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 #include "draw/draw_context.h"
  24 #include "draw/draw_vbuf.h"
  25
  26 #include "pipe/p_inlines.h"
  27
  28 #include "util/u_memory.h"
  29
  30 #include "r300_cs.h"
  31 #include "r300_context.h"
  32 #include "r300_emit.h"
  33 #include "r300_reg.h"
  34 #include "r300_render.h"
  35 #include "r300_state_derived.h"
  36
  37 /* r300_render: Vertex and index buffer primitive emission. */
  38 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  39
  40 static uint32_t r300_translate_primitive(unsigned prim)
  41 {
  42     switch (prim) {
  43         case PIPE_PRIM_POINTS:
  44             return R300_VAP_VF_CNTL__PRIM_POINTS;
  45         case PIPE_PRIM_LINES:
  46             return R300_VAP_VF_CNTL__PRIM_LINES;
  47         case PIPE_PRIM_LINE_LOOP:
  48             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  49         case PIPE_PRIM_LINE_STRIP:
  50             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  51         case PIPE_PRIM_TRIANGLES:
  52             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  53         case PIPE_PRIM_TRIANGLE_STRIP:
  54             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  55         case PIPE_PRIM_TRIANGLE_FAN:
  56             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  57         case PIPE_PRIM_QUADS:
  58             return R300_VAP_VF_CNTL__PRIM_QUADS;
  59         case PIPE_PRIM_QUAD_STRIP:
  60             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  61         case PIPE_PRIM_POLYGON:
  62             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  63         default:
  64             return 0;
  65     }
  66 }
  67
  68 /* This is the fast-path drawing & emission for HW TCL. */
  69 boolean r300_draw_range_elements(struct pipe_context* pipe,
  70                                  struct pipe_buffer* indexBuffer,
  71                                  unsigned indexSize,
  72                                  unsigned minIndex,
  73                                  unsigned maxIndex,
  74                                  unsigned mode,
  75                                  unsigned start,
  76                                  unsigned count)
  77 {
  78     struct r300_context* r300 = r300_context(pipe);
  79     CS_LOCALS(r300);
  80     uint32_t prim = r300_translate_primitive(mode);
  81     struct pipe_vertex_buffer* aos = r300->vertex_buffers;
  82     unsigned aos_count = r300->vertex_buffer_count;
  83     short* indices;
  84     unsigned packet_size;
  85     unsigned i;
  86     bool invalid = FALSE;
  87
  88 validate:
  89     for (i = 0; i < aos_count; i++) {
  90         if (!r300->winsys->add_buffer(r300->winsys, aos[i].buffer,
  91                     RADEON_GEM_DOMAIN_GTT, 0)) {
  92             pipe->flush(pipe, 0, NULL);
  93             goto validate;
  94         }
  95     }
  96     if (!r300->winsys->validate(r300->winsys)) {
  97         pipe->flush(pipe, 0, NULL);
  98         if (invalid) {
  99             /* Well, hell. */
 100             debug_printf("r300: Stuck in validation loop, gonna quit now.");
 101             exit(1);
 102         }
 103         invalid = TRUE;
 104         goto validate;
 105     }
 106
 107     r300_emit_dirty_state(r300);
 108
 109     packet_size = (aos_count >> 1) * 3 + (aos_count & 1) * 2;
 110
 111     BEGIN_CS(3 + packet_size + (aos_count * 2));
 112     OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
 113     OUT_CS(aos_count);
 114     for (i = 0; i < aos_count - 1; i += 2) {
 115         OUT_CS(aos[i].stride |
 116             (aos[i].stride << 8) |
 117             (aos[i + 1].stride << 16) |
 118             (aos[i + 1].stride << 24));
 119         OUT_CS(aos[i].buffer_offset + start * 4 * aos[i].stride);
 120         OUT_CS(aos[i + 1].buffer_offset + start * 4 * aos[i + 1].stride);
 121     }
 122     if (aos_count & 1) {
 123         OUT_CS(aos[i].stride | (aos[i].stride << 8));
 124         OUT_CS(aos[i].buffer_offset + start * 4 * aos[i].stride);
 125     }
 126     for (i = 0; i < aos_count; i++) {
 127         OUT_CS_RELOC(aos[i].buffer, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
 128     }
 129     END_CS;
 130
 131     if (indexBuffer) {
 132         indices = (short*)pipe_buffer_map(pipe->screen, indexBuffer,
 133                                           PIPE_BUFFER_USAGE_CPU_READ);
 134
 135         /* Set the starting point. */
 136         indices += start;
 137
 138         BEGIN_CS(2 + (count+1)/2);
 139         OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count + 1)/2);
 140         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | prim);
 141         for (i = 0; i < count - 1; i += 2) {
 142             OUT_CS(indices[i + 1] << 16 | indices[i]);
 143         }
 144         if (count % 2) {
 145             OUT_CS(indices[count - 1]);
 146         }
 147         END_CS;
 148     } else {
 149         BEGIN_CS(2);
 150         OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 151         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 152                prim);
 153         END_CS;
 154     }
 155
 156     return TRUE;
 157 }
 158
 159 /* Simple helpers for context setup. Should probably be moved to util. */
 160 boolean r300_draw_elements(struct pipe_context* pipe,
 161                            struct pipe_buffer* indexBuffer,
 162                            unsigned indexSize, unsigned mode,
 163                            unsigned start, unsigned count)
 164 {
 165     return pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 166                                      mode, start, count);
 167 }
 168
