1 /****************************************************************************
2 * Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved.
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:
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
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
25 * @brief Definitions for SWR_CONTEXT and DRAW_CONTEXT
26 * The SWR_CONTEXT is our global context and contains the DC ring,
29 * The DRAW_CONTEXT contains all state associated with a draw operation.
31 ******************************************************************************/
34 #include <condition_variable>
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"
46 // x.8 fixed point precision values
47 #define FIXED_POINT_SHIFT 8
48 #define FIXED_POINT_SCALE 256
50 // x.16 fixed point precision values
51 #define FIXED_POINT16_SHIFT 16
52 #define FIXED_POINT16_SCALE 65536
59 uint32_t frontFacing
: 1;
61 uint32_t coverageMask
: (SIMD_TILE_X_DIM
* SIMD_TILE_Y_DIM
);
62 uint32_t reserved
: 32 - 1 - 1 - (SIMD_TILE_X_DIM
* SIMD_TILE_Y_DIM
);
65 uint32_t renderTargetArrayIndex
;
68 //////////////////////////////////////////////////////////////////////////
70 /////////////////////////////////////////////////////////////////////////
71 struct SWR_TRIANGLE_DESC
83 float *pUserClipBuffer
;
85 uint64_t coverageMask
[SWR_MAX_NUM_MULTISAMPLES
];
86 uint64_t anyCoveredSamples
;
91 struct TRIANGLE_WORK_DESC
95 float *pUserClipBuffer
;
112 float clearRTColor
[4]; // RGBA_32F
113 float clearDepth
; // [0..1]
114 uint8_t clearStencil
;
117 struct DISCARD_INVALIDATE_TILES_DESC
119 uint32_t attachmentMask
;
121 SWR_TILE_STATE newTileState
;
128 PFN_CALLBACK_FUNC pfnCallbackFunc
;
139 struct STORE_TILES_DESC
141 SWR_RENDERTARGET_ATTACHMENT attachment
;
142 SWR_TILE_STATE postStoreTileState
;
147 uint32_t threadGroupCountX
;
148 uint32_t threadGroupCountY
;
149 uint32_t threadGroupCountZ
;
152 typedef void(*PFN_WORK_FUNC
)(DRAW_CONTEXT
* pDC
, uint32_t workerId
, uint32_t macroTile
, void* pDesc
);
159 DISCARDINVALIDATETILES
,
167 PFN_WORK_FUNC pfnWork
;
171 TRIANGLE_WORK_DESC tri
;
173 DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles
;
174 STORE_TILES_DESC storeTiles
;
175 QUERY_DESC queryStats
;
184 uint32_t numIndices
; // DrawIndexed: Number of indices for draw.
185 uint32_t numVerts
; // Draw: Number of verts (triangles, lines, etc)
189 const int32_t* pIB
; // DrawIndexed: App supplied indices
190 uint32_t startVertex
; // Draw: Starting vertex in VB to render from.
193 uint32_t numInstances
; // Number of instances
194 uint32_t startInstance
; // Instance offset
195 uint32_t startPrimID
; // starting primitiveID for this draw batch
196 uint32_t startVertexID
; // starting VertexID for this draw batch (only needed for non-indexed draws)
197 SWR_FORMAT type
; // index buffer type
200 typedef void(*PFN_FE_WORK_FUNC
)(SWR_CONTEXT
* pContext
, DRAW_CONTEXT
* pDC
, uint32_t workerId
, void* pDesc
);
204 PFN_FE_WORK_FUNC pfnWork
;
210 DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles
;
211 STORE_TILES_DESC storeTiles
;
212 QUERY_DESC queryStats
;
218 float left
, right
, top
, bottom
;
223 // function signature for pipeline stages that execute after primitive assembly
224 typedef void(*PFN_PROCESS_PRIMS
)(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simdvector prims
[],
225 uint32_t primMask
, simdscalari primID
);
227 OSALIGNLINE(struct) API_STATE
230 SWR_VERTEX_BUFFER_STATE vertexBuffers
[KNOB_NUM_STREAMS
];
233 SWR_INDEX_BUFFER_STATE indexBuffer
;
235 // FS - Fetch Shader State
236 PFN_FETCH_FUNC pfnFetchFunc
;
238 // VS - Vertex Shader State
239 PFN_VERTEX_FUNC pfnVertexFunc
;
241 // GS - Geometry Shader State
242 PFN_GS_FUNC pfnGsFunc
;
243 SWR_GS_STATE gsState
;
245 // CS - Compute Shader
246 PFN_CS_FUNC pfnCsFunc
;
247 uint32_t totalThreadsInGroup
;
248 uint32_t totalSpillFillSize
;
250 // FE - Frontend State
251 SWR_FRONTEND_STATE frontendState
;
253 // SOS - Streamout Shader State
254 PFN_SO_FUNC pfnSoFunc
[MAX_SO_STREAMS
];
257 SWR_STREAMOUT_STATE soState
;
258 mutable SWR_STREAMOUT_BUFFER soBuffer
[MAX_SO_STREAMS
];
260 // Tessellation State
261 PFN_HS_FUNC pfnHsFunc
;
262 PFN_DS_FUNC pfnDsFunc
;
263 SWR_TS_STATE tsState
;
265 // Specifies which VS outputs are sent to PS.
266 // Does not include position
267 uint32_t linkageMask
;
268 uint32_t linkageCount
;
269 uint8_t linkageMap
[MAX_ATTRIBUTES
];
271 // attrib mask, specifies the total set of attributes used
272 // by the frontend (vs, so, gs)
273 uint32_t feAttribMask
;
275 PRIMITIVE_TOPOLOGY topology
;
278 // RS - Rasterizer State
279 SWR_RASTSTATE rastState
;
280 // floating point multisample offsets
281 float samplePos
[SWR_MAX_NUM_MULTISAMPLES
* 2];
285 SWR_VIEWPORT vp
[KNOB_NUM_VIEWPORTS_SCISSORS
];
286 SWR_VIEWPORT_MATRIX vpMatrix
[KNOB_NUM_VIEWPORTS_SCISSORS
];
288 BBOX scissorRects
[KNOB_NUM_VIEWPORTS_SCISSORS
];
289 BBOX scissorInFixedPoint
;
292 SWR_BACKEND_STATE backendState
;
294 // PS - Pixel shader state
295 SWR_PS_STATE psState
;
297 SWR_DEPTH_STENCIL_STATE depthStencilState
;
299 // OM - Output Merger State
300 SWR_BLEND_STATE blendState
;
301 PFN_BLEND_JIT_FUNC pfnBlendFunc
[SWR_NUM_RENDERTARGETS
];
303 // Stats are incremented when this is true.
308 uint32_t colorHottileEnable
: 8;
309 uint32_t depthHottileEnable
: 1;
310 uint32_t stencilHottileEnable
: 1;
313 PFN_QUANTIZE_DEPTH pfnQuantizeDepth
;
319 struct RenderOutputBuffers
321 uint8_t* pColor
[SWR_NUM_RENDERTARGETS
];
326 // Plane equation A/B/C coeffs used to evaluate I/J barycentric coords
327 struct BarycentricCoeffs
341 simdscalar vRecipDet
;
343 simdscalar vAOneOverW
;
344 simdscalar vBOneOverW
;
345 simdscalar vCOneOverW
;
348 // pipeline function pointer types
349 typedef void(*PFN_BACKEND_FUNC
)(DRAW_CONTEXT
*, uint32_t, uint32_t, uint32_t, SWR_TRIANGLE_DESC
&, RenderOutputBuffers
&);
350 typedef void(*PFN_OUTPUT_MERGER
)(SWR_PS_CONTEXT
&, uint8_t* (&)[SWR_NUM_RENDERTARGETS
], uint32_t, const SWR_BLEND_STATE
*,
351 const PFN_BLEND_JIT_FUNC (&)[SWR_NUM_RENDERTARGETS
], simdscalar
&, simdscalar
);
352 typedef void(*PFN_CALC_PIXEL_BARYCENTRICS
)(const BarycentricCoeffs
&, SWR_PS_CONTEXT
&);
353 typedef void(*PFN_CALC_SAMPLE_BARYCENTRICS
)(const BarycentricCoeffs
&, SWR_PS_CONTEXT
&);
354 typedef void(*PFN_CALC_CENTROID_BARYCENTRICS
)(const BarycentricCoeffs
&, SWR_PS_CONTEXT
&, const uint64_t *const, const uint32_t,
355 const simdscalar
, const simdscalar
);
359 PFN_BACKEND_FUNC pfnBackend
;
360 PFN_CALC_PIXEL_BARYCENTRICS pfnCalcPixelBarycentrics
;
361 PFN_CALC_SAMPLE_BARYCENTRICS pfnCalcSampleBarycentrics
;
362 PFN_CALC_CENTROID_BARYCENTRICS pfnCalcCentroidBarycentrics
;
363 PFN_OUTPUT_MERGER pfnOutputMerger
;
372 void* pPrivateState
; // Its required the driver sets this up for each draw.
