[PATCH 1/4] lit.site.cfg.in: Use %polybenchOpts for '-O3 -loop-simplify -indvars -polly-prepare'

3 views
Skip to first unread message

ether zhhb

unread,
Apr 4, 2011, 10:06:34 AM4/4/11
to polly-commits
---
test/lit.site.cfg.in | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/test/lit.site.cfg.in b/test/lit.site.cfg.in
index b45da3b..fad2592 100644
--- a/test/lit.site.cfg.in
+++ b/test/lit.site.cfg.in
@@ -20,6 +20,7 @@ except KeyError,e:
config.substitutions.append(('%loadPolly', '-load '
+ config.llvm_libs_dir + '/LLVMPolly.so '))
config.substitutions.append(('%defaultOpts', ' -basicaa -scev-aa '))
+config.substitutions.append(('%polybenchOpts', ' -O3 -loop-simplify
-indvars -polly-prepare '))

# Let the main config do the real work.
lit.load_config(config, "@POLLY_SOURCE_DIR@/test/lit.cfg")
--
1.7.3.3

ether zhhb

unread,
Apr 4, 2011, 10:16:55 AM4/4/11
to polly-commits
hi,

i am going to push this out, too.

best regards
ether

Tobias Grosser

unread,
Apr 4, 2011, 10:22:22 AM4/4/11
to polly-...@googlegroups.com
On 04/04/2011 10:06 AM, ether zhhb wrote:
> ---
> test/lit.site.cfg.in | 1 +
> 1 files changed, 1 insertions(+), 0 deletions(-)
>
> diff --git a/test/lit.site.cfg.in b/test/lit.site.cfg.in
> index b45da3b..fad2592 100644
> --- a/test/lit.site.cfg.in
> +++ b/test/lit.site.cfg.in
> @@ -20,6 +20,7 @@ except KeyError,e:
> config.substitutions.append(('%loadPolly', '-load '
> + config.llvm_libs_dir + '/LLVMPolly.so '))
> config.substitutions.append(('%defaultOpts', ' -basicaa -scev-aa '))
> +config.substitutions.append(('%polybenchOpts', ' -O3 -loop-simplify
> -indvars -polly-prepare '))

Why are you planning to use -O3 for this? I think it is better, if we do
not run any other optimizations before Polly, but just store
preoptimized LLVM-IR. Like this the test cases do only depend on the
behaviour of Polly and only test Polly itself. If we really want to test
the transformations and canonicalization passes in -O3 we should add the
required test cases to those passes itself.

Cheers
Tobi

Tobias Grosser

unread,
Apr 6, 2011, 4:25:52 AM4/6/11
to polly-...@googlegroups.com, ether zhhb

Hey ether,

can you give your opinion on this?

Cheers
Tobi

ether zhhb

unread,
Apr 6, 2011, 6:00:53 AM4/6/11
to Tobias Grosser, polly-...@googlegroups.com
hi tobi,


>>
>> Why are you planning to use -O3 for this? I think it is better, if we do

Just because i see some have -O3 and some have -mem2reg and unified them to O3.


>> not run any other optimizations before Polly, but just store
>> preoptimized LLVM-IR. Like this the test cases do only depend on the
>> behaviour of Polly and only test Polly itself. If we really want to test
>> the transformations and canonicalization passes in -O3 we should add the
>> required test cases to those passes itself.

Yes you are right, i am going to change it to mem2reg.


>
> Hey ether,
>
> can you give your opinion on this?

hum, the original mail not appear in my box ...
>
> Cheers
> Tobi
>

Tobias Grosser

unread,
Apr 6, 2011, 6:02:46 AM4/6/11
to ether zhhb, polly-...@googlegroups.com
On 04/06/2011 12:00 PM, ether zhhb wrote:
> hi tobi,
>
>
>>>
>>> Why are you planning to use -O3 for this? I think it is better, if we do
> Just because i see some have -O3 and some have -mem2reg and unified them to O3.
>>> not run any other optimizations before Polly, but just store
>>> preoptimized LLVM-IR. Like this the test cases do only depend on the
>>> behaviour of Polly and only test Polly itself. If we really want to test
>>> the transformations and canonicalization passes in -O3 we should add the
>>> required test cases to those passes itself.
> Yes you are right, i am going to change it to mem2reg.
At best we remove them completely and the LLVM-IR is already the
preoptimized version.

>> Hey ether,
>>
>> can you give your opinion on this?
> hum, the original mail not appear in my box ...

Sorry. I think I did not CC you directly.

Cheers
Tobi

ether zhhb

unread,
Apr 6, 2011, 6:21:43 AM4/6/11
to Tobias Grosser, polly-...@googlegroups.com
hi tobi,

> At best we remove them completely and the LLVM-IR is already the
> preoptimized version.

This sounds good, i will try to figure this out.

> I think I did not CC you directly.

I just fixed my setting about the list.

best regards
ether

ether zhhb

unread,
Apr 6, 2011, 12:37:54 PM4/6/11
to Tobias Grosser, polly-...@googlegroups.com
Done with changing the result of several testcases.

Tobias Grosser

unread,
Apr 7, 2011, 3:21:27 AM4/7/11
to polly-...@googlegroups.com
On 04/06/2011 06:37 PM, ether zhhb wrote:
> Done with changing the result of several testcases.

Nice.

Thanks
Tobi

Reply all
Reply to author
Forward
0 new messages