[Mesa-dev] [PATCH v2 2/14] swr: [rasterizer memory] minify original sizes for block formats

Ilia Mirkin imirkin at alum.mit.edu
Tue Nov 15 19:40:28 UTC 2016


There's no guarantee that mip width/height will be a multiple of the
compressed block size. Doing a divide by the block size first yields
different results than GL expects, so we do the divide at the end.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---

This replaces the previous patch, "swr: [rasterizer memory] round up when
dividing by block sizes". My original comment was that I didn't need it, but
I do need this one since I'm back to using the tiling functions for determining
all the offset layouts. And since pretending it's really a RG32 format doesn't
work, I have to feed a compressed format to the swr backend.

 .../swr/rasterizer/memory/TilingFunctions.h        | 36 +++++++++++++++-------
 1 file changed, 25 insertions(+), 11 deletions(-)

diff --git a/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h b/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h
index 0694a99..11ed451 100644
--- a/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h
+++ b/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h
@@ -274,9 +274,12 @@ INLINE void ComputeLODOffset1D(
     else
     {
         uint32_t curWidth = baseWidth;
-        // translate mip width from pixels to blocks for block compressed formats
-        // @note hAlign is already in blocks for compressed formats so no need to convert
-        if (info.isBC) curWidth /= info.bcWidth;
+        // @note hAlign is already in blocks for compressed formats so upconvert
+        //       so that we have the desired alignment post-divide.
+        if (info.isBC)
+        {
+            hAlign *= info.bcWidth;
+        }
 
         offset = GFX_ALIGN(curWidth, hAlign);
         for (uint32_t l = 1; l < lod; ++l)
@@ -285,7 +288,7 @@ INLINE void ComputeLODOffset1D(
             offset += curWidth;
         }
 
-        if (info.isSubsampled)
+        if (info.isSubsampled || info.isBC)
         {
             offset /= info.bcWidth;
         }
@@ -312,14 +315,17 @@ INLINE void ComputeLODOffsetX(
     else
     {
         uint32_t curWidth = baseWidth;
-        // convert mip width from pixels to blocks for block compressed formats
-        // @note hAlign is already in blocks for compressed formats so no need to convert
-        if (info.isBC) curWidth /= info.bcWidth;
+        // @note hAlign is already in blocks for compressed formats so upconvert
+        //       so that we have the desired alignment post-divide.
+        if (info.isBC)
+        {
+            hAlign *= info.bcWidth;
+        }
 
         curWidth = std::max<uint32_t>(curWidth >> 1, 1U);
         curWidth = GFX_ALIGN(curWidth, hAlign);
 
-        if (info.isSubsampled)
+        if (info.isSubsampled || info.isBC)
         {
             curWidth /= info.bcWidth;
         }
@@ -350,9 +356,12 @@ INLINE void ComputeLODOffsetY(
         offset = 0;
         uint32_t mipHeight = baseHeight;
 
-        // translate mip height from pixels to blocks for block compressed formats
-        // @note VAlign is already in blocks for compressed formats so no need to convert
-        if (info.isBC) mipHeight /= info.bcHeight;
+        // @note vAlign is already in blocks for compressed formats so upconvert
+        //       so that we have the desired alignment post-divide.
+        if (info.isBC)
+        {
+            vAlign *= info.bcHeight;
+        }
 
         for (uint32_t l = 1; l <= lod; ++l)
         {
@@ -360,6 +369,11 @@ INLINE void ComputeLODOffsetY(
             offset += ((l != 2) ? alignedMipHeight : 0);
             mipHeight = std::max<uint32_t>(mipHeight >> 1, 1U);
         }
+
+        if (info.isBC)
+        {
+            offset /= info.bcHeight;
+        }
     }
 }
 
-- 
2.7.3



More information about the mesa-dev mailing list