Christoph Bumiller [Wed, 2 Feb 2011 21:04:53 +0000 (22:04 +0100)]
nvc0: fix stride of NVC0_3D_RT methods
Christoph Bumiller [Wed, 2 Feb 2011 15:35:21 +0000 (16:35 +0100)]
nvc0: correct storage type for 16 bit surface formats
Christoph Bumiller [Wed, 9 Feb 2011 13:26:14 +0000 (14:26 +0100)]
nvc0: make CSE work for ops with multiple results
Christoph Bumiller [Tue, 8 Feb 2011 15:55:06 +0000 (16:55 +0100)]
nvc0: replace branching with predicated insns where feasible
Christoph Bumiller [Mon, 7 Feb 2011 20:19:23 +0000 (21:19 +0100)]
nvc0: implement local memory load and store ops
Christoph Bumiller [Mon, 7 Feb 2011 20:17:37 +0000 (21:17 +0100)]
nvc0: make sure phi-ops really have one source per in-block
Christoph Bumiller [Mon, 7 Feb 2011 13:54:17 +0000 (14:54 +0100)]
nv50,nvc0: do not forget to apply sign mode to saved TGSI inputs
Christoph Bumiller [Sun, 6 Feb 2011 12:09:24 +0000 (13:09 +0100)]
nvc0: do not generate a backwards jump if a loop ends with BRK
Christoph Bumiller [Mon, 7 Feb 2011 18:05:55 +0000 (19:05 +0100)]
nvc0: store only one value per basic block for TGSI regs
Christoph Bumiller [Mon, 7 Feb 2011 18:03:09 +0000 (19:03 +0100)]
nv50,nvc0: fix condition code change when commuting SET sources
Christoph Bumiller [Mon, 7 Feb 2011 18:01:54 +0000 (19:01 +0100)]
nvc0: set basic block on manual instruction insertion
Christoph Bumiller [Mon, 7 Feb 2011 17:59:46 +0000 (18:59 +0100)]
nvc0: try to fix register conflicts for vector instructions
Vector here means using multiple 32 bit regs which are forced to be
consecutive in the register file.
This still isn't quite nice.
Christoph Bumiller [Wed, 2 Feb 2011 20:38:07 +0000 (21:38 +0100)]
nvc0: reset texture base address after read transfer
Christoph Bumiller [Wed, 2 Feb 2011 20:23:09 +0000 (21:23 +0100)]
nvc0: don't combine memory loads across block boundaries
Christoph Bumiller [Tue, 1 Feb 2011 19:52:49 +0000 (20:52 +0100)]
nvc0: detect no-op MIN/MAX, do CSE earlier to succeed more often
Thomas Hellstrom [Tue, 8 Feb 2011 10:36:56 +0000 (11:36 +0100)]
mesa/st: Clean up vertex buffer unreferencing
Avoid accessing draw module internal structures outside of the draw module.
Unreference vertex buffers in error path.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Marek Olšák [Wed, 9 Feb 2011 04:20:37 +0000 (05:20 +0100)]
gallium/docs: fix typo
Brian Paul [Wed, 9 Feb 2011 02:25:04 +0000 (19:25 -0700)]
r200: add cast to silence warning
Brian Paul [Wed, 9 Feb 2011 02:25:04 +0000 (19:25 -0700)]
mesa: remove unused BITFIELD64 macros
Brian Paul [Wed, 9 Feb 2011 02:25:04 +0000 (19:25 -0700)]
mesa: remove _mesa_create_context_for_api()
Just add the gl_api parameter to _mesa_create_context().
Brian Paul [Wed, 9 Feb 2011 02:25:04 +0000 (19:25 -0700)]
mesa: remove _mesa_initialize_context_for_api()
Just add the gl_api parameter to _mesa_initialize_context().
Brian Paul [Wed, 9 Feb 2011 02:19:34 +0000 (19:19 -0700)]
mesa: add/update VERBOSE_API logging
Brian Paul [Wed, 9 Feb 2011 01:21:50 +0000 (18:21 -0700)]
st/mesa: fix shader deletion regression
Fixes a regression from commit
5cbff0932e498f49b57cbb71037b93416bfe30e0.
The problem is *some* glDrawPixels fragment programs need to be deleted,
but not all. Use an explicit flag to indicate whether or not the program
needs to be deleted.
This should fix http://bugs.freedesktop.org/show_bug.cgi?id=34049
Kenneth Graunke [Wed, 9 Feb 2011 00:56:07 +0000 (16:56 -0800)]
i965: Add missing DEFINE_BITS for brw dirty bits.
