src/mesa/drivers/dri/r600/r700_render.c

   1 /*
   2  * Copyright (C) 2008-2009  Advanced Micro Devices, Inc.
   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  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included
  12  * in all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  15  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  18  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  19  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  20  */
  21
  22 /*
  23  * Authors:
  24  *   Richard Li <RichardZ.Li@amd.com>, <richardradeon@gmail.com>
  25  *   CooperYuan <cooper.yuan@amd.com>, <cooperyuan@gmail.com>
  26  */
  27
  28 #include "main/glheader.h"
  29 #include "main/state.h"
  30 #include "main/imports.h"
  31 #include "main/enums.h"
  32 #include "main/macros.h"
  33 #include "main/context.h"
  34 #include "main/dd.h"
  35 #include "main/simple_list.h"
  36 #include "main/api_arrayelt.h"
  37 #include "swrast/swrast.h"
  38 #include "swrast_setup/swrast_setup.h"
  39 #include "vbo/vbo.h"
  40
  41 #include "tnl/tnl.h"
  42 #include "tnl/t_vp_build.h"
  43 #include "tnl/t_context.h"
  44 #include "tnl/t_vertex.h"
  45 #include "tnl/t_pipeline.h"
  46
  47 #include "r600_context.h"
  48 #include "r600_cmdbuf.h"
  49
  50 #include "r600_tex.h"
  51
  52 #include "r700_vertprog.h"
  53 #include "r700_fragprog.h"
  54 #include "r700_state.h"
  55
  56 #include "radeon_common_context.h"
  57
  58 void r700WaitForIdle(context_t *context);
  59 void r700WaitForIdleClean(context_t *context);
  60 GLboolean r700SendTextureState(context_t *context);
  61 static unsigned int r700PrimitiveType(int prim);
  62 void r600UpdateTextureState(GLcontext * ctx);
  63 GLboolean r700SyncSurf(context_t *context,
  64                        struct radeon_bo *pbo,
  65                        uint32_t read_domain,
  66                        uint32_t write_domain,
  67                        uint32_t sync_type);
  68
  69 void r700WaitForIdle(context_t *context)
  70 {
  71     BATCH_LOCALS(&context->radeon);
  72     BEGIN_BATCH_NO_AUTOSTATE(3);
  73
  74     R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
  75     R600_OUT_BATCH(mmWAIT_UNTIL - ASIC_CONFIG_BASE_INDEX);
  76     R600_OUT_BATCH(WAIT_3D_IDLE_bit);
  77
  78     END_BATCH();
  79     COMMIT_BATCH();
  80 }
  81
  82 void r700WaitForIdleClean(context_t *context)
  83 {
  84     BATCH_LOCALS(&context->radeon);
  85     BEGIN_BATCH_NO_AUTOSTATE(5);
  86
  87     R600_OUT_BATCH(CP_PACKET3(R600_IT_EVENT_WRITE, 0));
  88     R600_OUT_BATCH(CACHE_FLUSH_AND_INV_EVENT);
  89
  90     R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
  91     R600_OUT_BATCH(mmWAIT_UNTIL - ASIC_CONFIG_BASE_INDEX);
  92     R600_OUT_BATCH(WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
  93
  94     END_BATCH();
  95     COMMIT_BATCH();
  96 }
  97
  98 void r700Start3D(context_t *context)
  99 {
 100     BATCH_LOCALS(&context->radeon);
 101     if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
 102     {
 103         BEGIN_BATCH_NO_AUTOSTATE(2);
 104         R600_OUT_BATCH(CP_PACKET3(R600_IT_START_3D_CMDBUF, 0));
 105         R600_OUT_BATCH(0);
 106         END_BATCH();
 107     }
 108
 109     BEGIN_BATCH_NO_AUTOSTATE(3);
 110     R600_OUT_BATCH(CP_PACKET3(R600_IT_CONTEXT_CONTROL, 1));
 111     R600_OUT_BATCH(0x80000000);
 112     R600_OUT_BATCH(0x80000000);
 113     END_BATCH();
 114
 115     COMMIT_BATCH();
 116
 117     r700WaitForIdleClean(context);
 118 }
 119
 120 GLboolean r700SyncSurf(context_t *context,
 121                        struct radeon_bo *pbo,
 122                        uint32_t read_domain,
 123                        uint32_t write_domain,
 124                        uint32_t sync_type)
 125 {
 126     BATCH_LOCALS(&context->radeon);
 127     uint32_t cp_coher_size;
 128
 129     if (!pbo)
 130             return GL_FALSE;
 131
 132     if (pbo->size == 0xffffffff)
 133             cp_coher_size = 0xffffffff;
 134     else
 135             cp_coher_size = ((pbo->size + 255) >> 8);
 136
 137     BEGIN_BATCH_NO_AUTOSTATE(5 + 2);
 138     R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_SYNC, 3));
 139     R600_OUT_BATCH(sync_type);
 140     R600_OUT_BATCH(cp_coher_size);
 141     R600_OUT_BATCH(0);
 142     R600_OUT_BATCH(10);
 143     R600_OUT_BATCH_RELOC(0,
 144                          pbo,
 145                          0,
 146                          read_domain, write_domain, 0); // ???
 147
 148     END_BATCH();
 149     COMMIT_BATCH();
 150
 151     return GL_TRUE;
 152 }
 153
 154 static unsigned int r700PrimitiveType(int prim)
 155 {
 156     switch (prim & PRIM_MODE_MASK)
 157     {
 158     case GL_POINTS:
 159         return DI_PT_POINTLIST;
 160         break;
 161     case GL_LINES:
 162         return DI_PT_LINELIST;
 163         break;
 164     case GL_LINE_STRIP:
 165         return DI_PT_LINESTRIP;
 166         break;
 167     case GL_LINE_LOOP:
 168         return DI_PT_LINELOOP;
 169         break;
 170     case GL_TRIANGLES:
 171         return DI_PT_TRILIST;
 172         break;
 173     case GL_TRIANGLE_STRIP:
 174         return DI_PT_TRISTRIP;
 175         break;
 176     case GL_TRIANGLE_FAN:
 177         return DI_PT_TRIFAN;
 178         break;
 179     case GL_QUADS:
 180         return DI_PT_QUADLIST;
 181         break;
 182     case GL_QUAD_STRIP:
 183         return DI_PT_QUADSTRIP;
 184         break;
 185     case GL_POLYGON:
 186         return DI_PT_POLYGON;
 187         break;
 188     default:
 189         assert(0);
 190         return -1;
 191         break;
 192     }
 193 }
 194
 195 static int r700NumVerts(int num_verts, int prim)
 196 {
 197         int verts_off = 0;
 198
 199         switch (prim & PRIM_MODE_MASK) {
 200         case GL_POINTS:
 201                 verts_off = 0;
 202                 break;
 203         case GL_LINES:
 204                 verts_off = num_verts % 2;
 205                 break;
 206         case GL_LINE_STRIP:
 207                 if (num_verts < 2)
 208                         verts_off = num_verts;
 209                 break;
 210         case GL_LINE_LOOP:
 211                 if (num_verts < 2)
 212                         verts_off = num_verts;
 213                 break;
 214         case GL_TRIANGLES:
 215                 verts_off = num_verts % 3;
 216                 break;
 217         case GL_TRIANGLE_STRIP:
 218                 if (num_verts < 3)
 219                         verts_off = num_verts;
 220                 break;
 221         case GL_TRIANGLE_FAN:
 222                 if (num_verts < 3)
 223                         verts_off = num_verts;
 224                 break;
 225         case GL_QUADS:
 226                 verts_off = num_verts % 4;
 227                 break;
 228         case GL_QUAD_STRIP:
 229                 if (num_verts < 4)
 230                         verts_off = num_verts;
 231                 else
 232                         verts_off = num_verts % 2;
 233                 break;
 234         case GL_POLYGON:
 235                 if (num_verts < 3)
 236                         verts_off = num_verts;
 237                 break;
 238         default:
 239                 assert(0);
 240                 return -1;
 241                 break;
 242         }
 243
 244         return num_verts - verts_off;
 245 }
 246
 247 static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim)
 248 {
 249         context_t *context = R700_CONTEXT(ctx);
 250         BATCH_LOCALS(&context->radeon);
 251         int type, i, total_emit;
 252         int num_indices;
 253         uint32_t vgt_draw_initiator = 0;
 254         uint32_t vgt_index_type     = 0;
 255         uint32_t vgt_primitive_type = 0;
 256         uint32_t vgt_num_indices    = 0;
 257
 258         type = r700PrimitiveType(prim);
 259         num_indices = r700NumVerts(end - start, prim);
 260
 261         if (type < 0 || num_indices <= 0)
 262                 return;
 263