 169 boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 170                          unsigned start, unsigned count)
 171 {
 172     return pipe->draw_elements(pipe, NULL, 0, mode, start, count);
 173 }
 174
 175 /****************************************************************************
 176  * The rest of this file is for SW TCL rendering only. Please be polite and *
 177  * keep these functions separated so that they are easier to locate. ~C.    *
 178  ***************************************************************************/
 179
 180 /* Draw-based drawing for SW TCL chipsets. */
 181 boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 182                                        struct pipe_buffer* indexBuffer,
 183                                        unsigned indexSize,
 184                                        unsigned minIndex,
 185                                        unsigned maxIndex,
 186                                        unsigned mode,
 187                                        unsigned start,
 188                                        unsigned count)
 189 {
 190     struct r300_context* r300 = r300_context(pipe);
 191     int i;
 192
 193     for (i = 0; i < r300->vertex_buffer_count; i++) {
 194         void* buf = pipe_buffer_map(pipe->screen,
 195                                     r300->vertex_buffers[i].buffer,
 196                                     PIPE_BUFFER_USAGE_CPU_READ);
 197         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 198     }
 199
 200     if (indexBuffer) {
 201         void* indices = pipe_buffer_map(pipe->screen, indexBuffer,
 202                                         PIPE_BUFFER_USAGE_CPU_READ);
 203         draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 204                                              minIndex, maxIndex, indices);
 205     } else {
 206         draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 207     }
 208
 209     draw_set_mapped_constant_buffer(r300->draw,
 210             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 211             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 212                 (sizeof(float) * 4));
 213
 214     draw_arrays(r300->draw, mode, start, count);
 215
 216     for (i = 0; i < r300->vertex_buffer_count; i++) {
 217         pipe_buffer_unmap(pipe->screen, r300->vertex_buffers[i].buffer);
 218         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 219     }
 220
 221     if (indexBuffer) {
 222         pipe_buffer_unmap(pipe->screen, indexBuffer);
 223         draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 224                                              start + count - 1, NULL);
 225     }
 226
 227     return TRUE;
 228 }
 229
 230 /* Object for rendering using Draw. */
 231 struct r300_render {
 232     /* Parent class */
 233     struct vbuf_render base;
 234
 235     /* Pipe context */
 236     struct r300_context* r300;
 237
 238     /* Vertex information */
 239     size_t vertex_size;
 240     unsigned prim;
 241     unsigned hwprim;
 242
 243     /* VBO */
 244     struct pipe_buffer* vbo;
 245     size_t vbo_size;
 246     size_t vbo_offset;
 247     size_t vbo_max_used;
 248     void * vbo_ptr;
 249 };
 250
 251 static INLINE struct r300_render*
 252 r300_render(struct vbuf_render* render)
 253 {
 254     return (struct r300_render*)render;
 255 }
 256
 257 static const struct vertex_info*
 258 r300_render_get_vertex_info(struct vbuf_render* render)
 259 {
 260     struct r300_render* r300render = r300_render(render);
 261     struct r300_context* r300 = r300render->r300;
 262
 263     r300_update_derived_state(r300);
 264
 265     return &r300->vertex_info->vinfo;
 266 }
 267
 268 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 269                                                    ushort vertex_size,
 270                                                    ushort count)
 271 {
 272     struct r300_render* r300render = r300_render(render);
 273     struct r300_context* r300 = r300render->r300;
 274     struct pipe_screen* screen = r300->context.screen;
 275     size_t size = (size_t)vertex_size * (size_t)count;
 276
 277     if (size + r300render->vbo_offset > r300render->vbo_size)
 278     {
 279         pipe_buffer_reference(&r300->vbo, NULL);
 280         r300render->vbo = pipe_buffer_create(screen,
 281                                              64,
 282                                              PIPE_BUFFER_USAGE_VERTEX,
 283                                              R300_MAX_VBO_SIZE);
 284         r300render->vbo_offset = 0;
 285         r300render->vbo_size = R300_MAX_VBO_SIZE;
 286     }
 287
 288     r300render->vertex_size = vertex_size;
 289     r300->vbo = r300render->vbo;
 290     r300->vbo_offset = r300render->vbo_offset;
 291
 292     return (r300render->vbo) ? TRUE : FALSE;
 293 }
 294
 295 static void* r300_render_map_vertices(struct vbuf_render* render)
 296 {
 297     struct r300_render* r300render = r300_render(render);
 298     struct pipe_screen* screen = r300render->r300->context.screen;
 299
 300     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 301                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 302
 303     return (r300render->vbo_ptr + r300render->vbo_offset);
 304 }
 305
 306 static void r300_render_unmap_vertices(struct vbuf_render* render,
 307                                              ushort min,
 308                                              ushort max)
 309 {
 310     struct r300_render* r300render = r300_render(render);
 311     struct pipe_screen* screen = r300render->r300->context.screen;
 312     CS_LOCALS(r300render->r300);
