x18
sxtw x18, w18
str w3, [x1, x18, lsl #2]
If you add -mattr=+reserve-x18, it does the right thing.
There's nothing very special about the attached ll - any sufficiently large
piece of ll should reproduce this.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
