Adding a verification system to verify benchmark results - #111

Open

thomasRoglin wants to merge 4517 commits into

ORNL:masterfrom

SiPearl:tr/verification

This pull request is big! We're only showing the most recent 250 commits

Commits on Dec 30, 2024

Merge remote-tracking branch 'refs/remotes/origin/master'
zjin-lcf
committed
587092c

Commits on Jan 6, 2025

Update Licence
zjin-lcf
committed
d301559

Commits on Jan 8, 2025

[attention] evaluate block, warp, and mixed reductions; increase the kernel iteration count to 1000
zjin-lcf
committed
45da25a
[addBiasResidualLayerNorm-sycl] replace custom reduction with sycl::reduce_over_group
zjin-lcf
committed
d2a7682
[attentionMultiHead-hip] replace custom reductions with hipcub reductions
zjin-lcf
committed
6161c50
[attentionMultiHead-cuda] replace custom reductions with cub reductions
zjin-lcf
committed
1faa508
[attentionMultiHead-sycl] replace custom reductions with sycl reductions
zjin-lcf
committed
f88db18
[attentionMultiHead] set a random seed for data initialization
zjin-lcf
committed
b880738
[atomicAggregate-hip] fix verification error with wave-specific kernels
zjin-lcf
committed
6b9d4df
[bscan-hip] add a program for a wavefront of 32 to fix verification error
Jin Z
committed
41af9bf
Merge remote-tracking branch 'refs/remotes/origin/master'
zjin-lcf
committed
3ed8f8e
[addBiasResidualLayerNorm-hip] update the block and warp reductions
zjin-lcf
committed
1f0f2fb
[addBiasResidualLayerNorm] remove a dimension of 768 in the problem set; update the checksum calculation
zjin-lcf
committed
98a72a4
[bsw-sycl] fix compile errors; TODO support a wavefront of 64
zjin-lcf
committed
52b2de5
[cc-hip] use the HIP warpSize
zjin-lcf
committed
96a74e3
[cc-sycl] make the program portable
zjin-lcf
committed
a2849a4
[fpdc-sycl] comment out the reqd_sub_group_size
zjin-lcf
committed
eb5ccfd
[fpdc-cuda] fix the deprecated warning
zjin-lcf
committed
ea789af
[vote-hip] make the program portable
zjin-lcf
committed
c317c07
[vote-cuda] specify the warp_size in the program instead of on the command line
zjin-lcf
committed
8729e64
[vote-sycl] make the program portable
zjin-lcf
committed
bb77db6

Commits on Jan 9, 2025

[gemv-hip] address compiler warnings; replace WARP_SIZE with warpSize
zjin-lcf
committed
27760fa
[gemv-sycl] comment out reqd_sub_group_size; TODO support a wavefront of 64
zjin-lcf
committed
ab2b9c3
[wedford-sycl] comment out reqd_sub_group_size; fix compile errors
zjin-lcf
committed
cad30d7
[softmax-sycl] comment out reqd_sub_group_size and remove warp_size check
zjin-lcf
committed
8314845
[simpleSpmv-cuda] remove the unreachable function call
zjin-lcf
committed
dfd5051
[simpleSpmv-sycl] comment out reqd_sub_group_size
zjin-lcf
committed
30e46c5
[warpsort-sycl] comment out reqd_sub_group_size
zjin-lcf
committed
168b0a9
[softmax-online-sycl] comment out reqd_sub_group_size
zjin-lcf
committed
579053c
[softmax-fused-sycl] remove reqd_sub_group_size
zjin-lcf
committed
d19f7d6
[gc-hip] a wavefront of 64 is not supported
zjin-lcf
committed
01a4344

Commits on Jan 10, 2025

[softmax-online] add an offline kernel as a baseline
zjin-lcf
committed
ecf2efc

Commits on Jan 11, 2025

Commits on Jan 13, 2025

Commits on Jan 14, 2025

[quantVLLM] add the examples in CUDA, HIP and SYCL
zjin-lcf
committed
24fd823
[quantVLLM] add the category and reference for the examples
zjin-lcf
committed
774f0ac

