src/gallium/drivers/r300/r300_render.c

   1 /*
   2  * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  22
  23 /* r300_render: Vertex and index buffer primitive emission. Contains both
  24  * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
  25
  26 #include "draw/draw_context.h"
  27 #include "draw/draw_vbuf.h"
  28
  29 #include "pipe/p_inlines.h"
  30
  31 #include "util/u_memory.h"
  32 #include "util/u_prim.h"
  33
  34 #include "r300_cs.h"
  35 #include "r300_context.h"
  36 #include "r300_emit.h"
  37 #include "r300_reg.h"
  38 #include "r300_render.h"
  39 #include "r300_state_derived.h"
  40 #include "r300_vbo.h"
  41
  42 /* r300_render: Vertex and index buffer primitive emission. */
  43 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  44
  45 uint32_t r300_translate_primitive(unsigned prim)
  46 {
  47     switch (prim) {
  48         case PIPE_PRIM_POINTS:
  49             return R300_VAP_VF_CNTL__PRIM_POINTS;
  50         case PIPE_PRIM_LINES:
  51             return R300_VAP_VF_CNTL__PRIM_LINES;
  52         case PIPE_PRIM_LINE_LOOP:
  53             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  54         case PIPE_PRIM_LINE_STRIP:
  55             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  56         case PIPE_PRIM_TRIANGLES:
  57             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  58         case PIPE_PRIM_TRIANGLE_STRIP:
  59             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  60         case PIPE_PRIM_TRIANGLE_FAN:
  61             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  62         case PIPE_PRIM_QUADS:
  63             return R300_VAP_VF_CNTL__PRIM_QUADS;
  64         case PIPE_PRIM_QUAD_STRIP:
  65             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  66         case PIPE_PRIM_POLYGON:
  67             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  68         default:
  69             return 0;
  70     }
  71 }
  72
  73 static void r300_emit_draw_arrays(struct r300_context *r300,
  74                                   unsigned mode,
  75                                   unsigned count)
  76 {
  77     CS_LOCALS(r300);
  78     assert(count < 65536);
  79
  80     BEGIN_CS(4);
  81     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count);
  82     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
  83     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
  84            r300_translate_primitive(mode));
  85     END_CS;
  86 }
  87
  88 static void r300_emit_draw_elements(struct r300_context *r300,
  89                                     struct pipe_buffer* indexBuffer,
  90                                     unsigned indexSize,
  91                                     unsigned minIndex,
  92                                     unsigned maxIndex,
  93                                     unsigned mode,
  94                                     unsigned start,
  95                                     unsigned count)
  96 {
  97     CS_LOCALS(r300);
  98     assert(indexSize == 4 || indexSize == 2);
  99     assert(count < 65536);
 100     assert((start * indexSize)  % 4 == 0);
 101
 102     uint32_t size_dwords;
 103     uint32_t skip_dwords = indexSize * start / sizeof(uint32_t);
 104     assert(skip_dwords == 0);
 105
 106     BEGIN_CS(10);
 107     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 108     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 109     if (indexSize == 4) {
 110         size_dwords = count + start;
 111         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 112                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 113                r300_translate_primitive(mode));
 114     } else {
 115         size_dwords = (count + start + 1) / 2;
 116         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 117                r300_translate_primitive(mode));
 118     }
 119
 120     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 121     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 122            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 123     OUT_CS(skip_dwords);
 124     OUT_CS(size_dwords);
 125     /* XXX hax */
 126     cs_winsys->write_cs_reloc(cs_winsys,
 127                               indexBuffer,
 128                               RADEON_GEM_DOMAIN_GTT,
 129                               0,
 130                               0);
 131     cs_count -= 2;
 132
 133     END_CS;
 134 }
 135
 136
 137 static boolean setup_vertex_buffers(struct r300_context *r300)
 138 {
 139     unsigned vbuf_count = r300->aos_count;
 140     struct pipe_vertex_buffer *vbuf= r300->vertex_buffer;
 141     struct pipe_vertex_element *velem= r300->vertex_element;
 142     bool invalid = false;
 143
 144 validate:
 145     for (int i = 0; i < vbuf_count; i++) {
 146         if (!r300->winsys->add_buffer(r300->winsys, vbuf[velem[i].vertex_buffer_index].buffer,
 147             RADEON_GEM_DOMAIN_GTT, 0)) {
 148             r300->context.flush(&r300->context, 0, NULL);
 149             goto validate;
 150         }
 151     }
 152
 153     if (!r300->winsys->validate(r300->winsys)) {
 154         r300->context.flush(&r300->context, 0, NULL);
 155         if (invalid) {
 156             /* Well, hell. */
 157             debug_printf("r300: Stuck in validation loop, gonna quit now.");
 158             exit(1);
 159         }
 160         invalid = true;
 161         goto validate;
 162     }
 163
 164     return invalid;
 165 }
 166
 167 /* This is the fast-path drawing & emission for HW TCL. */
 168 boolean r300_draw_range_elements(struct pipe_context* pipe,
 169                                  struct pipe_buffer* indexBuffer,
 170                                  unsigned indexSize,
