[Mesa-dev] [PATCH 4/5] swr: [rasterizer core] Frontend dependency work

[George Kyriazis]

Add frontend dependency concept in the DRAW_CONTEXT, which
allows serialization of frontend work if necessary.
---
 src/gallium/drivers/swr/rasterizer/core/api.cpp     |  3 +++
 src/gallium/drivers/swr/rasterizer/core/context.h   |  3 ++-
 src/gallium/drivers/swr/rasterizer/core/threads.cpp | 14 +++++++++++++-
 3 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/swr/rasterizer/core/api.cpp b/src/gallium/drivers/swr/rasterizer/core/api.cpp
index e67ede2..5f941e8 100644
--- a/src/gallium/drivers/swr/rasterizer/core/api.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/api.cpp
@@ -317,7 +317,10 @@ DRAW_CONTEXT* GetDrawContext(SWR_CONTEXT *pContext, bool isSplitDraw = false)
 
         SWR_ASSERT(pCurDrawContext->pArena->IsEmpty() == true);
 
+        // Reset dependency
         pCurDrawContext->dependent = false;
+        pCurDrawContext->dependentFE = false;
+
         pCurDrawContext->pContext = pContext;
         pCurDrawContext->isCompute = false; // Dispatch has to set this to true.
 
diff --git a/src/gallium/drivers/swr/rasterizer/core/context.h b/src/gallium/drivers/swr/rasterizer/core/context.h
index 9a26e33..a9de63b 100644
--- a/src/gallium/drivers/swr/rasterizer/core/context.h
+++ b/src/gallium/drivers/swr/rasterizer/core/context.h
@@ -404,7 +404,8 @@ struct DRAW_CONTEXT
     CachingArena*   pArena;
 
     uint32_t        drawId;
-    bool            dependent;
+    bool            dependentFE;    // Frontend work is dependent on all previous FE
+    bool            dependent;      // Backend work is dependent on all previous BE
     bool            isCompute;      // Is this DC a compute context?
     bool            cleanupState;   // True if this is the last draw using an entry in the state ring.
     volatile bool   doneFE;         // Is FE work done for this draw?
diff --git a/src/gallium/drivers/swr/rasterizer/core/threads.cpp b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
index ea5542a..701a550 100644
--- a/src/gallium/drivers/swr/rasterizer/core/threads.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/threads.cpp
@@ -313,6 +313,11 @@ bool CheckDependency(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t lastReti
     return pDC->dependent && IDComparesLess(lastRetiredDraw, pDC->drawId - 1);
 }
 
+bool CheckDependencyFE(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t lastRetiredDraw)
+{
+    return pDC->dependentFE && IDComparesLess(lastRetiredDraw, pDC->drawId - 1);
+}
+
 //////////////////////////////////////////////////////////////////////////
 /// @brief Update client stats.
 INLINE void UpdateClientStats(SWR_CONTEXT* pContext, uint32_t workerId, DRAW_CONTEXT* pDC)
@@ -595,6 +600,7 @@ INLINE void CompleteDrawFE(SWR_CONTEXT* pContext, uint32_t workerId, DRAW_CONTEX
     // Ensure all streaming writes are globally visible before marking this FE done
     _mm_mfence();
     pDC->doneFE = true;
+
     InterlockedDecrement((volatile LONG*)&pContext->drawsOutstandingFE);
 }
 
@@ -606,7 +612,7 @@ void WorkOnFifoFE(SWR_CONTEXT *pContext, uint32_t workerId, uint32_t &curDrawFE)
     {
         uint32_t dcSlot = curDrawFE % KNOB_MAX_DRAWS_IN_FLIGHT;
         DRAW_CONTEXT *pDC = &pContext->dcRing[dcSlot];
-        if (pDC->isCompute || pDC->doneFE || pDC->FeLock)
+        if (pDC->isCompute || pDC->doneFE)
         {
             CompleteDrawContextInl(pContext, workerId, pDC);
             curDrawFE++;
@@ -617,6 +623,7 @@ void WorkOnFifoFE(SWR_CONTEXT *pContext, uint32_t workerId, uint32_t &curDrawFE)
         }
     }
 
+    uint32_t lastRetiredFE = curDrawFE - 1;
     uint32_t curDraw = curDrawFE;
     while (IDComparesLess(curDraw, drawEnqueued))
     {
@@ -625,6 +632,11 @@ void WorkOnFifoFE(SWR_CONTEXT *pContext, uint32_t workerId, uint32_t &curDrawFE)
 
         if (!pDC->isCompute && !pDC->FeLock)
         {
+            if (CheckDependencyFE(pContext, pDC, lastRetiredFE))
+            {
+                return;
+            }
+
             uint32_t initial = InterlockedCompareExchange((volatile uint32_t*)&pDC->FeLock, 1, 0);
             if (initial == 0)
             {
-- 
2.7.4