OSDN Git Service
Eric Anholt [Mon, 23 Aug 2010 01:26:42 +0000 (18:26 -0700)]
glsl: Don't constant-fold in a constant in place of a function outval.
Eric Anholt [Mon, 23 Aug 2010 01:15:20 +0000 (18:15 -0700)]
glsl: Convert constant folding to the rvalue visitor.
This should be mostly a noop, except that a plain dereference of a
variable that is not part of a constant expression could now get
"constant folded". I expect that for all current backends this will
be either a noop, or possibly a win when it provokes more
ir_algebraic. It'll also ensure that when new features are added,
tree walking will work normally. Before this, constants weren't
getting folded inside of loops.
Eric Anholt [Mon, 23 Aug 2010 01:25:55 +0000 (18:25 -0700)]
glsl: Don't tree-graft in an expression in place of a function outval.
Fixes: glsl-constant-folding-call-1 (bug #29737)
Brian Paul [Mon, 23 Aug 2010 01:04:47 +0000 (19:04 -0600)]
st/mesa: fix ReadPixels crashes when reading depth/stencil from a FBO
This is based on a patch from Marek Olšák.
NOTE: This is a candidate for the Mesa 7.8 branch.
Brian Paul [Mon, 23 Aug 2010 00:54:50 +0000 (18:54 -0600)]
mesa: use driver hook for creating new renderbuffers
Brian Paul [Mon, 23 Aug 2010 00:48:28 +0000 (18:48 -0600)]
st/mesa: clean-up pipe_get_transfer() calls
Eric Anholt [Mon, 23 Aug 2010 00:34:18 +0000 (17:34 -0700)]
mesa: AC_SUBST the talloc libs/cflags so the ./configure results are saved.
I had used pkg-config from the Makefile because I didn't want to screw
around with the non-autoconf build, but that doesn't work because the
PKG_CONFIG_PATH or TALLOC_LIBS/TALLOC_CFLAGS that people set at
configure time needs to be respected and may not be present at build
time.
Bug #29585
Luca Barbieri [Sun, 22 Aug 2010 22:16:23 +0000 (00:16 +0200)]
nvfx: fix minor memory leak
Luca Barbieri [Sun, 22 Aug 2010 21:29:34 +0000 (23:29 +0200)]
nvfx: support both sprite coord origins
Now we lie less when claiming OpenGL 2 support.
Also, first piglit result group is now all green, except for
fdo25614-genmipmap, which seems mesa/st's fault.
Luca Barbieri [Sun, 22 Aug 2010 19:41:49 +0000 (21:41 +0200)]
nvfx: use 64-bit bitmasks for temps
Jerome Glisse [Sun, 22 Aug 2010 21:13:58 +0000 (17:13 -0400)]
r600g: fix DB decompression
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Vinson Lee [Sun, 22 Aug 2010 19:45:04 +0000 (12:45 -0700)]
nvfx: Include missing header in nvfx_vertprog.c.
Include draw_context.h for draw_*_vertex_shader symbols.
Fixes the following GCC warning.
nvfx_vertprog.c: In function 'nvfx_vp_state_create':
nvfx_vertprog.c:1276: warning: implicit declaration of function 'draw_create_vertex_shader'
nvfx_vertprog.c:1276: warning: assignment makes pointer from integer without a cast
nvfx_vertprog.c: In function 'nvfx_vp_state_delete':
nvfx_vertprog.c:1298: warning: implicit declaration of function 'draw_delete_vertex_shader'
Jakob Bornecrantz [Sun, 22 Aug 2010 17:58:57 +0000 (19:58 +0200)]
translate_sse: add R32G32B32A32_FLOAT -> X8X8X8X8_UNORM for EMIT_4UB
Changed by me to use movd instead of movss to avoid penalties.
