r300: add hw accelerated support for different vertex data formats
[mesa.git] / src / mesa / drivers / dri / r300 / r300_context.c
1 /*
2 Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
3
4 The Weather Channel (TM) funded Tungsten Graphics to develop the
5 initial release of the Radeon 8500 driver under the XFree86 license.
6 This notice must be preserved.
7
8 Permission is hereby granted, free of charge, to any person obtaining
9 a copy of this software and associated documentation files (the
10 "Software"), to deal in the Software without restriction, including
11 without limitation the rights to use, copy, modify, merge, publish,
12 distribute, sublicense, and/or sell copies of the Software, and to
13 permit persons to whom the Software is furnished to do so, subject to
14 the following conditions:
15
16 The above copyright notice and this permission notice (including the
17 next paragraph) shall be included in all copies or substantial
18 portions of the Software.
19
20 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
23 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
24 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
25 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
26 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27
28 **************************************************************************/
29
30 /**
31 * \file
32 *
33 * \author Keith Whitwell <keith@tungstengraphics.com>
34 *
35 * \author Nicolai Haehnle <prefect_@gmx.net>
36 */
37
38 #include "main/glheader.h"
39 #include "main/api_arrayelt.h"
40 #include "main/context.h"
41 #include "main/simple_list.h"
42 #include "main/imports.h"
43 #include "main/matrix.h"
44 #include "main/extensions.h"
45 #include "main/state.h"
46 #include "main/bufferobj.h"
47 #include "main/texobj.h"
48
49 #include "swrast/swrast.h"
50 #include "swrast_setup/swrast_setup.h"
51 #include "vbo/vbo.h"
52
53 #include "tnl/tnl.h"
54 #include "tnl/t_pipeline.h"
55 #include "tnl/t_vp_build.h"
56
57 #include "drivers/common/driverfuncs.h"
58
59 #include "r300_context.h"
60 #include "radeon_context.h"
61 #include "radeon_span.h"
62 #include "r300_cmdbuf.h"
63 #include "r300_state.h"
64 #include "r300_ioctl.h"
65 #include "r300_tex.h"
66 #include "r300_emit.h"
67 #include "r300_render.h"
68 #include "r300_swtcl.h"
69 #include "radeon_bocs_wrapper.h"
70
71
72 #include "vblank.h"
73 #include "utils.h"
74 #include "xmlpool.h" /* for symbolic values of enum-type options */
75
76 #define need_GL_VERSION_2_0
77 #define need_GL_ARB_point_parameters
78 #define need_GL_ARB_vertex_program
79 #define need_GL_EXT_blend_equation_separate
80 #define need_GL_EXT_blend_func_separate
81 #define need_GL_EXT_blend_minmax
82 #define need_GL_EXT_framebuffer_object
83 #define need_GL_EXT_fog_coord
84 #define need_GL_EXT_gpu_program_parameters
85 #define need_GL_EXT_secondary_color
86 #define need_GL_EXT_stencil_two_side
87 #define need_GL_ATI_separate_stencil
88 #define need_GL_NV_vertex_program
89
90 #include "extension_helper.h"
91
92
93 const struct dri_extension card_extensions[] = {
94 /* *INDENT-OFF* */
95 {"GL_ARB_depth_texture", NULL},
96 {"GL_ARB_fragment_program", NULL},
97 {"GL_ARB_multitexture", NULL},
98 {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions},
99 {"GL_ARB_shadow", NULL},
100 {"GL_ARB_shadow_ambient", NULL},
101 {"GL_ARB_texture_border_clamp", NULL},
102 {"GL_ARB_texture_cube_map", NULL},
103 {"GL_ARB_texture_env_add", NULL},
104 {"GL_ARB_texture_env_combine", NULL},
105 {"GL_ARB_texture_env_crossbar", NULL},
106 {"GL_ARB_texture_env_dot3", NULL},
107 {"GL_ARB_texture_mirrored_repeat", NULL},
108 {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
109 {"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions},
110 {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions},
111 {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
112 {"GL_EXT_blend_subtract", NULL},
113 {"GL_EXT_packed_depth_stencil", NULL},
114 {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions },
115 {"GL_EXT_gpu_program_parameters", GL_EXT_gpu_program_parameters_functions},
116 {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions},
117 {"GL_EXT_shadow_funcs", NULL},
118 {"GL_EXT_stencil_two_side", GL_EXT_stencil_two_side_functions},
119 {"GL_EXT_stencil_wrap", NULL},
120 {"GL_EXT_texture_edge_clamp", NULL},
121 {"GL_EXT_texture_env_combine", NULL},
122 {"GL_EXT_texture_env_dot3", NULL},
123 {"GL_EXT_texture_filter_anisotropic", NULL},
124 {"GL_EXT_texture_lod_bias", NULL},
125 {"GL_EXT_texture_mirror_clamp", NULL},
126 {"GL_EXT_texture_rectangle", NULL},
127 {"GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions},
128 {"GL_ATI_texture_env_combine3", NULL},
129 {"GL_ATI_texture_mirror_once", NULL},
130 {"GL_MESA_pack_invert", NULL},
131 {"GL_MESA_ycbcr_texture", NULL},
132 {"GL_MESAX_texture_float", NULL},
133 {"GL_NV_blend_square", NULL},
134 {"GL_NV_vertex_program", GL_NV_vertex_program_functions},
135 {"GL_SGIS_generate_mipmap", NULL},
136 {NULL, NULL}
137 /* *INDENT-ON* */
138 };
139
140
141 const struct dri_extension mm_extensions[] = {
142 { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions },
143 { NULL, NULL }
144 };
145
146 /**
147 * The GL 2.0 functions are needed to make display lists work with
148 * functions added by GL_ATI_separate_stencil.
