diff options
| author | Torbjorn Granlund <tege@gmplib.org> | 2011-11-02 15:02:47 +0100 |
|---|---|---|
| committer | Torbjorn Granlund <tege@gmplib.org> | 2011-11-02 15:02:47 +0100 |
| commit | 40ec51d08fd8bd939ee342e2249d867c1f268e9d (patch) | |
| tree | 6d7ac80f00e1a1e5eac500e50b63728610ba2ba2 | |
| parent | 0aa2759a1365b246cb90c079a623675deb061aa1 (diff) | |
| download | gmp-40ec51d08fd8bd939ee342e2249d867c1f268e9d.tar.gz | |
Correct a c/l number.
| -rw-r--r-- | mpn/s390_64/aorrlsh1_n.asm | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/mpn/s390_64/aorrlsh1_n.asm b/mpn/s390_64/aorrlsh1_n.asm index e1904e992..3bd9b1134 100644 --- a/mpn/s390_64/aorrlsh1_n.asm +++ b/mpn/s390_64/aorrlsh1_n.asm @@ -20,16 +20,16 @@ dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/. include(`../config.m4') C cycles/limb -C z900 10 -C z990 5 +C z900 9 +C z990 4.75 C z9 ? C z10 ? C z196 ? C TODO C * Optimise for small n, avoid 'la' like in aors_n.asm. -C * Tune to reach 4 c/l. For addlsh1, we could let the main alcgr propagate -C carry to the lsh1 alcgr. But even for sublsh1_n 5 c/l cannot be optimal. +C * Tune to reach 3.5 c/l. For addlsh1, we could let the main alcgr propagate +C carry to the lsh1 alcgr. C * Compute RETVAL for sublsh1_n less stupidly. C INPUT PARAMETERS |