[Mesa-dev] [PATCH] i965: Fast texture upload now supports all levels

[Chad Versace]

On 10/11/2013 10:17 AM, Courtney Goeltzenleuchter wrote:
> Support all levels of a supported texture format.
> ---
>   src/mesa/drivers/dri/i965/intel_tex_subimage.c | 13 +++++++++++--
>   1 file changed, 11 insertions(+), 2 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/intel_tex_subimage.c b/src/mesa/drivers/dri/i965/intel_tex_subimage.c
> index 4aec05d..5e46760 100644
> --- a/src/mesa/drivers/dri/i965/intel_tex_subimage.c
> +++ b/src/mesa/drivers/dri/i965/intel_tex_subimage.c
> @@ -541,14 +541,13 @@ intel_texsubimage_tiled_memcpy(struct gl_context * ctx,
>      uint32_t cpp;
>      mem_copy_fn mem_copy = NULL;
>
> -   /* This fastpath is restricted to specific texture types: level 0 of
> +   /* This fastpath is restricted to specific texture types:
>       * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support
>       * more types.
>       */
>      if (!brw->has_llc ||
>          type != GL_UNSIGNED_BYTE ||
>          texImage->TexObject->Target != GL_TEXTURE_2D ||
> -       texImage->Level != 0 ||
>          pixels == NULL ||
>          _mesa_is_bufferobj(packing->BufferObj) ||
>          packing->Alignment > 4 ||
> @@ -616,6 +615,16 @@ intel_texsubimage_tiled_memcpy(struct gl_context * ctx,
>      DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d)\n",
>          __FUNCTION__, texImage->Level, xoffset, yoffset, width, height);
>
> +   /* Adjust x and y offset based on miplevel
> +    */
> +   if (texImage->Level) {
> +      GLuint xlevel, ylevel;
> +      intel_miptree_get_image_offset(image->mt, texImage->Level, 0,
> +                                  &xlevel, &ylevel);
> +      xoffset += xlevel;
> +      yoffset += ylevel;
> +   }
> +
>      linear_to_tiled(
>         xoffset * cpp, (xoffset + width) * cpp,
>         yoffset, yoffset + height,
>

Usually when we commit performance patches like this, we state in the
commit message what the observed relative performance gain.

What gain did you see? Hardware? Benchmark? Kernel version? How many
runs?