src/gallium/drivers/swr/rasterizer/core/context.h

   1 /****************************************************************************
   2 * Copyright (C) 2014-2016 Intel Corporation.   All Rights Reserved.
   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 (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 NONINFRINGEMENT.  IN NO EVENT SHALL
  18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21 * IN THE SOFTWARE.
  22 *
  23 * @file context.h
  24 *
  25 * @brief Definitions for SWR_CONTEXT and DRAW_CONTEXT
  26 *        The SWR_CONTEXT is our global context and contains the DC ring,
  27 *        thread state, etc.
  28 *
  29 *        The DRAW_CONTEXT contains all state associated with a draw operation.
  30 *
  31 ******************************************************************************/
  32 #pragma once
  33
  34 #include <condition_variable>
  35 #include <algorithm>
  36
  37 #include "core/api.h"
  38 #include "core/utils.h"
  39 #include "core/arena.h"
  40 #include "core/fifo.hpp"
  41 #include "core/knobs.h"
  42 #include "common/simdintrin.h"
  43 #include "core/threads.h"
  44 #include "ringbuffer.h"
  45 #include "archrast/archrast.h"
  46
  47 // x.8 fixed point precision values
  48 #define FIXED_POINT_SHIFT 8
  49 #define FIXED_POINT_SCALE 256
  50
  51 // x.16 fixed point precision values
  52 #define FIXED_POINT16_SHIFT 16
  53 #define FIXED_POINT16_SCALE 65536
  54
  55 struct SWR_CONTEXT;
  56 struct DRAW_CONTEXT;
  57
  58 struct TRI_FLAGS
  59 {
  60     uint32_t frontFacing : 1;
  61     uint32_t yMajor : 1;
  62     uint32_t coverageMask : (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  63     uint32_t reserved : 32 - 1 - 1 - (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  64     float pointSize;
  65     uint32_t primID;
  66     uint32_t renderTargetArrayIndex;
  67     uint32_t viewportIndex;
  68 };
  69
  70 //////////////////////////////////////////////////////////////////////////
  71 /// SWR_TRIANGLE_DESC
  72 /////////////////////////////////////////////////////////////////////////
  73 struct SWR_TRIANGLE_DESC
  74 {
  75     float I[3];
  76     float J[3];
  77     float Z[3];
  78     float OneOverW[3];
  79     float recipDet;
  80
  81     float *pRecipW;
  82     float *pAttribs;
  83     float *pPerspAttribs;
  84     float *pSamplePos;
  85     float *pUserClipBuffer;
  86
  87     uint64_t coverageMask[SWR_MAX_NUM_MULTISAMPLES];
  88     uint64_t innerCoverageMask; // Conservative rasterization inner coverage: marked covered if entire pixel is covered
  89     uint64_t anyCoveredSamples;
  90
  91     TRI_FLAGS triFlags;
  92 };
  93
  94 struct TRIANGLE_WORK_DESC
  95 {
  96     float *pTriBuffer;
  97     float *pAttribs;
  98     float *pUserClipBuffer;
  99     uint32_t numAttribs;
 100     TRI_FLAGS triFlags;
 101 };
 102
 103 struct CLEAR_DESC
 104 {
 105     SWR_RECT rect;
 106     uint32_t attachmentMask;
 107     float clearRTColor[4];  // RGBA_32F
 108     float clearDepth;   // [0..1]
 109     uint8_t clearStencil;
 110 };
 111
 112 struct DISCARD_INVALIDATE_TILES_DESC
 113 {
 114     uint32_t attachmentMask;
 115     SWR_RECT rect;
 116     SWR_TILE_STATE newTileState;
 117     bool createNewTiles;
 118     bool fullTilesOnly;
 119 };
 120
 121 struct SYNC_DESC
 122 {
 123     PFN_CALLBACK_FUNC pfnCallbackFunc;
 124     uint64_t userData;
 125     uint64_t userData2;
 126     uint64_t userData3;
 127 };
 128
 129 struct STORE_TILES_DESC
 130 {
 131     uint32_t attachmentMask;
 132     SWR_TILE_STATE postStoreTileState;
 133     SWR_RECT rect;
 134 };
 135
 136 struct COMPUTE_DESC
 137 {
 138     uint32_t threadGroupCountX;
 139     uint32_t threadGroupCountY;
 140     uint32_t threadGroupCountZ;
 141 };
 142
 143 typedef void(*PFN_WORK_FUNC)(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t macroTile, void* pDesc);
 144
 145 enum WORK_TYPE
 146 {
 147     SYNC,
 148     DRAW,
 149     CLEAR,
 150     DISCARDINVALIDATETILES,
 151     STORETILES,
 152     SHUTDOWN,
 153 };
 154
 155 OSALIGNSIMD(struct) BE_WORK
 156 {
 157     WORK_TYPE type;
 158     PFN_WORK_FUNC pfnWork;
 159     union
 160     {
 161         SYNC_DESC sync;
 162         TRIANGLE_WORK_DESC tri;
 163         CLEAR_DESC clear;
 164         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 165         STORE_TILES_DESC storeTiles;
 166     } desc;
 167 };
 168
 169 struct DRAW_WORK
 170 {
 171     DRAW_CONTEXT*   pDC;
 172     union
 173     {
 174         uint32_t   numIndices;      // DrawIndexed: Number of indices for draw.