These are only used for debugging, but should be there.
Found by inspection.
Kenneth Graunke [Wed, 9 Feb 2011 00:55:20 +0000 (16:55 -0800)]
i965: Separate the BRW_NEW_(VS|WM)_CONSTBUF dirty bits.
These were incorrectly defined to the same value - likely due to a cut
and paste error. Found by inspection.
Kenneth Graunke [Wed, 9 Feb 2011 00:04:48 +0000 (16:04 -0800)]
i965: Rename a few more commands to match the documentation.
Benjamin Franzke [Tue, 8 Feb 2011 20:12:43 +0000 (21:12 +0100)]
st/egl: Fix platform selection
A break for case _EGL_PLATFORM_X11 is missing.
introduced by:
381ea0d67a6d84a34d23571c49bbf4339ffda364
Eric Anholt [Sat, 5 Feb 2011 17:59:21 +0000 (11:59 -0600)]
i965: Remove pointless keying of WM state on VUE size.
Eric Anholt [Fri, 4 Feb 2011 19:31:02 +0000 (13:31 -0600)]
mesa: Fix the Mesa IR copy propagation to not read past writes to the reg.
Fixes glsl-vs-post-increment-01.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Eric Anholt [Tue, 8 Feb 2011 07:27:03 +0000 (23:27 -0800)]
glsl: Disable the new copy propagation pass until it gets fixed.
It apparently regressed a bunch of ES2 cases.
Chad Versace [Fri, 4 Feb 2011 20:18:56 +0000 (12:18 -0800)]
glsl: Set operators '%' and '%=' to be reserved when GLSL < 1.30
From section 5.9 of the GLSL 1.20 spec:
The operator modulus (%) is reserved for future use.
From section 5.8 of the GLSL 1.20 spec:
The assignments modulus into (%=), left shift by (<<=), right shift by
(>>=), inclusive or into ( |=), and exclusive or into ( ^=). These
operators are reserved for future use.
The GLSL ES 1.00 spec and GLSL 1.10 spec have similiar language.
Fixes bug:
https://bugs.freedesktop.org//show_bug.cgi?id=33916
Fixes Piglit tests:
spec/glsl-1.00/compiler/arithmetic-operators/modulus-00.frag
spec/glsl-1.00/compiler/assignment-operators/modulus-assign-00.frag
spec/glsl-1.10/compiler/arithmetic-operators/modulus-00.frag
spec/glsl-1.10/compiler/assignment-operators/modulus-assign-00.frag
spec/glsl-1.20/compiler/arithmetic-operators/modulus-00.frag
spec/glsl-1.20/compiler/assignment-operators/modulus-assign-00.frag
Marek Olšák [Tue, 8 Feb 2011 17:18:13 +0000 (18:18 +0100)]
r600g: fixup assertion
Marek Olšák [Tue, 8 Feb 2011 16:36:48 +0000 (17:36 +0100)]
r600g: add a faster implementation of transfer_inline_write
u_default_transfer_inline_write uses util_copy_rect, which is kinda slow.
Marek Olšák [Tue, 8 Feb 2011 16:30:39 +0000 (17:30 +0100)]
r600g: slab-allocate buffer and transfer structures
Marek Olšák [Tue, 8 Feb 2011 15:34:22 +0000 (16:34 +0100)]
r300g: use the same upload buffer for vertices and indices
Marek Olšák [Tue, 8 Feb 2011 14:21:35 +0000 (15:21 +0100)]
r600g: use the same upload buffer for vertices, indices, and constants
This should reduce memory consumption.
Thomas Hellstrom [Tue, 8 Feb 2011 09:14:22 +0000 (01:14 -0800)]
mesa/st: Plug a fragment program variant parameter leak
Fixes a minor memory leak with the "engine" mesa demo.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Brian Paul <brianp@vmware.com>
Brian Paul [Tue, 8 Feb 2011 15:05:14 +0000 (08:05 -0700)]
st/mesa: free the temporary bitmap/drawpix shader code
Fixes a per-shader memory leak when drawing glBitmaps, glDrawPixels
or glCopyPixels.
NOTE: This is a candidate for the 7.9 and 7.10 branches.