 313     BEGIN_CS(2);
 314     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 315     END_CS;
 316
 317     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 318                                     r300render->vertex_size * (max + 1));
 319     pipe_buffer_unmap(screen, r300render->vbo);
 320 }
 321
 322 static void r300_render_release_vertices(struct vbuf_render* render)
 323 {
 324     struct r300_render* r300render = r300_render(render);
 325
 326     r300render->vbo_offset += r300render->vbo_max_used;
 327     r300render->vbo_max_used = 0;
 328 }
 329
 330 static boolean r300_render_set_primitive(struct vbuf_render* render,
 331                                                unsigned prim)
 332 {
 333     struct r300_render* r300render = r300_render(render);
 334     r300render->prim = prim;
 335
 336     switch (prim) {
 337         case PIPE_PRIM_POINTS:
 338             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_POINTS;
 339             break;
 340         case PIPE_PRIM_LINES:
 341             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINES;
 342             break;
 343         case PIPE_PRIM_LINE_LOOP:
 344             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
 345             break;
 346         case PIPE_PRIM_LINE_STRIP:
 347             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
 348             break;
 349         case PIPE_PRIM_TRIANGLES:
 350             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLES;
 351             break;
 352         case PIPE_PRIM_TRIANGLE_STRIP:
 353             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
 354             break;
 355         case PIPE_PRIM_TRIANGLE_FAN:
 356             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
 357             break;
 358         case PIPE_PRIM_QUADS:
 359             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_QUADS;
 360             break;
 361         case PIPE_PRIM_QUAD_STRIP:
 362             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
 363             break;
 364         case PIPE_PRIM_POLYGON:
 365             r300render->hwprim = R300_VAP_VF_CNTL__PRIM_POLYGON;
 366             break;
 367         default:
 368             return FALSE;
 369             break;
 370     }
 371
 372     return TRUE;
 373 }
 374
 375 static void r300_prepare_render(struct r300_render* render, unsigned count)
 376 {
 377     struct r300_context* r300 = render->r300;
 378
 379     r300_emit_dirty_state(r300);
 380 }
 381
 382 static void r300_render_draw_arrays(struct vbuf_render* render,
 383                                           unsigned start,
 384                                           unsigned count)
 385 {
 386     struct r300_render* r300render = r300_render(render);
 387     struct r300_context* r300 = r300render->r300;
 388
 389     CS_LOCALS(r300);
 390
 391     r300_prepare_render(r300render, count);
 392
 393     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 394
 395     BEGIN_CS(2);
 396     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 397     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 398            r300render->hwprim);
 399     END_CS;
 400 }
 401
 402 static void r300_render_draw(struct vbuf_render* render,
 403                                    const ushort* indices,
 404                                    uint count)
 405 {
 406     struct r300_render* r300render = r300_render(render);
 407     struct r300_context* r300 = r300render->r300;
 408     int i;
 409
 410     CS_LOCALS(r300);
 411
 412     r300_prepare_render(r300render, count);
 413
 414     BEGIN_CS(2 + (count+1)/2);
 415     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 416     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 417            r300render->hwprim);
 418     for (i = 0; i < count-1; i += 2) {
 419         OUT_CS(indices[i+1] << 16 | indices[i]);
 420     }
 421     if (count % 2) {
 422         OUT_CS(indices[count-1]);
 423     }
 424     END_CS;
 425 }
 426
 427 static void r300_render_destroy(struct vbuf_render* render)
 428 {
 429     FREE(render);
 430 }
 431
 432 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 433 {
 434     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 435
 436     r300render->r300 = r300;
 437
 438     /* XXX find real numbers plz */
 439     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 440     r300render->base.max_indices = 16 * 1024;
 441
 442     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 443     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 444     r300render->base.map_vertices = r300_render_map_vertices;
 445     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 446     r300render->base.set_primitive = r300_render_set_primitive;
 447     r300render->base.draw = r300_render_draw;
 448     r300render->base.draw_arrays = r300_render_draw_arrays;
 449     r300render->base.release_vertices = r300_render_release_vertices;
 450     r300render->base.destroy = r300_render_destroy;
 451
 452     r300render->vbo = NULL;
 453     r300render->vbo_size = 0;
 454     r300render->vbo_offset = 0;
 455
 456     return &r300render->base;
 457 }
 458
 459 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 460 {
 461     struct vbuf_render* render;
 462     struct draw_stage* stage;
 463
 464     render = r300_render_create(r300);
 465
 466     if (!render) {
 467         return NULL;
 468     }
 469
 470     stage = draw_vbuf_stage(r300->draw, render);
 471
 472     if (!stage) {
 473         render->destroy(render);
 474         return NULL;
 475     }
 476
 477     draw_set_render(r300->draw, render);
 478
 479     return stage;
 480 }