gallium/swr: spin-lock performance improvement

Currently, the worker threads are very aggresively
polling for new tasks. If the work is not constantly
fed into the pipeline (which is a case for most of
interactive applications), this creates unnecessary
memory pressure and is using CPU cycles that
could otherwise be used by the applications.
The change implements simple back off mechanism
to help with this problem

Change by Tomasz Pyra ([email protected])

Reviewed-by: Alok Hota <[email protected]>
Reviewed-by: Jan Zielinski <[email protected]>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4226>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4226>

1 files changed, 14 insertions, 1 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/core/threads.cpp b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
index c75fb568431..556e02e99ef 100644
--- a/src/gallium/drivers/swr/rasterizer/core/threads.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
@@ -592,17 +592,20 @@ bool WorkOnFifoBE(SWR_CONTEXT* pContext,
             pDC->pTileMgr->getTileIndices(tileID, x, y);
             if (((x ^ y) & numaMask) != numaNode)
             {
+                _mm_pause();
                 continue;
             }
 
             if (!tile->getNumQueued())
             {
+                _mm_pause();
                 continue;
             }
 
             // can only work on this draw if it's not in use by other threads
             if (lockedTiles.get(tileID))
             {
+                _mm_pause();
                 continue;
             }
 
@@ -663,6 +666,7 @@ bool WorkOnFifoBE(SWR_CONTEXT* pContext,
                 // This tile is already locked. So let's add it to our locked tiles set. This way we
                 // don't try locking this one again.
                 lockedTiles.set(tileID);
+                _mm_pause();
             }
         }
     }
@@ -750,7 +754,7 @@ void WorkOnFifoFE(SWR_CONTEXT* pContext, uint32_t workerId, uint32_t& curDrawFE)
         uint32_t      dcSlot = curDraw % pContext->MAX_DRAWS_IN_FLIGHT;
         DRAW_CONTEXT* pDC    = &pContext->dcRing[dcSlot];
 
-        if (!pDC->isCompute && !pDC->FeLock)
+        if (!pDC->FeLock && !pDC->isCompute)
         {
             if (CheckDependencyFE(pContext, pDC, lastRetiredFE))
             {
@@ -765,7 +769,16 @@ void WorkOnFifoFE(SWR_CONTEXT* pContext, uint32_t workerId, uint32_t& curDrawFE)
 
                 CompleteDrawFE(pContext, workerId, pDC);
             }
+            else
+            {
+                _mm_pause();
+            }
+        }
+        else
+        {
+            _mm_pause();
         }
+
         curDraw++;
     }
 }