This allows using the same codepath for both unix and windows for distributing new slices to code to threads. This also improves the performance on unix - instead of waiting for all the current threads to finish their current slice before handing out a new slice to each of them (where the threads that finish first will just wait instead of immediately getting a new slice to work on), we now use the same logic as on windows. In one setup, it improves the performance of encoding from ~920 fps to ~950 fps, and in another setup it goes from ~390 fps to ~660 fps. (These tests were done with the SM_ROWMB_SLICE mode, which heavily exercises the code for distributing new slices to the worker threads.) The extra WelsEventSignal call on windows where it isn't strictly necessary doesn't incur any measurable slowdown, so it is kept without any extra ifdefs to keep the code more readable and unified.
Showing the changes to codec/encoder/core/src/encoder_ext.cpp (1 of 3 files). Show all
Unified · Split
| @@ -3225,7 +3225,7 @@ int32_t WelsEncoderEncodeExt (sWelsEncCtx* pCtx, SFrameBSInfo * pFbi, const SSou · 3224 unmodified lines | |||
| 3225 | return ENC_RETURN_UNEXPECTED; | 3225 | return ENC_RETURN_UNEXPECTED; |
| 3226 | } | 3226 | } |
| 3227 | 3227 | ||
| 3228 | WelsMultipleEventsWaitAllBlocking (iSliceCount, &pCtx->pSliceThreading->pSliceCodedEvent[0]); | 3228 | WelsMultipleEventsWaitAllBlocking (iSliceCount, &pCtx->pSliceThreading->pSliceCodedEvent[0], &pCtx->pSliceThreading->pSliceCodedMasterEvent); |
| 3229 | 3229 | ||
| 3230 | 3230 | ||
| 3231 | // all slices are finished coding here | 3231 | // all slices are finished coding here |
| @@ -3266,12 +3266,12 @@ int32_t WelsEncoderEncodeExt (sWelsEncCtx* pCtx, SFrameBSInfo * pFbi, const SSou · 34 unmodified lines | |||
| 3266 | while (1) { | 3266 | while (1) { |
| 3267 | if (iIndexOfSliceToBeCoded >= iSliceCount && iNumThreadsRunning <= 0) | 3267 | if (iIndexOfSliceToBeCoded >= iSliceCount && iNumThreadsRunning <= 0) |
| 3268 | break; | 3268 | break; |
| 3269 | #ifdef _WIN32 | ||
| 3270 | WELS_THREAD_ERROR_CODE lwait = 0; | 3269 | WELS_THREAD_ERROR_CODE lwait = 0; |
| 3271 | int32_t iEventId = -1; | 3270 | int32_t iEventId = -1; |
| 3272 | 3271 | ||
| 3273 | lwait = WelsMultipleEventsWaitSingleBlocking (iNumThreadsScheduled, | 3272 | lwait = WelsMultipleEventsWaitSingleBlocking (iNumThreadsScheduled, |
| 3274 | &pCtx->pSliceThreading->pSliceCodedEvent[0]); | 3273 | &pCtx->pSliceThreading->pSliceCodedEvent[0], |
| 3274 | &pCtx->pSliceThreading->pSliceCodedMasterEvent); | ||
| 3275 | iEventId = (int32_t) (lwait - WELS_THREAD_ERROR_WAIT_OBJECT_0); | 3275 | iEventId = (int32_t) (lwait - WELS_THREAD_ERROR_WAIT_OBJECT_0); |
| 3276 | if (iEventId >= 0 && iEventId < iNumThreadsScheduled) { | 3276 | if (iEventId >= 0 && iEventId < iNumThreadsScheduled) { |
| 3277 | if (iIndexOfSliceToBeCoded < iSliceCount) { | 3277 | if (iIndexOfSliceToBeCoded < iSliceCount) { |
| @@ -3285,29 +3285,6 @@ int32_t WelsEncoderEncodeExt (sWelsEncCtx* pCtx, SFrameBSInfo * pFbi, const SSou · 7 unmodified lines | |||
| 3285 | -- iNumThreadsRunning; | 3285 | -- iNumThreadsRunning; |
| 3286 | } | 3286 | } |
| 3287 | } | 3287 | } |
| 3288 | #else | ||
| 3289 | // TODO for pthread platforms | ||
| 3290 | // alternate implementation using blocking due non-blocking with timeout mode not support at wels thread lib, tune back if available | ||
| 3291 | WelsMultipleEventsWaitAllBlocking (iNumThreadsRunning, &pCtx->pSliceThreading->pSliceCodedEvent[0]); | ||
| 3292 | WELS_VERIFY_RETURN_IFNEQ(pCtx->iEncoderError, ENC_RETURN_SUCCESS) | ||
| 3293 | if (iIndexOfSliceToBeCoded < iSliceCount) { | ||
| 3294 | int32_t iThreadIdx = 0; | ||
| 3295 | // pick up succeeding slices for threading if left | ||
| 3296 | while (iThreadIdx < iNumThreadsScheduled) { | ||
| 3297 | if (iIndexOfSliceToBeCoded >= iSliceCount) | ||
| 3298 | break; | ||
| 3299 | pCtx->pSliceThreading->pThreadPEncCtx[iThreadIdx].iSliceIndex = iIndexOfSliceToBeCoded; | ||
| 3300 | WelsEventSignal (&pCtx->pSliceThreading->pReadySliceCodingEvent[iThreadIdx]); | ||
| 3301 | |||
| 3302 | ++ iIndexOfSliceToBeCoded; | ||
| 3303 | ++ iThreadIdx; | ||
| 3304 | } | ||
| 3305 | // update iNumThreadsRunning | ||
| 3306 | iNumThreadsRunning = iThreadIdx; | ||
| 3307 | } else { | ||
| 3308 | iNumThreadsRunning = 0; | ||
| 3309 | } | ||
| 3310 | #endif//_WIN32 | ||
| 3311 | }//while(1) | 3288 | }//while(1) |
| 3312 | 3289 | ||
| 3313 | // all slices are finished coding here | 3290 | // all slices are finished coding here |
| @@ -3329,7 +3306,7 @@ int32_t WelsEncoderEncodeExt (sWelsEncCtx* pCtx, SFrameBSInfo * pFbi, const SSou · 15 unmodified lines | |||
| 3329 | return ENC_RETURN_UNEXPECTED; | 3306 | return ENC_RETURN_UNEXPECTED; |
| 3330 | } | 3307 | } |
| 3331 | 3308 | ||
| 3332 | WelsMultipleEventsWaitAllBlocking (kiPartitionCnt, &pCtx->pSliceThreading->pSliceCodedEvent[0]); | 3309 | WelsMultipleEventsWaitAllBlocking (kiPartitionCnt, &pCtx->pSliceThreading->pSliceCodedEvent[0], &pCtx->pSliceThreading->pSliceCodedMasterEvent); |
| 3333 | WELS_VERIFY_RETURN_IFNEQ(pCtx->iEncoderError, ENC_RETURN_SUCCESS) | 3310 | WELS_VERIFY_RETURN_IFNEQ(pCtx->iEncoderError, ENC_RETURN_SUCCESS) |
| 3334 | 3311 | ||
| 3335 | iLayerSize = AppendSliceToFrameBs (pCtx, pLayerBsInfo, kiPartitionCnt); | 3312 | iLayerSize = AppendSliceToFrameBs (pCtx, pLayerBsInfo, kiPartitionCnt); |