Add DPNP Framework Support to Selected Benchmarks #34

4Ran8jith · 2025-03-17T09:59:08Z

integrates the DPNP framework into selected benchmarks

…mark

…ly a handful BMs are too long, it keeps the full reservation for those 3 or so :'(. Maybe use -t 30, not 300...

… npbranch update

- /README1.md - /README.md

…es from clean

Changes were reverted so had missed this one

tbennun

Thank you for your contribution! Overall it looks good. However, there are some important notes on the readme that need to be addressed, as well as other comments that I left.

tbennun · 2025-03-17T14:00:47Z

README.md

 ## Quickstart

-To install NPBench, simply execute:
+IF Dpnp is critical to you, see below. Otherwise, to install NPBench, simply execute:


Dpnp is one framework out of several, please refrain from starting the quickstart section with that.

tbennun · 2025-03-17T14:00:51Z

README.md

 ## Quickstart

-To install NPBench, simply execute:
+IF Dpnp is critical to you, see below. Otherwise, to install NPBench, simply execute:


Dpnp is one framework out of several, please refrain from starting the quickstart section with that.

tbennun · 2025-03-17T14:02:08Z

README.md

+```bibtex
+@inproceedings{
+    npbench-dpnp,
+    author = {Ranjith et al.},
+    year = {2021},
+    title = {In prep.}
+}
+```
+


The only publication that needs to be cited if using NPBench is NPBench; if, e.g., Numba is used, then Numba will be cited separately.

tbennun · 2025-03-17T14:05:23Z

main.py

@@ -0,0 +1,72 @@
+import argparse


Renaming quickstart.py to main.py should be done as a separate PR.

I would just remove this file from the PR for now.

tbennun · 2025-03-17T14:31:13Z

environment.yml

+  - https://software.repos.intel.com/python/conda
+dependencies:
+  - python=3.10.14
+  - numpy=1.26.4    # dpnp theoretically requires numpy<=1.24.4, numba-dpex numpy>=1.26.4; seems that 1.26.4 works fine even dor dpnp


Suggested change

- numpy=1.26.4 # dpnp theoretically requires numpy<=1.24.4, numba-dpex numpy>=1.26.4; seems that 1.26.4 works fine even dor dpnp

- numpy=1.26.4 # dpnp theoretically requires numpy<=1.24.4, numba-dpex numpy>=1.26.4; seems that 1.26.4 works fine even for dpnp

tbennun · 2025-03-18T03:33:50Z

npbench/benchmarks/polybench/jacobi_1d/jacobi_1d_numpy.py

        B[1:-1] = 0.33333 * (A[:-2] + A[1:-1] + A[2:])
        A[1:-1] = 0.33333 * (B[:-2] + B[1:-1] + B[2:])
+
+    return A,B


Why was a return value added to this benchmark?

tbennun · 2025-03-18T03:33:55Z

npbench/benchmarks/polybench/k2mm/k2mm_numpy.py

 def kernel(alpha, beta, A, B, C, D):
-
    D[:] = alpha * A @ B @ C + beta * D
+    return D


Why was a return value added to this benchmark?

tbennun · 2025-03-18T03:36:32Z

README.md

+With `dpnp` it is strongly recommended to use `conda` instead of `pip` for its dependency on intel packages. 
+Refer to this 
+[LINK](https://intelpython.github.io/dpnp/quick_start_guide.html#building-for-custom-sycl-targets) to know more 
+about building custom SYCL targets or installing `dpnp` package from intel channel.


Suggested change

about building custom SYCL targets or installing `dpnp` package from intel channel.

about building custom SYCL targets or installing `dpnp` package from the `intel` channel.

README.md

tbennun · 2025-03-18T03:38:37Z

README.md

+If you are behind proxies, follow your system documentation. E.g. for pip you may have to specify:
+``` bash
+$ python -m pip --proxy=http://localhost:1234 install <...> # where "localhost:1234" is the value of the env var "HTTP_PROXY"
+```


This is unnecessary to include in the instructions. It is a pip issue rather than an NPBench issue.

4Ran8jith · 2025-04-01T09:57:14Z

README.md: Updated as requested to include clarification about main.py and its purpose.

main.py: This file is designed specifically for benchmarks supporting DPNP only. It is intentionally not renamed to quickstart. A line has been added to README.md explaining this distinction.

environment.yml: Updated to reflect the latest changes.

Benchmarks: Existing benchmarks have been updated based on suggestions, and a new benchmark covariance2 has been added.

Return Statements: Modified some functions to properly use return to ensure correct copying where applicable.