Marek Olšák [Tue, 8 Feb 2011 14:20:11 +0000 (15:20 +0100)]
u_vbuf_mgr: add a way to specify the BIND flag for the upload buffer
Marek Olšák [Tue, 8 Feb 2011 14:18:10 +0000 (15:18 +0100)]
u_vbuf_mgr: remove tabs
Marek Olšák [Tue, 8 Feb 2011 14:08:04 +0000 (15:08 +0100)]
u_vbuf_mgr: make the uploader public
Marek Olšák [Tue, 8 Feb 2011 11:09:29 +0000 (12:09 +0100)]
Revert "r600g: do not flush the uploader" (with comments)
This reverts commit
1c2a4f0820ff2272f993e6da28dcf8bcbbc3252a.
Brian Paterni [Tue, 8 Feb 2011 01:57:56 +0000 (19:57 -0600)]
r600g: silence a few valgrind warnings
Thomas Hellstrom [Mon, 7 Feb 2011 14:37:05 +0000 (15:37 +0100)]
mesa/st: Fix vertex buffer leak
Make sure we unreference the vertex buffer pointers in a local array.
This fixes huge vertex buffer / memory leaks in mesa demos "fire" and "engine".
NOTE: This is a candidate for the 7.9 and 7.10 branches.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Kristian Høgsberg [Tue, 8 Feb 2011 01:49:58 +0000 (20:49 -0500)]
wayland-egl: Add struct wl_egl_display argument to +wl_egl_window_create()
Henri Verbeet [Mon, 7 Feb 2011 14:22:08 +0000 (15:22 +0100)]
r600g: Add support for relative addressing on constant buffers.
Relative addressing of constant buffers can't work properly through the
kcache, since you can only address within the currently locked kcache window.
Instead, this patch binds the constant buffer as a shader resource, and then
explicitly fetches the constant using a vertex fetch with fetch type
VTX_FETCH_NO_INDEX_OFFSET from the shader. There's probably still some room
for improvement, doing the fetch right before the instruction that needs the
value may not be quite optimal for example.
Henri Verbeet [Mon, 7 Feb 2011 14:22:08 +0000 (15:22 +0100)]
r600g: Set the fetch type in r600_bc_vtx_build().
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Handle the ADD_INT instruction in r600_bc_get_num_operands().
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Generalize the pipe_add_vertex_attrib() functions.
This allows them to be used for VS or PS buffer resources as well.
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Remove vs_resource and ps_resource from the pipe context.
These are practically unused, only the vs_resource array is being abused for
fetch shader resources.
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Split constants in r600_shader_from_tgsi().
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Do the tgsi_full_src_register to r600_shader_src conversion in r600_shader_from_tgsi().
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Split r600_bc_alu_src.
The r600_bc_alu_src structure is used in two different ways, as a vector and
for the individual channels of that same vector. This is somewhat fragile,
and probably confusing.
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: Store literal values in the r600_bc_alu_src structure.
This is much easier to work with, and allows use to get rid of some of the
literal handling hacks.
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: tgsi_dst() can't fail.
Henri Verbeet [Mon, 7 Feb 2011 14:22:07 +0000 (15:22 +0100)]
r600g: tgsi_src() can't fail.
Kristian Høgsberg [Fri, 4 Feb 2011 20:37:51 +0000 (15:37 -0500)]
wayland-egl: Force roundtrips to get device name and authenticate correctly
If the client hasn't done the initial wl_display_iterate() at the time
we initialize the display, we have to do that in platform_wayland.c.
Make sure we detect that correctly instead of dup()ing fd=0, and use
the sync callback to make sure we don't wait forever for authorization that
won't happen.
Benjamin Franzke [Fri, 4 Feb 2011 11:39:40 +0000 (12:39 +0100)]
egl_dri2: Add wayland platform
Benjamin Franzke [Fri, 4 Feb 2011 11:38:58 +0000 (12:38 +0100)]
egl_dri2: Export dri2_get_driver_for_fd
Benjamin Franzke [Fri, 4 Feb 2011 11:32:30 +0000 (12:32 +0100)]
egl_dri2: Enable pixmap bind_to_texture according to the extension
Benjamin Franzke [Fri, 4 Feb 2011 11:32:05 +0000 (12:32 +0100)]
st/egl: Add wayland platform
Benjamin Franzke [Fri, 4 Feb 2011 11:27:18 +0000 (12:27 +0100)]
st/egl: drm_image: Check for MESA_drm_image
MESA_drm_image isnt limited to drm platform,
others can enable the extension too.
