r300g: Increase fragment shader limits for r400 cards
[mesa.git] / src / gallium / drivers / r300 / r300_screen.c
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24 #include "util/u_format.h"
25 #include "util/u_format_s3tc.h"
26 #include "util/u_memory.h"
27
28 #include "r300_context.h"
29 #include "r300_texture.h"
30 #include "r300_screen_buffer.h"
31 #include "r300_state_inlines.h"
32 #include "r300_winsys.h"
33 #include "r300_public.h"
34
35 #include "draw/draw_context.h"
36
37 #ifdef HAVE_LLVM
38 #include "gallivm/lp_bld_init.h"
39 #endif
40
41 /* Return the identifier behind whom the brave coders responsible for this
42 * amalgamation of code, sweat, and duct tape, routinely obscure their names.
43 *
44 * ...I should have just put "Corbin Simpson", but I'm not that cool.
45 *
46 * (Or egotistical. Yet.) */
47 static const char* r300_get_vendor(struct pipe_screen* pscreen)
48 {
49 return "X.Org R300 Project";
50 }
51
52 static const char* chip_families[] = {
53 "ATI R300",
54 "ATI R350",
55 "ATI R360",
56 "ATI RV350",
57 "ATI RV370",
58 "ATI RV380",
59 "ATI R420",
60 "ATI R423",
61 "ATI R430",
62 "ATI R480",
63 "ATI R481",
64 "ATI RV410",
65 "ATI RS400",
66 "ATI RC410",
67 "ATI RS480",
68 "ATI RS482",
69 "ATI RS600",
70 "ATI RS690",
71 "ATI RS740",
72 "ATI RV515",
73 "ATI R520",
74 "ATI RV530",
75 "ATI R580",
76 "ATI RV560",
77 "ATI RV570"
78 };
79
80 static const char* r300_get_name(struct pipe_screen* pscreen)
81 {
82 struct r300_screen* r300screen = r300_screen(pscreen);
83
84 return chip_families[r300screen->caps.family];
85 }
86
87 static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
88 {
89 struct r300_screen* r300screen = r300_screen(pscreen);
90 boolean is_r500 = r300screen->caps.is_r500;
91
92 switch (param) {
93 /* Supported features (boolean caps). */
94 case PIPE_CAP_NPOT_TEXTURES:
95 case PIPE_CAP_TWO_SIDED_STENCIL:
96 case PIPE_CAP_GLSL:
97 /* I'll be frank. This is a lie.
98 *
99 * We don't truly support GLSL on any of this driver's chipsets.
100 * To be fair, no chipset supports the full GLSL specification
101 * to the best of our knowledge, but some of the less esoteric
102 * features are still missing here.
103 *
104 * Rather than cripple ourselves intentionally, I'm going to set
105 * this flag, and as Gallium's interface continues to change, I
106 * hope that this single monolithic GLSL enable can slowly get
107 * split down into many different pieces and the state tracker
108 * will handle fallbacks transparently, like it should.
109 *
110 * ~ C.
111 */
112 case PIPE_CAP_ANISOTROPIC_FILTER:
113 case PIPE_CAP_POINT_SPRITE:
114 case PIPE_CAP_OCCLUSION_QUERY:
115 case PIPE_CAP_TEXTURE_SHADOW_MAP:
116 case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
117 case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
118 case PIPE_CAP_BLEND_EQUATION_SEPARATE:
119 return 1;
120 case PIPE_CAP_TEXTURE_SWIZZLE:
121 return util_format_s3tc_enabled ? r300screen->caps.dxtc_swizzle : 1;
122
123 /* Unsupported features (boolean caps). */
124 case PIPE_CAP_TIMER_QUERY:
125 case PIPE_CAP_DUAL_SOURCE_BLEND:
126 case PIPE_CAP_INDEP_BLEND_ENABLE:
127 case PIPE_CAP_INDEP_BLEND_FUNC:
128 case PIPE_CAP_DEPTH_CLAMP: /* XXX implemented, but breaks Regnum Online */
129 case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
130 case PIPE_CAP_SHADER_STENCIL_EXPORT:
131 case PIPE_CAP_STREAM_OUTPUT:
132 case PIPE_CAP_PRIMITIVE_RESTART:
133 return 0;
134
135 /* Texturing. */
136 case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
137 case PIPE_CAP_MAX_COMBINED_SAMPLERS:
138 return r300screen->caps.num_tex_units;
139 case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
140 return 0;
141 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
142 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
143 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
144 /* 13 == 4096, 12 == 2048 */
145 return is_r500 ? 13 : 12;
146
147 /* Render targets. */
148 case PIPE_CAP_MAX_RENDER_TARGETS:
149 return 4;
150
151 /* General shader limits and features. */
152 case PIPE_CAP_SM3:
153 return is_r500 ? 1 : 0;
154 /* Fragment coordinate conventions. */
155 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
156 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
157 return 1;
158 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
159 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
160 return 0;
161 default:
162 debug_printf("r300: Warning: Unknown CAP %d in get_param.\n",
163 param);
164 return 0;
165 }
166 }
167
168 static int r300_get_shader_param(struct pipe_screen *pscreen, unsigned shader, enum pipe_shader_cap param)
169 {
170 struct r300_screen* r300screen = r300_screen(pscreen);
171 boolean is_r400 = r300screen->caps.is_r400;
172 boolean is_r500 = r300screen->caps.is_r500;
173
174 switch (shader)
175 {
176 case PIPE_SHADER_FRAGMENT:
177 switch (param)
178 {
179 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
180 return is_r500 || is_r400 ? 512 : 96;
181 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
182 return is_r500 || is_r400 ? 512 : 64;
183 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
184 return is_r500 || is_r400 ? 512 : 32;
185 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
186 return is_r500 ? 511 : 4;
187 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
188 return is_r500 ? 64 : 0; /* Actually unlimited on r500. */
189 /* Fragment shader limits. */
190 case PIPE_SHADER_CAP_MAX_INPUTS:
191 /* 2 colors + 8 texcoords are always supported
192 * (minus fog and wpos).
193 *
194 * R500 has the ability to turn 3rd and 4th color into
195 * additional texcoords but there is no two-sided color
196 * selection then. However the facing bit can be used instead. */
197 return 10;
198 case PIPE_SHADER_CAP_MAX_CONSTS:
199 return is_r500 ? 256 : 32;
200 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
201 return 1;
202 case PIPE_SHADER_CAP_MAX_TEMPS:
203 return is_r500 ? 128 : is_r400 ? 64 : 32;
204 case PIPE_SHADER_CAP_MAX_ADDRS:
205 return 0;
206 case PIPE_SHADER_CAP_MAX_PREDS:
207 return is_r500 ? 1 : 0;
208 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
209 return 1;
210 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
211 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
212 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
213 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
214 return 0;
215 case PIPE_SHADER_CAP_SUBROUTINES:
216 return 0;
217 }
218 break;
219 case PIPE_SHADER_VERTEX:
220 if (!r300screen->caps.has_tcl) {
221 return draw_get_shader_param(shader, param);
222 }
223
224 switch (param)
225 {
226 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
227 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
228 return is_r500 ? 1024 : 256;
229 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
230 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
231 return 0;
232 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
233 return is_r500 ? 4 : 0; /* For loops; not sure about conditionals. */
234 case PIPE_SHADER_CAP_MAX_INPUTS:
235 return 16;
236 case PIPE_SHADER_CAP_MAX_CONSTS:
237 return 256;
238 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
239 return 1;
240 case PIPE_SHADER_CAP_MAX_TEMPS:
241 return 32;
242 case PIPE_SHADER_CAP_MAX_ADDRS:
243 return 1; /* XXX guessed */
244 case PIPE_SHADER_CAP_MAX_PREDS:
245 return is_r500 ? 4 : 0; /* XXX guessed. */
246 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
247 return 1;
248 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
249 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
250 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
251 return 0;
252 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
253 return 1;
254 case PIPE_SHADER_CAP_SUBROUTINES:
255 return 0;
256 default:
257 break;
258 }
259 break;
260 default:
261 break;
262 }
263 return 0;
264 }
265
266 static float r300_get_paramf(struct pipe_screen* pscreen, enum pipe_cap param)
267 {
268 struct r300_screen* r300screen = r300_screen(pscreen);
269
270 switch (param) {
271 case PIPE_CAP_MAX_LINE_WIDTH:
272 case PIPE_CAP_MAX_LINE_WIDTH_AA:
273 case PIPE_CAP_MAX_POINT_WIDTH:
274 case PIPE_CAP_MAX_POINT_WIDTH_AA:
275 /* The maximum dimensions of the colorbuffer are our practical
276 * rendering limits. 2048 pixels should be enough for anybody. */
277 if (r300screen->caps.is_r500) {
278 return 4096.0f;
279 } else if (r300screen->caps.is_r400) {
280 return 4021.0f;
281 } else {
282 return 2560.0f;
283 }
284 case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
285 return 16.0f;
286 case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
287 return 16.0f;
288 case PIPE_CAP_GUARD_BAND_LEFT:
289 case PIPE_CAP_GUARD_BAND_TOP:
290 case PIPE_CAP_GUARD_BAND_RIGHT:
291 case PIPE_CAP_GUARD_BAND_BOTTOM:
292 /* XXX I don't know what these should be but the least we can do is
293 * silence the potential error message */
294 return 0.0f;
295 default:
296 debug_printf("r300: Warning: Unknown CAP %d in get_paramf.\n",
297 param);
298 return 0.0f;
299 }
300 }
301
302 static boolean r300_is_format_supported(struct pipe_screen* screen,
303 enum pipe_format format,
304 enum pipe_texture_target target,
305 unsigned sample_count,
306 unsigned usage,
307 unsigned geom_flags)
308 {
309 struct r300_winsys_screen *rws = r300_screen(screen)->rws;
310 uint32_t retval = 0;
311 boolean drm_2_8_0 = rws->get_value(rws, R300_VID_DRM_2_8_0);
312 boolean is_r500 = r300_screen(screen)->caps.is_r500;
313 boolean is_r400 = r300_screen(screen)->caps.is_r400;
314 boolean is_color2101010 = format == PIPE_FORMAT_R10G10B10A2_UNORM ||
315 format == PIPE_FORMAT_R10G10B10X2_SNORM ||
316 format == PIPE_FORMAT_B10G10R10A2_UNORM ||
317 format == PIPE_FORMAT_R10SG10SB10SA2U_NORM;
318 boolean is_ati1n = format == PIPE_FORMAT_RGTC1_UNORM ||
319 format == PIPE_FORMAT_RGTC1_SNORM;
320 boolean is_ati2n = format == PIPE_FORMAT_RGTC2_UNORM ||
321 format == PIPE_FORMAT_RGTC2_SNORM;
322 boolean is_half_float = format == PIPE_FORMAT_R16_FLOAT ||
323 format == PIPE_FORMAT_R16G16_FLOAT ||
324 format == PIPE_FORMAT_R16G16B16_FLOAT ||
325 format == PIPE_FORMAT_R16G16B16A16_FLOAT;
326
327 /* Check multisampling support. */
328 switch (sample_count) {
329 case 0:
330 case 1:
331 break;
332 case 2:
333 case 3:
334 case 4:
335 case 6:
336 return FALSE;
337 #if 0
338 if (usage != PIPE_BIND_RENDER_TARGET ||
339 !util_format_is_rgba8_variant(
340 util_format_description(format))) {
341 return FALSE;
342 }
343 #endif
344 break;
345 default:
346 return FALSE;
347 }
348
349 /* Check sampler format support. */
350 if ((usage & PIPE_BIND_SAMPLER_VIEW) &&
351 /* ATI1N is r5xx-only. */
352 (is_r500 || !is_ati1n) &&
353 /* ATI2N is supported on r4xx-r5xx. */
354 (is_r400 || is_r500 || !is_ati2n) &&
355 r300_is_sampler_format_supported(format)) {
356 retval |= PIPE_BIND_SAMPLER_VIEW;
357 }
358
359 /* Check colorbuffer format support. */
360 if ((usage & (PIPE_BIND_RENDER_TARGET |
361 PIPE_BIND_DISPLAY_TARGET |
362 PIPE_BIND_SCANOUT |
363 PIPE_BIND_SHARED)) &&
364 /* 2101010 cannot be rendered to on non-r5xx. */
365 (!is_color2101010 || (is_r500 && drm_2_8_0)) &&
366 r300_is_colorbuffer_format_supported(format)) {
367 retval |= usage &
368 (PIPE_BIND_RENDER_TARGET |
