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 enhances vectorizer with support of loops
with unaligned stores. Both cases are handled:
a store which alignment is unknown in compile time
and a store with known unalignment.
The patch is base on loop transformation introduced
in vectorizer patch #2
(http://gcc.gnu.org/ml/gcc-patches/2004-09/msg01104.html,
hopefully to be reviewed!)
Tested on powerpc-apple-darwin7.0.0. All vectorizer test
pass. Testcases 26 and 28 are changed from
xfail to pass; additional two loops are vectorized in
testcase 31.
The diff includes also vectorizer patch #2, while
ChangeLog is only for this patch.
thanks,
olga
ChengeLog:
* tree-vectorizer.c (vect_get_loop_niters):
(vect_gen_niters_for_prolog_loop):
(vect_update_niters_after_peeling):
(vect_update_inits_of_dr):
(vect_update_inits_of_drs):
(vect_do_peeling_for_alignment): New functions.
(vect_transform_loop): Added unalignment support.
(vect_analyze_data_refs_alignment): Allowed one unaligned
store.
* tree-vectorizer.h (MAX_NUMBER_OF_UNALIGNED_DATA_REFS): New
define.
(do_peeling_for_alignment):
(unaligned_drs): New members of _loop_vec_info.
(LOOP_DO_PEELING_FOR_ALIGNMENT): New macro.
(See attached file: peeling_main1.txt)Attachment:
peeling_main1.txt
Description: Text document
| Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
|---|---|---|
| Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |