summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
diff options
context:
space:
mode:
authorJosé Fonseca <jfonseca@vmware.com>2009-08-22 22:26:55 +0100
committerJosé Fonseca <jfonseca@vmware.com>2009-08-29 09:21:40 +0100
commit5811ed87d732101ab8cfbd087bc99d8c6c963f30 (patch)
treed0167369e347bd22ccb6be7d668d5fd878593d5d /src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
parent3f36f4b0519f7a568d6de9919de1001880ab5c8a (diff)
llvmpipe: Add a bunch of comments.
Description/rationale/to-do items, while I still remember them...
Diffstat (limited to 'src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c')
-rw-r--r--src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c11
1 files changed, 10 insertions, 1 deletions
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
index 87ba456065..c11a9398f8 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
@@ -28,7 +28,16 @@
/**
* @file
- * Blend LLVM IR generation -- AOS form.
+ * Blend LLVM IR generation -- AoS layout.
+ *
+ * AoS blending is in general much slower than SoA, but there are some cases
+ * where it might be faster. In particular, if a pixel is rendered only once
+ * then the overhead of tiling and untiling will dominate over the speedup that
+ * SoA gives. So we might want to detect such cases and fallback to AoS in the
+ * future, but for now this function is here for historical/benchmarking
+ * purposes.
+ *
+ * Run lp_blend_test after any change to this file.
*
* @author Jose Fonseca <jfonseca@vmware.com>
*/