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 void r700WaitForIdle(context_t *context);
  57 void r700WaitForIdleClean(context_t *context);
  58 GLboolean r700SendTextureState(context_t *context);
  59 static unsigned int r700PrimitiveType(int prim);
  60 void r600UpdateTextureState(GLcontext * ctx);
  61 GLboolean r700SyncSurf(context_t *context,
  62                        struct radeon_bo *pbo,
  63                        uint32_t read_domain,
  64                        uint32_t write_domain,
  65                        uint32_t sync_type);
  66
  67 void r700WaitForIdle(context_t *context)
  68 {
  69     BATCH_LOCALS(&context->radeon);
  70     BEGIN_BATCH_NO_AUTOSTATE(3);
  71
  72     R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
  73     R600_OUT_BATCH(mmWAIT_UNTIL - ASIC_CONFIG_BASE_INDEX);
  74     R600_OUT_BATCH(WAIT_3D_IDLE_bit);
  75
  76     END_BATCH();
  77     COMMIT_BATCH();
  78 }
  79
  80 void r700WaitForIdleClean(context_t *context)
  81 {
  82     BATCH_LOCALS(&context->radeon);
  83     BEGIN_BATCH_NO_AUTOSTATE(5);
  84
  85     R600_OUT_BATCH(CP_PACKET3(R600_IT_EVENT_WRITE, 0));
  86     R600_OUT_BATCH(CACHE_FLUSH_AND_INV_EVENT);
  87
  88     R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
  89     R600_OUT_BATCH(mmWAIT_UNTIL - ASIC_CONFIG_BASE_INDEX);
  90     R600_OUT_BATCH(WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
  91
  92     END_BATCH();
  93     COMMIT_BATCH();
  94 }
  95
  96 void r700Start3D(context_t *context)
  97 {
  98     BATCH_LOCALS(&context->radeon);
  99     if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
 100     {
 101         BEGIN_BATCH_NO_AUTOSTATE(2);
 102         R600_OUT_BATCH(CP_PACKET3(R600_IT_START_3D_CMDBUF, 0));
 103         R600_OUT_BATCH(0);
 104         END_BATCH();
 105     }
 106
 107     BEGIN_BATCH_NO_AUTOSTATE(3);
 108     R600_OUT_BATCH(CP_PACKET3(R600_IT_CONTEXT_CONTROL, 1));
 109     R600_OUT_BATCH(0x80000000);
 110     R600_OUT_BATCH(0x80000000);
 111     END_BATCH();
 112
 113     COMMIT_BATCH();
 114
 115     r700WaitForIdleClean(context);
 116 }
 117
 118 GLboolean r700SyncSurf(context_t *context,
 119                        struct radeon_bo *pbo,
 120                        uint32_t read_domain,
 121                        uint32_t write_domain,
 122                        uint32_t sync_type)
 123 {
 124     BATCH_LOCALS(&context->radeon);
 125     uint32_t cp_coher_size;
 126
 127     if (!pbo)
 128             return GL_FALSE;
 129
 130     if (pbo->size == 0xffffffff)
 131             cp_coher_size = 0xffffffff;
 132     else
 133             cp_coher_size = ((pbo->size + 255) >> 8);
 134
 135     BEGIN_BATCH_NO_AUTOSTATE(5 + 2);
 136     R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_SYNC, 3));
 137     R600_OUT_BATCH(sync_type);
 138     R600_OUT_BATCH(cp_coher_size);
 139     R600_OUT_BATCH(0);
 140     R600_OUT_BATCH(10);
 141     R600_OUT_BATCH_RELOC(0,
 142                          pbo,
 143                          0,
 144                          read_domain, write_domain, 0); // ???
