[llvm-commits] [llvm] r43444 - /llvm/trunk/lib/Target/X86/README-SSE.txt
Chris Lattner
sabre at nondot.org
Sun Oct 28 23:19:48 PDT 2007
Author: lattner
Date: Mon Oct 29 01:19:48 2007
New Revision: 43444
URL: http://llvm.org/viewvc/llvm-project?rev=43444&view=rev
Log:
add a note.
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=43444&r1=43443&r2=43444&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/README-SSE.txt (original)
+++ llvm/trunk/lib/Target/X86/README-SSE.txt Mon Oct 29 01:19:48 2007
@@ -708,3 +708,36 @@
//===---------------------------------------------------------------------===//
+These functions:
+
+#include <xmmintrin.h>
+__m128i a;
+void x(unsigned short n) {
+ a = _mm_slli_epi32 (a, n);
+}
+void y(unsigned n) {
+ a = _mm_slli_epi32 (a, n);
+}
+
+compile to ( -O3 -static -fomit-frame-pointer):
+_x:
+ movzwl 4(%esp), %eax
+ movd %eax, %xmm0
+ movaps _a, %xmm1
+ pslld %xmm0, %xmm1
+ movaps %xmm1, _a
+ ret
+_y:
+ movd 4(%esp), %xmm0
+ movaps _a, %xmm1
+ pslld %xmm0, %xmm1
+ movaps %xmm1, _a
+ ret
+
+"y" looks good, but "x" does silly movzwl stuff around into a GPR. It seems
+like movd would be sufficient in both cases as the value is already zero
+extended in the 32-bit stack slot IIRC. For signed short, it should also be
+save, as a really-signed value would be undefined for pslld.
+
+
+//===---------------------------------------------------------------------===//
More information about the llvm-commits
mailing list