Add a note about SSE4.1 roundss/roundsd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@125438 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/README-SSE.txt b/lib/Target/X86/README-SSE.txt
index b2116e03b148ea222532c956e9dd78f49fea3a06..f16ec029e96ab3da31996d9b3d13e6a5650c5e4e 100644 (file)
--- a/lib/Target/X86/README-SSE.txt
+++ b/lib/Target/X86/README-SSE.txt
@@ -923,4 +923,15 @@ The insertps's of $0 are pointless complex copies.
 
 //===---------------------------------------------------------------------===//
 
+If SSE4.1 is available we should inline rounding functions instead of emitting
+a libcall.
 
+floor: roundsd $0x01, %xmm, %xmm
+ceil:  roundsd $0x02, %xmm, %xmm
+
+and likewise for the single precision versions.
+
+Currently, SelectionDAGBuilder doesn't turn calls to these functions into the
+corresponding nodes and some targets (including X86) aren't ready for them.
+
+//===---------------------------------------------------------------------===//