149 */
150 const struct dri_extension gl_20_extension[] = {
151 {"GL_VERSION_2_0", GL_VERSION_2_0_functions },
152 };
153
154 static const struct tnl_pipeline_stage *r300_pipeline[] = {
155
156 /* Catch any t&l fallbacks
157 */
158 &_tnl_vertex_transform_stage,
159 &_tnl_normal_transform_stage,
160 &_tnl_lighting_stage,
161 &_tnl_fog_coordinate_stage,
162 &_tnl_texgen_stage,
163 &_tnl_texture_transform_stage,
164 &_tnl_point_attenuation_stage,
165 &_tnl_vertex_program_stage,
166
167 /* Try again to go to tcl?
168 * - no good for asymmetric-twoside (do with multipass)
169 * - no good for asymmetric-unfilled (do with multipass)
170 * - good for material
171 * - good for texgen
172 * - need to manipulate a bit of state
173 *
174 * - worth it/not worth it?
175 */
176
177 /* Else do them here.
178 */
179 &_r300_render_stage,
180 &_tnl_render_stage, /* FALLBACK */
181 0,
182 };
183
184 static void r300_get_lock(radeonContextPtr rmesa)
185 {
186 drm_radeon_sarea_t *sarea = rmesa->sarea;
187
188 if (sarea->ctx_owner != rmesa->dri.hwContext) {
189 sarea->ctx_owner = rmesa->dri.hwContext;
190 if (!rmesa->radeonScreen->kernel_mm)
191 radeon_bo_legacy_texture_age(rmesa->radeonScreen->bom);
192 }
193 }
194
195 static void r300_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa)
196 {
197 /* please flush pipe do all pending work */
198 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
199 R300_SC_SCREENDOOR, 1));
200 radeon_cs_write_dword(cs, 0x0);
201 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
202 R300_SC_SCREENDOOR, 1));
203 radeon_cs_write_dword(cs, 0x00FFFFFF);
204 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
205 R300_SC_HYPERZ, 1));
206 radeon_cs_write_dword(cs, 0x0);
207 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
208 R300_US_CONFIG, 1));
209 radeon_cs_write_dword(cs, 0x0);
210 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
211 R300_ZB_CNTL, 1));
212 radeon_cs_write_dword(cs, 0x0);
213 radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen, R300_WAIT_3D));
214 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
215 R300_RB3D_DSTCACHE_CTLSTAT, 1));
216 radeon_cs_write_dword(cs, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
217 radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
218 R300_ZB_ZCACHE_CTLSTAT, 1));
219 radeon_cs_write_dword(cs, R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
220 radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen,
221 R300_WAIT_3D | R300_WAIT_3D_CLEAN));
222 }
223
224 static void r300_vtbl_pre_emit_atoms(radeonContextPtr radeon)
225 {
226 r300ContextPtr r300 = (r300ContextPtr)radeon;
227 BATCH_LOCALS(radeon);
228
229 r300->vap_flush_needed = GL_TRUE;
230
231 cp_wait(radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
232 BEGIN_BATCH_NO_AUTOSTATE(2);
233 OUT_BATCH_REGVAL(R300_TX_INVALTAGS, R300_TX_FLUSH);
234 END_BATCH();
235 end_3d(radeon);
236 }
237
238 static void r300_fallback(GLcontext *ctx, GLuint bit, GLboolean mode)
239 {
240 r300ContextPtr r300 = R300_CONTEXT(ctx);
241 if (mode)
242 r300->radeon.Fallback |= bit;
243 else
244 r300->radeon.Fallback &= ~bit;
245 }
246
247 static void r300_init_vtbl(radeonContextPtr radeon)
248 {
249 radeon->vtbl.get_lock = r300_get_lock;
250 radeon->vtbl.update_viewport_offset = r300UpdateViewportOffset;
251 radeon->vtbl.emit_cs_header = r300_vtbl_emit_cs_header;
252 radeon->vtbl.swtcl_flush = r300_swtcl_flush;
253 radeon->vtbl.pre_emit_atoms = r300_vtbl_pre_emit_atoms;
254 radeon->vtbl.fallback = r300_fallback;
255 }
256
257 static void r300InitConstValues(GLcontext *ctx, radeonScreenPtr screen)
258 {
