[Mesa-dev] [PATCH 0/10] i965: Implement hiz and separate stencil for window framebuffer
chad at chad-versace.us
Tue Jun 7 09:12:23 PDT 2011
I don't have performance data yet, but it's coming soon :)
I'm still unable to test real-world apps with hiz *fully* enabled because
i965 is still using packed detpth/stencil when the app requests a
texture or FBO with GL_DEPTH_STENCIL format. This should be fixed
in two or three days, and I'll give you some performance numbers then.
Even then, the performance improvement we will obtain then will be an optimistic upper bound on
hiz's *real* performance improvement. Without hiz/depth resolves
implemented yet, the numbers will be artificially high and rendering will be corrupt.
chad at chad-versace.us
On Tue, 7 Jun 2011 10:39:40 +0800, "Zou, Nanhai" <nanhai.zou at intel.com> wrote:
> Hi Chad,
> Glad to see the Hiz patches.
> Have you performance data for the Hiz patches, how much can this improve?
> Zou Nanhai
> -----Original Message-----
> From: mesa-dev-bounces+nanhai.zou=intel.com at lists.freedesktop.org [mailto:mesa-dev-bounces+nanhai.zou=intel.com at lists.freedesktop.org] On Behalf Of Chad Versace
> Sent: 2011年6月5日 8:46
> To: mesa-dev at lists.freedesktop.org
> Subject: [Mesa-dev] [PATCH 0/10] i965: Implement hiz and separate stencil for window framebuffer
> Patch 01: dri2proto
> Patch 02: xf86-video-intel
> Patch 03-10: mesa
> I'm aware that this patch series is painful to review. Thank you in advance
> for your reviewed-by's.
> There is a related patch series on the intel-gfx list that adds the necessary
> functionality to xf86-video-intel. See the thread with subject:
> ddx/dri: Add support for hiz and separate stencil buffers
> Give close attention to the comments in "intel: Define enum intel_dri2_has_hiz".
> That gives an overview of how the DRI2 handshake works and why it is
> I hear the grumbling already... "Chad significantly altered ancient, core
> driver routines! Did he break the world?" I can confidently say "No"; this
> patch series causes no regressions.
> I have performed full piglit runs that cover all (meaningful)
> combinations of the following:
> * xf86-video-intel: master vs my hiz branch
> * mesa: master vs my hiz branch
> * hiz enabled and disabled via environment var overrides
> You can find the test results here:
>  http://people.freedesktop.org/~chadversary/piglit-results/2011-06-04-1600-hiz-compare/summary/index.html
>  http://people.freedesktop.org/~chadversary/piglit-results/2011-06-04-1600-hiz-compare/summary/changes.html
> The column's titles should clearly mark what's what; the title format is
> "uuid--hostname--mesa at version--xf86-video-intel@version--hiz-[on|off]"
> Column 1 is the baseline (mesa at master, ddx at master). Columns 2 and 5 use my
> mesa hiz branch with hiz and separate stencil disabled.
> The diff from column 1 to 2 shows no regressions and one fix.
> CC: Eric Anholt <eric at anholt.net>
> CC: Ian Romanick <idr at freedesktop.org>
> CC: Kenneth Graunke <kenneth at whitecape.org>
> CC: Kristian Høgsberg <krh at bitplanet.net>
> Chad Versace (8):
> dri2: Add token for DRI2BufferHiz
> intel: Define enum intel_dri2_has_hiz
> intel: Add flags to intel_screen for hiz and separate stencil
> intel/intel_context.c: Remove unused functions
> intel: Add function intel_renderbuffer_set_hiz_region()
> intel: Refactor intel_update_renderbuffers()
> intel: Add assertions to intelCreateBuffer()
> intel: Request DRI2 buffers for separate stencil and hiz
> include/GL/internal/dri_interface.h | 1 +
> src/mesa/drivers/dri/intel/intel_context.c | 792 ++++++++++++++++++++++------
> src/mesa/drivers/dri/intel/intel_fbo.c | 12 +
> src/mesa/drivers/dri/intel/intel_fbo.h | 5 +
> src/mesa/drivers/dri/intel/intel_screen.c | 103 ++++-
> src/mesa/drivers/dri/intel/intel_screen.h | 62 +++
> 6 files changed, 803 insertions(+), 172 deletions(-)
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
More information about the mesa-dev