Luca Barbieri [Sun, 5 Sep 2010 15:57:12 +0000 (17:57 +0200)]
mesa: don't smash the stack in _mesa_find_used_registers
At some point this actually triggered, not sure if it still does.
Give a meaningful assert and refuse to smash the stack anyway.
Luca Barbieri [Tue, 24 Aug 2010 19:05:51 +0000 (21:05 +0200)]
mesa: don't expose unsupported GL_ARB_geometry_shader4 for now
The new GLSL compiler doesn't support it.
Advertising it prevents Unigine Heaven from working, since it attempts to
use it.
Luca Barbieri [Sun, 5 Sep 2010 08:34:55 +0000 (10:34 +0200)]
nouveau: delete nouveau_class.h, move nv50 regs to nv50_reg.h
nv50 should switch to rules-ng-ng too at some point.
The classic Mesa Nouveau driver also includes a copy of nouveau_class.h,
and should convert to rules-ng-ng too and remove it.
Luca Barbieri [Sun, 5 Sep 2010 08:45:55 +0000 (10:45 +0200)]
nvfx: move nv04_2d to rules-ng-ng
Luca Barbieri [Sun, 5 Sep 2010 08:10:09 +0000 (10:10 +0200)]
nvfx: switch to rules-ng-ng register headers
This is the new register generation toolkit in use by nouveau.
As far as I know, this is the best register description toolkit in
existence, and you should use it too for your hardware :)
Thanks to Marcin Kościelnicki for inventing it and performing
invaluable reverse engineering work of nVidia chips.
Luca Barbieri [Sun, 5 Sep 2010 07:17:32 +0000 (09:17 +0200)]
nvfx: remove remaining BEGIN_RING/eng3d uses
Luca Barbieri [Sun, 5 Sep 2010 06:27:12 +0000 (08:27 +0200)]
nvfx: pause occlusion queries during blitter usage
Thanks for Dave Airlie and Jerome Glisse for their code which made
me realize I need this too.
Luca Barbieri [Sun, 5 Sep 2010 04:42:04 +0000 (06:42 +0200)]
nvfx: properly return fogcoord.w == 1
Hardware sets it to 0, so we add an ADD to put an 1 there if the
application really wants the alpha channel.
Luca Barbieri [Sun, 5 Sep 2010 03:53:55 +0000 (05:53 +0200)]
nvfx: support saturate in vp
Completely untested, since Mesa apparently never uses this currently.
In particular, it might not work with scalar slot op.
Luca Barbieri [Sun, 5 Sep 2010 03:42:59 +0000 (05:42 +0200)]
nvfx: add rewritten swtnl support
The old swtnl code was broken by the new shader linkage support for
GLSL.
This is a rewrite of swtnl support, which should instead work properly,
be faster and more closer to the much more tested hardware pipeline.
Luca Barbieri [Sun, 5 Sep 2010 03:42:39 +0000 (05:42 +0200)]
nvfx: use a piglit-ignored format for unknown cap message
José Fonseca [Sun, 5 Sep 2010 09:40:38 +0000 (10:40 +0100)]
llvmpipe: Remove some broken MinGW hacks in the sin/cos reference code.
José Fonseca [Sun, 5 Sep 2010 09:16:30 +0000 (10:16 +0100)]
glsl: Add new files to sconscript.
José Fonseca [Sat, 4 Sep 2010 18:51:54 +0000 (19:51 +0100)]
llvmpipe: Fix perspective divide interpolation.
Intuition != mathematics, so this time I actually worked out the right
formula for first order approximation of perspective interpolation.
Ironically, per quad divide actually makes things slower when compared
with per pixel divide -- probably because the divide hardware unit is
rarely used, whereas the multiply unit is typically already saturated
and the first order approximation imply more multiplications.
José Fonseca [Sat, 4 Sep 2010 10:53:28 +0000 (11:53 +0100)]
util: Helper function to determined whether two formats can be memcpy'ed.
These are the non-trivial conversions that this function recognizes,
which was produced by u_format_compatible_test.c:
b8g8r8a8_unorm -> b8g8r8x8_unorm
a8r8g8b8_unorm -> x8r8g8b8_unorm
b5g5r5a1_unorm -> b5g5r5x1_unorm
b4g4r4a4_unorm -> b4g4r4x4_unorm
l8_unorm -> r8_unorm
i8_unorm -> l8_unorm
i8_unorm -> a8_unorm
i8_unorm -> r8_unorm
l16_unorm -> r16_unorm
z24_unorm_s8_uscaled -> z24x8_unorm
s8_uscaled_z24_unorm -> x8z24_unorm
r8g8b8a8_unorm -> r8g8b8x8_unorm
a8b8g8r8_srgb -> x8b8g8r8_srgb
b8g8r8a8_srgb -> b8g8r8x8_srgb
a8r8g8b8_srgb -> x8r8g8b8_srgb
a8b8g8r8_unorm -> x8b8g8r8_unorm
r10g10b10a2_uscaled -> r10g10b10x2_uscaled
r10sg10sb10sa2u_norm -> r10g10b10x2_snorm
State trackers and pipe drivers should be updated to take advantage of
this knowledge, e.g., in surface_copy.
José Fonseca [Fri, 3 Sep 2010 19:26:13 +0000 (20:26 +0100)]
llvmpipe: Relax the colormask constraint on opaque.
Also, include the color buffer in the key. Not having it there
causes a tight knots in the logic to determine when it is OK or not
to discard previous color buffer contents.
José Fonseca [Fri, 3 Sep 2010 19:21:33 +0000 (20:21 +0100)]
util: Utility function to determine the channels that can be written in a color format.
José Fonseca [Thu, 2 Sep 2010 12:44:29 +0000 (13:44 +0100)]
scons: Re-enable SSE on MinGW.
It seems to be working correctly with gcc 4.4, and enabling it allows to
test some of the llvmpipe instrinsics on Windows.
José Fonseca [Thu, 2 Sep 2010 10:32:09 +0000 (11:32 +0100)]
gallivm: Pass condition masks as an unsigned bitmask.
Much more convenient than boolean arrays.
José Fonseca [Thu, 2 Sep 2010 10:30:13 +0000 (11:30 +0100)]
gallivm: Cope with tgsi instruction reallocation failure.
Kenneth Graunke [Sun, 5 Sep 2010 07:58:34 +0000 (00:58 -0700)]
ir_reader: Only validate IR when a global 'debug' flag is set.
This extra validation is very useful when working on the built-ins, but
in general overkill - the results should stay the same unless the
built-ins or ir_validate have changed.
Also, validating all the built-in functions in every test case makes
piglit run unacceptably slow.
Marek Olšák [Sun, 5 Sep 2010 03:07:02 +0000 (05:07 +0200)]
r300g,r300c: memset the compiler struct to zeros
This should fix bogus reports "Too many temporaries." and maybe some others.
Tom Stellard [Sun, 5 Sep 2010 00:27:55 +0000 (17:27 -0700)]
r300/compiler: Remove stray break statement
This fixes glsl-fs-loop-nested.
Luca Barbieri [Sat, 4 Sep 2010 23:08:08 +0000 (01:08 +0200)]
nvfx: support unlimited constants and immediates in fp
Luca Barbieri [Sat, 4 Sep 2010 21:20:33 +0000 (23:20 +0200)]
nvfx: support using blitter to copy depth/stencil resources, fix Heaven
We might want to copy them as color ones though.
Also works around crash in Unigine Heaven due to failing to allocate
a 64 MB temporary in GART for a CPU copy.
Unigine Heaven now works on nv40, albeit with very heavy glitches (with
the floating branch with render_hdr 0).
Marek Olšák [Sat, 4 Sep 2010 22:43:34 +0000 (00:43 +0200)]
r300/compiler: fix the instruction limit in vertex shaders
Broken with commit
d774b0c710bb7d833d17bd12f5151a0176baad96.
Reported by Chris Rankin.
Luca Barbieri [Sat, 4 Sep 2010 19:29:43 +0000 (21:29 +0200)]
nvfx: support rendering to more formats
Luca Barbieri [Sat, 4 Sep 2010 18:16:54 +0000 (20:16 +0200)]
nvfx: move 2D format selection logic to 2D code
Luca Barbieri [Sat, 4 Sep 2010 15:12:02 +0000 (17:12 +0200)]
nvfx: fix swizzling of high bpp surfaces
Luca Barbieri [Sat, 4 Sep 2010 19:22:02 +0000 (21:22 +0200)]
nvfx: fix some subrectangle copies
Actually, we may want to get rid of the x/y coordinates for linear
surfaces, and realign the origin from scratch if necessary, instead
of doing this "on-demand realignment".
Luca Barbieri [Sat, 4 Sep 2010 19:21:34 +0000 (21:21 +0200)]
nvfx: fix inlinining in nv04_2d.c
Luca Barbieri [Sat, 4 Sep 2010 18:17:39 +0000 (20:17 +0200)]
nvfx: fix the temporary copying logic and add asserts
Luca Barbieri [Sat, 4 Sep 2010 15:48:19 +0000 (17:48 +0200)]
nvfx: prevent swizzled rendering into formats where it's not supported
Marek Olšák [Fri, 3 Sep 2010 22:42:36 +0000 (00:42 +0200)]
Revert "ir_to_mesa: Load all the STATE_VAR elements of a builtin uniform to a temp."
This reverts commit
5ad74779cea07cc6a19a52874cdaef8b018e2f1b.
Sorry, but I had to revert this.
Any commit which needlessly increases the number of temporaries is wrong.
More temporaries mean less shader performance because of reduced parallelism
and therefore less efficient latency hiding. In this case, there is possible
performance degradation of every shader which uses GL state variables.
I cannot accept this.
Marek Olšák [Sat, 4 Sep 2010 17:04:51 +0000 (19:04 +0200)]
Revert "r300g: refuse to create a texture with size 0"
This reverts commit
5cdedaaf295acae13ac10feeb3143d83bc53d314.
https://bugs.freedesktop.org/show_bug.cgi?id=30002
Conflicts:
src/gallium/drivers/r300/r300_texture.c
Marek Olšák [Tue, 31 Aug 2010 03:37:28 +0000 (05:37 +0200)]
r300g: remove unnecessary assignments
Marek Olšák [Fri, 3 Sep 2010 22:02:57 +0000 (00:02 +0200)]
r300/compiler: indent printed instructions according to the branch depth
Marek Olšák [Fri, 3 Sep 2010 19:43:36 +0000 (21:43 +0200)]
r300g: skip draw calls with no vertex elements, fixing hardlocks
Marek Olšák [Fri, 3 Sep 2010 18:43:48 +0000 (20:43 +0200)]
r300/compiler: use limits from the compiler input instead of inline constants
Marek Olšák [Fri, 3 Sep 2010 18:26:43 +0000 (20:26 +0200)]
r300/compiler: improve register allocation with indexable temporaries for VS
Register allocation can now reallocate temporaries right after the last indexed
source operand, instead of being disabled for the whole shader.
Marek Olšák [Thu, 2 Sep 2010 08:21:52 +0000 (10:21 +0200)]
r300/compiler: fix handling of indexed temporaries in peephole
Marek Olšák [Thu, 2 Sep 2010 08:21:23 +0000 (10:21 +0200)]
r300/compiler: disable deadcode elimination for indexed dst operands
Marek Olšák [Thu, 2 Sep 2010 05:01:36 +0000 (07:01 +0200)]
r300/compiler: allocate at least FS inputs if register allocation is disabled
Marek Olšák [Wed, 1 Sep 2010 06:12:51 +0000 (08:12 +0200)]
r300g: add a new debug option which disables compiler optimizations
Those are:
- dead-code elimination
- constant folding
- peephole (mainly copy propagation)
- register allocation
There are some bugs which I need to track down.
Also fix up the descriptions of all the debug options.
Marek Olšák [Wed, 1 Sep 2010 06:10:32 +0000 (08:10 +0200)]
r300/compiler: compute the final number of temporaries during translation
And not during the register allocation, which may be skipped for debugging
purposes. Also the predicate register is now added to the number of temps.
Marek Olšák [Wed, 1 Sep 2010 04:14:58 +0000 (06:14 +0200)]
r300/compiler: make optimizations not use 0.5 swizzles in vertex shaders
Marek Olšák [Wed, 1 Sep 2010 03:25:34 +0000 (05:25 +0200)]
r300/compiler: use peephole and constant folding for vertex shaders too
Marek Olšák [Thu, 2 Sep 2010 00:42:42 +0000 (02:42 +0200)]
r300/compiler: remove unused enum OPCODE_REPL_ALPHA
We use RC_OPCODE_REPL_ALPHA instead.
Marek Olšák [Wed, 1 Sep 2010 03:01:19 +0000 (05:01 +0200)]
r300/compiler: refactor fragment shader compilation
This cleans up the mess in r3xx_compile_fragment_program.
Marek Olšák [Wed, 1 Sep 2010 02:59:22 +0000 (04:59 +0200)]
r300/compiler: add new compiler parameter max_constants
Marek Olšák [Wed, 1 Sep 2010 01:19:05 +0000 (03:19 +0200)]
r300/compiler: refactor vertex shader compilation
First list compiler passes in an array, then run the new function rc_run_compiler.
Every backend may need a different set of passes.
This cleans up the mess in r3xx_compile_vertex_program.
Marek Olšák [Tue, 31 Aug 2010 23:55:26 +0000 (01:55 +0200)]
r300/compiler: remove a redundant parameter in rc_pair_regalloc
Marek Olšák [Tue, 31 Aug 2010 23:51:05 +0000 (01:51 +0200)]
r300/compiler: remove a redundant parameter in rc_dataflow_deadcode
&c->Base == c.
Marek Olšák [Tue, 31 Aug 2010 23:10:26 +0000 (01:10 +0200)]
r300/compiler: use null-terminated array of transformation functions
I need to reduce the number of parameters of each compiler pass function.
This is part of a larger cleanup.
Marek Olšák [Tue, 31 Aug 2010 22:59:52 +0000 (00:59 +0200)]
r300g: only check for an empty shader if there are no compile errors
Marek Olšák [Tue, 31 Aug 2010 22:56:57 +0000 (00:56 +0200)]
r300/compiler: add new compiler parameter max_alu_insts
Marek Olšák [Tue, 31 Aug 2010 18:51:37 +0000 (20:51 +0200)]
r300/compiler: put emulate_loop_state in radeon_compiler
Kenneth Graunke [Sat, 4 Sep 2010 08:09:43 +0000 (01:09 -0700)]
ir_reader: Run ir_validate on the generated IR.
It's just too easy to get something wrong in hand-written IR.
Kenneth Graunke [Sat, 4 Sep 2010 08:55:55 +0000 (01:55 -0700)]
ir_reader: Emit global variables at the top of the instruction list.
Since functions are emitted when scanning for prototypes, functions
always come first, even if the original IR listed the variable
declarations first.
Fixes an ir_validate error (to be turned on in the next commit).
Kenneth Graunke [Fri, 3 Sep 2010 23:14:40 +0000 (16:14 -0700)]
ir_reader: Drop support for reading the old assignment format.
Kenneth Graunke [Fri, 3 Sep 2010 23:21:08 +0000 (16:21 -0700)]
glsl: Regenerate autogenerated file builtin_function.cpp.
Kenneth Graunke [Fri, 3 Sep 2010 23:10:57 +0000 (16:10 -0700)]
glsl/builtins: Convert assignments to new format (with write mask).
Kenneth Graunke [Fri, 3 Sep 2010 06:54:40 +0000 (23:54 -0700)]
ir_reader: Read the new assignment format (with write mask).
This preserves the ability to read the old format, for momentary
compatibility with all the existing IR implementations of built-ins.
Kenneth Graunke [Sat, 4 Sep 2010 08:05:51 +0000 (01:05 -0700)]
ir_reader: Track the current function and report it in error messages.
Kenneth Graunke [Sat, 4 Sep 2010 07:49:23 +0000 (00:49 -0700)]
glsl/builtins: Actually print the info log if reading a builtin failed.
Luca Barbieri [Sat, 4 Sep 2010 03:30:27 +0000 (05:30 +0200)]
nvfx: consolidate tiny files
We probably want to reorganize the remaining files too, but that's
for later, maybe.
Luca Barbieri [Sat, 4 Sep 2010 03:13:06 +0000 (05:13 +0200)]
mesa/st: add missing _mesa_set_fetch_functions in st_get_tex_image
Fixes piglit fdo25614-genmipmap.
Luca Barbieri [Sat, 4 Sep 2010 02:43:02 +0000 (04:43 +0200)]
nvfx: fix vp DP2
Luca Barbieri [Sat, 4 Sep 2010 02:17:16 +0000 (04:17 +0200)]
nvfx: implement fp SSG properly
Luca Barbieri [Sat, 4 Sep 2010 01:40:49 +0000 (03:40 +0200)]
nvfx: don't claim we support preds since the driver doesn't
Luca Barbieri [Sat, 4 Sep 2010 01:35:22 +0000 (03:35 +0200)]
nv40: support all 10 texcoords
Luca Barbieri [Sat, 4 Sep 2010 01:05:28 +0000 (03:05 +0200)]
nvfx: add missing context init
Luca Barbieri [Sat, 4 Sep 2010 01:05:22 +0000 (03:05 +0200)]
nvfx: tidy up state_emit
Luca Barbieri [Sat, 4 Sep 2010 00:57:14 +0000 (02:57 +0200)]
nvfx: support all coord conventions in hardware
Luca Barbieri [Sat, 4 Sep 2010 00:37:41 +0000 (02:37 +0200)]
nvfx: add missing pushbuffer space check
Luca Barbieri [Sat, 4 Sep 2010 00:26:37 +0000 (02:26 +0200)]
nvfx: support all possible vs consts
We were incorrectly setting a register that limited the range of
constants accessible via indirect addressing.
Setting it correctly, we can address all the constants the GPU
supports.