259 r300ContextPtr r300 = R300_CONTEXT(ctx);
260
261 ctx->Const.MaxTextureImageUnits =
262 driQueryOptioni(&r300->radeon.optionCache, "texture_image_units");
263 ctx->Const.MaxTextureCoordUnits =
264 driQueryOptioni(&r300->radeon.optionCache, "texture_coord_units");
265 ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureImageUnits,
266 ctx->Const.MaxTextureCoordUnits);
267 /* FIXME: When no memory manager is available we should set this
268 * to some reasonable value based on texture memory pool size */
269 /* FIXME: r5xx limit is 4096 */
270 ctx->Const.MaxTextureLevels = 12;
271 ctx->Const.MaxTextureMaxAnisotropy = 16.0;
272 ctx->Const.MaxTextureLodBias = 16.0;
273
274 if (screen->chip_family >= CHIP_FAMILY_RV515)
275 ctx->Const.MaxTextureLevels = 13;
276 else
277 ctx->Const.MaxTextureLevels = 12;
278
279 ctx->Const.MinPointSize = 1.0;
280 ctx->Const.MinPointSizeAA = 1.0;
281 ctx->Const.MaxPointSize = R300_POINTSIZE_MAX;
282 ctx->Const.MaxPointSizeAA = R300_POINTSIZE_MAX;
283
284 ctx->Const.MinLineWidth = 1.0;
285 ctx->Const.MinLineWidthAA = 1.0;
286 ctx->Const.MaxLineWidth = R300_LINESIZE_MAX;
287 ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX;
288
289 ctx->Const.MaxDrawBuffers = 1;
290
291 /* currently bogus data */
292 if (r300->options.hw_tcl_enabled) {
293 ctx->Const.VertexProgram.MaxInstructions = VSF_MAX_FRAGMENT_LENGTH / 4;
294 ctx->Const.VertexProgram.MaxNativeInstructions =
295 VSF_MAX_FRAGMENT_LENGTH / 4;
296 ctx->Const.VertexProgram.MaxNativeAttribs = 16; /* r420 */
297 ctx->Const.VertexProgram.MaxTemps = 32;
298 ctx->Const.VertexProgram.MaxNativeTemps =
299 /*VSF_MAX_FRAGMENT_TEMPS */ 32;
300 ctx->Const.VertexProgram.MaxNativeParameters = 256; /* r420 */
301 ctx->Const.VertexProgram.MaxNativeAddressRegs = 1;
302 }
303
304 if (screen->chip_family >= CHIP_FAMILY_RV515) {
305 ctx->Const.FragmentProgram.MaxNativeTemps = R500_PFS_NUM_TEMP_REGS;
306 ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* copy i915... */
307 ctx->Const.FragmentProgram.MaxNativeParameters = R500_PFS_NUM_CONST_REGS;
308 ctx->Const.FragmentProgram.MaxNativeAluInstructions = R500_PFS_MAX_INST;
309 ctx->Const.FragmentProgram.MaxNativeTexInstructions = R500_PFS_MAX_INST;
310 ctx->Const.FragmentProgram.MaxNativeInstructions = R500_PFS_MAX_INST;
311 ctx->Const.FragmentProgram.MaxNativeTexIndirections = R500_PFS_MAX_INST;
312 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
313 } else {
314 ctx->Const.FragmentProgram.MaxNativeTemps = R300_PFS_NUM_TEMP_REGS;
315 ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* copy i915... */
316 ctx->Const.FragmentProgram.MaxNativeParameters = R300_PFS_NUM_CONST_REGS;
317 ctx->Const.FragmentProgram.MaxNativeAluInstructions = R300_PFS_MAX_ALU_INST;
318 ctx->Const.FragmentProgram.MaxNativeTexInstructions = R300_PFS_MAX_TEX_INST;
319 ctx->Const.FragmentProgram.MaxNativeInstructions = R300_PFS_MAX_ALU_INST + R300_PFS_MAX_TEX_INST;
320 ctx->Const.FragmentProgram.MaxNativeTexIndirections = R300_PFS_MAX_TEX_INDIRECT;
321 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
322 }
323 }
324
325 static void r300ParseOptions(r300ContextPtr r300, radeonScreenPtr screen)
326 {
327 struct r300_options options = { 0 };
328
329 driParseConfigFiles(&r300->radeon.optionCache, &screen->optionCache,
330 screen->driScreen->myNum, "r300");
331
332 r300->radeon.initialMaxAnisotropy = driQueryOptionf(&r300->radeon.optionCache, "def_max_anisotropy");
333
334 options.stencil_two_side_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_stencil_two_side");
335 options.s3tc_force_enabled = driQueryOptionb(&r300->radeon.optionCache, "force_s3tc_enable");
