This is always the way with real hardware and desktop OpenGL. Some
hardware can't do some formats natively. The alpha-only, luminance,
and intensity formats are usually the most problematic. Some sized
formats can also be problematic. This patch provides fall-back
formats for those that are not natively supported.
At some point it would be interesting to try providing
device-independent conversions using EXT_texture_swizzle. The drivers
that support EXT_texture_swizzle could, for example, see
GL_LUMINANCE16_SNORM as MESA_FORMAT_SIGNED_R16 with a { r, r, r, 1 }
swizzle. Care would need to be taken to prevent issues with using
those textures for FBO rendering.
This is the rest of the fix for glean's pixelFormats test on i965.
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This class of hardware can natively sample all of the snorm surface
formats that DX10 requires, but it can't do some of the legacy GL
formats. In particular, all of the alpha, luminance, and intensity
formats are unsupported.
This partially fixes the breakage in glean's pixelFormats test since
GL_EXT_texture_snorm support was added to Mesa.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We use this format to represent the accum buffer. No snorm texture
sampling or rendering takes place.
Fixes failed assertion with swrast and any app using the accum buffer
(and glxinfo).
Piglit tests:
- glsl-fs-shadow2d-01
- glsl-fs-shadow2d-02
- glsl-fs-shadow2d-03
- fs-shadow2d-red-01
- fs-shadow2d-red-02
- fs-shadow2d-red-03
NOTE: This is a candidate for the stable branches.
Various documentation mentions that "W" is handed to the WM stage,
but further digging seems to indicate that they really mean 1/W.
The code here is still unclear, but changing this fixes piglit
test "fragcoord_w" on Sandybridge as well as a Khronos ES2 conformance
test. I also tested 3DMarkMobile ES2.0's taiji and hoverjet demos, as
well as Nexuiz, just to be safe.
NOTE: This is a candidate for the 7.10 branch.
Fixes regressions caused by commit 9a21bc6401, namely GPU hangs when
running gnome-shell or compiz (Mesa bugs #35820 and #35853).
I incorrectly refactored the case that dealt with ARF_NULL; even in that
case, the source register needs to be changed to the MRF.
NOTE: This is a candidate for the 7.10 branch (if 9a21bc6401 is
cherry-picked, take this one too).
Branch emulation and loop unrolling are done in the GLSL frontend.
Transforming loops is no longer needed for fragment shaders, but it is still
necessary for vertex shaders.
Registers that are used inside of loops need to be considered live
starting with the first instruction of the outermost loop.
https://bugs.freedesktop.org/show_bug.cgi?id=34370
NOTE: This is a candidate for the 7.9 and 7.10 branches.
Reviewed-by: Eric Anholt <eric@anholt.net>
Oops, the mask was being used in the loop to determine whether to use
include the stencil || depth values. This began to fail when mask was
cleared at the beginning of the loop. So reorder the tests and do the
work up-front along with determining the depth_stencil value to use.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=35822
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
(some little changes by Marek Olšák)
Squashed commit of the following:
commit 737c0c6b7d591ac0fc969a7590e1691eeef0ce5e
Author: Luca Barbieri <luca@luca-barbieri.com>
Date: Fri Aug 27 02:13:57 2010 +0200
draw: disable SSE and PPC paths (use LLVM instead)
These paths don't support vertex clamping, and are anyway
obsoleted by LLVM.
If you want to re-enable them, add vertex clamping and test that it
works with the ARB_color_buffer_float piglit tests.
commit fed3486a7ca0683b403913604a26ee49a3ef48c7
Author: Luca Barbieri <luca@luca-barbieri.com>
Date: Thu Aug 26 18:27:38 2010 +0200
draw_llvm: respect vertex color clamp
commit ef0efe9f3d1d0f9b40ebab78940491d2154277a9
Author: Luca Barbieri <luca@luca-barbieri.com>
Date: Thu Aug 26 18:26:43 2010 +0200
draw: respect vertex clamping in interpreter path
Macro can lead to hard to debug list bugs. For instance consider
the following :
LIST_ADD(item, list->prev)
3 instruction of the macro became :
(list->prev)->next->prev = item
which is equivalent to :
list->prev = item
Thus list prev field changes and next instruction in the macro
(list->prev)->next = item
became :
item->next = item
And you endup with list corruption, other case lead to similar
list corruption. Inline function are not affected by this short
coming
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
When the condition
min_index == 0 && sizeof(ib[0]) == sizeof(draw_elts[0])
was true, we were wrongly ignoring istart and processing indices 0.
Reorder some statements to make the code easier to understand.
Now that we purposefully generate delta that point outside of the target
buffer, the assertion has outlived its usefulness.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Once more! This time without the unwarranted conversion from
drm_intel_bo_alloc_tiled.
Signed-off-by: [a very embarrassed] Chris Wilson <chris@chris-wilson.co.uk>
This reverts commit de7678ef52.
The conversion from using drm_intel_bo_alloc_tiled to a plain
drm_intel_bo_alloc forgot that the tiled variant adjusts the
allocation height even for TILING_NONE.
Reported-by: Dave Airlie <airlied@redhat.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=35786
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>