brw_draw.c revision 2fa2dd3908c783663ca421134cde82e9b6a38a0d
1/************************************************************************** 2 * 3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * 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, sub license, 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 portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28#include <stdlib.h> 29 30#include "glheader.h" 31#include "context.h" 32#include "state.h" 33#include "api_validate.h" 34#include "enums.h" 35 36#include "brw_draw.h" 37#include "brw_defines.h" 38#include "brw_context.h" 39#include "brw_state.h" 40#include "brw_fallback.h" 41 42#include "intel_ioctl.h" 43#include "intel_batchbuffer.h" 44#include "intel_buffer_objects.h" 45 46#include "tnl/tnl.h" 47#include "vbo/vbo_context.h" 48#include "swrast/swrast.h" 49#include "swrast_setup/swrast_setup.h" 50 51#define FILE_DEBUG_FLAG DEBUG_BATCH 52 53static GLuint hw_prim[GL_POLYGON+1] = { 54 _3DPRIM_POINTLIST, 55 _3DPRIM_LINELIST, 56 _3DPRIM_LINELOOP, 57 _3DPRIM_LINESTRIP, 58 _3DPRIM_TRILIST, 59 _3DPRIM_TRISTRIP, 60 _3DPRIM_TRIFAN, 61 _3DPRIM_QUADLIST, 62 _3DPRIM_QUADSTRIP, 63 _3DPRIM_POLYGON 64}; 65 66 67static const GLenum reduced_prim[GL_POLYGON+1] = { 68 GL_POINTS, 69 GL_LINES, 70 GL_LINES, 71 GL_LINES, 72 GL_TRIANGLES, 73 GL_TRIANGLES, 74 GL_TRIANGLES, 75 GL_TRIANGLES, 76 GL_TRIANGLES, 77 GL_TRIANGLES 78}; 79 80 81/* When the primitive changes, set a state bit and re-validate. Not 82 * the nicest and would rather deal with this by having all the 83 * programs be immune to the active primitive (ie. cope with all 84 * possibilities). That may not be realistic however. 85 */ 86static GLuint brw_set_prim(struct brw_context *brw, GLenum prim, GLboolean *need_flush) 87{ 88 int ret; 89 if (INTEL_DEBUG & DEBUG_PRIMS) 90 _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim)); 91 92 /* Slight optimization to avoid the GS program when not needed: 93 */ 94 if (prim == GL_QUAD_STRIP && 95 brw->attribs.Light->ShadeModel != GL_FLAT && 96 brw->attribs.Polygon->FrontMode == GL_FILL && 97 brw->attribs.Polygon->BackMode == GL_FILL) 98 prim = GL_TRIANGLE_STRIP; 99 100 if (prim != brw->primitive) { 101 brw->primitive = prim; 102 brw->state.dirty.brw |= BRW_NEW_PRIMITIVE; 103 104 if (reduced_prim[prim] != brw->intel.reduced_primitive) { 105 brw->intel.reduced_primitive = reduced_prim[prim]; 106 brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE; 107 } 108 109 ret = brw_validate_state(brw); 110 if (ret) 111 *need_flush = GL_TRUE; 112 } 113 114 return hw_prim[prim]; 115} 116 117 118static GLuint trim(GLenum prim, GLuint length) 119{ 120 if (prim == GL_QUAD_STRIP) 121 return length > 3 ? (length - length % 2) : 0; 122 else if (prim == GL_QUADS) 123 return length - length % 4; 124 else 125 return length; 126} 127 128 129static void brw_emit_prim( struct brw_context *brw, 130 const struct _mesa_prim *prim ) 131 132{ 133 struct brw_3d_primitive prim_packet; 134 GLboolean need_flush = GL_FALSE; 135 136 if (INTEL_DEBUG & DEBUG_PRIMS) 137 _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode), 138 prim->start, prim->count); 139 140 prim_packet.header.opcode = CMD_3D_PRIM; 141 prim_packet.header.length = sizeof(prim_packet)/4 - 2; 142 prim_packet.header.pad = 0; 143 prim_packet.header.topology = brw_set_prim(brw, prim->mode, &need_flush); 144 prim_packet.header.indexed = prim->indexed; 145 146 prim_packet.verts_per_instance = trim(prim->mode, prim->count); 147 prim_packet.start_vert_location = prim->start; 148 prim_packet.instance_count = 1; 149 prim_packet.start_instance_location = 0; 150 prim_packet.base_vert_location = 0; 151 152 if (prim_packet.verts_per_instance) { 153 intel_batchbuffer_data( brw->intel.batch, &prim_packet, 154 sizeof(prim_packet), LOOP_CLIPRECTS); 155 } 156 157 assert(need_flush == GL_FALSE); 158} 159 160static void brw_merge_inputs( struct brw_context *brw, 161 const struct gl_client_array *arrays[]) 162{ 163 struct brw_vertex_element *inputs = brw->vb.inputs; 164 struct brw_vertex_info old = brw->vb.info; 165 GLuint i; 166 167 memset(inputs, 0, sizeof(*inputs)); 168 memset(&brw->vb.info, 0, sizeof(brw->vb.info)); 169 170 for (i = 0; i < VERT_ATTRIB_MAX; i++) { 171 brw->vb.inputs[i].glarray = arrays[i]; 172 173 /* XXX: metaops passes null arrays */ 174 if (arrays[i]) { 175 if (arrays[i]->StrideB != 0) 176 brw->vb.info.varying |= 1 << i; 177 178 brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2); 179 } 180 } 181 182 /* Raise statechanges if input sizes and varying have changed: 183 */ 184 if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0) 185 brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS; 186 187 if (brw->vb.info.varying != old.varying) 188 brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING; 189} 190 191/* XXX: could split the primitive list to fallback only on the 192 * non-conformant primitives. 193 */ 194static GLboolean check_fallbacks( struct brw_context *brw, 195 const struct _mesa_prim *prim, 196 GLuint nr_prims ) 197{ 198 GLuint i; 199 200 if (!brw->intel.strict_conformance) 201 return GL_FALSE; 202 203 if (brw->attribs.Polygon->SmoothFlag) { 204 for (i = 0; i < nr_prims; i++) 205 if (reduced_prim[prim[i].mode] == GL_TRIANGLES) 206 return GL_TRUE; 207 } 208 209 /* BRW hardware will do AA lines, but they are non-conformant it 210 * seems. TBD whether we keep this fallback: 211 */ 212 if (brw->attribs.Line->SmoothFlag) { 213 for (i = 0; i < nr_prims; i++) 214 if (reduced_prim[prim[i].mode] == GL_LINES) 215 return GL_TRUE; 216 } 217 218 /* Stipple -- these fallbacks could be resolved with a little 219 * bit of work? 220 */ 221 if (brw->attribs.Line->StippleFlag) { 222 for (i = 0; i < nr_prims; i++) { 223 /* GS doesn't get enough information to know when to reset 224 * the stipple counter?!? 225 */ 226 if (prim[i].mode == GL_LINE_LOOP) 227 return GL_TRUE; 228 229 if (prim[i].mode == GL_POLYGON && 230 (brw->attribs.Polygon->FrontMode == GL_LINE || 231 brw->attribs.Polygon->BackMode == GL_LINE)) 232 return GL_TRUE; 233 } 234 } 235 236 237 if (brw->attribs.Point->SmoothFlag) { 238 for (i = 0; i < nr_prims; i++) 239 if (prim[i].mode == GL_POINTS) 240 return GL_TRUE; 241 } 242 243 return GL_FALSE; 244} 245 246/* May fail if out of video memory for texture or vbo upload, or on 247 * fallback conditions. 248 */ 249static GLboolean brw_try_draw_prims( GLcontext *ctx, 250 const struct gl_client_array *arrays[], 251 const struct _mesa_prim *prim, 252 GLuint nr_prims, 253 const struct _mesa_index_buffer *ib, 254 GLuint min_index, 255 GLuint max_index ) 256{ 257 struct intel_context *intel = intel_context(ctx); 258 struct brw_context *brw = brw_context(ctx); 259 GLboolean retval = GL_FALSE; 260 GLuint i; 261 GLuint ib_offset; 262 dri_bo *ib_bo; 263 GLboolean force_flush = GL_FALSE; 264 int ret; 265 266 if (ctx->NewState) 267 _mesa_update_state( ctx ); 268 269 brw_validate_textures( brw ); 270 271 /* Bind all inputs, derive varying and size information: 272 */ 273 brw_merge_inputs( brw, arrays ); 274 275 /* Have to validate state quite late. Will rebuild tnl_program, 276 * which depends on varying information. 277 * 278 * Note this is where brw->vs->prog_data.inputs_read is calculated, 279 * so can't access it earlier. 280 */ 281 282 LOCK_HARDWARE(intel); 283 284 if (brw->intel.numClipRects == 0) { 285 UNLOCK_HARDWARE(intel); 286 return GL_TRUE; 287 } 288 289 { 290 /* Flush the batch if it's approaching full, so that we don't wrap while 291 * we've got validated state that needs to be in the same batch as the 292 * primitives. This fraction is just a guess (minimal full state plus 293 * a primitive is around 512 bytes), and would be better if we had 294 * an upper bound of how much we might emit in a single 295 * brw_try_draw_prims(). 296 */ 297 flush: 298 if (force_flush) 299 brw->no_batch_wrap = GL_FALSE; 300 301 if (intel->batch->ptr - intel->batch->map > intel->batch->size * 3 / 4 302 /* brw_emit_prim may change the cliprect_mode to LOOP_CLIPRECTS */ 303 || intel->batch->cliprect_mode != LOOP_CLIPRECTS || (force_flush == GL_TRUE)) 304 intel_batchbuffer_flush(intel->batch); 305 306 force_flush = GL_FALSE; 307 brw->no_batch_wrap = GL_TRUE; 308 309 /* Set the first primitive early, ahead of validate_state: 310 */ 311 brw_set_prim(brw, prim[0].mode, &force_flush); 312 313 /* XXX: Need to separate validate and upload of state. 314 */ 315 ret = brw_validate_state( brw ); 316 if (ret) { 317 force_flush = GL_TRUE; 318 goto flush; 319 } 320 321 /* need to account for index buffer and vertex buffer */ 322 if (ib) { 323 ret = brw_prepare_indices( brw, ib , &ib_bo, &ib_offset); 324 if (ret) { 325 force_flush = GL_TRUE; 326 goto flush; 327 } 328 } 329 330 ret = brw_prepare_vertices( brw, min_index, max_index); 331 if (ret < 0) 332 goto out; 333 334 if (ret > 0) { 335 force_flush = GL_TRUE; 336 goto flush; 337 } 338 339 340 341 /* Various fallback checks: 342 */ 343 if (brw->intel.Fallback) 344 goto out; 345 346 if (check_fallbacks( brw, prim, nr_prims )) 347 goto out; 348 349 /* Upload index, vertex data: 350 */ 351 if (ib) 352 brw_emit_indices( brw, ib, ib_bo, ib_offset); 353 354 brw_emit_vertices( brw, min_index, max_index); 355 356 for (i = 0; i < nr_prims; i++) { 357 brw_emit_prim(brw, &prim[i]); 358 } 359 360 retval = GL_TRUE; 361 } 362 363 out: 364 365 brw->no_batch_wrap = GL_FALSE; 366 367 UNLOCK_HARDWARE(intel); 368 369 if (!retval) 370 DBG("%s failed\n", __FUNCTION__); 371 372 return retval; 373} 374 375static GLboolean brw_need_rebase( GLcontext *ctx, 376 const struct gl_client_array *arrays[], 377 const struct _mesa_index_buffer *ib, 378 GLuint min_index ) 379{ 380 if (min_index == 0) 381 return GL_FALSE; 382 383 if (ib) { 384 if (!vbo_all_varyings_in_vbos(arrays)) 385 return GL_TRUE; 386 else 387 return GL_FALSE; 388 } 389 else { 390 /* Hmm. This isn't quite what I wanted. BRW can actually 391 * handle the mixed case well enough that we shouldn't need to 392 * rebase. However, it's probably not very common, nor hugely 393 * expensive to do it this way: 394 */ 395 if (!vbo_all_varyings_in_vbos(arrays)) 396 return GL_TRUE; 397 else 398 return GL_FALSE; 399 } 400} 401 402 403void brw_draw_prims( GLcontext *ctx, 404 const struct gl_client_array *arrays[], 405 const struct _mesa_prim *prim, 406 GLuint nr_prims, 407 const struct _mesa_index_buffer *ib, 408 GLuint min_index, 409 GLuint max_index ) 410{ 411 GLboolean retval; 412 413 /* Decide if we want to rebase. If so we end up recursing once 414 * only into this function. 415 */ 416 if (brw_need_rebase( ctx, arrays, ib, min_index )) { 417 vbo_rebase_prims( ctx, arrays, 418 prim, nr_prims, 419 ib, min_index, max_index, 420 brw_draw_prims ); 421 422 return; 423 } 424 425 426 /* Make a first attempt at drawing: 427 */ 428 retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); 429 430 /* Otherwise, we really are out of memory. Pass the drawing 431 * command to the software tnl module and which will in turn call 432 * swrast to do the drawing. 433 */ 434 if (!retval) { 435 _swsetup_Wakeup(ctx); 436 _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); 437 } 438} 439 440void brw_draw_init( struct brw_context *brw ) 441{ 442 GLcontext *ctx = &brw->intel.ctx; 443 struct vbo_context *vbo = vbo_context(ctx); 444 445 /* Register our drawing function: 446 */ 447 vbo->draw_prims = brw_draw_prims; 448} 449 450void brw_draw_destroy( struct brw_context *brw ) 451{ 452 if (brw->vb.upload.bo != NULL) { 453 dri_bo_unreference(brw->vb.upload.bo); 454 brw->vb.upload.bo = NULL; 455 } 456} 457