swr/rast: Correctly allocate SWR_STATS memory as cacheline aligned

Cacheline alignment of SWR_STATS to prevent sharing of cachelines
between threads (performance).

Gets rid of gcc-7.1 warning about using c++17's over-aligned new
feature.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

diff --git a/src/gallium/drivers/swr/rasterizer/core/api.cpp b/src/gallium/drivers/swr/rasterizer/core/api.cpp
index d3d80e4a789ad05d4c32e468e1f1062dbca42831..087a24a95edf989b911aef80e1104765f70f5737 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/core/api.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/api.cpp
@@ -108,7 +108,7 @@ HANDLE SwrCreateContext(
     CreateThreadPool(pContext, &pContext->threadPool);
 
     pContext->ppScratch = new uint8_t*[pContext->NumWorkerThreads];
     CreateThreadPool(pContext, &pContext->threadPool);
 
     pContext->ppScratch = new uint8_t*[pContext->NumWorkerThreads];

-    pContext->pStats = new SWR_STATS[pContext->NumWorkerThreads];
+    pContext->pStats = (SWR_STATS*)AlignedMalloc(sizeof(SWR_STATS) * pContext->NumWorkerThreads, 64);

 
 #if defined(KNOB_ENABLE_AR)
     // Setup ArchRast thread contexts which includes +1 for API thread.
 
 #if defined(KNOB_ENABLE_AR)
     // Setup ArchRast thread contexts which includes +1 for API thread.


@@ -363,7 +363,7 @@ void SwrDestroyContext(HANDLE hContext)
     // free the fifos
     for (uint32_t i = 0; i < KNOB_MAX_DRAWS_IN_FLIGHT; ++i)
     {
     // free the fifos
     for (uint32_t i = 0; i < KNOB_MAX_DRAWS_IN_FLIGHT; ++i)
     {

-        delete[] pContext->dcRing[i].dynState.pStats;
+        AlignedFree(pContext->dcRing[i].dynState.pStats);

         delete pContext->dcRing[i].pArena;
         delete pContext->dsRing[i].pArena;
         pContext->pMacroTileManagerArray[i].~MacroTileMgr();
         delete pContext->dcRing[i].pArena;
         delete pContext->dsRing[i].pArena;
         pContext->pMacroTileManagerArray[i].~MacroTileMgr();


@@ -388,7 +388,7 @@ void SwrDestroyContext(HANDLE hContext)
     }
 
     delete[] pContext->ppScratch;
     }
 
     delete[] pContext->ppScratch;

-    delete[] pContext->pStats;
+    AlignedFree(pContext->pStats);

 
     delete(pContext->pHotTileMgr);
 
 
     delete(pContext->pHotTileMgr);
 

diff --git a/src/gallium/drivers/swr/rasterizer/core/threads.cpp b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
index e03632b443bfb8434baa7beb43e455be7c983727..36710bf0da56248a3ad4ce96d002fdcfde261ddf 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/core/threads.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
@@ -363,7 +363,7 @@ INLINE void UpdateClientStats(SWR_CONTEXT* pContext, uint32_t workerId, DRAW_CON
     }
 
     DRAW_DYNAMIC_STATE& dynState = pDC->dynState;
     }
 
     DRAW_DYNAMIC_STATE& dynState = pDC->dynState;

-    SWR_STATS stats{ 0 };
+    OSALIGNLINE(SWR_STATS) stats{ 0 };

 
     // Sum up stats across all workers before sending to client.
     for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i)
 
     // Sum up stats across all workers before sending to client.
     for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i)


@@ -986,7 +986,7 @@ void CreateThreadPool(SWR_CONTEXT* pContext, THREAD_POOL* pPool)
     // Initialize DRAW_CONTEXT's per-thread stats
     for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc)
     {
     // Initialize DRAW_CONTEXT's per-thread stats
     for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc)
     {

-        pContext->dcRing[dc].dynState.pStats = new SWR_STATS[numThreads];
+        pContext->dcRing[dc].dynState.pStats = (SWR_STATS*)AlignedMalloc(sizeof(SWR_STATS) * numThreads, 64);

         memset(pContext->dcRing[dc].dynState.pStats, 0, sizeof(SWR_STATS) * numThreads);
     }
 
         memset(pContext->dcRing[dc].dynState.pStats, 0, sizeof(SWR_STATS) * numThreads);
     }