336 options.s3tc_force_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_s3tc");
337
338 if (!(screen->chip_flags & RADEON_CHIPSET_TCL) || driQueryOptioni(&r300->radeon.optionCache, "tcl_mode") == DRI_CONF_TCL_SW)
339 options.hw_tcl_enabled = 0;
340 else
341 options.hw_tcl_enabled = 1;
342
343 options.conformance_mode = !driQueryOptionb(&r300->radeon.optionCache, "disable_lowimpact_fallback");
344
345 r300->options = options;
346 }
347
348 static void r300InitGLExtensions(GLcontext *ctx)
349 {
350 r300ContextPtr r300 = R300_CONTEXT(ctx);
351
352 driInitExtensions(ctx, card_extensions, GL_TRUE);
353 if (r300->radeon.radeonScreen->kernel_mm)
354 driInitExtensions(ctx, mm_extensions, GL_FALSE);
355
356 if (r300->options.stencil_two_side_disabled)
357 _mesa_disable_extension(ctx, "GL_EXT_stencil_two_side");
358
359 if (ctx->Mesa_DXTn && !r300->options.s3tc_force_enabled) {
360 _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
361 _mesa_enable_extension(ctx, "GL_S3_s3tc");
362 } else if (r300->options.s3tc_force_disabled) {
363 _mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc");
364 }
365 }
366
367 /* Create the device specific rendering context.
368 */
369 GLboolean r300CreateContext(const __GLcontextModes * glVisual,
370 __DRIcontextPrivate * driContextPriv,
371 void *sharedContextPrivate)
372 {
373 __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
374 radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private);
375 struct dd_function_table functions;
376 r300ContextPtr r300;
377 GLcontext *ctx;
378
379 assert(glVisual);
380 assert(driContextPriv);
381 assert(screen);
382
383 r300 = (r300ContextPtr) CALLOC(sizeof(*r300));
384 if (!r300)
385 return GL_FALSE;
386
387 r300ParseOptions(r300, screen);
388
389 r300_init_vtbl(&r300->radeon);
390
391 _mesa_init_driver_functions(&functions);
392 r300InitIoctlFuncs(&functions);
393 r300InitStateFuncs(&functions);
394 r300InitTextureFuncs(&functions);
395 r300InitShaderFuncs(&functions);
396
397 if (!radeonInitContext(&r300->radeon, &functions,
398 glVisual, driContextPriv,
399 sharedContextPrivate)) {
400 FREE(r300);
401 return GL_FALSE;
402 }
403
404 ctx = r300->radeon.glCtx;
405
406 r300->fallback = 0;
407 if (r300->options.hw_tcl_enabled)
408 ctx->VertexProgram._MaintainTnlProgram = GL_TRUE;
409
410 ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE;
411
412 r300InitConstValues(ctx, screen);
413
414 /* Initialize the software rasterizer and helper modules.
415 */
416 _swrast_CreateContext(ctx);
417 _vbo_CreateContext(ctx);
418 _tnl_CreateContext(ctx);
419 _swsetup_CreateContext(ctx);
420 _swsetup_Wakeup(ctx);
421
422 /* Install the customized pipeline:
423 */
424 _tnl_destroy_pipeline(ctx);
425 _tnl_install_pipeline(ctx, r300_pipeline);
426 TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
427
428 /* Configure swrast and TNL to match hardware characteristics:
429 */
430 _swrast_allow_pixel_fog(ctx, GL_FALSE);
431 _swrast_allow_vertex_fog(ctx, GL_TRUE);
432 _tnl_allow_pixel_fog(ctx, GL_FALSE);
433 _tnl_allow_vertex_fog(ctx, GL_TRUE);
434
435 if (r300->options.hw_tcl_enabled) {
436 r300InitDraw(ctx);
437 } else {
438 r300InitSwtcl(ctx);
439 }
440
441 radeon_fbo_init(&r300->radeon);
442 radeonInitSpanFuncs( ctx );
443 r300InitCmdBuf(r300);
444 r300InitState(r300);
445 r300InitShaderFunctions(r300);
446
447 if (screen->chip_family == CHIP_FAMILY_RS600 || screen->chip_family == CHIP_FAMILY_RS690 ||
448 screen->chip_family == CHIP_FAMILY_RS740) {
449 r300->radeon.texture_row_align = 64;
450 }
451
452 r300InitGLExtensions(ctx);
453
454 return GL_TRUE;
455 }
456