[llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
25 messages Options
12
Reply | Threaded
Open this post in threaded view
|

[llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

Any comments/concerns?

Thanks,
Dehao

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Besides speccpu, any other real-world applications benefit from this option?

Regards,
chenwj

2017-05-19 6:30 GMT+08:00 Dehao Chen via llvm-dev <[hidden email]>:
Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

Any comments/concerns?

Thanks,
Dehao

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev




--
Wei-Ren Chen (陳韋任)
Homepage: https://people.cs.nctu.edu.tw/~chenwj

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev
This sounds good to me. Enabling this by default has been mentioned a few times already. I've tested this feature in the past on AArch64 (Kryo and Falkor) and found it to be beneficial for mixed-type loops. Thanks!

On Thu, May 18, 2017 at 6:30 PM, Dehao Chen via llvm-dev <[hidden email]> wrote:
Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

Any comments/concerns?

Thanks,
Dehao

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev



_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev
Yes, we do see performance benefits for this change on some google internal benchmarks.

Dehao

On Fri, May 19, 2017 at 6:18 AM, 陳韋任 <[hidden email]> wrote:
Besides speccpu, any other real-world applications benefit from this option?

Regards,
chenwj

2017-05-19 6:30 GMT+08:00 Dehao Chen via llvm-dev <[hidden email]>:
Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

Any comments/concerns?

Thanks,
Dehao

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev




--
Wei-Ren Chen (陳韋任)
Homepage: https://people.cs.nctu.edu.tw/~chenwj


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev

On May 18, 2017, at 3:30 PM, Dehao Chen via llvm-dev <[hidden email]> wrote:

Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

Can you please describe the config for the runs (optimization level, PGO/no-PGO, etc).

It would be good to provide analysis for the changes >1%. I.e. we need to make sure that the improvements are not noise either ;).


I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

I will run it on Cyclone/AArch64 next week.

Adam


Any comments/concerns?

Thanks,
Dehao
_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev


On Fri, May 19, 2017 at 4:01 PM, Adam Nemet <[hidden email]> wrote:

On May 18, 2017, at 3:30 PM, Dehao Chen via llvm-dev <[hidden email]> wrote:

Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

Can you please describe the config for the runs (optimization level, PGO/no-PGO, etc).

This is O2 build without PGO.
 

It would be good to provide analysis for the changes >1%. I.e. we need to make sure that the improvements are not noise either ;).

Good point. I just examined all benchmarks with >1% "improvement". Turns out they are all noises: the hot functions (with >1% total cycles) are all identical. So the conclusion is: this change does not affect speccpu2006 performance.

Thanks,
Dehao
 


I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

I will run it on Cyclone/AArch64 next week.

Adam


Any comments/concerns?

Thanks,
Dehao
_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev



_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Thanks all for the comment. Any other comments on how we should proceed with this?

Thanks,
Dehao

On Mon, May 22, 2017 at 9:57 AM, Dehao Chen <[hidden email]> wrote:


On Fri, May 19, 2017 at 4:01 PM, Adam Nemet <[hidden email]> wrote:

On May 18, 2017, at 3:30 PM, Dehao Chen via llvm-dev <[hidden email]> wrote:

Hi,

I'm proposing to make vectorizer-maximize-bandwidth on by default for loop vectorizer because it should generally help performance.

I've tested the performance impact on Intel sandybridge machine with speccpu benchmarks:

           Benchmark             Base:Reference   (1)  
-------------------------------------------------------
spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

  Scores are benchmark specific.

We do have regression on 453.povray, but it's due to secondary effects as all hot functions are the same. I've also tested the code size impact, it does not change for tested speccpu benchmarks.

Can you please describe the config for the runs (optimization level, PGO/no-PGO, etc).

This is O2 build without PGO.
 

It would be good to provide analysis for the changes >1%. I.e. we need to make sure that the improvements are not noise either ;).

Good point. I just examined all benchmarks with >1% "improvement". Turns out they are all noises: the hot functions (with >1% total cycles) are all identical. So the conclusion is: this change does not affect speccpu2006 performance.

Thanks,
Dehao
 


I've prepared https://reviews.llvm.org/D33341 to do this.

I really appreciate if the community can help test the performance impact of this change on other architectures so that we can decide if this should go target-dependent.

