[llvm-bugs] [Bug 49357] New: [AArch64/FastIsel] Misoptimization of sext from function argument in the presence of spills.

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=49357

            Bug ID: 49357
           Summary: [AArch64/FastIsel] Misoptimization of sext from
                    function argument in the presence of spills.
           Product: libraries
           Version: trunk
          Hardware: Macintosh
                OS: MacOS X
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: AArch64
          Assignee: unassignedbugs at nondot.org
          Reporter: keno at alumni.harvard.edu
                CC: arnaud.degrandmaison at arm.com,
                    llvm-bugs at lists.llvm.org, smithp352 at googlemail.com,
                    Ties.Stuij at arm.com

Created attachment 24573
  --> https://bugs.llvm.org/attachment.cgi?id=24573&action=edit
Testcase showing the issue.

Consider the attached testcase.ll, reduced from a downstream issue report at
https://github.com/JuliaLang/julia/issues/39823.

The relevant snippet of the testcase is:
```
define nonnull {}* @julia_f_1854(i32 signext %0) #0 {
[snip]
  call void inttoptr (i64 4342098656 to void ()*)()
[snip]
  %10 = sext i32 %0 to i64
  %11 = icmp slt i64 %10, 0
```

When run as `llc -mtriple arm64-apple-darwin20.0.0 -mcpu apple-a12
-code-model=small -O0 -fast-isel -o - testcase.ll`, the produced assembly is in
relevant part:

```
        str     w0, [sp, #8]                    ; 4-byte Folded Spill
        blr     x8
        ldr     w9, [sp, #8]                    ; 4-byte Folded Reload
[snip]
        mov     x8, x9
        cmp     x8, #0  

```
Note that the sext got isel'ed to a `mov        x8, x9`. This is because of
fast isel decided at [1] that this sext did not need an actual instruction
because of the arguments signext attribute. However, the implementation does
not work, because upper bits of the SUBREG_TO_REG'd register are undefined and
subsequent passes may clobber them (e.g. as in this case by a stack spill).

[1]
https://github.com/llvm/llvm-project/blob/2c56776a319edf33505ca6c7f9be59657cdaf52b/llvm/lib/Target/AArch64/AArch64FastISel.cpp#L4598-L4618

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210225/54b3f7b3/attachment.html>