Commits on Jan 18, 2025

[quantAQLM] add the examples in CUDA, HIP and SYCL
zjin-lcf
committed
949ef08
[quantAQLM] add the category and reference for the examples
zjin-lcf
committed
431a70a

Commits on Jan 20, 2025

[scatter] add the examples in CUDA, HIP and SYCL
zjin-lcf
committed
f043826
[scatter] add the category and reference for the examples
zjin-lcf
committed
cf581dd

Commits on Jan 23, 2025

[scatter] update the license
zjin-lcf
committed
c4420d7

Commits on Jan 24, 2025

Commits on Jan 26, 2025

[kiss] add the CUDA, HIP, and SYCL examples
zjin-lcf
committed
508deb2
[kiss] add the category and reference for the examples
zjin-lcf
committed
253ced4

Commits on Jan 27, 2025

[blockAccess-sycl] main2.cpp contains a kernel that implements block accesses using the experimental SYCL group load/store functions
zjin-lcf
committed
2276cec

Commits on Feb 1, 2025

[blockAccess-sycl] add more properties to the group load/store functions
zjin-lcf
committed
d624551

Commits on Feb 3, 2025

[readme] add a missing comma
zjin-lcf
committed
3963453

Commits on Feb 6, 2025

Commits on Feb 7, 2025

[heat2d] remove the return statement in the CUDA/HIP kernels and factor out the kernel function in the SYCL program
zjin-lcf
committed
f445186

Commits on Feb 12, 2025

[logprob-sycl] fix compile errors and warnings
zjin-lcf
committed
e459608

Commits on Feb 13, 2025

fix hybridsort segfault
thomasRoglin
committed
f3ea54b
replace non standard calls to sycl::pow<>
thomasRoglin
committed
3d0189b
Remove fsycl flag when vendor is AdaptiveCpp
thomasRoglin
committed
75a5679

Commits on Feb 14, 2025

replace sycl::abs() by sycl::fabs()
thomasRoglin
committed
4f702bb

Commits on Feb 15, 2025

[laplace3d] slightly update the kernels for consistence
zjin-lcf
committed
9a505ca

Commits on Feb 17, 2025

gitignore main
pvelesko
committed
8d2e18b
rename 2 targets to main
pvelesko
committed
73649a8
rename ans- exec to main
pvelesko
committed
02ae378
ans- rename exec to main
pvelesko
committed
0587267
Relax precision on f16max
pvelesko
committed
ec8e7c8
Relax precision haccmk
pvelesko
committed
ec654d1
head2d use python instead of python2
pvelesko
committed
d0e4dcc
Relax precision on hellinger-cuda
pvelesko
committed
8e70d0b
is-cuda default to S class (C class causes OOM on some GPUs)
pvelesko
committed
96010de
keogh - relax precision limits
pvelesko
committed
c60eec0
relax precision lombscargle
pvelesko
committed
f4de831
wordcount-hip default to CPP headers
pvelesko
committed
fd937d8
reduce wordcount problem size to better fit igpu
pvelesko
committed
79e7ed6

Commits on Feb 20, 2025

Commits on Feb 21, 2025

Commits on Feb 22, 2025

Commits on Feb 24, 2025

Commits on Feb 25, 2025

[channelSum] simplify the CUB block reduce calls
zjin-lcf
committed
3859bdf
[geodesic] change the time unit to `us`
zjin-lcf
committed
51e47eb

Commits on Feb 26, 2025

[scatter] improve readability of the `update` function
zjin-lcf
committed
1f38128
[p4, clink] rename sigmoid() as sigmoidf()
zjin-lcf
committed
c1c9af2

Commits on Feb 27, 2025

[vanGenuchten] rename abs() as fabs() in the kernels; call delete[] to deallocate memory
zjin-lcf
committed
a16e742

Commits on Feb 28, 2025

[lif] call the fmin/fmax functions for the min/max comparisons in the kernels
zjin-lcf
committed
068223a

Commits on Mar 2, 2025

[accuracy] factor out the block reduce function
zjin-lcf
committed
7f62982
[channelSum] factor out the block reduce function
zjin-lcf
committed
173d050
[fdtd3d] reorder the indexing sequence
zjin-lcf
committed
3dcf525

Commits on Mar 3, 2025

[lif] fix the issue that the min/max functions were not called in the CUDA/HIP kernels; remove the redundant type conversion
zjin-lcf
committed
2be139d

Commits on Mar 4, 2025

Commits on Mar 5, 2025

Commits on Mar 6, 2025

[attentionMultiHead] verify the kernel results with a reference implementation
zjin-lcf
committed
fdf3eca

Commits on Mar 7, 2025

Commits on Mar 8, 2025

Commits on Mar 9, 2025

[wedford-omp] add the OpenMP version
zjin-lcf
committed
6810a23

Commits on Mar 10, 2025

[zerocopy-cuda] add a README file
zjin-lcf
committed
ddd8b72

Commits on Mar 17, 2025

[geodesic] add the math header
zjin-lcf
committed
d74684d

Commits on Mar 22, 2025

[moe] add the examples in CUDA, HIP and SYCL
zjin-lcf
committed
ced508e
[moe] add the category and reference for the examples
zjin-lcf
committed
7cefaf6

Commits on Mar 23, 2025

[p4] add the math header and inline the sigmoid function
zjin-lcf
committed
10b3bae
[permute] update the array names
zjin-lcf
committed
92669b5

Commits on Mar 24, 2025

Commits on Mar 25, 2025

[tqs] use a `next` reference in the kernel
zjin-lcf
committed
e78e60d
[tqs] avoid memory leaks on verification failure
zjin-lcf
committed
c7f112d

Commits on Mar 26, 2025

Commits on Mar 27, 2025

Commits on Mar 28, 2025

[pathfinder] add a space in the timing message
zjin-lcf
committed
c4173e4
[haccmk] rename the executable to main
zjin-lcf
committed
7826537

Commits on Mar 30, 2025

[aidw] relax the accuracy from 0.1 to 1 when the fast math option is used across compilers; enable the verification of the kernel results using the large data size
zjin-lcf
committed
f8937bf

Commits on Mar 31, 2025

[meanshift-cuda] use a short name for threadIdx.x
zjin-lcf
committed
883989e

Commits on Apr 1, 2025

[heat] rename the executable
zjin-lcf
committed
27ecbb7
[hwt1d] remove a character unsupported by utf-8 encoding
zjin-lcf
committed
cd2bc43

Commits on Apr 2, 2025

[chemv] move the ppcg macros out of the kernel functions
zjin-lcf
committed
7da23ee

Commits on Apr 3, 2025

[axhelm] change the name of the kernel argument
zjin-lcf
committed
fcdf154
[md-cuda] update the indexing slightly
zjin-lcf
committed
427ce8b
[maxpool3d-cuda] update the indexing slightly
zjin-lcf
committed
5a2e794

Commits on Apr 4, 2025

[heat] inline the PI macro in the kernels and replace PI with M_PI in the host function
zjin-lcf
committed
d1f1335

Commits on Apr 6, 2025

Commits on Apr 9, 2025

[maxpool3d] remove the constraint of the image dimensions; change the work-group dimension from (16,16,1) to (8,8,4) in the CUDA, HIP, and SYCL programs; add the boundary checks in the kernels
zjin-lcf
committed
78a1b46

Commits on Apr 10, 2025

[adamw] verify the device results using reference implementations; replace the custom reductions with CUB reductions in the CUDA/HIP kernels
zjin-lcf
committed
c19d025

Commits on Apr 11, 2025

[gelu] add the baseline kernels for performance comparison
zjin-lcf
committed
2291750

