[llvm] r312873 - [X86] Add v2i2 test case (PR20011)

[Mikael Holmén via llvm-commits]

Hi Simon,

I'm a little curious about this, see comment below.

On 09/09/2017 10:22 PM, Simon Pilgrim via llvm-commits wrote:
> Author: rksimon
> Date: Sat Sep  9 13:22:35 2017
> New Revision: 312873
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=312873&view=rev
> Log:
> [X86] Add v2i2 test case (PR20011)
> 
> Added:
>      llvm/trunk/test/CodeGen/X86/pr20011.ll
> 
> Added: llvm/trunk/test/CodeGen/X86/pr20011.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr20011.ll?rev=312873&view=auto
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/pr20011.ll (added)
> +++ llvm/trunk/test/CodeGen/X86/pr20011.ll Sat Sep  9 13:22:35 2017
> @@ -0,0 +1,33 @@
> +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
> +; RUN: llc < %s -mtriple=i686-unknown-linux-gnu | FileCheck %s --check-prefix=X86
> +; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu | FileCheck %s --check-prefix=X64
> +
> +%destTy = type { i2, i2 }
> +
> +define void @crash(i64 %x0, i64 %y0, %destTy* nocapture %dest) nounwind {
> +; X86-LABEL: crash:
> +; X86:       # BB#0:
> +; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    movb {{[0-9]+}}(%esp), %dl
> +; X86-NEXT:    andb $3, %dl
> +; X86-NEXT:    movb %dl, (%ecx)
> +; X86-NEXT:    andb $3, %al
> +; X86-NEXT:    movb %al, (%ecx)

So the two stores will go to exactly the same byte, so the
  movb %dl, (%ecx)
will be overwritten by
  movb %al, (%ecx)
?

I.e. in the end result we will only have written one of the input 
arguments to the v2i2 vector?

Is this really how it's supposed to work?

Regards,
Mikael

> +; X86-NEXT:    retl
> +;
> +; X64-LABEL: crash:
> +; X64:       # BB#0:
> +; X64-NEXT:    andl $3, %esi
> +; X64-NEXT:    movb %sil, (%rdx)
> +; X64-NEXT:    andl $3, %edi
> +; X64-NEXT:    movb %dil, (%rdx)
> +; X64-NEXT:    retq
> +  %x1 = trunc i64 %x0 to i2
> +  %y1 = trunc i64 %y0 to i2
> +  %1 = bitcast %destTy* %dest to <2 x i2>*
> +  %2 = insertelement <2 x i2> undef, i2 %x1, i32 0
> +  %3 = insertelement <2 x i2> %2, i2 %y1, i32 1
> +  store <2 x i2> %3, <2 x i2>* %1, align 1
> +  ret void
> +}
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>