 145
 146     END_BATCH();
 147     COMMIT_BATCH();
 148
 149     return GL_TRUE;
 150 }
 151
 152 static unsigned int r700PrimitiveType(int prim)
 153 {
 154     switch (prim & PRIM_MODE_MASK)
 155     {
 156     case GL_POINTS:
 157         return DI_PT_POINTLIST;
 158         break;
 159     case GL_LINES:
 160         return DI_PT_LINELIST;
 161         break;
 162     case GL_LINE_STRIP:
 163         return DI_PT_LINESTRIP;
 164         break;
 165     case GL_LINE_LOOP:
 166         return DI_PT_LINELOOP;
 167         break;
 168     case GL_TRIANGLES:
 169         return DI_PT_TRILIST;
 170         break;
 171     case GL_TRIANGLE_STRIP:
 172         return DI_PT_TRISTRIP;
 173         break;
 174     case GL_TRIANGLE_FAN:
 175         return DI_PT_TRIFAN;
 176         break;
 177     case GL_QUADS:
 178         return DI_PT_QUADLIST;
 179         break;
 180     case GL_QUAD_STRIP:
 181         return DI_PT_QUADSTRIP;
 182         break;
 183     case GL_POLYGON:
 184         return DI_PT_POLYGON;
 185         break;
 186     default:
 187         assert(0);
 188         return -1;
 189         break;
 190     }
 191 }
 192
 193 static int r700NumVerts(int num_verts, int prim)
 194 {
 195         int verts_off = 0;
 196
 197         switch (prim & PRIM_MODE_MASK) {
 198         case GL_POINTS:
 199                 verts_off = 0;
 200                 break;
 201         case GL_LINES:
 202                 verts_off = num_verts % 2;
 203                 break;
 204         case GL_LINE_STRIP:
 205                 if (num_verts < 2)
 206                         verts_off = num_verts;
 207                 break;
 208         case GL_LINE_LOOP:
 209                 if (num_verts < 2)
 210                         verts_off = num_verts;
 211                 break;
 212         case GL_TRIANGLES:
 213                 verts_off = num_verts % 3;
 214                 break;
 215         case GL_TRIANGLE_STRIP:
 216                 if (num_verts < 3)
 217                         verts_off = num_verts;
 218                 break;
 219         case GL_TRIANGLE_FAN:
 220                 if (num_verts < 3)
 221                         verts_off = num_verts;
 222                 break;
 223         case GL_QUADS:
 224                 verts_off = num_verts % 4;
 225                 break;
 226         case GL_QUAD_STRIP:
 227                 if (num_verts < 4)
 228                         verts_off = num_verts;
 229                 else
 230                         verts_off = num_verts % 2;
 231                 break;
 232         case GL_POLYGON:
 233                 if (num_verts < 3)
 234                         verts_off = num_verts;
 235                 break;
 236         default:
 237                 assert(0);
 238                 return -1;
 239                 break;
 240         }
 241
 242         return num_verts - verts_off;
 243 }
 244
 245 static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim)
 246 {
 247         context_t *context = R700_CONTEXT(ctx);
 248         BATCH_LOCALS(&context->radeon);
 249         int type, i, total_emit;
 250         int num_indices;
 251         uint32_t vgt_draw_initiator = 0;
 252         uint32_t vgt_index_type     = 0;
 253         uint32_t vgt_primitive_type = 0;
 254         uint32_t vgt_num_indices    = 0;
 255
 256         type = r700PrimitiveType(prim);
 257         num_indices = r700NumVerts(end - start, prim);
 258
 259         if (type < 0 || num_indices <= 0)
 260                 return;
 261
 262         total_emit =   3 /* VGT_PRIMITIVE_TYPE */
 263                      + 2 /* VGT_INDEX_TYPE */
 264                      + 2 /* NUM_INSTANCES */
 265                      + num_indices + 3; /* DRAW_INDEX_IMMD */
 266
 267         BEGIN_BATCH_NO_AUTOSTATE(total_emit);
 268         // prim
 269         SETfield(vgt_primitive_type, type,
 270                  VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask);
 271         R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
 272         R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX);
 273         R600_OUT_BATCH(vgt_primitive_type);
 274
 275         // index type
 276         SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask);
 277         R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0));
 278         R600_OUT_BATCH(vgt_index_type);
 279
 280         // num instances
 281         R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0));
 282         R600_OUT_BATCH(1);
 283
 284         // draw packet
 285         vgt_num_indices = num_indices;
 286         SETfield(vgt_draw_initiator, DI_SRC_SEL_IMMEDIATE, SOURCE_SELECT_shift, SOURCE_SELECT_mask);
 287         SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask);
 288
 289         R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (num_indices + 1)));
 290         R600_OUT_BATCH(vgt_num_indices);
 291         R600_OUT_BATCH(vgt_draw_initiator);
 292
 293         for (i = start; i < (start + num_indices); i++) {
 294             R600_OUT_BATCH(i);
 295         }
 296         END_BATCH();
 297         COMMIT_BATCH();
 298
 299 }
 300
 301 /* start 3d, idle, cb/db flush */
