1a6c8218fd1305446349085f6e549225b3d394df
[mesa.git] / src / mesa / drivers / dri / i965 / brw_sf_state.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
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:
13
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.
17
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.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33
34 #include "brw_context.h"
35 #include "brw_state.h"
36 #include "brw_defines.h"
37 #include "main/macros.h"
38
39 static void upload_sf_vp(struct brw_context *brw)
40 {
41 GLcontext *ctx = &brw->intel.ctx;
42 const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
43 struct brw_sf_viewport sfv;
44 GLfloat y_scale, y_bias;
45 const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0);
46 const GLfloat *v = ctx->Viewport._WindowMap.m;
47
48 memset(&sfv, 0, sizeof(sfv));
49
50 if (render_to_fbo) {
51 y_scale = 1.0;
52 y_bias = 0;
53 }
54 else {
55 y_scale = -1.0;
56 y_bias = ctx->DrawBuffer->Height;
57 }
58
59 /* _NEW_VIEWPORT */
60
61 sfv.viewport.m00 = v[MAT_SX];
62 sfv.viewport.m11 = v[MAT_SY] * y_scale;
63 sfv.viewport.m22 = v[MAT_SZ] * depth_scale;
64 sfv.viewport.m30 = v[MAT_TX];
65 sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias;
66 sfv.viewport.m32 = v[MAT_TZ] * depth_scale;
67
68 /* _NEW_SCISSOR | _NEW_BUFFERS | _NEW_VIEWPORT
69 * for DrawBuffer->_[XY]{min,max}
70 */
71
72 /* The scissor only needs to handle the intersection of drawable
73 * and scissor rect, since there are no longer cliprects for shared
74 * buffers with DRI2.
75 *
76 * Note that the hardware's coordinates are inclusive, while Mesa's min is
77 * inclusive but max is exclusive.
78 */
79
80 if (ctx->DrawBuffer->_Xmin == ctx->DrawBuffer->_Xmax ||
81 ctx->DrawBuffer->_Ymin == ctx->DrawBuffer->_Ymax) {
82 /* If the scissor was out of bounds and got clamped to 0
83 * width/height at the bounds, the subtraction of 1 from
84 * maximums could produce a negative number and thus not clip
85 * anything. Instead, just provide a min > max scissor inside
86 * the bounds, which produces the expected no rendering.
87 */
88 sfv.scissor.xmin = 1;
89 sfv.scissor.xmax = 0;
90 sfv.scissor.ymin = 1;
91 sfv.scissor.ymax = 0;
92 } else if (render_to_fbo) {
93 /* texmemory: Y=0=bottom */
94 sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
95 sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
96 sfv.scissor.ymin = ctx->DrawBuffer->_Ymin;
97 sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1;
98 }
99 else {
100 /* memory: Y=0=top */
101 sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
102 sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
103 sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
104 sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
105 }
106
107 dri_bo_unreference(brw->sf.vp_bo);
108 brw->sf.vp_bo = brw_cache_data(&brw->cache, BRW_SF_VP, &sfv, sizeof(sfv),
109 NULL, 0);
110 }
111
112 const struct brw_tracked_state brw_sf_vp = {
113 .dirty = {
114 .mesa = (_NEW_VIEWPORT |
115 _NEW_SCISSOR |
116 _NEW_BUFFERS),
117 .brw = 0,
118 .cache = 0
119 },
120 .prepare = upload_sf_vp
121 };
122
123 struct brw_sf_unit_key {
124 unsigned int total_grf;
125 unsigned int urb_entry_read_length;
126
127 unsigned int nr_urb_entries, urb_size, sfsize;
128
129 GLenum front_face, cull_face;
130 unsigned pv_first:1;
131 unsigned scissor:1;
132 unsigned line_smooth:1;
133 unsigned point_sprite:1;
134 unsigned point_attenuated:1;
135 unsigned render_to_fbo:1;
136 float line_width;
137 float point_size;
138 };
139
140 static void
141 sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
142 {
143 GLcontext *ctx = &brw->intel.ctx;
144 memset(key, 0, sizeof(*key));
145
146 /* CACHE_NEW_SF_PROG */
147 key->total_grf = brw->sf.prog_data->total_grf;
148 key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
149
150 /* BRW_NEW_URB_FENCE */
151 key->nr_urb_entries = brw->urb.nr_sf_entries;
152 key->urb_size = brw->urb.vsize;
153 key->sfsize = brw->urb.sfsize;
154
155 key->scissor = ctx->Scissor.Enabled;
156 key->front_face = ctx->Polygon.FrontFace;
157
158 if (ctx->Polygon.CullFlag)
159 key->cull_face = ctx->Polygon.CullFaceMode;
160 else
161 key->cull_face = GL_NONE;
162
163 key->line_width = ctx->Line.Width;
164 key->line_smooth = ctx->Line.SmoothFlag;
165
166 key->point_sprite = ctx->Point.PointSprite;
167 key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
168 key->point_attenuated = ctx->Point._Attenuated;
169
170 /* _NEW_LIGHT */
171 key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
172
173 key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
174 }
175
176 static dri_bo *
177 sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
178 dri_bo **reloc_bufs)
179 {
180 struct intel_context *intel = &brw->intel;
181 struct brw_sf_unit_state sf;
182 dri_bo *bo;
183 int chipset_max_threads;
184 memset(&sf, 0, sizeof(sf));
185
186 sf.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
187 sf.thread0.kernel_start_pointer = brw->sf.prog_bo->offset >> 6; /* reloc */
188
189 sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
190
191 sf.thread3.dispatch_grf_start_reg = 3;
192
193 if (intel->gen == 5)
194 sf.thread3.urb_entry_read_offset = 3;
195 else
196 sf.thread3.urb_entry_read_offset = 1;
197
198 sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
199
200 sf.thread4.nr_urb_entries = key->nr_urb_entries;
201 sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
202
203 /* Each SF thread produces 1 PUE, and there can be up to 24 (Pre-Ironlake) or
204 * 48 (Ironlake) threads.
205 */
206 if (intel->gen == 5)
207 chipset_max_threads = 48;
208 else
209 chipset_max_threads = 24;
210
211 sf.thread4.max_threads = MIN2(chipset_max_threads, key->nr_urb_entries) - 1;
212
213 if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
214 sf.thread4.max_threads = 0;
215
216 if (INTEL_DEBUG & DEBUG_STATS)
217 sf.thread4.stats_enable = 1;
218
219 /* CACHE_NEW_SF_VP */
220 sf.sf5.sf_viewport_state_offset = brw->sf.vp_bo->offset >> 5; /* reloc */
221
222 sf.sf5.viewport_transform = 1;
223
224 /* _NEW_SCISSOR */
225 if (key->scissor)
226 sf.sf6.scissor = 1;
227
228 /* _NEW_POLYGON */
229 if (key->front_face == GL_CCW)
230 sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
231 else
232 sf.sf5.front_winding = BRW_FRONTWINDING_CW;
233
234 /* The viewport is inverted for rendering to a FBO, and that inverts
235 * polygon front/back orientation.
236 */
237 sf.sf5.front_winding ^= key->render_to_fbo;
238
239 switch (key->cull_face) {
240 case GL_FRONT:
241 sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
242 break;
243 case GL_BACK:
244 sf.sf6.cull_mode = BRW_CULLMODE_BACK;
245 break;
246 case GL_FRONT_AND_BACK:
247 sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
248 break;
249 case GL_NONE:
250 sf.sf6.cull_mode = BRW_CULLMODE_NONE;
251 break;
252 default:
253 assert(0);
254 break;
255 }
256
257 /* _NEW_LINE */
258 /* XXX use ctx->Const.Min/MaxLineWidth here */
259 sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
260
261 sf.sf6.line_endcap_aa_region_width = 1;
262 if (key->line_smooth)
263 sf.sf6.aa_enable = 1;
264 else if (sf.sf6.line_width <= 0x2)
265 sf.sf6.line_width = 0;
266
267 /* _NEW_BUFFERS */
268 key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
269 if (!key->render_to_fbo) {
270 /* Rendering to an OpenGL window */
271 sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
272 }
273 else {
274 /* If rendering to an FBO, the pixel coordinate system is
275 * inverted with respect to the normal OpenGL coordinate
276 * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
277 * But this value is listed as "Reserved, but not seen as useful"
278 * in Intel documentation (page 212, "Point Rasterization Rule",
279 * section 7.4 "SF Pipeline State Summary", of document
280 * "IntelĀ® 965 Express Chipset Family and IntelĀ® G35 Express
281 * Chipset Graphics Controller Programmer's Reference Manual,
282 * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
283 * available at
284 * http://intellinuxgraphics.org/documentation.html
285 * at the time of this writing).
286 *
287 * It does work on at least some devices, if not all;
288 * if devices that don't support it can be identified,
289 * the likely failure case is that points are rasterized
290 * incorrectly, which is no worse than occurs without
291 * the value, so we're using it here.
292 */
293 sf.sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
294 }
295 /* XXX clamp max depends on AA vs. non-AA */
296
297 /* _NEW_POINT */
298 sf.sf7.sprite_point = key->point_sprite;
299 sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
300 sf.sf7.use_point_size_state = !key->point_attenuated;
301 sf.sf7.aa_line_distance_mode = 0;
302
303 /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
304 */
305 if (!key->pv_first) {
306 sf.sf7.trifan_pv = 2;
307 sf.sf7.linestrip_pv = 1;
308 sf.sf7.tristrip_pv = 2;
309 } else {
310 sf.sf7.trifan_pv = 1;
311 sf.sf7.linestrip_pv = 0;
312 sf.sf7.tristrip_pv = 0;
313 }
314 sf.sf7.line_last_pixel_enable = 0;
315
316 /* Set bias for OpenGL rasterization rules:
317 */
318 sf.sf6.dest_org_vbias = 0x8;
319 sf.sf6.dest_org_hbias = 0x8;
320
321 bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
322 key, sizeof(*key),
323 reloc_bufs, 2,
324 &sf, sizeof(sf));
325
326 /* STATE_PREFETCH command description describes this state as being
327 * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain.
328 */
329 /* Emit SF program relocation */
330 dri_bo_emit_reloc(bo,
331 I915_GEM_DOMAIN_INSTRUCTION, 0,
332 sf.thread0.grf_reg_count << 1,
333 offsetof(struct brw_sf_unit_state, thread0),
334 brw->sf.prog_bo);
335
336 /* Emit SF viewport relocation */
337 dri_bo_emit_reloc(bo,
338 I915_GEM_DOMAIN_INSTRUCTION, 0,
339 sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
340 offsetof(struct brw_sf_unit_state, sf5),
341 brw->sf.vp_bo);
342
343 return bo;
344 }
345
346 static void upload_sf_unit( struct brw_context *brw )
347 {
348 struct brw_sf_unit_key key;
349 dri_bo *reloc_bufs[2];
350
351 sf_unit_populate_key(brw, &key);
352
353 reloc_bufs[0] = brw->sf.prog_bo;
354 reloc_bufs[1] = brw->sf.vp_bo;
355
356 dri_bo_unreference(brw->sf.state_bo);
357 brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT,
358 &key, sizeof(key),
359 reloc_bufs, 2,
360 NULL);
361 if (brw->sf.state_bo == NULL) {
362 brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
363 }
364 }
365
366 const struct brw_tracked_state brw_sf_unit = {
367 .dirty = {
368 .mesa = (_NEW_POLYGON |
369 _NEW_LIGHT |
370 _NEW_LINE |
371 _NEW_POINT |
372 _NEW_SCISSOR |
373 _NEW_BUFFERS),
374 .brw = BRW_NEW_URB_FENCE,
375 .cache = (CACHE_NEW_SF_VP |
376 CACHE_NEW_SF_PROG)
377 },
378 .prepare = upload_sf_unit,
379 };