Jonathan Adamczewski [Wed, 14 Jan 2009 01:37:46 +0000 (12:37 +1100)]
cell: Specify constant as float for CEILF().
Without the f, the constant is treated as a double, resulting in
slower arithmetic and libgcc conversion calls each time CEILF()
is used.
Alan Hourihane [Tue, 13 Jan 2009 23:59:18 +0000 (23:59 +0000)]
glsl: fix a comment typo
Alan Hourihane [Tue, 13 Jan 2009 23:54:46 +0000 (23:54 +0000)]
glsl: support sampler arrays.
Ben Skeggs [Tue, 13 Jan 2009 03:25:14 +0000 (13:25 +1000)]
nv50: fix progs/tests/manytex
Previously all squares were textured with the same texture.. not quite what
the demo was supposed to look like!
Ben Skeggs [Tue, 13 Jan 2009 03:19:22 +0000 (13:19 +1000)]
nv50: get glxgears showing all 3 gears instead of 1!!
This fixes a lot of other things where not all the geometry got drawn
also.
Ben Skeggs [Tue, 13 Jan 2009 02:49:53 +0000 (12:49 +1000)]
nv50: change some magic reg, makes more things work
No real idea what this does.. but a lot of things that misrendered and
made the GPU throw a DATA_ERROR now work.. I'm wondering what side-effects
we'll see from this :)
Ben Skeggs [Tue, 13 Jan 2009 01:44:30 +0000 (11:44 +1000)]
nv50: add DXTn formats
Ben Skeggs [Tue, 13 Jan 2009 00:58:17 +0000 (10:58 +1000)]
nv50: shadow mapping
Ben Skeggs [Tue, 13 Jan 2009 00:55:06 +0000 (10:55 +1000)]
nv50: aniso
Ben Skeggs [Tue, 13 Jan 2009 00:44:52 +0000 (10:44 +1000)]
nv50: occlusion queries
Not quite working, but the general idea is right I think.
Ben Skeggs [Mon, 12 Jan 2009 23:56:40 +0000 (09:56 +1000)]
nv50: implement KIL enough for progs/fp/kil to work
Jonathan Adamczewski [Tue, 13 Jan 2009 03:02:18 +0000 (14:02 +1100)]
cell: Add missing suffix to SHUFFLE macro
Jonathan Adamczewski [Mon, 12 Jan 2009 23:24:49 +0000 (16:24 -0700)]
cell: allocate batch buffers w/ 16-byte alignment
Replace cell_batch{align,alloc)*() with cell_batch_alloc16(), allocating
multiples of 16 bytes that are 16 byte aligned.
Opcodes are stored in preferred slot of SPU machine word.
Various structures are explicitly padded to 16 byte multiples.
Added STATIC_ASSERT().
Michel Dänzer [Mon, 12 Jan 2009 14:05:05 +0000 (15:05 +0100)]
gallivm: Make sure the bitcode buffer is followed by a 0 byte.
May fail to parse otherwise.
Michel Dänzer [Mon, 12 Jan 2009 11:39:31 +0000 (12:39 +0100)]
gallivm: Print error message from ParseBitcodeFile() in case it fails.
Michel Dänzer [Mon, 12 Jan 2009 11:37:13 +0000 (12:37 +0100)]
gallivm: Explicitly specify the LLVM components we need.
Michel Dänzer [Mon, 12 Jan 2009 11:34:27 +0000 (12:34 +0100)]
gallivm: Adapt to header file move in LLVM 2.4.
Ben Skeggs [Mon, 12 Jan 2009 06:47:17 +0000 (16:47 +1000)]
nv50: make rtt work again
Ben Skeggs [Mon, 12 Jan 2009 06:32:49 +0000 (16:32 +1000)]
nv50: fix handling of depth textures
Ben Skeggs [Mon, 12 Jan 2009 06:24:42 +0000 (16:24 +1000)]
nv50: another typo..
Ben Skeggs [Mon, 12 Jan 2009 06:15:58 +0000 (16:15 +1000)]
nouveau: fix warning
Ben Skeggs [Mon, 12 Jan 2009 05:42:20 +0000 (15:42 +1000)]
nv50: fix assertion failure
Ben Skeggs [Mon, 12 Jan 2009 05:19:35 +0000 (15:19 +1000)]
nv50: remove previous hack to manage tiled surfaces
Ben Skeggs [Mon, 12 Jan 2009 05:06:15 +0000 (15:06 +1000)]
nv50: fix a typo and a thinko
Ben Skeggs [Mon, 12 Jan 2009 04:27:51 +0000 (14:27 +1000)]
nv50: enable npot textures
Ben Skeggs [Mon, 12 Jan 2009 04:26:15 +0000 (14:26 +1000)]
nv50: disable shader debug
Ben Skeggs [Mon, 12 Jan 2009 04:10:24 +0000 (14:10 +1000)]
nv50: any cpu access to a texture is done on its backing images
Still a little dodgy:
- RTT will hit an assertion (hopefully!) and fail
- 3D textures with depth >= 32 will cause bad things to happen
Ben Skeggs [Mon, 12 Jan 2009 03:42:19 +0000 (13:42 +1000)]
nv50: create buffers for each image that makes up a texture
Ben Skeggs [Mon, 12 Jan 2009 03:27:13 +0000 (13:27 +1000)]
nouveau: return buffer map to something sane.