 171                                  unsigned minIndex,
 172                                  unsigned maxIndex,
 173                                  unsigned mode,
 174                                  unsigned start,
 175                                  unsigned count)
 176 {
 177     struct r300_context* r300 = r300_context(pipe);
 178
 179     if (!u_trim_pipe_prim(mode, &count))
 180         return false;
 181
 182     r300_update_derived_state(r300);
 183
 184     setup_vertex_buffers(r300);
 185
 186     setup_vertex_attributes(r300);
 187
 188     setup_index_buffer(r300, indexBuffer, indexSize);
 189
 190     r300_emit_dirty_state(r300);
 191
 192     r300_emit_aos(r300, 0);
 193
 194     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 195                             mode, start, count);
 196
 197     return TRUE;
 198 }
 199
 200 /* Simple helpers for context setup. Should probably be moved to util. */
 201 boolean r300_draw_elements(struct pipe_context* pipe,
 202                            struct pipe_buffer* indexBuffer,
 203                            unsigned indexSize, unsigned mode,
 204                            unsigned start, unsigned count)
 205 {
 206     return pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 207                                      mode, start, count);
 208 }
 209
 210 boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 211                          unsigned start, unsigned count)
 212 {
 213     struct r300_context* r300 = r300_context(pipe);
 214
 215     if (!u_trim_pipe_prim(mode, &count))
 216         return false;
 217
 218     r300_update_derived_state(r300);
 219
 220     setup_vertex_buffers(r300);
 221
 222     setup_vertex_attributes(r300);
 223
 224     r300_emit_dirty_state(r300);
 225
 226     r300_emit_aos(r300, start);
 227
 228     r300_emit_draw_arrays(r300, mode, count);
 229
 230     return TRUE;
 231 }
 232
 233 /****************************************************************************
 234  * The rest of this file is for SW TCL rendering only. Please be polite and *
 235  * keep these functions separated so that they are easier to locate. ~C.    *
 236  ***************************************************************************/
 237
 238 /* Draw-based drawing for SW TCL chipsets. */
 239 boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 240                                        struct pipe_buffer* indexBuffer,
 241                                        unsigned indexSize,
 242                                        unsigned minIndex,
 243                                        unsigned maxIndex,
 244                                        unsigned mode,
 245                                        unsigned start,
 246                                        unsigned count)
 247 {
 248     assert(0);
 249 #if 0
 250     struct r300_context* r300 = r300_context(pipe);
 251     int i;
 252
 253     if (!u_trim_pipe_prim(mode, &count)) {
 254         return FALSE;
 255     }
 256
 257     for (i = 0; i < r300->vertex_buffer_count; i++) {
 258         void* buf = pipe_buffer_map(pipe->screen,
 259                                     r300->vertex_buffers[i].buffer,
 260                                     PIPE_BUFFER_USAGE_CPU_READ);
 261         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 262     }
 263
 264     if (indexBuffer) {
 265         void* indices = pipe_buffer_map(pipe->screen, indexBuffer,
 266                                         PIPE_BUFFER_USAGE_CPU_READ);
 267         draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 268                                              minIndex, maxIndex, indices);
 269     } else {
 270         draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 271     }
 272
 273     draw_set_mapped_constant_buffer(r300->draw,
 274             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 275             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 276                 (sizeof(float) * 4));
 277
 278     draw_arrays(r300->draw, mode, start, count);
 279
 280     for (i = 0; i < r300->vertex_buffer_count; i++) {
 281         pipe_buffer_unmap(pipe->screen, r300->vertex_buffers[i].buffer);
 282         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 283     }
 284
 285     if (indexBuffer) {
 286         pipe_buffer_unmap(pipe->screen, indexBuffer);
 287         draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 288                                              start + count - 1, NULL);
 289     }
 290 #endif
 291     return TRUE;
 292 }
 293
 294 /* Object for rendering using Draw. */
 295 struct r300_render {
 296     /* Parent class */
 297     struct vbuf_render base;
 298
 299     /* Pipe context */
 300     struct r300_context* r300;
 301
 302     /* Vertex information */
 303     size_t vertex_size;
 304     unsigned prim;
 305     unsigned hwprim;
 306
 307     /* VBO */
 308     struct pipe_buffer* vbo;
 309     size_t vbo_size;
 310     size_t vbo_offset;
 311     size_t vbo_max_used;
 312     void * vbo_ptr;
 313 };
 314
 315 static INLINE struct r300_render*
 316 r300_render(struct vbuf_render* render)
 317 {
 318     return (struct r300_render*)render;
 319 }
 320
 321 static const struct vertex_info*
 322 r300_render_get_vertex_info(struct vbuf_render* render)
 323 {
 324     struct r300_render* r300render = r300_render(render);
 325     struct r300_context* r300 = r300render->r300;
 326
 327     r300_update_derived_state(r300);
 328
 329     return &r300->vertex_info->vinfo;
 330 }
 331
 332 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 333                                                    ushort vertex_size,
 334                                                    ushort count)
 335 {
 336     struct r300_render* r300render = r300_render(render);
