[Mesa-dev] [PATCH 78/78] nir: Fix output swizzle in get_mul_for_src

Jason Ekstrand jason at jlekstrand.net
Tue Jun 30 16:41:29 PDT 2015


On Fri, Jun 26, 2015 at 1:07 AM, Eduardo Lima Mitev <elima at igalia.com> wrote:
> From: Samuel Iglesias Gonsalvez <siglesias at igalia.com>
>
> Avoid copying an overwritten swizzle, use the original values.
>
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89580
> Signed-off-by: Samuel Iglesias Gonsalvez <siglesias at igalia.com>
> ---
>  src/glsl/nir/nir_opt_peephole_ffma.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/src/glsl/nir/nir_opt_peephole_ffma.c b/src/glsl/nir/nir_opt_peephole_ffma.c
> index 798506b..7c9ac12 100644
> --- a/src/glsl/nir/nir_opt_peephole_ffma.c
> +++ b/src/glsl/nir/nir_opt_peephole_ffma.c
> @@ -76,6 +76,7 @@ static nir_alu_instr *
>  get_mul_for_src(nir_alu_src *src, int num_components,
>                  uint8_t swizzle[4], bool *negate, bool *abs)
>  {
> +   uint8_t swizzle_tmp[4];
>     assert(src->src.is_ssa && !src->abs && !src->negate);
>
>     nir_instr *instr = src->src.ssa->parent_instr;
> @@ -116,8 +117,10 @@ get_mul_for_src(nir_alu_src *src, int num_components,
>     if (!alu)
>        return NULL;
>
> +   /* Copy swizzle data before overwriting it */

I'd like the comment to be a little more detailed.  Specifically, an
example of when this is a problem would be really good.  Looking at
it, I *think* that if src->swizzle is xyxx, it would trip the problem.

> +   memcpy(swizzle_tmp, swizzle, 4*sizeof(uint8_t));
>     for (unsigned i = 0; i < num_components; i++)
> -      swizzle[i] = swizzle[src->swizzle[i]];
> +      swizzle[i] = swizzle_tmp[src->swizzle[i]];
>
>     return alu;
>  }
> --
> 2.1.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list