I will run it on Cyclone/AArch64 next week.

Adam


Any comments/concerns?

Thanks,
Dehao
_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev




_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev
On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:
I will run it on Cyclone/AArch64 next week.

FYI, we're still waiting on these Adam...

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Thank you for running these.

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.
I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

Kevin

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:
On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:
I will run it on Cyclone/AArch64 next week.

FYI, we're still waiting on these Adam...

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev



_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
If you care about such hardware, please run benchmarks with the flag?

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:
Thank you for running these.

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.
I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

Kevin

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:
On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:
I will run it on Cyclone/AArch64 next week.

FYI, we're still waiting on these Adam...

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev

We’re seeing nice improvements but also significant degradations on IA, which we would like to investigate before the patch is committed.

 

Major degradations we see:

 

networking

   ip_pktcheckb1m          -6.80 %

   ip_pktcheckb2m          -6.74 %

   ip_pktcheckb4m          -7.57 % 

   ip_pktcheckb512k       -6.58 %

Telecom

   autcor00data_1          -78.02 %

   autcor00data_2          -76.80 %

   autcor00data_3          -77.00 %

(on Atom)

 

We still working on creating reproducers.

In general we support this patch, just want to have a chance to investigate the issues. We need a few days for that.

 

 

 

From: llvm-dev [[hidden email]] On Behalf Of Chandler Carruth via llvm-dev
Sent: Tuesday, May 30, 2017 11:27
To: Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

If you care about such hardware, please run benchmarks with the flag?

 

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

 

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:

Thank you for running these.

 

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.

I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

 

Kevin

 

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam...

 

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev

BTW, we also tested on AVX2, and saw a few smaller degradations there:

 

denbench

   cjpegv2data1   -5.02 %

   cjpegv2data2   -4.29 %

   cjpegv2data3   -4.89 %

   cjpegv2data4   -5.44 %

   cjpegv2data5   -4.80 %

   cjpegv2data6   -3.98 %

   cjpegv2data7   -5.48 %

coremark-pro

   cjpeg-rose7-preset   -4.91 %

   core                 -6.04 %

telecom

   autcor00data_1       -2.00 %

 

 

From: Agabaria, Mohammed
Sent: Tuesday, May 30, 2017 11:40
To: 'Flamedoge' <[hidden email]>; 'Dehao Chen' <[hidden email]>
Cc: 'llvm-dev' <[hidden email]>
Subject: RE: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

We’re seeing nice improvements but also significant degradations on IA, which we would like to investigate before the patch is committed.

 

Major degradations we see:

 

networking

   ip_pktcheckb1m          -6.80 %

   ip_pktcheckb2m          -6.74 %

   ip_pktcheckb4m          -7.57 % 

   ip_pktcheckb512k       -6.58 %

Telecom

   autcor00data_1          -78.02 %

   autcor00data_2          -76.80 %

   autcor00data_3          -77.00 %

(on Atom)

 

We still working on creating reproducers.

In general we support this patch, just want to have a chance to investigate the issues. We need a few days for that.

 

 

 

From: llvm-dev [[hidden email]] On Behalf Of Chandler Carruth via llvm-dev
Sent: Tuesday, May 30, 2017 11:27
To: Flamedoge <
[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <
[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

If you care about such hardware, please run benchmarks with the flag?

 

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

 

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:

Thank you for running these.

 

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.

I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

 

Kevin

 

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam...

 

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev

On May 30, 2017, at 12:58 AM, Chandler Carruth <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:
I will run it on Cyclone/AArch64 next week.

FYI, we're still waiting on these Adam…

I was going to test spec but this does not seem to trigger on spec according to Dehao, so there is really no reason for me to test this.  We have some spec perf bots that test trunk; if there is some unexpected regression we should pick it up.  Sorry for not being explicit about this.

Adam


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
On Tue, May 30, 2017 at 9:17 AM Adam Nemet <[hidden email]> wrote:
On May 30, 2017, at 12:58 AM, Chandler Carruth <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:
I will run it on Cyclone/AArch64 next week.

FYI, we're still waiting on these Adam…

I was going to test spec but this does not seem to trigger on spec according to Dehao, so there is really no reason for me to test this.  We have some spec perf bots that test trunk; if there is some unexpected regression we should pick it up.  Sorry for not being explicit about this.

