This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[gccgo] Don't force SSE2
- From: Ian Lance Taylor <iant at google dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Tue, 24 Nov 2009 09:12:17 -0800
- Subject: [gccgo] Don't force SSE2
I used a horrible hack in the Go compiler to force the use of SSE2, in
order to get stable floating point results on 32-bit x86.
Fortunately, thanks to Joseph's work on -fexcess-precision, that hack
is no longer necessary. I committed this patch to remove it.
Ian
Index: gospec.c
===================================================================
--- gospec.c (revision 154387)
+++ gospec.c (working copy)
@@ -222,19 +222,6 @@ lang_specific_driver (int *in_argc, cons
/* Copy the 0th argument, i.e., the name of the program itself. */
arglist[i++] = argv[j++];
-#ifdef TARGET_SSE2
- /* This is a major hack. On the x86, we want to use SSE floating
- point, because it gives more consistent results. So we insert
- arguments to select it. Doing this means that anybody who wants
- to run on a Pentium 3 or earlier will need to use explicit -march
- and -mfpmath options. */
- arglist[j++] = "-mfpmath=sse";
- arglist[j++] = "-msse2";
-#if !defined(TARGET_64BIT_DEFAULT) || !TARGET_64BIT_DEFAULT
- arglist[j++] = "-march=pentium4";
-#endif
-#endif /* defined(TARGET_SSE2) */
-
/* We always combine all input files. */
arglist[j++] = "-combine";