[llvm-bugs] [Bug 48248] New: Crash on win64 vectorcall with large

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=48248

            Bug ID: 48248
           Summary: Crash on win64 vectorcall with large
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Windows XP
            Status: NEW
          Severity: normal
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: christoffer.lerno at gmail.com
                CC: llvm-bugs at lists.llvm.org, neeilans at live.com,
                    richard-llvm at metafoo.co.uk

I found a possible bug in TargetInfo.cpp, WinX86_64ABIInfo::classify

On rows 4215-4216 we have the following:

 } else if (IsReturnType) {
   return ABIArgInfo::getExpand();

This would suggest that we use "Expand" as strategy for a return value on a
vector call. However, "Expand" is not a valid strategy for return values, as it
is only for parameters. Hoping to expose this bug, I created large vectors
embedded in structs, which would then hit the "homogenous aggregate" path in
said function.

I finally struck gold with the following:

typedef double foo __attribute__ ((vector_size (64)));
struct Test {
    foo a[2];
};

struct Test __vectorcall bar(struct Test a) {
    return (struct Test) {};
}

However I am not sure that I'm actually triggering the bug, or if there are
multiple bugs, as this only crashes when "struct Test a" is there, which makes
me suspect that maybe we're hitting some other bug here.

Godbolt link: https://www.godbolt.org/z/EWWaor

-- 
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/20201121/5efc1c4f/attachment-0001.html>