[Mesa-dev] radeonsi bug in fmask calculation?

[Marek Olšák]

You're right. It's a bug.

Marek

On Fri, Sep 18, 2015 at 6:47 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> Hi Marek,
>
> You weren't on IRC, and I'll forget if I don't write this down now.
> Take a look at the code at:
>
> http://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/radeonsi/si_shader.c#n2352
>
> fmask_ptr = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, SI_PARAM_RESOURCE);
> fmask_ptr = build_indexed_load_const(si_shader_ctx, res_ptr, ind_index);
>
> It seems odd that the second fmask_ptr ignores the setting of the
> first. Is the first one not needed, or should fmask_ptr be used
> instead of res_ptr, or is there something more subtle going on?
>
> Cheers,
>
>   -ilia
>
> P.S. To remind you, I have no knowledge of what any of these things
> actually mean. Just looking to refactor a bit to facilitate TXQS
> implementation, and happened to notice this bit of oddness.