Sorry, but no, we're not doing this.. Correctness always takes precedence
over speed. Implement this higher up where you know it's safe to do so,
and doesn't break other things in the process.
Ben Skeggs [Mon, 12 Jan 2009 03:25:28 +0000 (13:25 +1000)]
nouveau: use usage, not uninitialised flags value...
Brian Paul [Sun, 11 Jan 2009 22:28:38 +0000 (15:28 -0700)]
cell: optimize unpack_colors() function, saving 12 cycles
Brian Paul [Sun, 11 Jan 2009 22:18:28 +0000 (15:18 -0700)]
cell: move color unpacking code into separate function
Brian Paul [Sun, 11 Jan 2009 22:11:00 +0000 (15:11 -0700)]
cell: re-order the z/stencil fetch/extract/convert instructions for better perf
The new instruction order is 10 cycles faster.
Brian Paul [Sun, 11 Jan 2009 21:22:00 +0000 (14:22 -0700)]
cell: datatype clean-ups in SPE rtasm
Brian Paul [Sun, 11 Jan 2009 21:06:39 +0000 (14:06 -0700)]
cell: simplify the 'optional register' code
Brian Paul [Sun, 11 Jan 2009 20:52:58 +0000 (13:52 -0700)]
cell: asst datatype clean-ups
Brian Paul [Sun, 11 Jan 2009 20:40:28 +0000 (13:40 -0700)]
cell: move depth/stencil code into separate function
Brian Paul [Sun, 11 Jan 2009 20:23:44 +0000 (13:23 -0700)]
cell: clean-up, re-indent, comments
Brian Paul [Sat, 10 Jan 2009 05:32:11 +0000 (22:32 -0700)]
cell: use tgsi_dump_instruction() instead of spe_comment()
Brian Paul [Sat, 10 Jan 2009 05:03:48 +0000 (22:03 -0700)]
gallium: use tgsi_dump_instruction() instead of ppc_comment()
Brian Paul [Sat, 10 Jan 2009 04:51:22 +0000 (21:51 -0700)]
gallium: remove unused struct type
Brian Paul [Sat, 10 Jan 2009 04:48:54 +0000 (21:48 -0700)]
gallium: fix register clobber bug in TGSI->PPC codegen
When negating a src vector that's stored in a altivec register, need to put
negated value into a new register so we don't upset the original value.
This solves the dark colors in the mandelbrot GLSL demo.
Also, use new predicate functions to check if a TGSI temp is stored in
an altivec register.
Brian Paul [Sat, 10 Jan 2009 04:46:08 +0000 (21:46 -0700)]
gallium: emit comments in TGSI->PPC codegen
Brian Paul [Sat, 10 Jan 2009 04:42:58 +0000 (21:42 -0700)]
gallium: code to dump/debug PPC code (disabled)
Brian Paul [Sat, 10 Jan 2009 04:42:17 +0000 (21:42 -0700)]
gallium: added comment/annotation support to PPC rtasm
Brian Paul [Sat, 10 Jan 2009 04:00:15 +0000 (21:00 -0700)]
cell: added rule to produce .s assembly files
Brian Paul [Sat, 10 Jan 2009 03:57:14 +0000 (20:57 -0700)]
gallium: s/false/FALSE/
Younes Manton [Sat, 10 Jan 2009 19:11:30 +0000 (14:11 -0500)]
mesa: Update .gitignore
Younes Manton [Sat, 10 Jan 2009 19:09:59 +0000 (14:09 -0500)]
nouveau: Update nv30 swizzling.
Younes Manton [Sat, 10 Jan 2009 18:30:29 +0000 (13:30 -0500)]
nouveau: Factor out common winsys bits into libnouveaudrm.a
Younes Manton [Sat, 20 Dec 2008 22:37:24 +0000 (17:37 -0500)]
g3dvl: Use Gallium thread wrappers.
Younes Manton [Sat, 20 Dec 2008 22:09:55 +0000 (17:09 -0500)]
g3dvl: Use Gallium MALLOC wrappers.