 302 #define PRE_EMIT_STATE_BUFSZ 10 + 5 + 14
 303
 304 static GLuint r700PredictRenderSize(GLcontext* ctx)
 305 {
 306     context_t *context = R700_CONTEXT(ctx);
 307     TNLcontext *tnl = TNL_CONTEXT(ctx);
 308      struct r700_vertex_program *vpc
 309              = (struct r700_vertex_program *)ctx->VertexProgram._Current;
 310     struct vertex_buffer *vb = &tnl->vb;
 311     GLboolean flushed;
 312     GLuint dwords, i;
 313     GLuint state_size;
 314     /* pre calculate aos count so state prediction works */
 315     context->radeon.tcl.aos_count = _mesa_bitcount(vpc->mesa_program.Base.InputsRead);
 316
 317     dwords = PRE_EMIT_STATE_BUFSZ;
 318     for (i = 0; i < vb->PrimitiveCount; i++)
 319             dwords += vb->Primitive[i].count + 10;
 320     state_size = radeonCountStateEmitSize(&context->radeon);
 321     flushed = rcommonEnsureCmdBufSpace(&context->radeon,
 322                              dwords + state_size, __FUNCTION__);
 323
 324     if (flushed)
 325         dwords += radeonCountStateEmitSize(&context->radeon);
 326     else
 327         dwords += state_size;
 328
 329     if (RADEON_DEBUG & DEBUG_PRIMS)
 330         fprintf(stderr, "%s: total prediction size is %d.\n", __FUNCTION__, dwords);
 331     return dwords;
 332 }
 333
 334 static GLboolean r700RunRender(GLcontext * ctx,
 335                                struct tnl_pipeline_stage *stage)
 336 {
 337     context_t *context = R700_CONTEXT(ctx);
 338     radeonContextPtr radeon = &context->radeon;
 339     unsigned int i, id = 0;
 340     TNLcontext *tnl = TNL_CONTEXT(ctx);
 341     struct vertex_buffer *vb = &tnl->vb;
 342     struct radeon_renderbuffer *rrb;
 343
 344         if (RADEON_DEBUG & DEBUG_PRIMS)
 345                 fprintf(stderr, "%s: cs begin at %d\n",
 346                 __func__, context->radeon.cmdbuf.cs->cdw);
 347
 348     r700UpdateShaders(ctx);
 349     r700SetScissor(context);
 350     r700SetupVertexProgram(ctx);
 351     r700SetupFragmentProgram(ctx);
 352     r600UpdateTextureState(ctx);
 353
 354     GLuint emit_end = r700PredictRenderSize(ctx)
 355         + context->radeon.cmdbuf.cs->cdw;
 356     r700SetupStreams(ctx);
 357
 358     radeonEmitState(radeon);
 359
 360     /* richard test code */
 361     for (i = 0; i < vb->PrimitiveCount; i++) {
 362         GLuint prim = _tnl_translate_prim(&vb->Primitive[i]);
 363         GLuint start = vb->Primitive[i].start;
 364         GLuint end = vb->Primitive[i].start + vb->Primitive[i].count;
 365         r700RunRenderPrimitive(ctx, start, end, prim);
 366     }
 367
 368     /* Flush render op cached for last several quads. */
 369     r700WaitForIdleClean(context);
 370
 371     rrb = radeon_get_colorbuffer(&context->radeon);
 372     if (!rrb || !rrb->bo)
 373             r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
 374                          CB_ACTION_ENA_bit | (1 << (id + 6)));
 375
 376     rrb = radeon_get_depthbuffer(&context->radeon);
 377     if (!rrb || !rrb->bo)
 378             r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
 379                          DB_ACTION_ENA_bit | DB_DEST_BASE_ENA_bit);
 380
 381     radeonReleaseArrays(ctx, ~0);
 382     assert(context->radeon.cmdbuf.cs->cdw <= emit_end);
 383
 384     return GL_FALSE;
 385 }
 386
 387 static GLboolean r700RunNonTCLRender(GLcontext * ctx,
 388                                      struct tnl_pipeline_stage *stage) /* -------------------- */
 389 {
 390         GLboolean bRet = GL_TRUE;
 391
 392         return bRet;
 393 }
 394
 395 static GLboolean r700RunTCLRender(GLcontext * ctx,  /*----------------------*/
 396                                   struct tnl_pipeline_stage *stage)
 397 {
 398         GLboolean bRet = GL_FALSE;
 399
 400     /* TODO : sw fallback */
 401
 402     /**
 403     * Ensure all enabled and complete textures are uploaded along with any buffers being used.
 404     */
 405     if(!r600ValidateBuffers(ctx))
 406     {
 407         return GL_TRUE;
 408     }
 409
 410     bRet = r700RunRender(ctx, stage);
 411
 412     return bRet;
 413         //GL_FALSE will stop to do other pipe stage in _tnl_run_pipeline
 414     //The render here DOES finish the whole pipe, so GL_FALSE should be returned for success.
 415 }
 416
 417 const struct tnl_pipeline_stage _r700_render_stage = {
 418         "r700 Hardware Rasterization",
 419         NULL,
 420         NULL,
 421         NULL,
 422         NULL,
 423         r700RunNonTCLRender
 424 };
 425
 426 const struct tnl_pipeline_stage _r700_tcl_stage = {
 427         "r700 Hardware Transform, Clipping and Lighting",
 428         NULL,
 429         NULL,
 430         NULL,
 431         NULL,
 432         r700RunTCLRender
 433 };
 434
 435 const struct tnl_pipeline_stage *r700_pipeline[] =
 436 {
 437     &_r700_tcl_stage,
 438     &_tnl_vertex_transform_stage,
 439         &_tnl_normal_transform_stage,
 440         &_tnl_lighting_stage,
 441         &_tnl_fog_coordinate_stage,
 442         &_tnl_texgen_stage,
 443         &_tnl_texture_transform_stage,
 444         &_tnl_vertex_program_stage,
 445
 446     &_r700_render_stage,
 447     &_tnl_render_stage,
 448     0,
 449 };
 450
 451