Luca Barbieri [Sun, 22 Aug 2010 16:11:22 +0000 (17:11 +0100)]
translate_sse: refactor constant management
Luca Barbieri [Sun, 22 Aug 2010 14:15:51 +0000 (16:15 +0200)]
nvfx: refactor to support multiple fragment program versions
Luca Barbieri [Sun, 22 Aug 2010 13:48:41 +0000 (15:48 +0200)]
nvfx: move stuff around
Jerome Glisse [Sun, 22 Aug 2010 18:22:00 +0000 (14:22 -0400)]
r600g: depth buffer likely needs decompression when used as texture
Before using depth buffer as texture, it needs to be decompressed
(tile pattern of db are different from one used for colorbuffer
like texture)
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Keith Whitwell [Sun, 22 Aug 2010 13:14:55 +0000 (14:14 +0100)]
glx/xlib: configurable strict/non-strict buffer size invalidate
Introduce a new configuration option XMESA_STRICT_INVALIDATE to switch
between swapbuffers-based and glViewport-based buffer invalidation.
Default strict invalidate to false, ie glViewport-based invalidation,
aka ST_MANAGER_BROKEN_INVALIDATE.
This means we will not call XGetGeometry after every swapbuffers,
which allows swapbuffers to remain asynchronous. For apps running at
100fps with synchronous swapping, a 10% boost is typical. For gears,
I see closer to 20% speedup.
Note that the work of copying data on swapbuffers doesn't disappear -
this change just allows the X server to execute the PutImage
asynchronously without us effectively blocked until its completion.
This applies even to llvmpipe's threaded rasterization as the
swapbuffers operation was a large part of the serial component of an
llvmpipe frame.
The downside of this is correctness - applications which don't call
glViewport on window resizes will get incorrect rendering, unless
XMESA_STRICT_INVALIDATE is set.
The ultimate solution would be to have per-frame but asynchronous
invalidation. Xcb almost looks as if it could provide this, but the
API doesn't seem to quite be there.
Keith Whitwell [Sun, 22 Aug 2010 11:31:18 +0000 (12:31 +0100)]
llvmpipe: reduce size of fragment shader variant key
Don't spend as much time comparing them.
Keith Whitwell [Sun, 22 Aug 2010 11:16:45 +0000 (12:16 +0100)]
llvmpipe: remove unused member from lp_fragment_shader_variant_key
Keith Whitwell [Sun, 22 Aug 2010 10:43:01 +0000 (11:43 +0100)]
llvmpipe: don't clear unused bins
If bins outside the current scene bounds are being corrupted, we'll
need to fix that separately. Currently seems ok though.
Keith Whitwell [Sat, 21 Aug 2010 21:51:38 +0000 (22:51 +0100)]
draw: reduce the size of the llvm variant key
Keith Whitwell [Thu, 19 Aug 2010 23:14:47 +0000 (00:14 +0100)]
glx/xlib: remove another XSync
With this change, xmesa_get_window_size still does one round trip, but
that's better than doing two.
Keith Whitwell [Thu, 19 Aug 2010 23:08:22 +0000 (00:08 +0100)]
glx/xlib: no need to call XSync from XMesaFlush
Try to eliminate some unnecessary X server round trips.
Luca Barbieri [Sun, 22 Aug 2010 10:02:41 +0000 (12:02 +0200)]
nvfx: simplify and correct fragment program update logic
This version should hopefully be much clearer and thus less likely
to be subtly broken.
Also fixes point sprites on nv40 and possibly some other bugs too.
Luca Barbieri [Sun, 22 Aug 2010 09:58:54 +0000 (11:58 +0200)]
nvfx: make stipple setting independent of enable
Luca Barbieri [Sun, 22 Aug 2010 12:53:49 +0000 (14:53 +0200)]
nvfx: fix vertex programs
Luca Barbieri [Sat, 21 Aug 2010 22:21:55 +0000 (00:21 +0200)]
nvfx: use relocations array for vp constants
Henri Verbeet [Mon, 16 Aug 2010 20:18:37 +0000 (22:18 +0200)]
r600g: Don't blindly unmap NULL->size.
There may actually be something mapped in that range, especially for large
buffers like e.g. the GL Drawable.
José Fonseca [Sun, 15 Aug 2010 12:36:02 +0000 (13:36 +0100)]
svga: Do not shortcut NULL surface relocations with SVGA3D_INVALID_ID.
How to cope with NULL surface relocations should be entirely at winsys'
discretion.
Eric Anholt [Sun, 22 Aug 2010 07:47:45 +0000 (00:47 -0700)]
i965: Fix 8-wide FB writes on gen6.
My merge of Zhenyu's patch on top of my previous patches broke it by
my code expecting simd16 single write and Zhenyu's simd8 path being
disabled by mine. Merge the two for success.
Eric Anholt [Sun, 22 Aug 2010 07:44:28 +0000 (00:44 -0700)]
i965: Fix brw_math1 with scalar argument in gen6 FS.
The docs claim two conflicting things: One, that a scalar source is
supported. Two, source hstride must be 1 and width must be exec size.
So splat a constant argument out into a full reg to operate on, since
violating the second set of constraints is clearly failing.
The alternative here might be to do a 1-wide exec on a constant
argument for math1. It would probably save cycles too. But I'll
leave that for the glsl2-965 branch.
Fixes glsl-algebraic-div-one-2.shader_test.
Eric Anholt [Sun, 22 Aug 2010 07:26:09 +0000 (00:26 -0700)]
i965: Fix up WM push constant setup on gen6.
Fixes glsl-algebraic-add-add-1.
Eric Anholt [Sun, 22 Aug 2010 06:47:06 +0000 (23:47 -0700)]
i965: Use intel->gen >= 6 instead of IS_GEN6.
Vinson Lee [Sun, 22 Aug 2010 07:30:47 +0000 (00:30 -0700)]
libgl-xlib: Include missing header in xlib.c.
Include st_api.h for st_api_create_OpenGL symbol.
Vinson Lee [Sun, 22 Aug 2010 07:16:54 +0000 (00:16 -0700)]
nvfx: Silence unused variable warning.
The variable is used but only in the body of an assert.
Vinson Lee [Sun, 22 Aug 2010 07:09:43 +0000 (00:09 -0700)]
mesa: Initialize member variables in ir_to_mesa_src_reg constructor.
The default constructor did not initialize some member variables.
Vinson Lee [Sun, 22 Aug 2010 06:56:24 +0000 (23:56 -0700)]
mesa: Initialize variables in mesa_src_reg_from_ir_src_reg.
Vinson Lee [Sun, 22 Aug 2010 06:36:30 +0000 (23:36 -0700)]
util: Use #ifdef instead of #if.
This is a typo fix of earlier commit
0f3b3751b8643352dcc242567b3696bd1505df1d.
Vinson Lee [Sun, 22 Aug 2010 06:28:52 +0000 (23:28 -0700)]
util: Define dump_cpu only for DEBUG builds.
dump_cpu is used only when DEBUG is defined.
Fixes the following GCC warning on builds without DEBUG defined.
util/u_cpu_detect.c:76: warning: 'debug_get_option_dump_cpu' defined but not used
Vinson Lee [Sun, 22 Aug 2010 06:24:28 +0000 (23:24 -0700)]
translate_sse: Silence uninitialized variable warnings.
Initialize variables on error paths.
Vinson Lee [Sun, 22 Aug 2010 05:59:46 +0000 (22:59 -0700)]
nvfx: Silence uninitialized variable warnings.
Variables weren't initialized on the error paths.
Vinson Lee [Sun, 22 Aug 2010 05:45:09 +0000 (22:45 -0700)]
i965g: Silence printf format warnings on 64-bit builds.
Vinson Lee [Sun, 22 Aug 2010 05:09:47 +0000 (22:09 -0700)]
nvfx: Silence uninitialized variable warnings.
Silence the following i686-apple-darwin10-gcc-4.2.1 warnings.
nv04_2d.c: In function 'nv04_region_copy_cpu':
nv04_2d.c:560: warning: 'dswy' may be used uninitialized in this function
nv04_2d.c:559: warning: 'dswx' may be used uninitialized in this function
nv04_2d.c:562: warning: 'sswy' may be used uninitialized in this function
nv04_2d.c:561: warning: 'sswx' may be used uninitialized in this function
Vinson Lee [Sun, 22 Aug 2010 05:01:04 +0000 (22:01 -0700)]
nv50: Silence incompatible pointer type initialization warning.
Silence the following GCC warning.
warning: initialization from incompatible pointer type
Vinson Lee [Sun, 22 Aug 2010 04:42:17 +0000 (21:42 -0700)]
nv50: Disable unused code.
Disable release_hw and emit_mov_from_pred functions as they are
currently not being used.
Vinson Lee [Sun, 22 Aug 2010 04:27:43 +0000 (21:27 -0700)]
i965g: Fix printf format warning on 32-bit platforms.
Fixes the following GCC warning on 32-bit platforms.
warning: format '%li' expects type 'long int', but argument 4 has type 'int'
Vinson Lee [Sun, 22 Aug 2010 03:38:07 +0000 (20:38 -0700)]
glsl: Silence uninitialized variable warning.
i686-apple-darwin10-gcc-4.2.1 generated the following warning.
warning: 'score' may be used uninitialized in this function
GCC 4.4.3 on Linux didn't generate the above warning.
Jerome Glisse [Sun, 22 Aug 2010 02:49:22 +0000 (22:49 -0400)]
r600g: partialy fix texturing from depth buffer + initial support for untiling
Partialy fix texturing from depth buffer, depth buffer is tiled
following different tile organisation that color buffer. This
properly set the tile type & array mode field of texture sampler
when sampling from db resource.
Add initial support to untiling buffer when transfering them,
it's kind of broken by corruption the vertex buffer of previous
draw.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
José Fonseca [Sun, 22 Aug 2010 01:26:09 +0000 (02:26 +0100)]
draw: Don't assert if indices point outside vertex buffer.
This is valid input, and asserting here does causes the test suites that
verify this to crash.
Also, the assert was wrongly accepting the case
max_index == vert_info->count
which, IIUC, is the first vertex outside the buffer. Assuming the
vert_info->count is precise (which often is not the case).
José Fonseca [Sat, 21 Aug 2010 14:04:47 +0000 (15:04 +0100)]
mesa: Removed another unused variable.
Vinson Lee [Sat, 21 Aug 2010 23:21:41 +0000 (16:21 -0700)]
glsl: Silence unused variable warning.
The variable is actually used but only in the body of an assert.
Vinson Lee [Sat, 21 Aug 2010 22:48:25 +0000 (15:48 -0700)]
util: Silence uninitialized variable warnings.
Kenneth Graunke [Sat, 21 Aug 2010 22:30:34 +0000 (15:30 -0700)]
glsl: Handle array declarations in function parameters.
The 'vec4[12] foo' style already worked, but the 'vec4 foo[12]' style
did not. Also, 'vec4[] foo' was wrongly accepted.
Fixes piglit test cases array-19.vert and array-21.vert.
May fix fd.o bug #29684 (or at least part of it).
Luca Barbieri [Sat, 21 Aug 2010 21:53:39 +0000 (23:53 +0200)]
nvfx: actually fix it properly
Luca Barbieri [Sat, 21 Aug 2010 21:33:51 +0000 (23:33 +0200)]
nvfx: fix incorrect assert
Vinson Lee [Sat, 21 Aug 2010 21:36:29 +0000 (14:36 -0700)]
util: Move loop variable declaration outside for loop.
Fixes build error with MSVC.
Vinson Lee [Sat, 21 Aug 2010 21:29:50 +0000 (14:29 -0700)]
nvfx: Fix SCons build.
Move declarations before code.
Fix void pointer arithmetic.
Luca Barbieri [Sat, 21 Aug 2010 20:48:29 +0000 (22:48 +0200)]
nvfx: fix warnings
José Fonseca [Sat, 21 Aug 2010 20:58:22 +0000 (21:58 +0100)]
gallivm: Emit DIVPS instead of RCPPS.
See comments for detailed rationale.
Thanks to Michal Krol and Zack Rusin for detecting and investigating this
in detail.
Luca Barbieri [Sat, 21 Aug 2010 19:29:18 +0000 (21:29 +0200)]
nvfx: enable translate_sse
Vinson Lee [Sat, 21 Aug 2010 19:32:17 +0000 (12:32 -0700)]
auxiliary: Add missing files to SCons build.
Add u_linear.c and u_linkages.c to SCons build.
Reorder list of files to be more alphabetical.
Vinson Lee [Sat, 21 Aug 2010 19:21:59 +0000 (12:21 -0700)]
auxiliary: Reorder list of files in Makefile.
This patch reorders the list of files so that the order is more alphabetic.
Vinson Lee [Sat, 21 Aug 2010 19:00:57 +0000 (12:00 -0700)]
scons: Fix nvfx build.
Luca Barbieri [Sat, 21 Aug 2010 18:23:41 +0000 (20:23 +0200)]
nvfx: slightly improve handling of overlong vps
Luca Barbieri [Sat, 21 Aug 2010 18:14:35 +0000 (20:14 +0200)]
nvfx: tweak CMP in fp
Luca Barbieri [Sat, 21 Aug 2010 18:14:16 +0000 (20:14 +0200)]
nvfx: implement CMP in vp
Luca Barbieri [Sat, 21 Aug 2010 18:07:48 +0000 (20:07 +0200)]
nvfx: implement TXL in fp
Luca Barbieri [Sat, 21 Aug 2010 18:05:04 +0000 (20:05 +0200)]
nvfx: implement SSG in fp
Luca Barbieri [Sat, 21 Aug 2010 17:43:46 +0000 (19:43 +0200)]
nvfx: implement DP2 in vp and fp
Luca Barbieri [Sat, 21 Aug 2010 17:35:06 +0000 (19:35 +0200)]
nvfx: implement TRUNC in vp and fp
Luca Barbieri [Sat, 21 Aug 2010 17:45:06 +0000 (19:45 +0200)]
nvfx: implement NOP
Luca Barbieri [Sat, 21 Aug 2010 16:37:21 +0000 (18:37 +0200)]
nvfx: add vertex program control flow
Luca Barbieri [Sat, 21 Aug 2010 16:37:01 +0000 (18:37 +0200)]
nvfx: fix vertex shader headers
Luca Barbieri [Fri, 20 Aug 2010 19:16:49 +0000 (21:16 +0200)]
nv40: add fragment program control flow
Luca Barbieri [Sat, 21 Aug 2010 10:32:59 +0000 (12:32 +0200)]
nvfx: refactor shader assembler
Luca Barbieri [Sat, 21 Aug 2010 11:28:38 +0000 (13:28 +0200)]
nvfx: add option to dump shaders in TGSI and native code
Luca Barbieri [Thu, 25 Feb 2010 16:46:37 +0000 (17:46 +0100)]
nvfx: improve and correct nvfx_shader.h
Luca Barbieri [Thu, 19 Aug 2010 20:47:03 +0000 (22:47 +0200)]
nvfx: fix lodbias
Luca Barbieri [Thu, 19 Aug 2010 20:36:00 +0000 (22:36 +0200)]
nvfx: mostly fix inline corruption magically
Not sure why this mostly works.
Luca Barbieri [Thu, 19 Aug 2010 10:58:14 +0000 (12:58 +0200)]
nvfx: fix GPU hardlocks when depth buffer is absent
Luca Barbieri [Mon, 16 Aug 2010 23:01:42 +0000 (01:01 +0200)]
nvfx: fire ring after transfers
Might reduce the risk of running out of memory
Luca Barbieri [Mon, 16 Aug 2010 14:55:00 +0000 (16:55 +0200)]
nv30: band-aid viewport issues
For some reason nv30 seems to like to reset the viewport, even though
attempts to isolate where exactly it does that have currently been
inconclusive.
Luca Barbieri [Sun, 15 Aug 2010 08:15:40 +0000 (10:15 +0200)]
nvfx: support flatshade_first
Luca Barbieri [Sat, 13 Mar 2010 01:28:59 +0000 (02:28 +0100)]
nvfx: expose GLSL
Still no control flow support, but basic stuff works.
Luca Barbieri [Tue, 10 Aug 2010 21:09:53 +0000 (23:09 +0200)]
nvfx: support proper shader linkage - adds glsl support
Luca Barbieri [Sat, 7 Aug 2010 03:39:18 +0000 (05:39 +0200)]
nvfx: rewrite draw code and buffer code
This is a full rewrite of the drawing and buffer management logic.
It offers a lot of improvements:
1. A copy of buffers is now always kept in system memory. This is
necessary to allow software processing of them, which is necessary
or improves performance in many cases.
2. Support for pushing vertices on the FIFO, with index lookup if necessary.
3. "Smart" draw code that tries to intelligently choose the cheapest
way to draw something: whether to use inline vertices or hardware
vertex buffer, and whether to use hardware index buffers
4. Support for all vertex formats supported by the hardware
5. Usage of translate to push vertices, supporting all formats that are
sensible to use as vertex formats
6. Support for base vertex
7. Usage of Ben Skeggs' primitive splitter originally for nv50, allowing
correct splitting of line loops, triangle fans, etc.
8. Support for instancing
9. Precomputation using the vertex elements CSO
Thanks to Ben Skeggs for his primitive splitter originally for nv50.
Thanks to Christoph Bumiller for his nv50 push code, that was the basis
of this work, even though I changed his code dramatically, in particular
to replace his ad-hoc vertex data emitter with translate.
The changes could also go into nv50 too, but there are substantial
differences due to the additional nv50 hardware features.
Luca Barbieri [Sat, 7 Aug 2010 01:47:25 +0000 (03:47 +0200)]
nvfx: refactor sampling code, add support for swizzles and depth tex
This is a significant refactoring of the sampling code that:
- Moves all generic functions in nvfx_fragtex.c
- Adds a driver-specific sampler view structure and uses it to
precompute texture setup as it should be done
- Unifies a bit more of code between nv30 and nv40
- Adds support for sampler view swizzles
- Support for specifying as sampler view format different from the
resource one (only trivially)
- Support for sampler view specification of first and last level
- Support for depth textures on nv30, both for reading depth and
for compare
- Support for sRGB textures
- Unifies the format table between nv30 and nv40
- Expands the format table to include essentially all supportable formats
except mixed sign and "autonormal" formats
- Fixes the "is format supported" logic, which was quite broken, and
makes it use the format table
Only tested on nv30 currently.
Luca Barbieri [Tue, 3 Aug 2010 20:49:19 +0000 (22:49 +0200)]
nvfx: new 2D: unify textures and buffers
Stop using the vtbl, and use real transfers for buffers too.
Luca Barbieri [Sun, 18 Apr 2010 14:43:19 +0000 (16:43 +0200)]
nvfx: new 2D: use a CPU copy for up to 4 pixels, up from 0
Seems a reasonable threshold for now.
Significantly speeds up Piglit's 1x1 glReadPixels (but, you know,
reading pixels in 1x1 blocks is NOT a good idea, especially if you
might be running on a less-than-perfect driver).
Luca Barbieri [Tue, 3 Aug 2010 03:47:41 +0000 (05:47 +0200)]
nvfx: new 2D: new render temporaries with resources
This patch adds support for creating temporary surfaces to allow
rendering to surfaces that cannot be rendered to.
It uses the _second_ version of the render temporary infrastructure.
This is necessary for swizzled 3D textures and small mipmaps of
swizzled 2D textures.
This version of the patch creates a resource to use as a temporary
instead of a raw BO, making the code simpler.
Luca Barbieri [Thu, 11 Mar 2010 17:06:28 +0000 (18:06 +0100)]
nv30: new 2D: support ARB_texture_rectangle
This uses nv30's _RECT formats.
Luca Barbieri [Tue, 23 Feb 2010 11:47:45 +0000 (12:47 +0100)]
nvfx: new 2D: optimize fragtex format lookup
Use an array indexed by the pipe format instead of doing a linear scan.
Luca Barbieri [Mon, 18 Jan 2010 22:40:22 +0000 (23:40 +0100)]
nvfx: new 2D: enable swizzling for all surfaces
Now that the new 2D code is in place, swizzling can be safely enabled.
Render temporaries are needed in some cases, so this may degrade nv30
a bit until it gets render temporaries too.
Luca Barbieri [Tue, 3 Aug 2010 04:24:22 +0000 (06:24 +0200)]
nvfx: new 2D: use new 2D engine in Gallium
This patch implements nv04_surface_copy/fill using the new 2D engine module.
It supports falling back to the 3D engine using the u_blitter module, which will be
added in a later patch.
Also adds support for using the 3D engine, reusing the u_blitter module
created for r300.
This is used for unswizzling and copies between swizzled surfaces.
Luca Barbieri [Tue, 19 Jan 2010 17:51:10 +0000 (18:51 +0100)]
nv04-nv40: new 2D: add new Gallium-independent 2D engine
This patch add a brand new nv04-nv40 2D engine module.
It should correctly implement all operations involving swizzled, and 3D-swizzled surfaces.
