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] |
Hi, This patch models the vector float-to-int conversion for altivec (int-to-float already existed) in the rs6000 implementation of the target builtin vectorize_builtin_conversion. A minor change for the definition of the builtin was required: builtin_conversion (which is used for both int-to-float and float-to-int) used to take as argument the rhs of the conversion. Instead, it now takes as argument the type of the int side of the conversion, because we need to know whether it's a conversion to/from signed or unsigned. Testcases for this already exist in the testsuite ( vect-floatint-conversion-1.c, vect-floatint-conversion-2.c). Passed bootstrap w vectorization, and the vectorizer testsuite with no regressions, on powerpc970, Passed bootstrap, and full regression testing with no regressions, on i386. I need approval on the rs6000.c changes (the i386.c change is obvious). thanks, dorit 2008-08-06 Dorit Nuzman <dorit@il.ibm.com> * tree-vect-transform.c (vectorizable_conversion): Pass the integral type to vectorize.builtin_conversion. (vectorizable_conversion): Likewise.. * config/i386/i386.c (ix86_vectorize_builtin_conversion): Always takes integral type as input. * config/rs6000/rs6000.c (rs6000_builtin_conversion): Add case for FIX_TRUNC_EXPR. 2008-08-06 Dorit Nuzman <dorit@il.ibm.com> * lib/target-supports.exp (check_effective_target_vect_floatint_cvt): Add powerpc. (See attached file: fp2int.txt)
Attachment:
fp2int.txt
Description: Text document
Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
---|---|---|
Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |