95fc270b06
Merge branch 'master' of git+ssh://git.freedesktop.org/git/mesa/mesa
2007-01-21 20:47:10 -05:00
ea441355d3
nouveau: Fixes for nv30.
2007-01-21 20:30:31 -05:00
d1f0a55af4
nouveau: fix some bugs in the nv10 swtcl.
2007-01-22 00:12:05 +01:00
3c0961d299
nouveau: add nv04 state support, and small nv04 fixes.
2007-01-21 04:06:57 +01:00
4f800c5bac
nouveau: rename pass0_arb to pass0.
...
I was expecting to have 2 frontends for the shader code (asm, glsl).
With Brian's work on GLSL this is unnecessary :)
2007-01-21 09:48:33 +11:00
029b81742a
nouveau: kill off shader pass1.
...
It sucks, and we have someone who can do a much better job than I can starting
work on it soon. alloc_temp/free_temp is left in pass2 to workaround fragprog
temps/outputs overlapping, but this all belongs in the optimiser.
2007-01-21 09:31:00 +11:00
ee3d0617f9
nouveau: shader backend branching support for all cards that support it.
2007-01-21 09:13:27 +11:00
059e901446
Add a simple mechanism for annotating instructions for easier debugging.
2007-01-20 10:47:10 -07:00
f116fcc328
optimization: emit MAD instructions when possible
2007-01-20 09:46:02 -07:00
1f208c3546
remove unused swizzle parameter to -storage_to_src_reg()
2007-01-20 09:29:04 -07:00
5ae49cf3ed
Initial implementation of OPCODE_IF/ELSE/ENDIF instructions.
2007-01-20 09:27:40 -07:00
a3e938b8da
Reimplement && and || to do short-circuit evaluation.
...
Improved shader error handling.
2007-01-20 09:22:13 -07:00
9f07ed00e4
rewrite more __postIncr functions
2007-01-20 07:56:00 -07:00
3cec66512d
Implement do/while loops. Replace IR_CJUMP with IR_CJUMP0 and IR_CJUMP1 so
...
we can either jump on zero, or non-zero predicate.
2007-01-19 13:13:15 -07:00
eff9690351
Rewrite normalize(vec3/vec4) to use one less register.
2007-01-19 12:19:38 -07:00
dceae2829e
report error message when something fails
2007-01-19 12:02:31 -07:00
203946e1f9
print error msg when there's a problem
2007-01-19 12:02:06 -07:00
74b27674a8
remove stray tab
2007-01-19 10:31:52 -07:00
a0092c51b1
Implement fragment discard/kill.
2007-01-19 10:15:34 -07:00
8374ccb66f
change while-loop to create new scope for loop body, per spec
2007-01-19 10:02:58 -07:00
669fefd3da
nouveau: bump drm patchlevel
2007-01-19 15:39:36 +11:00
16183e6430
Implement constant sharing so that 4 float constants can share a single
...
float[4] register slot.
2007-01-18 17:29:23 -07:00
ce6640001d
_mesa_add_unnamed_constant() now tries to re-use constants already in
...
the parameter list.
2007-01-18 17:23:48 -07:00
d70771752f
Reimplement code for swizzling so that expressions like (p+q).x for vectors p and q works correctly.
2007-01-18 15:35:44 -07:00
d885ff470a
rewrite a bunch of assignment operators (like +=)
2007-01-18 15:33:13 -07:00
3c59483ed7
nouveau: preliminary nv50 state
2007-01-18 22:07:17 +01:00
f43eac3ed8
remove swizzle param from new_var()
2007-01-18 09:22:44 -07:00
c171166987
1. Fix bug #155
...
2. I notice multiple ARB_occlusion_query should be able to overlap according to spec.
3. Declaring extern variables in a .c file is evil, fix it.
2007-01-18 14:41:38 +08:00
62efc4ba3e
support as much of GL_EXT_stencil_two_side as we can. untested.
2007-01-18 05:56:13 +02:00
feeed10dfd
typo fix. this case still fails.
2007-01-18 05:30:20 +02:00
a3b4b11739
Fix invalid enums passed to MapBuffer
2007-01-18 04:46:04 +02:00
2cb3594af9
Fix invalid enums passed to MapBuffer
2007-01-18 04:17:56 +02:00
8a2368edce
rewrite additional matrix-related functions to reduce register needs
2007-01-17 18:37:20 -07:00
89a3bafe04
rewrite mat4 * mat4 operator to use fewer temps
2007-01-17 18:30:53 -07:00
272622a2e0
do bounds check in _slang_push_var_table(), added comment
2007-01-17 18:21:00 -07:00
2585b74e19
rewrite matrix constructors
2007-01-17 18:06:40 -07:00
4864aaeb02
handle var size > 4 in _slang_pop_var_table()
2007-01-17 18:02:44 -07:00
eb0c478b17
Rewrite a bunch of constructors. It's now important that the first
...
constructor for any given type be the one that takes the most parameters
as this is the constructor that'll be used when there's no perfect match
to the caller's arguments.
See the _slang_adapt_call() function for details.
2007-01-17 16:29:51 -07:00
3a3bb953b6
_slang_gen_function_call_name() now tries to adapt function calls (expand
...
vectors, insert casts) when a perfect match isn't found.
2007-01-17 16:14:03 -07:00
eabb7e66bd
New _slang_adapt_call() function.
...
This is used to modify function calls (when possible) to make the arguments
map to the function parameters. This includes "unrolling" vector types
and doing casts.
Example:
vec2 v2 = vec2(1.2, 3.4)
ivec3 iv = ivec3(false, v2);
Is converted into:
ivec3 iv = ivec3(int(false), int(v2[0]), int(v2[1]))
2007-01-17 16:11:50 -07:00
397b807ad5
code clean-up, re-org. Added IR_F_TO_I support.
2007-01-17 16:05:03 -07:00
bb53124fca
added IR_F_TO_I, update comments
2007-01-17 15:58:24 -07:00
0bad236cfb
Added OPCODE_INT to convert 4 floats to 4 ints.
2007-01-17 15:54:14 -07:00
811f54fa75
Fix/clean-up a number of things related to variable/temporary allocation.
2007-01-17 09:54:31 -07:00
747c9129c0
I965: fix bug#9625-get the correct PV for quardstrip
...
The order of vertices in payload for quardstrip is (0, 1, 3, 2),
so the PV for quardstrip is c->reg.vertex[2].
2007-01-17 10:39:50 +08:00
552a65e454
Implement codegen for the selection operator ( b ? x : y )
2007-01-16 17:38:39 -07:00
3596903068
fix typo
2007-01-16 16:53:41 -07:00
9b5dc1358a
when automatically binding vertex attributes, start with attrib 1, not 0
2007-01-16 14:52:22 -07:00
4f027a33b9
remove dead code
2007-01-16 14:19:37 -07:00
0f91310bf3
comments
2007-01-16 14:17:57 -07:00
Jeremy Kolb