Luca Barbieri [Sat, 4 Sep 2010 00:05:14 +0000 (02:05 +0200)]
nvfx: set magic bit to round NPOT mipmap sizes down and not up
Does any API even use rounding-up?
Luca Barbieri [Fri, 3 Sep 2010 21:27:49 +0000 (23:27 +0200)]
nvfx: allow nested blitter usage, fixing bug in clear
Brian Paul [Fri, 3 Sep 2010 22:35:07 +0000 (16:35 -0600)]
galahad: do map/unmap counting for resources
Brian Paul [Fri, 3 Sep 2010 22:33:17 +0000 (16:33 -0600)]
libgl-xlib: enable galahad support
If the GALLIUM_GALAHAD env var is 1 we'll wrap the regular driver with
the galahad validation driver.
Brian Paul [Fri, 3 Sep 2010 22:25:44 +0000 (16:25 -0600)]
scons: added galahad to driver list
Brian Paul [Fri, 3 Sep 2010 21:57:48 +0000 (15:57 -0600)]
mesa: also build galahad driver
Brian Paul [Fri, 3 Sep 2010 21:25:50 +0000 (15:25 -0600)]
exec_list: replace class with struct
To match the definition below.
Brian Paul [Fri, 3 Sep 2010 20:39:43 +0000 (14:39 -0600)]
mesa: fix up a comment
Brian Paul [Thu, 2 Sep 2010 20:11:53 +0000 (14:11 -0600)]
st/glx: added some comments
Luca Barbieri [Fri, 3 Sep 2010 20:06:41 +0000 (22:06 +0200)]
nvfx: implement LIT in fp
Ian Romanick [Thu, 2 Sep 2010 21:53:17 +0000 (14:53 -0700)]
glsl2: Use as_constant some places instead of constant_expression_value
The places where constant_expression_value are still used in loop
analysis are places where a new expression tree is created and
constant folding won't have happened. This is used, for example, when
we try to determine the maximal loop iteration count.
Based on review comments by Eric. "...rely on constant folding to
have done its job, instead of going all through the subtree again when
it wasn't a constant."
Ian Romanick [Fri, 27 Aug 2010 23:22:36 +0000 (16:22 -0700)]
glsl2: Allow copy / constant propagation into array indices
Ian Romanick [Fri, 27 Aug 2010 20:59:49 +0000 (13:59 -0700)]
glsl2: Add module to perform simple loop unrolling
Ian Romanick [Fri, 27 Aug 2010 22:41:20 +0000 (15:41 -0700)]
glsl2: Track the number of ir_loop_jump instructions that are in a loop
Ian Romanick [Fri, 27 Aug 2010 20:53:25 +0000 (13:53 -0700)]
ir_expression: Add static operator_string method
I've used this in quite a few debug commits that never reached an
up-stream tree.
Ian Romanick [Fri, 27 Aug 2010 20:53:56 +0000 (13:53 -0700)]
exec_node: Add insert_before that inserts an entire list
Ian Romanick [Fri, 27 Aug 2010 18:26:08 +0000 (11:26 -0700)]
glsl2: Eliminate zero-iteration loops
Ian Romanick [Thu, 26 Aug 2010 23:45:22 +0000 (16:45 -0700)]
glsl2: Perform initial bits of loop analysis during compilation
Ian Romanick [Thu, 26 Aug 2010 23:43:57 +0000 (16:43 -0700)]
glsl2: Add module to suss out loop control variables from loop analysis data
This is the next step on the road to loop unrolling
Ian Romanick [Thu, 26 Aug 2010 22:58:33 +0000 (15:58 -0700)]
glsl2: Add module to analyze variables used in loops
This is the first step eventually leading to loop unrolling.
Ian Romanick [Thu, 26 Aug 2010 22:49:33 +0000 (15:49 -0700)]
ir_to_mesa: Handle loops with loop controls set
The downside of our talloc usage is that we can't really make static
(i.e., not created with new) instances of our IR types. This leads to
a lot of unnecessary dynamic allocation in this patch.
Ian Romanick [Thu, 26 Aug 2010 22:22:06 +0000 (15:22 -0700)]
ir_validate: Validate loop control fields in ir_loop
Ian Romanick [Thu, 26 Aug 2010 22:11:26 +0000 (15:11 -0700)]
glsl2: Add cmp field to ir_loop
This reprents the type of comparison between the loop induction
variable and the loop termination value.
Ian Romanick [Thu, 5 Aug 2010 22:29:24 +0000 (15:29 -0700)]
glsl2: Set a flag when visiting the assignee of an assignment