Ok, if you're fine with that, cool.

Only reason I asked was because I wasn't sure that it fundamentally didn't fire on SPEC or just didn't with the particular cost model and might fire/behave differently with a different cost model.

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev
On Tue, May 30, 2017 at 1:40 AM Agabaria, Mohammed via llvm-dev <[hidden email]> wrote:

We’re seeing nice improvements but also significant degradations on IA, which we would like to investigate before the patch is committed.

 

Major degradations we see:

 

networking

   ip_pktcheckb1m          -6.80 %

   ip_pktcheckb2m          -6.74 %

   ip_pktcheckb4m          -7.57 % 

   ip_pktcheckb512k       -6.58 %

Telecom

   autcor00data_1          -78.02 %

   autcor00data_2          -76.80 %

   autcor00data_3          -77.00 %

(on Atom)

 

We still working on creating reproducers.

In general we support this patch, just want to have a chance to investigate the issues. We need a few days for that.


I mean, OK... but keep in mind that Dehao's original email went out over a week ago, so this patch has already been held up a while. As these benchmarks aren't readily available, we also can't do anything to help until a test case is posted.

Have you considered contributing these benchmarks to the LLVM test suite?
 

 

 

 

From: llvm-dev [[hidden email]] On Behalf Of Chandler Carruth via llvm-dev
Sent: Tuesday, May 30, 2017 11:27
To: Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

If you care about such hardware, please run benchmarks with the flag?

 

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

 

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:

Thank you for running these.

 

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.

I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

 

Kevin

 

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam...

 

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev

Hi,

 

We enabled “vectorizer-maximize-bandwidth” and ran SPEC CPU2006 (base,rate) on Ryzen 8 core, 16 copies with below config:

 

Base: -m64 -O3 -march=znver1 -mavx2

Base + VMB: -m64 -O3 -march=znver1 -mavx2  -mllvm -vectorizer-maximize-bandwidth

 

There’s a small uplift for gcc and some small regression for sjeng. Others are within noise levels.

 

CPU2006 Results:

Benchmark

Base/(Base + VMB)

400.perlbench

1

401.bzip2    

1

403.gcc      

1.01517

429.mcf      

1.00222

445.gobmk    

1

456.hmmer     

1

458.sjeng    

0.98641

462.libquantum

1

464.h264ref  

1.01005

471.omnetpp  

1.00187

473.astar    

1

483.xalancbmk

1.00149

433.milc    

1

444.namd    

1

447.dealII  

1

450.soplex  

1

453.povray  

1.00515

470.lbm     

1

482.sphinx3

1

 

*  Ratio more than 1 indicates vectorizer-maximize-bandwidth is improving the performance

*  Ratio less than 1 indicates vectorizer-maximize-bandwidth is regressing the performance.

*  Ratio 1 indicates no change.

 

Regards,

Ashutosh

 

From: llvm-dev [mailto:[hidden email]] On Behalf Of Adam Nemet via llvm-dev
Sent: Tuesday, May 30, 2017 9:47 PM
To: Chandler Carruth <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

 

On May 30, 2017, at 12:58 AM, Chandler Carruth <[hidden email]> wrote:

 

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam…

 

I was going to test spec but this does not seem to trigger on spec according to Dehao, so there is really no reason for me to test this.  We have some spec perf bots that test trunk; if there is some unexpected regression we should pick it up.  Sorry for not being explicit about this.

 

Adam

 


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Does the regression seem acceptable to you? Have you done any analysis of what changed and why it regresses?

On Thu, Jun 1, 2017 at 4:47 AM Nema, Ashutosh <[hidden email]> wrote:

Hi,

 

We enabled “vectorizer-maximize-bandwidth” and ran SPEC CPU2006 (base,rate) on Ryzen 8 core, 16 copies with below config:

 

Base: -m64 -O3 -march=znver1 -mavx2

Base + VMB: -m64 -O3 -march=znver1 -mavx2  -mllvm -vectorizer-maximize-bandwidth

 

There’s a small uplift for gcc and some small regression for sjeng. Others are within noise levels.

