On 9 July 2013 18:12, Anuj Phogat <anuj.pho...@gmail.com> wrote:
> For HSW GT3 clear rectangle must be aligned to two times the number of
> pixels in the table shown in Ivy Bridge PRM, Vol2 Part1 11.7.
> It should be safe to do this for all gen7 systems unless we see any
> performance regressions. I observed no piglit, gles3conform regressions
> with this patch.
>
I realize it was my idea to do this for all Gen7 systems, but on further
reflection, it's so easy to restrict this workaround to just HSW GT3 that I
think we should go ahead and do that. Along the lines of Chris Forbes'
suggestion, the code would look something like:
if (brw->is_haswell && brw->gt == 3) {
/* rationale ... */
x0 = ROUND_DOWN_TO(x0, 2 * x_align);
y0 = ROUND_DOWN_TO(y0, 2 * y_align);
x1 = ALIGN(x1, 2 * x_align);
y1 = ALIGN(y1, 2 * y_align);
} else {
x0 = ROUND_DOWN_TO(x0, x_align);
y0 = ROUND_DOWN_TO(y0, y_align);
x1 = ALIGN(x1, x_align);
y1 = ALIGN(y1, y_align);
}
>
> It fixes: https://bugs.freedesktop.org/show_bug.cgi?id=65744
>
> Signed-off-by: Anuj Phogat <anuj.pho...@gmail.com>
> ---
> Thanks Paul for your great help in figuring out the fix.
>
> src/mesa/drivers/dri/i965/brw_blorp_clear.cpp | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> index bf11135..2bebe8e 100644
> --- a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> @@ -261,10 +261,14 @@
> brw_blorp_clear_params::brw_blorp_clear_params(struct brw_context *brw,
> intel_get_non_msrt_mcs_alignment(intel, irb->mt, &x_align,
> &y_align);
> x_align *= 16;
> y_align *= 32;
> - x0 = ROUND_DOWN_TO(x0, x_align);
> - y0 = ROUND_DOWN_TO(y0, y_align);
> - x1 = ALIGN(x1, x_align);
> - y1 = ALIGN(y1, y_align);
> + /* For HSW GT3 clear rectangle must be aligned to two times the
> number
> + * of pixels in the table.
> + */
> +
> + x0 = ROUND_DOWN_TO(x0, (intel->gen > 6 ? 2 * x_align : x_align));
> + y0 = ROUND_DOWN_TO(y0, (intel->gen > 6 ? 2 * y_align : y_align));
> + x1 = ALIGN(x1, (intel->gen > 6 ? 2 * x_align : x_align));
> + y1 = ALIGN(y1, (intel->gen > 6 ? 2 * y_align : y_align));
>
> /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
> * Target(s)", beneath the "Fast Color Clear" bullet (p327):
> --
> 1.8.1.4
>
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
