[llvm-commits] [llvm] r51491 - /llvm/trunk/lib/Target/X86/README-SSE.txt
Dan Gohman
gohman at apple.com
Fri May 23 11:05:39 PDT 2008
Author: djg
Date: Fri May 23 13:05:39 2008
New Revision: 51491
URL: http://llvm.org/viewvc/llvm-project?rev=51491&view=rev
Log:
Elaborate on the entry on integer vector multiplication by constants.
Modified:
llvm/trunk/lib/Target/X86/README-SSE.txt
Modified: llvm/trunk/lib/Target/X86/README-SSE.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/README-SSE.txt?rev=51491&r1=51490&r2=51491&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/README-SSE.txt (original)
+++ llvm/trunk/lib/Target/X86/README-SSE.txt Fri May 23 13:05:39 2008
@@ -814,7 +814,7 @@
ret <4 x i32> %A
}
-Compiles into:
+On targets without SSE4.1, this compiles into:
LCPI1_0: ## <4 x i32>
.long 10
@@ -846,6 +846,11 @@
punpckldq %xmm2, %xmm0
ret
+It would be better to synthesize integer vector multiplication by constants
+using shifts and adds, pslld and paddd here. And even on targets with SSE4.1,
+simple cases such as multiplication by powers of two would be better as
+vector shifts than as multiplications.
+
//===---------------------------------------------------------------------===//
We compile this:
More information about the llvm-commits
mailing list