 337     struct r300_context* r300 = r300render->r300;
 338     struct pipe_screen* screen = r300->context.screen;
 339     size_t size = (size_t)vertex_size * (size_t)count;
 340
 341     if (size + r300render->vbo_offset > r300render->vbo_size)
 342     {
 343         pipe_buffer_reference(&r300->vbo, NULL);
 344         r300render->vbo = pipe_buffer_create(screen,
 345                                              64,
 346                                              PIPE_BUFFER_USAGE_VERTEX,
 347                                              R300_MAX_VBO_SIZE);
 348         r300render->vbo_offset = 0;
 349         r300render->vbo_size = R300_MAX_VBO_SIZE;
 350     }
 351
 352     r300render->vertex_size = vertex_size;
 353     r300->vbo = r300render->vbo;
 354     r300->vbo_offset = r300render->vbo_offset;
 355
 356     return (r300render->vbo) ? TRUE : FALSE;
 357 }
 358
 359 static void* r300_render_map_vertices(struct vbuf_render* render)
 360 {
 361     struct r300_render* r300render = r300_render(render);
 362     struct pipe_screen* screen = r300render->r300->context.screen;
 363
 364     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 365                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 366
 367     return (r300render->vbo_ptr + r300render->vbo_offset);
 368 }
 369
 370 static void r300_render_unmap_vertices(struct vbuf_render* render,
 371                                              ushort min,
 372                                              ushort max)
 373 {
 374     struct r300_render* r300render = r300_render(render);
 375     struct pipe_screen* screen = r300render->r300->context.screen;
 376     CS_LOCALS(r300render->r300);
 377     BEGIN_CS(2);
 378     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 379     END_CS;
 380
 381     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 382                                     r300render->vertex_size * (max + 1));
 383     pipe_buffer_unmap(screen, r300render->vbo);
 384 }
 385
 386 static void r300_render_release_vertices(struct vbuf_render* render)
 387 {
 388     struct r300_render* r300render = r300_render(render);
 389
 390     r300render->vbo_offset += r300render->vbo_max_used;
 391     r300render->vbo_max_used = 0;
 392 }
 393
 394 static boolean r300_render_set_primitive(struct vbuf_render* render,
 395                                                unsigned prim)
 396 {
 397     struct r300_render* r300render = r300_render(render);
 398
 399     r300render->prim = prim;
 400     r300render->hwprim = r300_translate_primitive(prim);
 401
 402     return TRUE;
 403 }
 404
 405 static void r300_render_draw_arrays(struct vbuf_render* render,
 406                                           unsigned start,
 407                                           unsigned count)
 408 {
 409     struct r300_render* r300render = r300_render(render);
 410     struct r300_context* r300 = r300render->r300;
 411
 412     CS_LOCALS(r300);
 413
 414     r300_emit_dirty_state(r300);
 415
 416     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 417
 418     BEGIN_CS(2);
 419     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 420     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 421            r300render->hwprim);
 422     END_CS;
 423 }
 424
 425 static void r300_render_draw(struct vbuf_render* render,
 426                                    const ushort* indices,
 427                                    uint count)
 428 {
 429     struct r300_render* r300render = r300_render(render);
 430     struct r300_context* r300 = r300render->r300;
 431     int i;
 432
 433     CS_LOCALS(r300);
 434
 435     r300_emit_dirty_state(r300);
 436
 437     BEGIN_CS(2 + (count+1)/2);
 438     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 439     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 440            r300render->hwprim);
 441     for (i = 0; i < count-1; i += 2) {
 442         OUT_CS(indices[i+1] << 16 | indices[i]);
 443     }
 444     if (count % 2) {
 445         OUT_CS(indices[count-1]);
 446     }
 447     END_CS;
 448 }
 449
 450 static void r300_render_destroy(struct vbuf_render* render)
 451 {
 452     FREE(render);
 453 }
 454
 455 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 456 {
 457     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 458
 459     r300render->r300 = r300;
 460
 461     /* XXX find real numbers plz */
 462     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 463     r300render->base.max_indices = 16 * 1024;
 464
 465     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 466     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 467     r300render->base.map_vertices = r300_render_map_vertices;
 468     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 469     r300render->base.set_primitive = r300_render_set_primitive;
 470     r300render->base.draw = r300_render_draw;
 471     r300render->base.draw_arrays = r300_render_draw_arrays;
 472     r300render->base.release_vertices = r300_render_release_vertices;
 473     r300render->base.destroy = r300_render_destroy;
 474
 475     r300render->vbo = NULL;
 476     r300render->vbo_size = 0;
 477     r300render->vbo_offset = 0;
 478
 479     return &r300render->base;
 480 }
 481
 482 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 483 {
 484     struct vbuf_render* render;
 485     struct draw_stage* stage;
 486
 487     render = r300_render_create(r300);
 488
 489     if (!render) {
 490         return NULL;
 491     }
 492
 493     stage = draw_vbuf_stage(r300->draw, render);
 494
 495     if (!stage) {
 496         render->destroy(render);
 497         return NULL;
 498     }
 499
 500     draw_set_render(r300->draw, render);
 501
 502     return stage;
 503 }