 175         uint32_t   numVerts;        // Draw: Number of verts (triangles, lines, etc)
 176     };
 177     union
 178     {
 179         const int32_t* pIB;        // DrawIndexed: App supplied indices
 180         uint32_t   startVertex;    // Draw: Starting vertex in VB to render from.
 181     };
 182     int32_t    baseVertex;
 183     uint32_t   numInstances;        // Number of instances
 184     uint32_t   startInstance;       // Instance offset
 185     uint32_t   startPrimID;         // starting primitiveID for this draw batch
 186     uint32_t   startVertexID;       // starting VertexID for this draw batch (only needed for non-indexed draws)
 187     SWR_FORMAT type;                // index buffer type
 188 };
 189
 190 typedef void(*PFN_FE_WORK_FUNC)(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t workerId, void* pDesc);
 191 struct FE_WORK
 192 {
 193     WORK_TYPE type;
 194     PFN_FE_WORK_FUNC pfnWork;
 195     union
 196     {
 197         SYNC_DESC sync;
 198         DRAW_WORK draw;
 199         CLEAR_DESC clear;
 200         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 201         STORE_TILES_DESC storeTiles;
 202     } desc;
 203 };
 204
 205 struct GUARDBANDS
 206 {
 207     float left[KNOB_NUM_VIEWPORTS_SCISSORS];
 208     float right[KNOB_NUM_VIEWPORTS_SCISSORS];
 209     float top[KNOB_NUM_VIEWPORTS_SCISSORS];
 210     float bottom[KNOB_NUM_VIEWPORTS_SCISSORS];
 211 };
 212
 213 struct PA_STATE;
 214
 215 // function signature for pipeline stages that execute after primitive assembly
 216 typedef void(*PFN_PROCESS_PRIMS)(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[],
 217     uint32_t primMask, simdscalari primID, simdscalari viewportIdx);
 218
 219 OSALIGNLINE(struct) API_STATE
 220 {
 221     // Vertex Buffers
 222     SWR_VERTEX_BUFFER_STATE vertexBuffers[KNOB_NUM_STREAMS];
 223
 224     // Index Buffer
 225     SWR_INDEX_BUFFER_STATE  indexBuffer;
 226
 227     // FS - Fetch Shader State
 228     PFN_FETCH_FUNC          pfnFetchFunc;
 229
 230     // VS - Vertex Shader State
 231     PFN_VERTEX_FUNC         pfnVertexFunc;
 232
 233     // GS - Geometry Shader State
 234     PFN_GS_FUNC             pfnGsFunc;
 235     SWR_GS_STATE            gsState;
 236
 237     // CS - Compute Shader
 238     PFN_CS_FUNC             pfnCsFunc;
 239     uint32_t                totalThreadsInGroup;
 240     uint32_t                totalSpillFillSize;
 241
 242     // FE - Frontend State
 243     SWR_FRONTEND_STATE      frontendState;
 244
 245     // SOS - Streamout Shader State
 246     PFN_SO_FUNC             pfnSoFunc[MAX_SO_STREAMS];
 247
 248     // Streamout state
 249     SWR_STREAMOUT_STATE     soState;
 250     mutable SWR_STREAMOUT_BUFFER soBuffer[MAX_SO_STREAMS];
 251
 252     // Tessellation State
 253     PFN_HS_FUNC             pfnHsFunc;
 254     PFN_DS_FUNC             pfnDsFunc;
 255     SWR_TS_STATE            tsState;
 256
 257     // Number of attributes used by the frontend (vs, so, gs)
 258     uint32_t                feNumAttributes;
 259
 260     PRIMITIVE_TOPOLOGY      topology;
 261     bool                    forceFront;
 262
 263     // RS - Rasterizer State
 264     SWR_RASTSTATE           rastState;