Benjamin Franzke [Fri, 4 Feb 2011 11:25:04 +0000 (12:25 +0100)]
st/egl: native_helper: Add resource_surface_import_resource
Benjamin Franzke [Fri, 4 Feb 2011 11:24:08 +0000 (12:24 +0100)]
egl: Implement libwayland-egl
This library is required and defined by wayland for
EGL implementations supporting wayland.
Benjamin Franzke [Fri, 4 Feb 2011 11:22:58 +0000 (12:22 +0100)]
egl: Add wayland platform
Benjamin Franzke [Fri, 4 Feb 2011 11:01:31 +0000 (12:01 +0100)]
intel: Implement dri2::{Allocate,Release}Buffer
Benjamin Franzke [Fri, 4 Feb 2011 11:00:46 +0000 (12:00 +0100)]
st/dri: Implement dri2::{Allocate,Release}Buffer
Benjamin Franzke [Fri, 4 Feb 2011 10:59:12 +0000 (11:59 +0100)]
Add dri2::{Allocate,Release}Buffer extension
Marek Olšák [Mon, 7 Feb 2011 02:46:25 +0000 (03:46 +0100)]
r600g: correctly report supported vertex formats
Marek Olšák [Mon, 7 Feb 2011 01:00:44 +0000 (02:00 +0100)]
r300g: use the new vertex buffer manager
Marek Olšák [Mon, 7 Feb 2011 01:00:56 +0000 (02:00 +0100)]
r600g: use the new vertex buffer manager
Marek Olšák [Sat, 29 Jan 2011 19:53:57 +0000 (20:53 +0100)]
util: import a new vertex buffer manager
This code has originally matured in r300g and was ported to r600g several
times. It was obvious it's a code duplication.
See also comments in the header file.
Marek Olšák [Sun, 6 Feb 2011 20:12:19 +0000 (21:12 +0100)]
r600g: do not flush the uploader
Marek Olšák [Sun, 6 Feb 2011 20:09:38 +0000 (21:09 +0100)]
r300g: do not flush the uploader
We don't have to unmap and recreate the upload buffer when a flush occurs.
This should also prevent buffer allocations from failing.
Marek Olšák [Sun, 6 Feb 2011 14:42:55 +0000 (15:42 +0100)]
configure.ac: correctly check for libdrm_radeon version
Marek Olšák [Wed, 2 Feb 2011 05:46:24 +0000 (06:46 +0100)]
r300g: RS400 doesn't have ZMASK
Dave Airlie [Sun, 6 Feb 2011 08:57:11 +0000 (18:57 +1000)]
r600g: use surface format not underlying texture format
This uses the surface format to set the CB up not the underlying texture
format, since these can and do differ.
Fixes piglit fbo-srgb.
Tom Stellard [Sun, 6 Feb 2011 06:39:58 +0000 (22:39 -0800)]
r300/compiler: Disable register rename pass on r500
The scheduler and the register allocator are not good enough yet to deal
with the effects of the register rename pass. This was causing a 50%
performance drop in Lightsmark. The pass can be re-enabled once the
scheduler and the register allocator are more mature. r300 and r400
still need this pass, because it prevents a lot of shaders from using
too many texture indirections.
NOTE: This is a candidate for the 7.10 branch.
Tom Stellard [Sat, 5 Feb 2011 08:27:24 +0000 (00:27 -0800)]
r300/compiler: Don't count BEGIN_TEX instructions in the compiler stats
Dave Airlie [Fri, 28 Jan 2011 04:16:00 +0000 (14:16 +1000)]
mesa/965: add support for GL_EXT_framebuffer_sRGB (v2)
This adds i965 support for GL_EXT_framebuffer_sRGB, it introduces a new
constant to say that the driver can support sRGB enabled FBOs since enabling
the extension doesn't mean the driver can actually support sRGB.
Also adds the suggested state flush in the core code suggested by Brian.
fix the ARB_fbo color encoding.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Ian Romanick [Fri, 4 Feb 2011 01:56:44 +0000 (17:56 -0800)]
mesa: Fix error checks in GetVertexAttrib functions
Querying index zero is not an error in OpenGL ES 2.0.
Querying an index larger than the value returned by
GL_MAX_VERTEX_ATTRIBS is an error in all APIs.
Fixes bugzilla #32375.
Ian Romanick [Fri, 4 Feb 2011 01:10:14 +0000 (17:10 -0800)]
linker: Generate link errors when ES shaders are missing stages
ES requires that a vertex shader and a fragment shader be present.
Fixes bugzilla #32214.
Vinson Lee [Fri, 4 Feb 2011 19:47:01 +0000 (11:47 -0800)]
glsl: Add opt_copy_propagation_elements.cpp to SConscript.