 

CPU2006 Results:

Benchmark

Base/(Base + VMB)

400.perlbench

1

401.bzip2    

1

403.gcc      

1.01517

429.mcf      

1.00222

445.gobmk    

1

456.hmmer     

1

458.sjeng    

0.98641

462.libquantum

1

464.h264ref  

1.01005

471.omnetpp  

1.00187

473.astar    

1

483.xalancbmk

1.00149

433.milc    

1

444.namd    

1

447.dealII  

1

450.soplex  

1

453.povray  

1.00515

470.lbm     

1

482.sphinx3

1

 

*  Ratio more than 1 indicates vectorizer-maximize-bandwidth is improving the performance

*  Ratio less than 1 indicates vectorizer-maximize-bandwidth is regressing the performance.

*  Ratio 1 indicates no change.

 

Regards,

Ashutosh

 

From: llvm-dev [mailto:[hidden email]] On Behalf Of Adam Nemet via llvm-dev
Sent: Tuesday, May 30, 2017 9:47 PM
To: Chandler Carruth <[hidden email]>


Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

 

On May 30, 2017, at 12:58 AM, Chandler Carruth <[hidden email]> wrote:

 

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam…

I was going to test spec but this does not seem to trigger on spec according to Dehao, so there is really no reason for me to test this.  We have some spec perf bots that test trunk; if there is some unexpected regression we should pick it up.  Sorry for not being explicit about this.

 

Adam

 


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev

Hi Chandler,

 

We haven’t analyze the failure yet, need some time to investigate.

 

In general this patch looks OK.

 

Regards,

Ashutosh

 

From: Chandler Carruth [mailto:[hidden email]]
Sent: Thursday, June 1, 2017 5:27 PM
To: Nema, Ashutosh <[hidden email]>; Dehao Chen <[hidden email]>; Adam Nemet <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

Does the regression seem acceptable to you? Have you done any analysis of what changed and why it regresses?

 

On Thu, Jun 1, 2017 at 4:47 AM Nema, Ashutosh <[hidden email]> wrote:

Hi,

 

We enabled “vectorizer-maximize-bandwidth” and ran SPEC CPU2006 (base,rate) on Ryzen 8 core, 16 copies with below config:

 

Base: -m64 -O3 -march=znver1 -mavx2

Base + VMB: -m64 -O3 -march=znver1 -mavx2  -mllvm -vectorizer-maximize-bandwidth

 

There’s a small uplift for gcc and some small regression for sjeng. Others are within noise levels.

 

CPU2006 Results:

Benchmark

Base/(Base + VMB)

400.perlbench

1

401.bzip2    

1

403.gcc      

1.01517

429.mcf      

1.00222

445.gobmk    

1

456.hmmer     

1

458.sjeng    

0.98641

462.libquantum

1

464.h264ref  

1.01005

471.omnetpp  

1.00187

473.astar    

1

483.xalancbmk

1.00149

433.milc    

1

444.namd    

1

447.dealII  

1

450.soplex  

1

453.povray  

1.00515

470.lbm     

1

482.sphinx3

1

 

*  Ratio more than 1 indicates vectorizer-maximize-bandwidth is improving the performance

*  Ratio less than 1 indicates vectorizer-maximize-bandwidth is regressing the performance.

*  Ratio 1 indicates no change.

 

Regards,

Ashutosh

 

From: llvm-dev [mailto:[hidden email]] On Behalf Of Adam Nemet via llvm-dev
Sent: Tuesday, May 30, 2017 9:47 PM
To: Chandler Carruth <[hidden email]>


Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

 

On May 30, 2017, at 12:58 AM, Chandler Carruth <[hidden email]> wrote:

 

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam…

I was going to test spec but this does not seem to trigger on spec according to Dehao, so there is really no reason for me to test this.  We have some spec perf bots that test trunk; if there is some unexpected regression we should pick it up.  Sorry for not being explicit about this.

 

Adam

 


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
In reply to this post by Gerolf Hoflehner via llvm-dev

Attached the first reproducer of the Atom\SLM arch. degradation.  (~70% degradation).

Chandler, those are part of EEMBC benchmarks.

 

From: Chandler Carruth [mailto:[hidden email]]
Sent: Tuesday, May 30, 2017 23:07
To: Agabaria, Mohammed <[hidden email]>; Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

On Tue, May 30, 2017 at 1:40 AM Agabaria, Mohammed via llvm-dev <[hidden email]> wrote:

We’re seeing nice improvements but also significant degradations on IA, which we would like to investigate before the patch is committed.

 

Major degradations we see:

 

networking

   ip_pktcheckb1m          -6.80 %

   ip_pktcheckb2m          -6.74 %

   ip_pktcheckb4m          -7.57 % 

   ip_pktcheckb512k       -6.58 %

Telecom

   autcor00data_1          -78.02 %

   autcor00data_2          -76.80 %

   autcor00data_3          -77.00 %

(on Atom)

 

We still working on creating reproducers.

In general we support this patch, just want to have a chance to investigate the issues. We need a few days for that.

 

I mean, OK... but keep in mind that Dehao's original email went out over a week ago, so this patch has already been held up a while. As these benchmarks aren't readily available, we also can't do anything to help until a test case is posted.

 

Have you considered contributing these benchmarks to the LLVM test suite?

 

 

 

 

From: llvm-dev [[hidden email]] On Behalf Of Chandler Carruth via llvm-dev
Sent: Tuesday, May 30, 2017 11:27
To: Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

If you care about such hardware, please run benchmarks with the flag?

 

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

 

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:

Thank you for running these.

 

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.

I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

 

Kevin

 

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam...

 

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

slm-no-vectorize.ll (2K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

Gerolf Hoflehner via llvm-dev
Thanks for the testcase. Could you add some more details about the regression?

* How to build/run the testcase to reproduce the degradation?
* Is the degradation only exist in ATOM/SLM arch? If yes, could you help analyze why maximize the bandwidth would make performance worse (as we do not have these arches to run the perf tests).

Thanks,
Dehao

On Thu, Jun 1, 2017 at 6:12 AM, Agabaria, Mohammed <[hidden email]> wrote:

Attached the first reproducer of the Atom\SLM arch. degradation.  (~70% degradation).

Chandler, those are part of EEMBC benchmarks.

 

From: Chandler Carruth [mailto:[hidden email]]
Sent: Tuesday, May 30, 2017 23:07
To: Agabaria, Mohammed <[hidden email]>; Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>


Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

On Tue, May 30, 2017 at 1:40 AM Agabaria, Mohammed via llvm-dev <[hidden email]> wrote:

We’re seeing nice improvements but also significant degradations on IA, which we would like to investigate before the patch is committed.

 

Major degradations we see:

 

networking

   ip_pktcheckb1m          -6.80 %

   ip_pktcheckb2m          -6.74 %

   ip_pktcheckb4m          -7.57 % 

   ip_pktcheckb512k       -6.58 %

Telecom

   autcor00data_1          -78.02 %

   autcor00data_2          -76.80 %

   autcor00data_3          -77.00 %

(on Atom)

 

We still working on creating reproducers.

In general we support this patch, just want to have a chance to investigate the issues. We need a few days for that.

 

I mean, OK... but keep in mind that Dehao's original email went out over a week ago, so this patch has already been held up a while. As these benchmarks aren't readily available, we also can't do anything to help until a test case is posted.

 

Have you considered contributing these benchmarks to the LLVM test suite?

 

 

 

 

From: llvm-dev [[hidden email]] On Behalf Of Chandler Carruth via llvm-dev
Sent: Tuesday, May 30, 2017 11:27
To: Flamedoge <[hidden email]>; Dehao Chen <[hidden email]>
Cc: llvm-dev <[hidden email]>
Subject: Re: [llvm-dev] Enable vectorizer-maximize-bandwidth by default?

 

If you care about such hardware, please run benchmarks with the flag?

 

Dehao has made this flag available. It is important that those who care about particular hardware provid ebenchmark results. Not everyone in the community will have access to particular hardware variants.

 

On Tue, May 30, 2017 at 1:14 AM Flamedoge <[hidden email]> wrote:

Thank you for running these.

 

May I suggest testing on AVX2 capable hardware? That would be Intel Haswell, AMD Carrizo and up.

I'm not sure what "vectorizer-maximize-bandwidth" implies, but doubling the vector lanes may help light up parallel regions.

 

Kevin

 

On Tue, May 30, 2017 at 12:58 AM, Chandler Carruth via llvm-dev <[hidden email]> wrote:

On Fri, May 19, 2017 at 4:01 PM Adam Nemet via llvm-dev <[hidden email]> wrote:

I will run it on Cyclone/AArch64 next week.

 

FYI, we're still waiting on these Adam...

 

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.



_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
12