 265     // floating point multisample offsets
 266     float samplePos[SWR_MAX_NUM_MULTISAMPLES * 2];
 267
 268     GUARDBANDS               gbState;
 269
 270     SWR_VIEWPORT            vp[KNOB_NUM_VIEWPORTS_SCISSORS];
 271     SWR_VIEWPORT_MATRICES   vpMatrices;
 272
 273     SWR_RECT                scissorRects[KNOB_NUM_VIEWPORTS_SCISSORS];
 274     SWR_RECT                scissorsInFixedPoint[KNOB_NUM_VIEWPORTS_SCISSORS];
 275     bool                    scissorsTileAligned;
 276
 277     // Backend state
 278     SWR_BACKEND_STATE       backendState;
 279
 280     SWR_DEPTH_BOUNDS_STATE  depthBoundsState;
 281
 282     // PS - Pixel shader state
 283     SWR_PS_STATE            psState;
 284
 285     SWR_DEPTH_STENCIL_STATE depthStencilState;
 286
 287     // OM - Output Merger State
 288     SWR_BLEND_STATE         blendState;
 289     PFN_BLEND_JIT_FUNC      pfnBlendFunc[SWR_NUM_RENDERTARGETS];
 290
 291     struct
 292     {
 293         uint32_t enableStatsFE : 1;             // Enable frontend pipeline stats
 294         uint32_t enableStatsBE : 1;             // Enable backend pipeline stats
 295         uint32_t colorHottileEnable : 8;        // Bitmask of enabled color hottiles
 296         uint32_t depthHottileEnable: 1;         // Enable depth buffer hottile
 297         uint32_t stencilHottileEnable : 1;      // Enable stencil buffer hottile
 298     };
 299
 300     PFN_QUANTIZE_DEPTH      pfnQuantizeDepth;
 301 };
 302
 303 class MacroTileMgr;
 304 class DispatchQueue;
 305
 306 struct RenderOutputBuffers
 307 {
 308     uint8_t* pColor[SWR_NUM_RENDERTARGETS];
 309     uint8_t* pDepth;
 310     uint8_t* pStencil;
 311 };
 312
 313 // Plane equation A/B/C coeffs used to evaluate I/J barycentric coords
 314 struct BarycentricCoeffs
 315 {
 316     simdscalar vIa;
 317     simdscalar vIb;
 318     simdscalar vIc;
 319
 320     simdscalar vJa;
 321     simdscalar vJb;
 322     simdscalar vJc;
 323
 324     simdscalar vZa;
 325     simdscalar vZb;
 326     simdscalar vZc;
 327
 328     simdscalar vRecipDet;
 329
 330     simdscalar vAOneOverW;
 331     simdscalar vBOneOverW;
 332     simdscalar vCOneOverW;
 333 };
 334
 335 // pipeline function pointer types
 336 typedef void(*PFN_BACKEND_FUNC)(DRAW_CONTEXT*, uint32_t, uint32_t, uint32_t, SWR_TRIANGLE_DESC&, RenderOutputBuffers&);
 337 typedef void(*PFN_OUTPUT_MERGER)(SWR_PS_CONTEXT &, uint8_t* (&)[SWR_NUM_RENDERTARGETS], uint32_t, const SWR_BLEND_STATE*,
 338                                  const PFN_BLEND_JIT_FUNC (&)[SWR_NUM_RENDERTARGETS], simdscalar&, simdscalar);
 339 typedef void(*PFN_CALC_PIXEL_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &);
 340 typedef void(*PFN_CALC_SAMPLE_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT&);
 341 typedef void(*PFN_CALC_CENTROID_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &, const uint64_t *const, const uint32_t,
 342                                               const simdscalar, const simdscalar);
 343
 344 struct BACKEND_FUNCS
 345 {
 346     PFN_BACKEND_FUNC pfnBackend;
 347 };
 348
 349 // Draw State
 350 struct DRAW_STATE
 351 {
 352     API_STATE state;
 353
 354     void* pPrivateState;  // Its required the driver sets this up for each draw.
 355
 356     // pipeline function pointers, filled in by API thread when setting up the draw
 357     BACKEND_FUNCS backendFuncs;
 358     PFN_PROCESS_PRIMS pfnProcessPrims;
 359
 360     CachingArena* pArena;     // This should only be used by API thread.
 361 };
 362
 363 struct DRAW_DYNAMIC_STATE
 364 {
 365     void Reset(uint32_t numThreads)
 366     {
 367         SWR_STATS* pSavePtr = pStats;
 368         memset(this, 0, sizeof(*this));
 369         pStats = pSavePtr;
 370         memset(pStats, 0, sizeof(SWR_STATS) * numThreads);
 371     }
 372     ///@todo Currently assumes only a single FE can do stream output for a draw.
 373     uint32_t SoWriteOffset[4];
 374     bool     SoWriteOffsetDirty[4];
 375
 376     SWR_STATS_FE statsFE;   // Only one FE thread per DC.
 377     SWR_STATS*   pStats;
 378 };
 379
 380 // Draw Context
 381 //    The api thread sets up a draw context that exists for the life of the draw.
 382 //    This draw context maintains all of the state needed for the draw operation.
 383 struct DRAW_CONTEXT
 384 {
 385     SWR_CONTEXT*    pContext;
 386     union
 387     {
 388         MacroTileMgr*   pTileMgr;
 389         DispatchQueue*  pDispatch;      // Queue for thread groups. (isCompute)
 390     };
 391     DRAW_STATE*     pState;             // Read-only state. Core should not update this outside of API thread.
 392     DRAW_DYNAMIC_STATE dynState;
 393
 394     CachingArena*   pArena;
 395
 396     uint32_t        drawId;
 397     bool            dependentFE;    // Frontend work is dependent on all previous FE
 398     bool            dependent;      // Backend work is dependent on all previous BE
 399     bool            isCompute;      // Is this DC a compute context?
 400     bool            cleanupState;   // True if this is the last draw using an entry in the state ring.
 401     volatile bool   doneFE;         // Is FE work done for this draw?
 402
 403     FE_WORK         FeWork;
 404
 405     volatile OSALIGNLINE(uint32_t)   FeLock;
 406     volatile int32_t    threadsDone;
 407
 408     SYNC_DESC       retireCallback; // Call this func when this DC is retired.
 409
 410
 411 };
 412
 413 static_assert((sizeof(DRAW_CONTEXT) & 63) == 0, "Invalid size for DRAW_CONTEXT");
 414
 415 INLINE const API_STATE& GetApiState(const DRAW_CONTEXT* pDC)
 416 {
 417     SWR_ASSERT(pDC != nullptr);
 418     SWR_ASSERT(pDC->pState != nullptr);
 419
 420     return pDC->pState->state;
 421 }
 422
 423 INLINE void* GetPrivateState(const DRAW_CONTEXT* pDC)
 424 {
 425     SWR_ASSERT(pDC != nullptr);
 426     SWR_ASSERT(pDC->pState != nullptr);
 427
 428     return pDC->pState->pPrivateState;
 429 }
 430
 431 class HotTileMgr;
 432
 433 struct SWR_CONTEXT
 434 {
 435     // Draw Context Ring
 436     //  Each draw needs its own state in order to support mulitple draws in flight across multiple threads.
 437     //  We maintain N draw contexts configured as a ring. The size of the ring limits the maximum number
 438     //  of draws that can be in flight at any given time.
 439     //
 440     //  Description:
 441     //  1. State - When an application first sets state we'll request a new draw context to use.
 442     //     a. If there are no available draw contexts then we'll have to wait until one becomes free.
 443     //     b. If one is available then set pCurDrawContext to point to it and mark it in use.
 444     //     c. All state calls set state on pCurDrawContext.
 445     //  2. Draw - Creates submits a work item that is associated with current draw context.
 446     //     a. Set pPrevDrawContext = pCurDrawContext
 447     //     b. Set pCurDrawContext to NULL.
 448     //  3. State - When an applications sets state after draw
 449     //     a. Same as step 1.
 450     //     b. State is copied from prev draw context to current.
 451     RingBuffer<DRAW_CONTEXT> dcRing;
 452
 453     DRAW_CONTEXT *pCurDrawContext;    // This points to DC entry in ring for an unsubmitted draw.
 454     DRAW_CONTEXT *pPrevDrawContext;   // This points to DC entry for the previous context submitted that we can copy state from.