Fixes SCons build.
Eric Anholt [Mon, 31 Jan 2011 05:13:17 +0000 (21:13 -0800)]
i965: Drop the dead tracking of color_regions[].
We pull the draw regions right out of the renderbuffers these days.
Eric Anholt [Thu, 20 Jan 2011 03:52:15 +0000 (19:52 -0800)]
i965: Drop the INTEL_DEBUG=state spam about the cache size check.
There's way more interesting info in INTEL_DEBUG=state if you could find
it among the state size checks.
Eric Anholt [Sat, 29 Jan 2011 21:59:14 +0000 (07:59 +1000)]
glsl: Remove extra checks for constant true assignment conditions.
These are already stripped by opt_constant_folding.cpp.
Eric Anholt [Sat, 29 Jan 2011 21:50:21 +0000 (07:50 +1000)]
glsl: Fix a comment typo in copy propagation.
Eric Anholt [Tue, 25 Jan 2011 00:28:13 +0000 (10:28 +1000)]
glsl: Add a new opt_copy_propagation variant that does it channel-wise.
This patch cleans up many of the extra copies in GLSL IR introduced by
i965's scalarizing passes. It doesn't result in a statistically
significant performance difference on nexuiz high settings (n=3) or my
demo (n=10), due to brw_fs.cpp's register coalescing covering most of
those extra moves anyway. However, it does make the debug of wine's
GLSL shaders much more tractable, and reduces instruction count of
glsl-fs-convolution-2 from 376 to 288.
Vinson Lee [Fri, 4 Feb 2011 06:10:16 +0000 (22:10 -0800)]
ralloc: Add missing va_end following va_copy.
Dave Airlie [Fri, 4 Feb 2011 05:26:41 +0000 (15:26 +1000)]
r600g: don't set tile_type on evergreen.
Since we never bind the actual DB to the CB/texture only the flushed one
we don't need to track the tile type at the moment.
Dave Airlie [Fri, 4 Feb 2011 05:26:09 +0000 (15:26 +1000)]
r600g: fix evergreen sampler view + depth interaction
Vinson Lee [Fri, 4 Feb 2011 04:15:25 +0000 (20:15 -0800)]
util: Change u_get_transfer_vtbl usage argument type to match prototype.
The type of u_get_transfer_vtbl of the usage argument in u_transfer.h is
unsigned and not enum pipe_transfer_usage. This patch changes the type
of usage to unsigned to match the prototype in the header file.
Vinson Lee [Fri, 4 Feb 2011 03:19:12 +0000 (19:19 -0800)]
glsl: Add using statements for standard library functions.
Standard library functions in C++ are in the std namespace. When using
C++-style header files for the standard library, some compilers, such as
Sun Studio, provide symbols only for the std namespace and not for the
global namespace.
This patch adds using statements for standard library functions. Another
option could have been to prepend standard library function calls with
'std::'.
This patch fixes several compilation errors with Sun Studio.
Dave Airlie [Thu, 3 Feb 2011 23:38:01 +0000 (09:38 +1000)]
r600g: get offset for correct texture when setting up CB.
this fixes the mipmap tests with tiling forced on.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Thu, 3 Feb 2011 23:36:02 +0000 (09:36 +1000)]
r600g: avoid trying to flush the flushing texture.
Since these textures still have the depth bit set.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Thu, 3 Feb 2011 23:34:32 +0000 (09:34 +1000)]
r600g: evergreen CB check for flushed texture
Dave Airlie [Thu, 3 Feb 2011 23:08:32 +0000 (09:08 +1000)]
r600g: flushing texture needs all levels.
For mipmap generation we need all levels in the flushing texture.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Thu, 3 Feb 2011 23:08:18 +0000 (09:08 +1000)]
r600g: remove unused variables
Dave Airlie [Thu, 3 Feb 2011 23:07:08 +0000 (09:07 +1000)]
r600g: add a flag to just create flushed texture without flushing.
This just adds a flag to create the texture without doing any
flushing to it. Flushing occurs in the draw function. This avoids
unnecessary flushes when we end up rebinding a CB/DB/texture due
to the blitter just restoring state.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Thu, 3 Feb 2011 23:06:02 +0000 (09:06 +1000)]
r600g: also check CB bindings for textures to depth flush.
This checks the color buffer bindings to make sure there is something
to flush.
Signed-off-by: Dave Airlie <airlied@redhat.com>