Younes Manton [Sat, 20 Dec 2008 21:31:29 +0000 (16:31 -0500)]
g3dvl: Get rid of old unbuffered motion compensation code.
Younes Manton [Sat, 20 Dec 2008 21:30:33 +0000 (16:30 -0500)]
g3dvl: Map vertex bufs once per frame/flush.
Younes Manton [Sat, 20 Dec 2008 19:42:29 +0000 (14:42 -0500)]
nouveau: Catch some more leaks.
Younes Manton [Fri, 12 Dec 2008 04:04:51 +0000 (23:04 -0500)]
g3dvl: Fix some memory leaks.
Younes Manton [Thu, 11 Dec 2008 23:05:59 +0000 (18:05 -0500)]
g3dvl: Allocate one set of bufs, let winsys rename them as necessary.
Younes Manton [Thu, 11 Dec 2008 22:55:16 +0000 (17:55 -0500)]
nouveau: Add busy() query, determines if BOs can be mapped immediately.
Younes Manton [Sat, 6 Dec 2008 20:45:00 +0000 (15:45 -0500)]
nouveau: Use PIPE_BUFFER_USAGE_CPU_* instead of custom.
Younes Manton [Fri, 5 Dec 2008 07:27:35 +0000 (02:27 -0500)]
g3dvl: Expand YCbCr to full RGB range by default.
Younes Manton [Fri, 5 Dec 2008 07:27:10 +0000 (02:27 -0500)]
g3dvl: Work around mplayer weirdness in XvMCPutSurface().
Younes Manton [Fri, 5 Dec 2008 07:26:47 +0000 (02:26 -0500)]
g3dvl: Flag textures holding incoming data as dynamic.
Younes Manton [Fri, 5 Dec 2008 07:26:07 +0000 (02:26 -0500)]
nouveau: Put dynamic textures in GART for CPU access and don't swizzle.
Also flag shadows as dynamic since they're for CPU access as well.
Younes Manton [Fri, 5 Dec 2008 07:22:56 +0000 (02:22 -0500)]
gallium: Define PIPE_TEXTURE_USAGE_DYNAMIC.
Knowing how the client intends to use the texture will give the
driver the opportunity to optimize for such cases.
Younes Manton [Fri, 5 Dec 2008 07:21:55 +0000 (02:21 -0500)]
nouveau: Swizzle textures larger than nv04 SIFM limit in parts.
Limit of SIFM on nv40 is 1024x1024, not sure about others.
José Fonseca [Fri, 9 Jan 2009 19:08:56 +0000 (19:08 +0000)]
draw: Add missing include.
Alan Hourihane [Fri, 9 Jan 2009 11:23:47 +0000 (11:23 +0000)]
st: build fix
Brian Paul [Fri, 9 Jan 2009 00:21:20 +0000 (17:21 -0700)]
mesa: 7.3-rc-1 Makefile changes
Brian Paul [Fri, 9 Jan 2009 00:20:41 +0000 (17:20 -0700)]
mesa: latest glxext.h header, no version change
Brian Paul [Fri, 9 Jan 2009 00:20:18 +0000 (17:20 -0700)]
mesa: import glext.h version 44
Brian Paul [Fri, 9 Jan 2009 00:19:51 +0000 (17:19 -0700)]
docs: dri2proto, libdrm tweaks
Brian Paul [Fri, 9 Jan 2009 00:07:28 +0000 (17:07 -0700)]
glsl: fix typo in the vec2 += operator function
Brian Paul [Thu, 8 Jan 2009 23:16:36 +0000 (16:16 -0700)]
mesa: set version string to 7.3-rc1
Brian Paul [Thu, 8 Jan 2009 23:14:19 +0000 (16:14 -0700)]
docs: more 7.3 doc updates
Brian Paul [Thu, 8 Jan 2009 23:12:23 +0000 (16:12 -0700)]
docs: import 7.2 relnotes, start on 7.3 relnotes
Brian Paul [Thu, 8 Jan 2009 22:32:02 +0000 (15:32 -0700)]
glsl: fix broken +=, -=, *=, /= operators
These functions need to return the final computed value.
Now expressions such as a = (b += c) work properly.
Also, no need to use __asm intrinsics in these functions. The resulting
code is the same when using ordinary arithmetic operators and is more legible.
Brian Paul [Wed, 7 Jan 2009 19:31:14 +0000 (12:31 -0700)]
mesa: fix off-by-one bug in _mesa_delete_instructions()
Eric Anholt [Thu, 8 Jan 2009 00:56:02 +0000 (16:56 -0800)]
i965: Fix GLSL FS DPH to return the right value instead of src0.w * src1.w.
Eric Anholt [Wed, 7 Jan 2009 22:26:11 +0000 (14:26 -0800)]
i965: Remove worrisome comment about _NEW_PROGRAM signaling fp change.
