amdgpu/addrlib: add matchStencilTileCfg and tcCompatible fixes

The usage should be client first call AddrComputeSurfaceInfo() on
depth surface with flag "matchStencilTilecfg", AddrLib will use
2DThin1 tile index for depth as much as possible and do not down grade
unless alignment requirement cannot be met.

1. If there is a matched 2DThin1 tile index for stencil which make
sure they will share same tile config parameters, then return the
stencil 2DThin1 tile index as well.
2. If using 2DThin1 tile mode cannot make sure such thing happen, and
TcCompatible flag was set, then ignore this flag then try 2DThin1 tile
mode for depth and stencil again.
3. If 2DThin1 tile mode cannot make sure depth and stencil to have
same tile config parameters, then down grade depth surface tile mode
to 1DThin1.
4. If depth surface's tile mode was 1DThin1, then return 1DThin1 tile
index for stencil.
5. If depth surface's tile mode is PRT, then return invalid tile index
to stencil since their tile config parameters will never be met.

Client driver then check the returned tile index of stencil -- if it
is not invalid tile index, then call AddrComputeSurfaceInfo() on
stencil surface with the returned stencil tile index to get full
output information. Please note, client needs to set flag
"useTileIndex" when AddrLib get created.

4 files changed, 152 insertions, 20 deletions

diff --git a/src/amd/addrlib/addrinterface.h b/src/amd/addrlib/addrinterface.h
index c68cacf0952..2a61b2b33f2 100644
--- a/src/amd/addrlib/addrinterface.h
+++ b/src/amd/addrlib/addrinterface.h
@@ -519,7 +519,9 @@ typedef union _ADDR_SURFACE_FLAGS
         UINT_32 rotateDisplay        : 1; ///< Rotate micro tile type
         UINT_32 minimizeAlignment    : 1; ///< Minimize alignment
         UINT_32 preferEquation       : 1; ///< Return equation index without adjusting tile mode
-        UINT_32 reserved             : 4; ///< Reserved bits
+        UINT_32 matchStencilTileCfg  : 1; ///< Select tile index of stencil as well as depth surface
+                                          ///  to make sure they share same tile config parameters
+        UINT_32 reserved             : 3; ///< Reserved bits
     };
 
     UINT_32 value;
@@ -628,6 +630,8 @@ typedef struct _ADDR_COMPUTE_SURFACE_INFO_OUTPUT
 
     /// Stereo info
     ADDR_QBSTEREOINFO*  pStereoInfo;///< Stereo information, needed when .qbStereo flag is TRUE
+
+    INT_32          stencilTileIdx; ///< stencil tile index output when matchStencilTileCfg was set
 } ADDR_COMPUTE_SURFACE_INFO_OUTPUT;
 
 /**
diff --git a/src/amd/addrlib/core/addrlib1.cpp b/src/amd/addrlib/core/addrlib1.cpp
index 9cdc459867b..fb33145f88d 100644
--- a/src/amd/addrlib/core/addrlib1.cpp
+++ b/src/amd/addrlib/core/addrlib1.cpp
@@ -3557,10 +3557,13 @@ VOID Lib::OptimizeTileMode(
     ) const
 {
     AddrTileMode tileMode = pInOut->tileMode;
+
     BOOL_32 doOpt = (pInOut->flags.opt4Space == TRUE) ||
                     (pInOut->flags.minimizeAlignment == TRUE) ||
                     (pInOut->maxBaseAlign != 0);
 
+    BOOL_32 convertToPrt = FALSE;
+
     // Optimization can only be done on level 0 and samples <= 1
     if ((doOpt == TRUE)                     &&
         (pInOut->mipLevel == 0)             &&
@@ -3570,7 +3573,6 @@ VOID Lib::OptimizeTileMode(
         UINT_32 width = pInOut->width;
         UINT_32 height = pInOut->height;
         UINT_32 thickness = Thickness(tileMode);
-        BOOL_32 convertToPrt = FALSE;
         BOOL_32 macroTiledOK = TRUE;
         UINT_32 macroWidthAlign = 0;
         UINT_32 macroHeightAlign = 0;
@@ -3680,16 +3682,23 @@ VOID Lib::OptimizeTileMode(
                 }
             }
         }
+    }
 
-        if (convertToPrt)
+    if (convertToPrt)
+    {
+        if ((pInOut->flags.matchStencilTileCfg == TRUE) && (pInOut->numSamples <= 1))
         {
-            HwlSetPrtTileMode(pInOut);
+            pInOut->tileMode = ADDR_TM_1D_TILED_THIN1;
         }
-        else if (tileMode != pInOut->tileMode)
+        else
         {
-            pInOut->tileMode = tileMode;
+            HwlSetPrtTileMode(pInOut);
         }
     }
+    else if (tileMode != pInOut->tileMode)
+    {
+        pInOut->tileMode = tileMode;
+    }
 
     HwlOptimizeTileMode(pInOut);
 }
diff --git a/src/amd/addrlib/r800/ciaddrlib.cpp b/src/amd/addrlib/r800/ciaddrlib.cpp
index 2c62979619a..5ccc5da4197 100644
--- a/src/amd/addrlib/r800/ciaddrlib.cpp
+++ b/src/amd/addrlib/r800/ciaddrlib.cpp
@@ -709,16 +709,60 @@ ADDR_E_RETURNCODE CiLib::HwlComputeSurfaceInfo(
         pOut->macroModeIndex = TileIndexInvalid;
     }
 
-    // Pass tcCompatible flag from input to output; and turn off it if tile split occurs
-    pOut->tcCompatible = pIn->flags.tcCompatible;
-
-    ADDR_E_RETURNCODE retCode = SiLib::HwlComputeSurfaceInfo(pIn,pOut);
+    ADDR_E_RETURNCODE retCode = SiLib::HwlComputeSurfaceInfo(pIn, pOut);
 
     if (pOut->macroModeIndex == TileIndexNoMacroIndex)
     {
         pOut->macroModeIndex = TileIndexInvalid;
     }
 
+    if ((pIn->flags.matchStencilTileCfg == TRUE) &&
+        (pIn->flags.depth == TRUE))
+    {
+        pOut->stencilTileIdx = TileIndexInvalid;
+
+        if ((MinDepth2DThinIndex <= pOut->tileIndex) &&
+            (MaxDepth2DThinIndex >= pOut->tileIndex))
+        {
+            BOOL_32 depthStencil2DTileConfigMatch = DepthStencilTileCfgMatch(pIn, pOut);
+
+            if ((depthStencil2DTileConfigMatch == FALSE) &&
+                (pOut->tcCompatible == TRUE))
+            {
+                pOut->macroModeIndex = TileIndexInvalid;
+
+                ADDR_COMPUTE_SURFACE_INFO_INPUT localIn = *pIn;
+                localIn.tileIndex = TileIndexInvalid;
+                localIn.pTileInfo = NULL;
+                localIn.flags.tcCompatible = FALSE;
+
+                SiLib::HwlComputeSurfaceInfo(&localIn, pOut);
+
+                ADDR_ASSERT((MinDepth2DThinIndex <= pOut->tileIndex) && (MaxDepth2DThinIndex >= pOut->tileIndex));
+
+                depthStencil2DTileConfigMatch = DepthStencilTileCfgMatch(pIn, pOut);
+            }
+
+            if ((depthStencil2DTileConfigMatch == FALSE) &&
+                (pIn->numSamples <= 1))
+            {
+                pOut->macroModeIndex = TileIndexInvalid;
+
+                ADDR_COMPUTE_SURFACE_INFO_INPUT localIn = *pIn;
+                localIn.tileMode = ADDR_TM_1D_TILED_THIN1;
+                localIn.tileIndex = TileIndexInvalid;
+                localIn.pTileInfo = NULL;
+
+                retCode = SiLib::HwlComputeSurfaceInfo(&localIn, pOut);
+            }
+        }
+
+        if (pOut->tileIndex == Depth1DThinIndex)
+        {
+            pOut->stencilTileIdx = Depth1DThinIndex;
+        }
+    }
+
     return retCode;
 }
 
@@ -1150,11 +1194,11 @@ VOID CiLib::HwlSelectTileMode(
     {
         pInOut->flags.opt4Space = TRUE;
         pInOut->maxBaseAlign = Block64K;
-
-        // Optimize tile mode if possible
-        OptimizeTileMode(pInOut);
     }
 
+    // Optimize tile mode if possible
+    OptimizeTileMode(pInOut);
+
     HwlOverrideTileMode(pInOut);
 }
 
@@ -1256,6 +1300,12 @@ VOID CiLib::HwlSetupTileInfo(
         }
     }
 
+    // tcCompatible flag is only meaningful for gfx8.
+    if (m_settings.isVolcanicIslands == FALSE)
+    {
+        flags.tcCompatible = FALSE;
+    }
+
     if (IsTileInfoAllZero(pTileInfo))
     {
         // See table entries 0-4
@@ -1268,14 +1318,16 @@ VOID CiLib::HwlSetupTileInfo(
             if (m_rowSize < tileSize)
             {
                 flags.tcCompatible = FALSE;
-                pOut->tcCompatible = FALSE;
             }
 
-            if (flags.depth && (flags.nonSplit || flags.tcCompatible || flags.needEquation))
+            if (flags.nonSplit | flags.tcCompatible | flags.needEquation)
             {
                 // Texture readable depth surface should not be split
                 switch (tileSize)
                 {
+                    case 64:
+                        index = 0;
+                        break;
                     case 128:
                         index = 1;
                         break;
@@ -1451,7 +1503,7 @@ VOID CiLib::HwlSetupTileInfo(
 
                     ADDR_ASSERT(macroTileBytes == PrtTileBytes);
 
-                    pOut->tcCompatible = FALSE;
+                    flags.tcCompatible = FALSE;
                     pOut->dccUnsupport = TRUE;
                 }
             }
@@ -1475,7 +1527,6 @@ VOID CiLib::HwlSetupTileInfo(
             if (m_rowSize < tileSize)
             {
                 flags.tcCompatible = FALSE;
-                pOut->tcCompatible = FALSE;
             }
         }
 