Commits on Apr 15, 2025

[pointwise] inline the sigmoid function
zjin-lcf
committed
ee1ba33

Commits on Apr 16, 2025

[gd-omp] update the time unit
zjin-lcf
committed
57fcb21

Commits on Apr 17, 2025

[adamw] add a constant template parameter in the CUDA/HIP kernels
zjin-lcf
committed
6a79914
[adamw-omp] add an OpenMP version
zjin-lcf
committed
a325209

Commits on Apr 23, 2025

[pointwise] rename the kernel
zjin-lcf
committed
117d7a3
Merge branch 'master' of https://github.com/zjin-lcf/HeCBench
zjin-lcf
committed
3609739
[chi2-omp] rename the executable in Makefiles
zjin-lcf
committed
6d4d2a1
[assert] add the OpenMP version
zjin-lcf
committed
0627f6d
[softmax-cuda/hip] update the Makefiles
zjin-lcf
committed
508701d
[softmax-omp] adjust the execution time unit
zjin-lcf
committed
506f728

Commits on Apr 24, 2025

[tissue-omp] relax the error bound; disable -O3 in nvc++ due to incorrect reference result
zjin-lcf
committed
1a9ddb3
[scripts] upgrade to C++17 in Makefile.aomp
zjin-lcf
committed
9674f48
[pointwise-omp] fix memory map error
zjin-lcf
committed
0396e5e

Commits on Apr 25, 2025

[p4-omp] fix a memory map issue
zjin-lcf
committed
93aeaac

Commits on Apr 28, 2025

[md-sycl] define a macro for zero
zjin-lcf
committed
7ef9082
Add missing cuda.h includes
illuhad
committed
6a853ad

Commits on Apr 29, 2025

Commits on May 2, 2025

Commits on May 9, 2025

Commits on May 12, 2025

Commits on May 13, 2025

Commits on May 14, 2025

[readme] acknowledge Leveraging LLMs to Automate Energy-Aware Refactoring of Parallel Scientific Codes
zjin-lcf
committed
70617ad

Commits on May 15, 2025

[memcpy-omp] fix compile errors
zjin-lcf
committed
c662f70
[cm-omp] fix the compile errors
zjin-lcf
committed
690c7eb
[slit] add the reference codes to fix compile errors
zjin-lcf
committed
9863904
[lzss] update the sample data name
zjin-lcf
committed
685bbb7
Merge branch 'master' of https://github.com/zjin-lcf/HeCBench
zjin-lcf
committed
6911d12
Merge branch 'master' of https://github.com/zjin-lcf/HeCBench
zjin-lcf
committed
e712ed9
[graphExecution] add the C++ chrono header
zjin-lcf
committed
b9ee0bd
[grep] update the timer header
zjin-lcf
committed
1bf3b2c
[softmax-online-cuda] add the C++ chrono header
zjin-lcf
committed
46e9019
[readme] acknowledge Can Large Language Models Predict Parallel Code Performance?
zjin-lcf
committed
8d88f22
[debayer-omp] fix memory map error
zjin-lcf
committed
dbe2271

Commits on May 29, 2025

Revise the copyright statement
zjin-lcf
committed
8e28892
Add a License section
zjin-lcf
committed
e32aba5

Commits on Jun 1, 2025

[bwt-cuda] add the CUDA header
zjin-lcf
committed
32b7ac2
[bwt-omp] add the OpenMP header
zjin-lcf
committed
8c81664

Commits on Jun 8, 2025

[adjacent] add the CUDA and HIP examples
zjin-lcf
committed
8a7cf3c

Commits on Jun 9, 2025

[geglu] add the GeGLU examples
zjin-lcf
committed
64c40f7
[geglu] add the category and reference for the examples
zjin-lcf
committed
ffd7f62

Commits on Jul 3, 2025

Add a verification system to autohecbench script
thomasRoglin
committed
10e1303
Use enums for benchmark status instead of string
thomasRoglin
committed
ed66181
Add verification info to subset.json
thomasRoglin
committed
2344bf5