Everything now depends on either BRW_NEW_FRAGMENT_PROGRAM or
BRW_NEW_VERTEX_PROGRAM.
Eric Anholt [Wed, 7 Jan 2009 22:09:07 +0000 (14:09 -0800)]
mesa: Remove _Active and _UseTexEnvProgram flags from fragment programs.
There was a note in state.c about _Active deserving to die, and there were
potential issues with it due to i965 forgetting to set _UseTexEnvProgram.
Removing both simplifies things.
Reviewed-by: Brian Paul <brianp@vmware.com>
Eric Anholt [Wed, 7 Jan 2009 21:52:51 +0000 (13:52 -0800)]
i965: Remove dead brw_vs_tnl.c
Brian Paul [Thu, 8 Jan 2009 01:45:49 +0000 (18:45 -0700)]
i965: allow gl_FragData[0] usage when there's only one color buffer
If gl_FragData[0] is written but not gl_FragCOlor, use the former.
Brian Paul [Thu, 8 Jan 2009 01:44:41 +0000 (18:44 -0700)]
mesa: additional case in file_string()
Brian Paul [Thu, 8 Jan 2009 01:44:00 +0000 (18:44 -0700)]
glsl: pass GLcontext::Extension info down into GLSL preprocessor
Now the #extension directives can be handled properly.
Brian Paul [Thu, 8 Jan 2009 01:41:54 +0000 (18:41 -0700)]
glsl: bump up MAX_FOR_LOOP_UNROLL_COMPLEXITY
Brian Paul [Thu, 8 Jan 2009 01:22:56 +0000 (18:22 -0700)]
glsl: check that the fragment shader does not write both gl_FragColor and gl_FragData[]
Brian Paul [Wed, 7 Jan 2009 22:06:06 +0000 (15:06 -0700)]
i965: init dst reg RelAddr field to zero
Eric Anholt [Wed, 7 Jan 2009 20:38:34 +0000 (12:38 -0800)]
i965: Note when we drop saturate mode on the floor in a VP.
Eric Anholt [Wed, 7 Jan 2009 20:37:58 +0000 (12:37 -0800)]
i965: Add support for LRP in VPs.
Bug #19226.
Brian Paul [Wed, 7 Jan 2009 15:56:10 +0000 (08:56 -0700)]
glsl: disable some unused functions (but don't remove just yet)
Brian Paul [Wed, 7 Jan 2009 15:54:09 +0000 (08:54 -0700)]
glsl: also unroll loops with variable declarations such as "for (int i = 0; ..."
Brian Paul [Wed, 7 Jan 2009 15:32:21 +0000 (08:32 -0700)]
glsl: remove dead code
Brian Paul [Wed, 7 Jan 2009 15:25:59 +0000 (08:25 -0700)]
mesa: OSMesa Makefile fixes (use LIB_DIR)
Brian Paul [Wed, 7 Jan 2009 00:36:20 +0000 (17:36 -0700)]
glsl: loop unroll adjustments
Add a "max complexity" heuristic to allow unrolling long loops with small
bodies and short loops with large bodies.
The loop unroll limits may need further tweaking...
Brian Paul [Wed, 7 Jan 2009 00:24:23 +0000 (17:24 -0700)]
glsl: implement loop unrolling for simple 'for' loops
Loops such as this will be unrolled:
for (i = 0; i < 4; ++i) {
body;
}
where 'body' isn't too large.
This also helps to fix the issue reported in bug #19190. The problem there
is indexing vector types with a variable index. For example:
vec4 v;
v[2] = 1.0; // equivalent to v.z = 1.0
v[i] = 2.0; // variable index into vector!!
Since the for-i loop can be unrolled, we can avoid the problems associated
with variable indexing into a vector (at least in this case).
Brian Paul [Tue, 6 Jan 2009 21:28:49 +0000 (14:28 -0700)]
mesa: Move var declaration to top of scope.
(cherry picked from commit
3740a06e28f4cd09e2a3dce2da60320aa9304df1)
José Fonseca [Thu, 8 Jan 2009 12:42:25 +0000 (12:42 +0000)]
glut: Unlist file.
Again..
José Fonseca [Thu, 8 Jan 2009 12:42:00 +0000 (12:42 +0000)]
gdi: Add newline at end of file.
José Fonseca [Thu, 8 Jan 2009 12:41:45 +0000 (12:41 +0000)]
gallium: Replace uint64 by standard uint64_t.
uint64 is not (so?) standard, and often redefined by third parties,
causing name clashes.
José Fonseca [Thu, 8 Jan 2009 12:31:36 +0000 (12:31 +0000)]
draw: Predeclare struct.