Hi Maintainers,
Please fine below patch that sets preferred vectorization mode for btver2 as
avx128.
We tested some benchmarks and found that enabling 128-bit AVX instruction
generation by the auto vectorizer performs better than
default 256-bit AVX.
Completed bootstrap and make -k check passes.
Ok for trunk?
Gcc/ChangeLog
-------------------
2013-01-09 Venkataramanan Kumar <venkataramanan.ku...@amd.com>
* config/i386/i386.c (initial_ix86_tune_features): Set up
X86_TUNE_AVX128_OPTIMAL for m_BTVER2.
Index: i386.c
===================================================================
--- i386.c (revision 194833)
+++ i386.c (working copy)
@@ -2014,7 +2014,7 @@
/* X86_TUNE_AVX128_OPTIMAL: Enable 128-bit AVX instruction generation for
the auto-vectorizer. */
- m_BDVER,
+ m_BDVER | m_BTVER2,
/* X86_TUNE_REASSOC_INT_TO_PARALLEL: Try to produce parallel computations
during reassociation of integer computation. */
Regards,
Venkat.
