[Mesa-dev] [PATCH 2/2] i965: don't drop const initializers in vector splitting

[Jason Ekstrand]

On Jun 25, 2016 5:39 AM, "Rob Clark" <robdclark at gmail.com> wrote:
>
> From: Rob Clark <robclark at freedesktop.org>
>
> Not entirely sure how we didn't hit this before, but dropping the const
> initializer on the floor is obviously not correct.
>
> Signed-off-by: Rob Clark <robclark at freedesktop.org>
> ---
> No idea why i965 even still uses this pass, vs nir scalarizing pass.
> So might want to drop this.  But I figured fixing it first gives
> something that can be cherry-picked to release branches, so this
> patch should land before deleting the ir pass.
>
>  src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp
b/src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp
> index 5fe24de..7c58089 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_vector_splitting.cpp
> @@ -372,6 +372,18 @@ brw_do_vector_splitting(exec_list *instructions)
>
>           ralloc_free(name);
>
> +         if (entry->var->constant_initializer) {
> +            ir_constant_data data = {0};
> +            assert(entry->var->data.has_initializer);
> +            if (entry->var->type->is_double()) {
> +               data.d[i] = entry->var->constant_initializer->value.d[i];

I think you want data.d[0] on the l left hands side.

> +            } else {
> +               data.u[i] = entry->var->constant_initializer->value.u[i];

Same here

> +            }
> +            entry->components[i]->data.has_initializer = true;
> +            entry->components[i]->constant_initializer =
new(entry->components[i]) ir_constant(type, &data);
> +         }
> +
>          entry->var->insert_before(entry->components[i]);
>        }
>
> --
> 2.7.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160625/20e67691/attachment.html>