Skip to content

Exit with error code 1 when verification fails#76

Open
linehill wants to merge 3968 commits intoORNL:masterfrom
linehill:errexit-on-fail
Open

Exit with error code 1 when verification fails#76
linehill wants to merge 3968 commits intoORNL:masterfrom
linehill:errexit-on-fail

Conversation

@linehill
Copy link
Copy Markdown

Make a lot of benchmarks to exit with error code 1 when the verification fails and make autohecbench.py skip result recording for the failing benchmarks.

(This commit is extended version of the commit started by @franz.)

zjin-lcf and others added 30 commits July 16, 2023 07:18
[word2vec] rename word2vec variants to make the naming consistent
Jin Z and others added 27 commits December 28, 2023 18:03
… header files; all GPU thread block sizes are 256; delete any BLAS library codes; the compiler option fastmath is enabled by default; the SYCL queue is an out-of-order queue; call the SYCL sincos() function instead of the sin() and cos() functions for performance portability in nbodyfft.cpp
…e argument; fix compiler warnings about comparison of integer expressions of different signedness; update the Makefile for the icpx compiler
…__host__ function `malloc` in __host__ __device__ function
…lvm#4595 for compiling the SYCL program; optimize the CUDA and HIP programs with the compile-time values of M, N and K variables mentioned in https://github.com/zjin-lcf/HeCBench/pull/70; change the variable name from P to K
…than needed reported in https://github.com/zjin-lcf/HeCBench/pull/72; slightly modify the loop count for computing the maximum value
Fix bspline-vgh thread block count in CUDA and HIP versions. Their
block count was 256 times more than needed.
[bspline-vgh] fix bspline-vgh's thread block count in CUDA and HIP versions. Their block count was 256 times more than needed.
…s because the SYCL version is non-blocking copies. The issue is reported in https://github.com/zjin-lcf/HeCBench/pull/74
Make a lot of benchmarks to exit with error code 1 when the
verification fails and make autohecbench.py skip result recording for
the failing benchmarks.

This commit is extended version of the
[commit](franz@7a7f935)
started by @franz.

Co-authored-by: Henry Linjamäki <henry.mikael.linjamaki@intel.com>
@zjin-lcf
Copy link
Copy Markdown
Collaborator

zjin-lcf commented Feb 19, 2024

I appreciate your changes. I am concerned about memory leaks for early exit. I will try to update all programs.

@linehill
Copy link
Copy Markdown
Author

I am concerned about memory leaks for early exit.

OS should take care of reclaiming the memory that is not freed by programs (link).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants