September 23, 2016 Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Mir is LLVM-accelerated Generic Numerical Library for Science and Machine Learning. Benchmark: http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html Mir v0.18.0 release notes: https://github.com/libmir/mir/releases/tag/v0.18.0 The release includes Mir's D Foundation GSoC project. Do not forget to star the project: https://github.com/libmir/mir Best regards, Ilya |
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Ilya Yaroshenko | Could someone with big reddit rating please post the link there? (maybe Andrei?) |
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Ilya Yaroshenko | On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
> Benchmark:
> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>
Nice but you should explain how to reproduce the results, if only to give more credit to the post.
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Guillaume Piolat | On Friday, 23 September 2016 at 13:31:35 UTC, Guillaume Piolat wrote: > On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote: >> Benchmark: >> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html >> > > Nice but you should explain how to reproduce the results, if only to give more credit to the post. The benchmark source code can be found here https://github.com/libmir/mir/blob/master/benchmarks/glas/gemm_report.d. The post will be updated with the link soon. |
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Ilya Yaroshenko | On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
> Mir is LLVM-accelerated Generic Numerical Library for Science and Machine Learning.
>
> Benchmark:
> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>
> Mir v0.18.0 release notes:
> https://github.com/libmir/mir/releases/tag/v0.18.0
> The release includes Mir's D Foundation GSoC project.
Nice work. The blog post is a bit short though: it might benefit from some usage samples showing how the equivalent D code is much nicer to write and read. The benchmark source shows that the C API calls are much longer than D, highlight that in the post.
People don't just want speed and portability, they want ease of use. Show them that D will be easier.
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Ilya Yaroshenko | On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
> Mir is LLVM-accelerated Generic Numerical Library for Science and Machine Learning.
>
> Benchmark:
> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>
> Mir v0.18.0 release notes:
> https://github.com/libmir/mir/releases/tag/v0.18.0
> The release includes Mir's D Foundation GSoC project.
>
> Do not forget to star the project:
> https://github.com/libmir/mir
>
> Best regards,
> Ilya
Looks excellent. I tested compared to Apple's Accelerate BLAS and got similar results.
Environment variables to set single thread for cblas:
for openBLAS: OPENBLAS_NUM_THREADS=1
for Accelerate (Apple): VECLIB_MAXIMUM_THREADS=1
for intel MKL: MKL_NUM_THREADS=1
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to John Colvin | On Friday, 23 September 2016 at 14:55:02 UTC, John Colvin wrote:
> On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
>> Mir is LLVM-accelerated Generic Numerical Library for Science and Machine Learning.
>>
>> Benchmark:
>> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>>
>> Mir v0.18.0 release notes:
>> https://github.com/libmir/mir/releases/tag/v0.18.0
>> The release includes Mir's D Foundation GSoC project.
>>
>> Do not forget to star the project:
>> https://github.com/libmir/mir
>>
>> Best regards,
>> Ilya
>
> Looks excellent. I tested compared to Apple's Accelerate BLAS and got similar results.
>
> Environment variables to set single thread for cblas:
> for openBLAS: OPENBLAS_NUM_THREADS=1
> for Accelerate (Apple): VECLIB_MAXIMUM_THREADS=1
> for intel MKL: MKL_NUM_THREADS=1
Thank you, John. Will add them to the blog
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to Ilya Yaroshenko | On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
> Benchmark:
> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>
Great work.
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to jmh530 | On Friday, 23 September 2016 at 15:01:28 UTC, jmh530 wrote:
> On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
>> Benchmark:
>> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>>
>
> Great work.
+1. Please also provide the used compilers for reproduceability and credibility, e.g., its's not clear how OpenBLAS was compiled.
|
September 23, 2016 Re: Numerical age for D: Mir v0.18.0 is faster then OpenBLAS | ||||
---|---|---|---|---|
| ||||
Posted in reply to kinke | On Friday, 23 September 2016 at 16:48:33 UTC, kinke wrote:
> On Friday, 23 September 2016 at 15:01:28 UTC, jmh530 wrote:
>> On Friday, 23 September 2016 at 13:25:30 UTC, Ilya Yaroshenko wrote:
>>> Benchmark:
>>> http://blog.mir.dlang.io/glas/benchmark/openblas/2016/09/23/glas-gemm-benchmark.html
>>>
>>
>> Great work.
>
> +1. Please also provide the used compilers for reproduceability and credibility, e.g., its's not clear how OpenBLAS was compiled.
clang and recent ldc beta. C compiler does not matter because OpenBLAS Haswell kernels was written in assembler. Mir support only LDC compiler.
|
Copyright © 1999-2021 by the D Language Foundation