 264         total_emit =   3 /* VGT_PRIMITIVE_TYPE */
 265                      + 2 /* VGT_INDEX_TYPE */
 266                      + 2 /* NUM_INSTANCES */
 267                      + num_indices + 3; /* DRAW_INDEX_IMMD */
 268
 269         BEGIN_BATCH_NO_AUTOSTATE(total_emit);
 270         // prim
 271         SETfield(vgt_primitive_type, type,
 272                  VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask);
 273         R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
 274         R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX);
 275         R600_OUT_BATCH(vgt_primitive_type);
 276
 277         // index type
 278         SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask);
 279         R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0));
 280         R600_OUT_BATCH(vgt_index_type);
 281
 282         // num instances
 283         R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0));
 284         R600_OUT_BATCH(1);
 285
 286         // draw packet
 287         vgt_num_indices = num_indices;
 288         SETfield(vgt_draw_initiator, DI_SRC_SEL_IMMEDIATE, SOURCE_SELECT_shift, SOURCE_SELECT_mask);
 289         SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask);
 290
 291         R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (num_indices + 1)));
 292         R600_OUT_BATCH(vgt_num_indices);
 293         R600_OUT_BATCH(vgt_draw_initiator);
 294
 295         for (i = start; i < (start + num_indices); i++) {
 296             R600_OUT_BATCH(i);
 297         }
 298         END_BATCH();
 299         COMMIT_BATCH();
 300
 301 }
 302
 303 /* start 3d, idle, cb/db flush */
 304 #define PRE_EMIT_STATE_BUFSZ 10 + 5 + 14
 305
 306 static GLuint r700PredictRenderSize(GLcontext* ctx)
 307 {
 308     context_t *context = R700_CONTEXT(ctx);
 309     TNLcontext *tnl = TNL_CONTEXT(ctx);
 310     struct r700_vertex_program *vpc
 311         = (struct r700_vertex_program *)ctx->VertexProgram._Current;
 312     struct vertex_buffer *vb = &tnl->vb;
 313     GLboolean flushed;
 314     GLuint dwords, i;
 315     GLuint state_size;
 316     /* pre calculate aos count so state prediction works */
 317     context->radeon.tcl.aos_count = _mesa_bitcount(vpc->mesa_program.Base.InputsRead);
 318
 319     dwords = PRE_EMIT_STATE_BUFSZ;
 320     for (i = 0; i < vb->PrimitiveCount; i++)
 321         dwords += vb->Primitive[i].count + 10;
 322     state_size = radeonCountStateEmitSize(&context->radeon);
 323     flushed = rcommonEnsureCmdBufSpace(&context->radeon,
 324             dwords + state_size, __FUNCTION__);
 325
 326     if (flushed)
 327         dwords += radeonCountStateEmitSize(&context->radeon);
 328     else
 329         dwords += state_size;
 330
 331     if (RADEON_DEBUG & DEBUG_PRIMS)
 332         fprintf(stderr, "%s: total prediction size is %d.\n", __FUNCTION__, dwords);
 333     return dwords;
 334 }
 335
 336 static GLboolean r700RunRender(GLcontext * ctx,
 337                                struct tnl_pipeline_stage *stage)
 338 {
 339     context_t *context = R700_CONTEXT(ctx);
 340     radeonContextPtr radeon = &context->radeon;
 341     unsigned int i, id = 0;
 342     TNLcontext *tnl = TNL_CONTEXT(ctx);
 343     struct vertex_buffer *vb = &tnl->vb;
 344     struct radeon_renderbuffer *rrb;
 345
 346     if (RADEON_DEBUG & DEBUG_PRIMS)
 347         fprintf(stderr, "%s: cs begin at %d\n",
 348                 __func__, context->radeon.cmdbuf.cs->cdw);
 349
 350     /* always emit CB base to prevent
 351      * lock ups on some chips.
 352      */
 353     R600_STATECHANGE(context, cb_target);
 354     /* mark vtx as dirty since it changes per-draw */
 355     R600_STATECHANGE(context, vtx);
 356
 357     r700UpdateShaders(ctx);
 358     r700SetScissor(context);
 359     r700SetupVertexProgram(ctx);
 360     r700SetupFragmentProgram(ctx);
 361     r600UpdateTextureState(ctx);
 362
 363     GLuint emit_end = r700PredictRenderSize(ctx)
 364         + context->radeon.cmdbuf.cs->cdw;
 365     r700SetupStreams(ctx);
 366
 367     radeonEmitState(radeon);
 368
 369     /* richard test code */
 370     for (i = 0; i < vb->PrimitiveCount; i++) {
 371         GLuint prim = _tnl_translate_prim(&vb->Primitive[i]);
 372         GLuint start = vb->Primitive[i].start;
 373         GLuint end = vb->Primitive[i].start + vb->Primitive[i].count;
 374         r700RunRenderPrimitive(ctx, start, end, prim);
 375     }
 376
 377     /* Flush render op cached for last several quads. */
 378     r700WaitForIdleClean(context);
 379
 380     rrb = radeon_get_colorbuffer(&context->radeon);
 381     if (rrb && rrb->bo)
 382             r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
 383                          CB_ACTION_ENA_bit | (1 << (id + 6)));
 384
 385     rrb = radeon_get_depthbuffer(&context->radeon);
 386     if (rrb && rrb->bo)
 387             r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
 388                          DB_ACTION_ENA_bit | DB_DEST_BASE_ENA_bit);
 389
 390     radeonReleaseArrays(ctx, ~0);
 391
 392     if (RADEON_DEBUG & DEBUG_PRIMS)
 393         fprintf(stderr, "%s: cs end at %d\n",
 394                 __func__, context->radeon.cmdbuf.cs->cdw);
 395
 396     if ( emit_end < context->radeon.cmdbuf.cs->cdw )
 397        WARN_ONCE("Rendering was %d commands larger than predicted size."
 398                " We might overflow  command buffer.\n", context->radeon.cmdbuf.cs->cdw - emit_end);
 399
 400     return GL_FALSE;
 401 }
 402
 403 static GLboolean r700RunNonTCLRender(GLcontext * ctx,
 404                                      struct tnl_pipeline_stage *stage) /* -------------------- */
 405 {
 406         GLboolean bRet = GL_TRUE;
 407
 408         return bRet;
 409 }
 410
 411 static GLboolean r700RunTCLRender(GLcontext * ctx,  /*----------------------*/
 412                                   struct tnl_pipeline_stage *stage)
 413 {
 414         GLboolean bRet = GL_FALSE;
 415
 416     /* TODO : sw fallback */
 417
 418     /**
 419     * Ensure all enabled and complete textures are uploaded along with any buffers being used.
 420     */
 421     if(!r600ValidateBuffers(ctx))
 422     {
 423         return GL_TRUE;
 424     }
 425
 426     bRet = r700RunRender(ctx, stage);
 427
 428     return bRet;
 429         //GL_FALSE will stop to do other pipe stage in _tnl_run_pipeline
 430     //The render here DOES finish the whole pipe, so GL_FALSE should be returned for success.
 431 }
 432
 433 const struct tnl_pipeline_stage _r700_render_stage = {
 434         "r700 Hardware Rasterization",
 435         NULL,
 436         NULL,
 437         NULL,
 438         NULL,
 439         r700RunNonTCLRender
 440 };
 441
 442 const struct tnl_pipeline_stage _r700_tcl_stage = {
 443         "r700 Hardware Transform, Clipping and Lighting",
 444         NULL,
 445         NULL,
 446         NULL,
 447         NULL,
 448         r700RunTCLRender
 449 };
 450
 451 const struct tnl_pipeline_stage *r700_pipeline[] =
 452 {
 453     &_r700_tcl_stage,
 454     &_tnl_vertex_transform_stage,
 455         &_tnl_normal_transform_stage,
 456         &_tnl_lighting_stage,
 457         &_tnl_fog_coordinate_stage,
 458         &_tnl_texgen_stage,
 459         &_tnl_texture_transform_stage,
 460         &_tnl_vertex_program_stage,
 461
 462     &_r700_render_stage,
 463     &_tnl_render_stage,
 464     0,
 465 };
 466
 467