 455
 456     MacroTileMgr* pMacroTileManagerArray;
 457     DispatchQueue* pDispatchQueueArray;
 458
 459     // Draw State Ring
 460     //  When draw are very large (lots of primitives) then the API thread will break these up.
 461     //  These split draws all have identical state. So instead of storing the state directly
 462     //  in the Draw Context (DC) we instead store it in a Draw State (DS). This allows multiple DCs
 463     //  to reference a single entry in the DS ring.
 464     RingBuffer<DRAW_STATE> dsRing;
 465
 466     uint32_t curStateId;               // Current index to the next available entry in the DS ring.
 467
 468     uint32_t NumWorkerThreads;
 469     uint32_t NumFEThreads;
 470     uint32_t NumBEThreads;
 471
 472     THREAD_POOL threadPool; // Thread pool associated with this context
 473     SWR_THREADING_INFO threadInfo;
 474
 475     std::condition_variable FifosNotEmpty;
 476     std::mutex WaitLock;
 477
 478     uint32_t privateStateSize;
 479
 480     HotTileMgr *pHotTileMgr;
 481
 482     // Callback functions, passed in at create context time
 483     PFN_LOAD_TILE               pfnLoadTile;
 484     PFN_STORE_TILE              pfnStoreTile;
 485     PFN_CLEAR_TILE              pfnClearTile;
 486     PFN_UPDATE_SO_WRITE_OFFSET  pfnUpdateSoWriteOffset;
 487     PFN_UPDATE_STATS            pfnUpdateStats;
 488     PFN_UPDATE_STATS_FE         pfnUpdateStatsFE;
 489
 490
 491     // Global Stats
 492     SWR_STATS* pStats;
 493
 494     // Scratch space for workers.
 495     uint8_t** ppScratch;
 496
 497     volatile int32_t  drawsOutstandingFE;
 498
 499     CachingAllocator cachingArenaAllocator;
 500     uint32_t frameCount;
 501
 502     uint32_t lastFrameChecked;
 503     uint64_t lastDrawChecked;
 504     TileSet singleThreadLockedTiles;
 505
 506     // ArchRast thread contexts.
 507     HANDLE* pArContext;
 508 };
 509
 510 #define UPDATE_STAT_BE(name, count) if (GetApiState(pDC).enableStatsBE) { pDC->dynState.pStats[workerId].name += count; }
 511 #define UPDATE_STAT_FE(name, count) if (GetApiState(pDC).enableStatsFE) { pDC->dynState.statsFE.name += count; }
 512
 513 // ArchRast instrumentation framework
 514 #define AR_WORKER_CTX  pContext->pArContext[workerId]
 515 #define AR_API_CTX     pContext->pArContext[pContext->NumWorkerThreads]
 516
 517 #ifdef KNOB_ENABLE_AR
 518     #define _AR_BEGIN(ctx, type, id)    ArchRast::Dispatch(ctx, ArchRast::Start(ArchRast::type, id))
 519     #define _AR_END(ctx, type, count)   ArchRast::Dispatch(ctx, ArchRast::End(ArchRast::type, count))
 520     #define _AR_EVENT(ctx, event)       ArchRast::Dispatch(ctx, ArchRast::event)
 521 #else
 522     #ifdef KNOB_ENABLE_RDTSC
 523         #define _AR_BEGIN(ctx, type, id) (void)ctx; RDTSC_START(type)
 524         #define _AR_END(ctx, type, id)   RDTSC_STOP(type, id, 0)
 525     #else
 526         #define _AR_BEGIN(ctx, type, id) (void)ctx
 527         #define _AR_END(ctx, type, id)
 528     #endif
 529     #define _AR_EVENT(ctx, event)
 530 #endif
 531
 532 // Use these macros for api thread.
 533 #define AR_API_BEGIN(type, id) _AR_BEGIN(AR_API_CTX, type, id)
 534 #define AR_API_END(type, count) _AR_END(AR_API_CTX, type, count)
 535 #define AR_API_EVENT(event) _AR_EVENT(AR_API_CTX, event)
 536
 537 // Use these macros for worker threads.
 538 #define AR_BEGIN(type, id) _AR_BEGIN(AR_WORKER_CTX, type, id)
 539 #define AR_END(type, count) _AR_END(AR_WORKER_CTX, type, count)
 540 #define AR_EVENT(event) _AR_EVENT(AR_WORKER_CTX, event)