This code is independent from the Gallium framework and can thus be reused in the DDX and classic Mesa drivers (it's only likely to be useful in the latter, though).
Currently, surface_copy and surface_fill are broken for 3D textures, for swizzled source textures and possibly for some misaligned cases
The code is based around the new nv04_region structure, which encapsulates the information from pipe_surface needed for the 2D engine and CPU copies.
The use of nv04_region makes the code independent of the Gallium framework and allows to transform the nv04_region without clobbering the nv04_region.
The existing M2MF, blitter, and SWIZZLED_SURFACE paths have been improved and a new CPU path has been added.
There is also support to tell the caller to use the 3D engine.
The main feature of the copy/fill setup algorithm is linearization/contiguous-linearization of swizzled surfaces.
The idea of linearization is that some swizzled surfaces are laid out like linear ones (1xN, 2xN, Nx1) and can thus be used as such (e.g. useful for copying single pixels).
Also, some rectangles (e.g. the whole surface) are contiguous in memory. If both the source and destination rectangles are swizzled but contiguous, then they can be regarded as both linear: this is the idea of "contiguous linearization".
This, for instance, allows to use the 2D engine to duplicate the content of a swizzled surface to another swizzled surface, by pretending they are actually linear.
After linearization, the result may not be 64-byte aligned. Another transformation is done to enlarge the linear surface so that it becomes 64-byte aligned.
This is also used to 64-byte align swizzled texture mipmaps.
The inner loop of the CPU path is as optimized as possible without using SSE/SSE2.
Future improvements could include SSE/SSE2 support, and possibly a faster coordinate swizzling algorithm (which is however not used in the inner loop).
It may be a good idea to autogenerate swizzling code at least for all possible POT 2D texture dimensions (less than 256), maybe for all 3D ones too (less than 4096).
Also, it woud be a very good idea to make a copy with the GPU first if the source surface is in uncached memory.
Luca Barbieri [Tue, 3 Aug 2010 19:32:42 +0000 (21:32 +0200)]
nvfx: new 2D: rewrite transfer code to use staging transfers
This greatly simplifies the code, and avoids ad-hoc copy code.
Also, these new transfers work for buffers too, even though they
are still used for miptrees only.
Luca Barbieri [Fri, 15 Jan 2010 09:11:11 +0000 (10:11 +0100)]
nvfx: new 2D: rewrite miptree code, adapt transfers
Changes:
- Disable swizzling on non-RGBA 2D textures, since the current 2D
code is mostly broken in those cases. A later patch will fix this.
Thanks to Andrew Randrianasulu who reported this.
- Fix compressed texture transfers and hack around the current 2D
code inability to copy compressed textures by using direct access.
Thanks to Andrew Randrianasulu who reported this.
This patch rewrites all the miptree layout and transfer code in the
nvfx driver.
The current code is broken in several ways:
1. 3D textures are laid out first by face, then by level, which is
incorrect
2. Cube maps should have 128-byte aligned faces
3. Swizzled textures have a strange alignment test that seems
unnecessary
4. We store the image_offsets for each face/slice but they can be
easily computed instead
5. "Swizzling" is not supported for compressed formats. They can be
"swizzled" but swizzling only means that there are no gaps (pitch is
level-dependant) and the layout is still linear
6. Swizzling is not supported for non-RGBA formats. All formats (except
possibly depth) can be swizzled according to my testing.
The miptree layout is rewritten based on my empirical testing, which I
posted in the "miptree findings" mail.
The image_offset array is removed, since it can be calculated with a
simple multiplication; the only array in the miptree structure is now
the one for mipmap level starts, which it seems cannot be easily
computed in constant time.
Also, we now directly store a nouveau_bo instead of a pipe_buffer in
the miptree structure, like nv50 does.
Support for render temporaries is removed, and will be readded in a
later patch.
Note that the current temporary code is broken, because it does not
copy the temporary back on render cache flushes.
Luca Barbieri [Fri, 6 Aug 2010 23:56:01 +0000 (01:56 +0200)]
nvfx: add nouveau_resource_on_gpu
Add a function to get whether a resource is likely on the GPU or not.
Currently always returns TRUE.
Luca Barbieri [Tue, 3 Aug 2010 20:50:19 +0000 (22:50 +0200)]
nvfx: add linear flag for buffers