2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
31 #include "pipe/p_error.h"
33 #include "tgsi/tgsi_info.h"
35 #include "brw_context.h"
36 #include "brw_screen.h"
39 #include "brw_state.h"
40 #include "brw_debug.h"
41 #include "brw_pipe_rast.h"
44 /** Return number of src args for given instruction */
45 GLuint
brw_wm_nr_args( GLuint opcode
)
62 assert(opcode
< MAX_OPCODE
);
63 return tgsi_get_opcode_info(opcode
)->num_src
;
68 GLuint
brw_wm_is_scalar_result( GLuint opcode
)
91 * Do GPU code generation for shaders without flow control. Shaders
92 * without flow control instructions can more readily be analysed for
93 * SSA-style optimizations.
96 brw_wm_linear_shader_emit(struct brw_context
*brw
, struct brw_wm_compile
*c
)
98 /* Augment fragment program. Add instructions for pre- and
99 * post-fragment-program tasks such as interpolation and fogging.
103 /* Translate to intermediate representation. Build register usage
108 /* Dead code removal.
112 /* Register allocation.
113 * Divide by two because we operate on 16 pixels at a time and require
114 * two GRF entries for each logical shader register.
116 c
->grf_limit
= BRW_WM_MAX_GRF
/ 2;
120 /* how many general-purpose registers are used */
121 c
->prog_data
.total_grf
= c
->max_wm_grf
;
123 /* Scratch space is used for register spilling */
124 if (c
->last_scratch
) {
125 c
->prog_data
.total_scratch
= c
->last_scratch
+ 0x40;
128 c
->prog_data
.total_scratch
= 0;
138 * All Mesa program -> GPU code generation goes through this function.
139 * Depending on the instructions used (i.e. flow control instructions)
140 * we'll use one of two code generators.
142 static int do_wm_prog( struct brw_context
*brw
,
143 struct brw_fragment_shader
*fp
,
144 struct brw_wm_prog_key
*key
)
146 struct brw_wm_compile
*c
;
147 const GLuint
*program
;
150 c
= brw
->wm
.compile_data
;
152 brw
->wm
.compile_data
= calloc(1, sizeof(*brw
->wm
.compile_data
));
153 c
= brw
->wm
.compile_data
;
155 /* Ouch - big out of memory problem. Can't continue
156 * without triggering a segfault, no way to signal,
159 return PIPE_ERROR_OUT_OF_MEMORY
;
162 memset(c
, 0, sizeof(*brw
->wm
.compile_data
));
164 memcpy(&c
->key
, key
, sizeof(*key
));
167 c
->env_param
= NULL
; /*brw->intel.ctx.FragmentProgram.Parameters;*/
169 brw_init_compile(brw
, &c
->func
);
171 /* temporary sanity check assertion */
172 assert(fp
->has_flow_control
== brw_wm_has_flow_control(c
->fp
));
175 * Shader which use GLSL features such as flow control are handled
176 * differently from "simple" shaders.
178 if (fp
->has_flow_control
) {
179 c
->dispatch_width
= 8;
180 brw_wm_branching_shader_emit(brw
, c
);
183 c
->dispatch_width
= 16;
184 brw_wm_linear_shader_emit(brw
, c
);
187 if (BRW_DEBUG
& DEBUG_WM
)
192 program
= brw_get_program(&c
->func
, &program_size
);
194 brw
->sws
->bo_unreference(brw
->wm
.prog_bo
);
195 brw
->wm
.prog_bo
= brw_upload_cache( &brw
->cache
, BRW_WM_PROG
,
196 &c
->key
, sizeof(c
->key
),
198 program
, program_size
,
200 &brw
->wm
.prog_data
);
207 static void brw_wm_populate_key( struct brw_context
*brw
,
208 struct brw_wm_prog_key
*key
)
210 unsigned lookup
, line_aa
;
213 memset(key
, 0, sizeof(*key
));
215 /* PIPE_NEW_FRAGMENT_SHADER
216 * PIPE_NEW_DEPTH_STENCIL_ALPHA
218 lookup
= (brw
->curr
.zstencil
->iz_lookup
|
219 brw
->curr
.fragment_shader
->iz_lookup
);
223 * BRW_NEW_REDUCED_PRIMITIVE
225 switch (brw
->reduced_primitive
) {
226 case PIPE_PRIM_POINTS
:
229 case PIPE_PRIM_LINES
:
233 line_aa
= brw
->curr
.rast
->unfilled_aa_line
;
237 brw_wm_lookup_iz(line_aa
,
239 brw
->curr
.fragment_shader
->uses_depth
,
242 /* Revisit this, figure out if it's really useful, and either push
243 * it into the state tracker so that everyone benefits (use to
244 * create fs varients with TEX rather than TXP), or discard.
246 key
->proj_attrib_mask
= ~0; /*brw->wm.input_size_masks[4-1];*/
249 key
->flat_shade
= brw
->curr
.rast
->templ
.flatshade
;
251 /* This can be determined by looking at the INTERP mode each input decl.
253 key
->linear_attrib_mask
= 0;
255 /* PIPE_NEW_BOUND_TEXTURES */
256 for (i
= 0; i
< brw
->curr
.num_textures
; i
++) {
257 const struct brw_texture
*tex
= brw
->curr
.texture
[i
];
259 if (tex
->base
.format
== PIPE_FORMAT_YCBCR
)
260 key
->yuvtex_mask
|= 1 << i
;
262 if (tex
->base
.format
== PIPE_FORMAT_YCBCR_REV
)
263 key
->yuvtex_swap_mask
|= 1 << i
;
265 /* XXX: shadow texture
267 /* key->shadowtex_mask |= 1<<i; */
270 /* CACHE_NEW_VS_PROG */
271 key
->vp_nr_outputs
= brw
->vs
.prog_data
->nr_outputs
;
273 /* The unique fragment program ID */
274 key
->program_string_id
= brw
->curr
.fragment_shader
->id
;
278 static int brw_prepare_wm_prog(struct brw_context
*brw
)
280 struct brw_wm_prog_key key
;
281 struct brw_fragment_shader
*fs
= brw
->curr
.fragment_shader
;
283 brw_wm_populate_key(brw
, &key
);
285 /* Make an early check for the key.
287 brw
->sws
->bo_unreference(brw
->wm
.prog_bo
);
288 brw
->wm
.prog_bo
= brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
292 if (brw
->wm
.prog_bo
== NULL
)
293 return do_wm_prog(brw
, fs
, &key
);
299 const struct brw_tracked_state brw_wm_prog
= {
301 .mesa
= (PIPE_NEW_FRAGMENT_SHADER
|
302 PIPE_NEW_DEPTH_STENCIL_ALPHA
|
304 PIPE_NEW_BOUND_TEXTURES
),
305 .brw
= (BRW_NEW_WM_INPUT_DIMENSIONS
|
306 BRW_NEW_REDUCED_PRIMITIVE
),
307 .cache
= CACHE_NEW_VS_PROG
,
309 .prepare
= brw_prepare_wm_prog