369 PIPE_BIND_DISPLAY_TARGET |
370 PIPE_BIND_SCANOUT |
371 PIPE_BIND_SHARED);
372 }
373
374 /* Check depth-stencil format support. */
375 if (usage & PIPE_BIND_DEPTH_STENCIL &&
376 r300_is_zs_format_supported(format)) {
377 retval |= PIPE_BIND_DEPTH_STENCIL;
378 }
379
380 /* Check vertex buffer format support. */
381 if (usage & PIPE_BIND_VERTEX_BUFFER &&
382 /* Half float is supported on >= RV350. */
383 (is_r400 || is_r500 || !is_half_float) &&
384 r300_translate_vertex_data_type(format) != R300_INVALID_FORMAT) {
385 retval |= PIPE_BIND_VERTEX_BUFFER;
386 }
387
388 /* Transfers are always supported. */
389 if (usage & PIPE_BIND_TRANSFER_READ)
390 retval |= PIPE_BIND_TRANSFER_READ;
391 if (usage & PIPE_BIND_TRANSFER_WRITE)
392 retval |= PIPE_BIND_TRANSFER_WRITE;
393
394 return retval == usage;
395 }
396
397 static void r300_destroy_screen(struct pipe_screen* pscreen)
398 {
399 struct r300_screen* r300screen = r300_screen(pscreen);
400 struct r300_winsys_screen *rws = r300_winsys_screen(pscreen);
401
402 util_slab_destroy(&r300screen->pool_buffers);
403
404 if (rws)
405 rws->destroy(rws);
406
407 FREE(r300screen);
408 }
409
410 static void r300_fence_reference(struct pipe_screen *screen,
411 struct pipe_fence_handle **ptr,
412 struct pipe_fence_handle *fence)
413 {
414 struct r300_fence **oldf = (struct r300_fence**)ptr;
415 struct r300_fence *newf = (struct r300_fence*)fence;
416
417 if (pipe_reference(&(*oldf)->reference, &newf->reference))
418 FREE(*oldf);
419
420 *ptr = fence;
421 }
422
423 static int r300_fence_signalled(struct pipe_screen *screen,
424 struct pipe_fence_handle *fence,
425 unsigned flags)
426 {
427 struct r300_fence *rfence = (struct r300_fence*)fence;
428
429 return rfence->signalled ? 0 : 1; /* 0 == success */
430 }
431
432 static int r300_fence_finish(struct pipe_screen *screen,
433 struct pipe_fence_handle *fence,
434 unsigned flags)
435 {
436 struct r300_fence *rfence = (struct r300_fence*)fence;
437
438 r300_finish(rfence->ctx);
439 rfence->signalled = TRUE;
440 return 0; /* 0 == success */
441 }
442
443 struct pipe_screen* r300_screen_create(struct r300_winsys_screen *rws)
444 {
445 struct r300_screen *r300screen = CALLOC_STRUCT(r300_screen);
446
447 if (!r300screen) {
448 FREE(r300screen);
449 return NULL;
450 }
451
452 r300screen->caps.pci_id = rws->get_value(rws, R300_VID_PCI_ID);
453 r300screen->caps.num_frag_pipes = rws->get_value(rws, R300_VID_GB_PIPES);
454 r300screen->caps.num_z_pipes = rws->get_value(rws, R300_VID_Z_PIPES);
455
456 r300_init_debug(r300screen);
457 r300_parse_chipset(&r300screen->caps);
458
459 r300screen->caps.index_bias_supported =
460 r300screen->caps.is_r500 &&
461 rws->get_value(rws, R300_VID_DRM_2_3_0);
462
463 util_slab_create(&r300screen->pool_buffers,
464 sizeof(struct r300_buffer), 64,
465 UTIL_SLAB_SINGLETHREADED);
466
467 r300screen->rws = rws;
468 r300screen->screen.winsys = (struct pipe_winsys*)rws;
469 r300screen->screen.destroy = r300_destroy_screen;
470 r300screen->screen.get_name = r300_get_name;
471 r300screen->screen.get_vendor = r300_get_vendor;
472 r300screen->screen.get_param = r300_get_param;
473 r300screen->screen.get_shader_param = r300_get_shader_param;
474 r300screen->screen.get_paramf = r300_get_paramf;
475 r300screen->screen.is_format_supported = r300_is_format_supported;
476 r300screen->screen.context_create = r300_create_context;
477
478 r300screen->screen.fence_reference = r300_fence_reference;
479 r300screen->screen.fence_signalled = r300_fence_signalled;
480 r300screen->screen.fence_finish = r300_fence_finish;
481
482 r300_init_screen_resource_functions(r300screen);
483
484 util_format_s3tc_init();
485
486 #ifdef HAVE_LLVM
487 lp_build_init();
488 #endif
489
490 return &r300screen->screen;
491 }