[llvm-dev] LLC Segmentation Fault in Matrix Multiplication

[Craig Topper via llvm-dev]

Can you attach your source code or .ll file? Have you tried a debug build
of llc which might give a more readable stack dump or even an assertion
message.

~Craig

On Fri, Jun 30, 2017 at 10:51 PM, hameeza ahmed via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> It happens only when i keep my vector width to >=64, why is that so?
>
> I have to keep it >=64. Please suggest some way?
>
> On Sat, Jul 1, 2017 at 10:47 AM, hameeza ahmed <hahmed2305 at gmail.com>
> wrote:
>
>> Hello,
>>
>> when i used clang for matrix multiplication code with i=j=k=1000351, it
>> successfully generates IR. then i run opt as follows
>> opt  -S -O3 -force-vector-width=64 -force-vector-interleave=32 matn.ll -o
>> pimmatnl_64_u32_o3.ll
>>
>> it also worked fine. but when i run llc on the optimized .ll file
>>
>> llc -x86-asm-syntax=intel pimmatnl_64_u32_o3.ll -o matnl-knl4_intel.s
>>
>> i get the following segmentation fault error:
>>
>> #0 0x0000000001826ad8 (llc+0x1826ad8)
>> #1 0x0000000001824aae (llc+0x1824aae)
>> #2 0x0000000001824bfc (llc+0x1824bfc)
>> #3 0x00007fa6bdb52d10 __restore_rt (/lib/x86_64-linux-gnu/libpthr
>> ead.so.0+0x10d10)
>> #4 0x00000000014d0974 (llc+0x14d0974)
>> #5 0x000000000169bdb5 (llc+0x169bdb5)
>> #6 0x00000000016bcfdc (llc+0x16bcfdc)
>> #7 0x00000000016d8684 (llc+0x16d8684)
>> #8 0x000000000170a4c0 (llc+0x170a4c0)
>> #9 0x000000000171199d (llc+0x171199d)
>> #10 0x0000000001714258 (llc+0x1714258)
>> #11 0x0000000000e530f4 (llc+0xe530f4)
>> #12 0x000000000122c931 (llc+0x122c931)
>> #13 0x0000000001499b2a (llc+0x1499b2a)
>> #14 0x0000000001499bd3 (llc+0x1499bd3)
>> #15 0x0000000001499694 (llc+0x1499694)
>> #16 0x000000000068a68c (llc+0x68a68c)
>> #17 0x0000000000626a80 (llc+0x626a80)
>> #18 0x00007fa6bccf3a40 __libc_start_main (/lib/x86_64-linux-gnu/libc.so
>> .6+0x20a40)
>> #19 0x000000000067f8e9 (llc+0x67f8e9)
>> Stack dump:
>> 0. Program arguments: llc -x86-asm-syntax=intel pimmatnl_64_u32_o3.ll -o
>> matnl-knl4_intel.s
>> 1. Running pass 'Function Pass Manager' on module
>> 'pimmatnl_64_u32_o3.ll'.
>> 2. Running pass 'X86 DAG->DAG Instruction Selection' on function
>> '@multiply'
>> Segmentation fault (core dumped)
>>
>> Why does it happen? please correct me.
>>
>> Thank You
>>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170630/d8ee6f92/attachment.html>