374 // pipeline function pointers, filled in by API thread when setting up the draw
375 BACKEND_FUNCS backendFuncs
;
376 PFN_PROCESS_PRIMS pfnProcessPrims
;
378 CachingArena
* pArena
; // This should only be used by API thread.
382 // The api thread sets up a draw context that exists for the life of the draw.
383 // This draw context maintains all of the state needed for the draw operation.
386 SWR_CONTEXT
* pContext
;
390 MacroTileMgr
* pTileMgr
;
391 DispatchQueue
* pDispatch
; // Queue for thread groups. (isCompute)
395 CachingArena
* pArena
;
397 bool isCompute
; // Is this DC a compute context?
398 bool cleanupState
; // True if this is the last draw using an entry in the state ring.
399 volatile bool doneFE
; // Is FE work done for this draw?
403 volatile OSALIGNLINE(uint32_t) FeLock
;
404 volatile int64_t threadsDone
;
407 static_assert((sizeof(DRAW_CONTEXT
) & 63) == 0, "Invalid size for DRAW_CONTEXT");
409 INLINE
const API_STATE
& GetApiState(const DRAW_CONTEXT
* pDC
)
411 SWR_ASSERT(pDC
!= nullptr);
412 SWR_ASSERT(pDC
->pState
!= nullptr);
414 return pDC
->pState
->state
;
417 INLINE
void* GetPrivateState(const DRAW_CONTEXT
* pDC
)
419 SWR_ASSERT(pDC
!= nullptr);
420 SWR_ASSERT(pDC
->pState
!= nullptr);
422 return pDC
->pState
->pPrivateState
;
430 // Each draw needs its own state in order to support mulitple draws in flight across multiple threads.
431 // We maintain N draw contexts configured as a ring. The size of the ring limits the maximum number
432 // of draws that can be in flight at any given time.
435 // 1. State - When an application first sets state we'll request a new draw context to use.
436 // a. If there are no available draw contexts then we'll have to wait until one becomes free.
437 // b. If one is available then set pCurDrawContext to point to it and mark it in use.
438 // c. All state calls set state on pCurDrawContext.
439 // 2. Draw - Creates submits a work item that is associated with current draw context.
440 // a. Set pPrevDrawContext = pCurDrawContext
441 // b. Set pCurDrawContext to NULL.
442 // 3. State - When an applications sets state after draw
443 // a. Same as step 1.
444 // b. State is copied from prev draw context to current.
445 RingBuffer
<DRAW_CONTEXT
> dcRing
;
447 DRAW_CONTEXT
*pCurDrawContext
; // This points to DC entry in ring for an unsubmitted draw.
448 DRAW_CONTEXT
*pPrevDrawContext
; // This points to DC entry for the previous context submitted that we can copy state from.
450 MacroTileMgr
* pMacroTileManagerArray
;
451 DispatchQueue
* pDispatchQueueArray
;
454 // When draw are very large (lots of primitives) then the API thread will break these up.
455 // These split draws all have identical state. So instead of storing the state directly
456 // in the Draw Context (DC) we instead store it in a Draw State (DS). This allows multiple DCs
457 // to reference a single entry in the DS ring.
458 RingBuffer
<DRAW_STATE
> dsRing
;
460 uint32_t curStateId
; // Current index to the next available entry in the DS ring.
462 uint32_t NumWorkerThreads
;
463 uint32_t NumFEThreads
;
464 uint32_t NumBEThreads
;
466 THREAD_POOL threadPool
; // Thread pool associated with this context
468 std::condition_variable FifosNotEmpty
;
471 DRIVER_TYPE driverType
;
473 uint32_t privateStateSize
;
475 HotTileMgr
*pHotTileMgr
;
477 // tile load/store functions, passed in at create context time
478 PFN_LOAD_TILE pfnLoadTile
;
479 PFN_STORE_TILE pfnStoreTile
;
480 PFN_CLEAR_TILE pfnClearTile
;
483 SWR_STATS stats
[KNOB_MAX_NUM_THREADS
];
485 // Scratch space for workers.
486 uint8_t* pScratch
[KNOB_MAX_NUM_THREADS
];
488 CachingAllocator cachingArenaAllocator
;
492 void WaitForDependencies(SWR_CONTEXT
*pContext
, uint64_t drawId
);
493 void WakeAllThreads(SWR_CONTEXT
*pContext
);
495 #define UPDATE_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name += count; }
496 #define SET_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name = count; }