This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

PATCH: PR testsuite/32062: A MASK is non-efficient in certain sse4_1 tests


On Thu, May 24, 2007 at 09:15:29AM +0200, Uros Bizjak wrote:
> Hello!
> 
> There is a test "problem" with sse4_1-blendpd.c and sse4_1-blendps.c
> [1]. The MASK of 0x03 or 0x0f is non-efficient to test blendpd and
> blendps instruction. gcc will figure out that this insn can be
> substituted with simple move and won't generate expected insn.
> 
> The solution is to change MASK to 0x02 for blendpd and some
> appropriate value for blendps. Could you fix this issue?
> 
> [1]: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32062

I was compiling SSE4.1 testsuite with -DMASK=0x2 and did generate
SSE4.1 instructions for all SSE4.1 testcases.  I am checking in this
patch as an obvious fix. Tested on Linux/Intel64.


H.J.
---
2007-05-24  H.J. Lu  <hongjiu.lu@intel.com>

	PR testsuite/32062
	* gcc.target/i386/sse4_1-check.h (MASK): New.

--- gcc/testsuite/gcc.target/i386/sse4_1-check.h.mask	2007-05-23 09:42:16.000000000 -0700
+++ gcc/testsuite/gcc.target/i386/sse4_1-check.h	2007-05-24 07:09:19.000000000 -0700
@@ -4,6 +4,8 @@
 
 static void sse4_1_test (void);
 
+#define MASK 0x2
+
 int
 main ()
 {


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]