@@ -1510,7 +1561,7 @@ VOID CiLib::HwlSetupTileInfo(
         *pTileInfo = m_tileTable[8].info;
     }
 
-    if (pOut->tcCompatible)
+    if (flags.tcCompatible)
     {
         if (IsMacroTiled(tileMode))
         {
@@ -1535,7 +1586,7 @@ VOID CiLib::HwlSetupTileInfo(
 
                     if (m_rowSize < colorTileSplit)
                     {
-                        pOut->tcCompatible = FALSE;
+                        flags.tcCompatible = FALSE;
                     }
                 }
             }
@@ -1543,9 +1594,11 @@ VOID CiLib::HwlSetupTileInfo(
         else
         {
             // Client should not enable tc compatible for linear and 1D tile modes.
-            pOut->tcCompatible = FALSE;
+            flags.tcCompatible = FALSE;
         }
     }
+
+    pOut->tcCompatible = flags.tcCompatible;
 }
 
 /**
@@ -2177,5 +2230,64 @@ ADDR_E_RETURNCODE CiLib::HwlGetMaxAlignments(
     return ADDR_OK;
 }
 
+/**
+****************************************************************************************************
+*   CiLib::DepthStencilTileCfgMatch
+*
+*   @brief
+*       Try to find a tile index for stencil which makes its tile config parameters matches to depth
+*   @return
+*       TRUE if such tile index for stencil can be found
+****************************************************************************************************
+*/
+BOOL_32 CiLib::DepthStencilTileCfgMatch(
+    const ADDR_COMPUTE_SURFACE_INFO_INPUT*  pIn,    ///< [in] input structure
+    ADDR_COMPUTE_SURFACE_INFO_OUTPUT*       pOut    ///< [out] output structure
+    ) const
+{
+    BOOL_32 depthStencil2DTileConfigMatch = FALSE;
+
+    for (INT_32 stencilTileIndex = MinDepth2DThinIndex;
+         stencilTileIndex <= MaxDepth2DThinIndex;
+         stencilTileIndex++)
+    {
+        ADDR_TILEINFO tileInfo = {0};
+        INT_32 stencilMacroIndex = HwlComputeMacroModeIndex(stencilTileIndex,
+                                                            pIn->flags,
+                                                            8,
+                                                            pIn->numSamples,
+                                                            &tileInfo);
+
+        if (stencilMacroIndex != TileIndexNoMacroIndex)
+        {
+            if ((m_macroTileTable[stencilMacroIndex].banks ==
+                 m_macroTileTable[pOut->macroModeIndex].banks) &&
+                (m_macroTileTable[stencilMacroIndex].bankWidth ==
+                 m_macroTileTable[pOut->macroModeIndex].bankWidth) &&
+                (m_macroTileTable[stencilMacroIndex].bankHeight ==
+                 m_macroTileTable[pOut->macroModeIndex].bankHeight) &&
+                (m_macroTileTable[stencilMacroIndex].macroAspectRatio ==
+                 m_macroTileTable[pOut->macroModeIndex].macroAspectRatio) &&
+                (m_macroTileTable[stencilMacroIndex].pipeConfig ==
+                 m_macroTileTable[pOut->macroModeIndex].pipeConfig))
+            {
+                if ((pOut->tcCompatible == FALSE) ||
+                    (tileInfo.tileSplitBytes >= MicroTileWidth * MicroTileHeight * pIn->numSamples))
+                {
+                    depthStencil2DTileConfigMatch = TRUE;
+                    pOut->stencilTileIdx = stencilTileIndex;
+                    break;
+                }
+            }
+        }
+        else
+        {
+            ADDR_ASSERT_ALWAYS();
+        }
+    }
+
+    return depthStencil2DTileConfigMatch;
+}
+
 } // V1
 } // Addr
diff --git a/src/amd/addrlib/r800/ciaddrlib.h b/src/amd/addrlib/r800/ciaddrlib.h
index 93b2853d066..48835b32822 100644
--- a/src/amd/addrlib/r800/ciaddrlib.h
+++ b/src/amd/addrlib/r800/ciaddrlib.h
@@ -204,8 +204,15 @@ private:
         UINT_32 numOfBanks,
         UINT_32 numOfSamplesPerSplit) const;
 
+    BOOL_32 DepthStencilTileCfgMatch(
+        const ADDR_COMPUTE_SURFACE_INFO_INPUT*  pIn,
+        ADDR_COMPUTE_SURFACE_INFO_OUTPUT*       pOut) const;
+
     static const UINT_32    MacroTileTableSize = 16;
     static const UINT_32    PrtMacroModeOffset = MacroTileTableSize / 2;
+    static const INT_32     MinDepth2DThinIndex = 0;
+    static const INT_32     MaxDepth2DThinIndex = 4;
+    static const INT_32     Depth1DThinIndex = 5;
 
     ADDR_TILEINFO           m_macroTileTable[MacroTileTableSize];
     UINT_32                 m_noOfMacroEntries;