[Mesa-dev] [RFC 3/3] u_vbuf: use single vertex buffer if needed

[Nicolai Hähnle]

On 11.06.2016 21:21, Christian Gmeiner wrote:
> From: "Wladimir J. van der Laan" <laanwj at gmail.com>
>
> CONST, VERTEX and INSTANCE attributes into one vertex buffer if
> necessary due to hardware constraints.
>
> Signed-off-by: Wladimir J. van der Laan <laanwj at gmail.com>
> ---
>   src/gallium/auxiliary/util/u_vbuf.c | 28 ++++++++++++++++++++++++----
>   1 file changed, 24 insertions(+), 4 deletions(-)
>
> diff --git a/src/gallium/auxiliary/util/u_vbuf.c b/src/gallium/auxiliary/util/u_vbuf.c
> index 464c279..d35f3b0 100644
> --- a/src/gallium/auxiliary/util/u_vbuf.c
> +++ b/src/gallium/auxiliary/util/u_vbuf.c
> @@ -539,25 +539,45 @@ u_vbuf_translate_find_free_vb_slots(struct u_vbuf *mgr,
>      uint32_t unused_vb_mask =
>         (mgr->ve->incompatible_vb_mask_all | mgr->incompatible_vb_mask |
>         ~mgr->enabled_vb_mask | extra_free_vb_mask) & mgr->allowed_vb_mask;
> +   uint32_t unused_vb_mask_temp;
> +   boolean insufficient_buffers = false;
> +
> +   /* No vertex buffers available at all */
> +   if(!unused_vb_mask)
> +      return FALSE;
>
>      memset(fallback_vbs, ~0, sizeof(fallback_vbs));
>
>      /* Find free slots for each type if needed. */
> +   unused_vb_mask_temp = unused_vb_mask;
>      for (type = 0; type < VB_NUM; type++) {
>         if (mask[type]) {
>            uint32_t index;
>
> -         if (!unused_vb_mask) {
> -            return FALSE;
> +         if (!unused_vb_mask_temp) {
> +            insufficient_buffers = TRUE;
> +            break;
>            }
>
> -         index = ffs(unused_vb_mask) - 1;
> +         index = ffs(unused_vb_mask_temp) - 1;
>            fallback_vbs[type] = index;
> -         unused_vb_mask &= ~(1 << index);
> +         unused_vb_mask_temp &= ~(1 << index);
>            /*printf("found slot=%i for type=%i\n", index, type);*/
>         }
>      }
>
> +   if (insufficient_buffers) {
> +      /* not enough vbs for all types supported by the hardware, they will have to
> +       * share one buffer */
> +      uint32_t index = ffs(unused_vb_mask) - 1;
> +
> +      /* When sharing one vertex buffer use per-vertex frequency for everything. */
> +      fallback_vbs[VB_VERTEX] = index;
> +      mask[VB_VERTEX] = mask[VB_VERTEX] | mask[VB_CONST] | mask[VB_INSTANCE];
> +      mask[VB_CONST] = 0;
> +      mask[VB_INSTANCE] = 0;
> +   }
> +
>      for (type = 0; type < VB_NUM; type++) {
>         if (mask[type]) {
>            mgr->dirty_real_vb_mask |= 1 << fallback_vbs[type];
>

I have a slight preference to make an unused_vb_mask_orig copy that is 
used in the insufficient_buffers fallback case rather than change 
everything to unused_vb_mask_temp.

Cheers,
Nicolai