[LLVMdev] Question on Inline Assembly Code generation for X64

[mats petersson]

I'm fairly sure you need the register definition to be "Y0" or
something like that for the input operand to be XMM0.

--
Mats

On 21 March 2015 at 06:37, Tae Jun Ham <ham.taejun at gmail.com> wrote:
> Hi,
>
> I have a question on inline assembly code generation for X64.
>
> I have following LLVM IR :
>
> %c = fsub fast float %a, %b
>  tail call void asm sideeffect "mov $0, %eax \0A\09  movd $1, %xmm0 \0A\09
> xchg %dx, %dx\0A", "ir,r,~{ax}"(i32 100, float %c)
>
> The assembly that I get for this part of the code is as follows (x64
> target).
> *xchgw %dx, %dx is a nop that I am using to do some stuff in binary
> translator.
>
> subss %xmm1, %xmm0
> movd %xmm0, %edi
> #APP
> movl $25, %eax
> movd %edi, %xmm0
> xchgw %dx, %dx
>
> What I wanted/expected was following :
>
> subss %xmm1, %xmm0
> #APP
> movl $25, %eax
> movd %xmm0, %xmm0
> xchgw %dx, %dx
>
> What would be the proper way to get the output I want ?
>
> Thank you
> --
> Tae Jun Ham
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>