Let me know if it's alright!

alexnick83

Looks OK to me. I suggest that you remove this main.py file; we can amend quickstart separately at a later stage.

alexnick83 · 2025-04-03T07:22:08Z

main.py

@@ -0,0 +1,72 @@
+import argparse


I would just remove this file from the PR for now.

svtcli and others added 30 commits August 19, 2024 15:30

Made main.py. tested pythran and dace_cpu

4bf0fe3

Added jobfiles and first plots

1c27dcd

Now pythran is using openblas

54dd46c

Made plot_result.py safer, and print .csv grouped by preset and bench…

0c26e49

…mark

run in cloudgpu for paper

8e3c55d

job paper on selected

6b2dc8b

paper_results

c710e4d

error and output

46f2b92

Added prototype jobFarm & loop.sh

36abda3

Merge branch 'npbranch' of gitlab.lrz.de:di52vum/npb-lrz into npbranch

e50028a

New jobFarm and jobFarmAll work well, and isolate errors. Unfortunate…

5d0a8c6

…ly a handful BMs are too long, it keeps the full reservation for those 3 or so :'(. Maybe use -t 30, not 300...

Add dpnp

53d2ca9

Update dpnp.json

84d4e59

toolkit

cf7aa33

add dpnp

ba727c1

add benchmarks for dpnp

b3f753b

delete file

5ad35b0

plot dpnp

7be1403

dpnp on gpu

fcde0b7

modified dpnp

ea2fd69

dpnpmap

bd3fb70

dpntool

78ec9b1

Add numba_dpex

3aa1da2

Update numba_dpex.json

f7f775a

Add environment.yml with conda package specifications

c424db2

Merge branch 'npbranch' of https://gitlab.lrz.de/di52vum/npb-lrz into…

f65f461

… npbranch update

Add requiremets_conda.txt

c454bcb

change name add numba dpex

300bed6

Update environment.yml

6218b20

Added some jobfile, ran once with clean environment, adjusted README

1ef6e2c

4Ran8jith and others added 11 commits February 25, 2025 13:31

Update 2 files

fbce409

- /README1.md - /README.md

Merged clean into main, keeping README.md from main and restoring fil…

449ca89

…es from clean

Add new file

f421b72

Add new file

cfd8f8d

Add cavityflow

b79fc42

Add c_flow

4ac82fe

add missing files

e8ee4cd

Merge branch 'main' of https://gitlab.lrz.de/di52vum/npb-lrz

f197d45

Copyback test.py

24a5a16

Update_framework.py

0c712ec

Changes were reverted so had missed this one

Update framework.py

4a14fb0

tbennun requested changes Mar 18, 2025

View reviewed changes

4Ran8jith and others added 10 commits March 18, 2025 11:34

update Readme,con_int and .yml

69c3d75

update crc16,main.py,Readme

e5f0b4c

Update Readme.md

47b290e

Add covariance2

6cb0c63

new benchmark added modified cholosky and covariance

0f7c72b

Edit covariance2.json

19be73e

add jacobi2_d

b8c32f3

Update gemm_numpy.py

04b9000

Update jacobi_1d_numpy.py

207d6ab

Update k2mm_numpy.py

39ebcd9

4Ran8jith requested a review from tbennun April 1, 2025 10:01

alexnick83 reviewed Apr 3, 2025

View reviewed changes

main.py Outdated

@@ -0,0 +1,72 @@

import argparse

Copy link

Contributor

alexnick83 Apr 3, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would just remove this file from the PR for now.

remove main.py

3b681f0

alexnick83 self-requested a review April 4, 2025 06:57

alexnick83 approved these changes Apr 4, 2025

View reviewed changes

Merge branch 'main' into gitlab-main

7a33924

alexnick83 merged commit 13e95c6 into spcl:main Apr 4, 2025
1 check passed

	- numpy=1.26.4 # dpnp theoretically requires numpy<=1.24.4, numba-dpex numpy>=1.26.4; seems that 1.26.4 works fine even dor dpnp
	- numpy=1.26.4 # dpnp theoretically requires numpy<=1.24.4, numba-dpex numpy>=1.26.4; seems that 1.26.4 works fine even for dpnp

	about building custom SYCL targets or installing `dpnp` package from intel channel.
	about building custom SYCL targets or installing `dpnp` package from the `intel` channel.

Add DPNP Framework Support to Selected Benchmarks #34

Add DPNP Framework Support to Selected Benchmarks #34

Uh oh!

Conversation

4Ran8jith commented Mar 17, 2025

Uh oh!

tbennun left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

4Ran8jith commented Apr 1, 2025

Uh oh!

alexnick83 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants