[x86] Code generation is wrong: insertElement( v3i1, i1 )

[llvmbot]

Bugzilla Link	12618
Version	trunk
OS	Windows NT
Blocks	llvm/llvm-bugzilla-archive#31265
Reporter	LLVM Bugzilla Contributor
CC	@asl

Extended Description

Hello guys,

following LLVM ir

define inlinehint <3 x i1> @"Mbool3@@QBB@@QBB@@QBB@@"(i1 %.v0, i1 %.v1, i1 %.v2) {
.body:
  %0 = insertelement <3 x i1> undef, i1 %.v0, i32 0
  %1 = insertelement <3 x i1> %0, i1 %.v1, i32 1
  %2 = insertelement <3 x i1> %1, i1 %.v2, i32 2
  ret <3 x i1> %2
}

generates ASMs:

005400C0  mov         cl,byte ptr [esp+10h]  
005400C4  and         cl,1  
005400C7  mov         eax,dword ptr [esp+4]  
005400CB  mov         byte ptr [eax],cl  
005400CD  mov         cl,byte ptr [esp+0Ch]  
005400D1  and         cl,1  
005400D4  mov         byte ptr [eax],cl  
005400D6  mov         cl,byte ptr [esp+8]  
005400DA  and         cl,1  
005400DD  mov         byte ptr [eax],cl  
005400DF  ret  

And this is not correct. In my opinion, following code would be generated:

005400C0  mov         cl,byte ptr [esp+10h]  
005400C4  and         cl,1  
005400C7  mov         eax,dword ptr [esp+4]  
005400CB  mov         byte ptr [eax+1],cl   ; offset
005400CD  mov         cl,byte ptr [esp+0Ch]  
005400D1  and         cl,1  
005400D4  mov         byte ptr [eax+2],cl  
005400D6  mov         cl,byte ptr [esp+8]  ; offset
005400DA  and         cl,1  
005400DD  mov         byte ptr [eax],cl    ; offset
005400DF  ret  

Am I right?
And another question, why the argument order is +10h +0Ch and +8h rather than ascending or descending ?

[RKSimon]

mentioned in issue llvm/llvm-bugzilla-archive#31265