Prepare for MGARD 1.7.0 release - #259

Merged

JieyangChen7 merged 238 commits into

CODARcode:masterfrom

JieyangChen7:mgard-x

Sep 21, 2026

Commits on Mar 24, 2025

fix memcpy for new gcc

Jieyang Chen
authored and
JieyangChen7
committed
8ac6544
Adjust timing strategies and add throughput calculations

Jieyang Chen
authored and
JieyangChen7
committed
373881c
Add batched BP encoder interface

Jieyang Chen
authored and
JieyangChen7
committed
d3eb815
fix huffman for mdr
JieyangChen7
committed
87595fe
Make pitched/mamaned template variables for Array/SubArray
JieyangChen7
committed
f4eb435
add BPEncoderOptV1 and modify MDR-X lossless interface
JieyangChen7
committed
f0fd447
optimized huffman deflate for MDR and MDR low-level pipeline performance optimization
JieyangChen7
committed
ef18815
fix decompression CR output
JieyangChen7
committed
829d432
Add nagabinary to BPEncoderOptV1
JieyangChen7
committed
d58103b
Add run length encoding
JieyangChen7
committed
892aa94
add time profile for MDR lossless
JieyangChen7
committed
dcaf385
Add hybrid compressor for MDR
JieyangChen7
committed
8e67618
remove reduce memory footprint option (default on now) and add hybrid compressor for MDR
JieyangChen7
committed
3ca57c8

Commits on Mar 27, 2025

Fix compiling serial issue with clang
Jieyang Chen
committed
a96c5a5
Fix minor issues with MDR on HIP
Jieyang Chen
committed
ab27747
Add adaptive lossless for MDR
Jieyang Chen
committed
fb3f723
Adjust allocation and output estimazation for MDR refactoring
Jieyang Chen
committed
9f288b2
Adjust allocation and output estimazation for MDR reconstruct
Jieyang Chen
committed
20f2c71
Fix domain decomposition for MDR
Jieyang Chen
committed
c9c3de1

Commits on Mar 28, 2025

add QOI pipeline
Jieyang Chen
committed
79bac72
fix lossless signature verify segfault
JieyangChen7
committed
8a4c8cf
clang format
JieyangChen7
committed
a950c17
Add mdr-qoi.cpp
JieyangChen7
committed
01b1e47

Commits on Mar 29, 2025

Move BP encoder sign reset inside kernel
JieyangChen7
committed
9260e42
Add experimental BP encoders
JieyangChen7
committed
7106240

Commits on Mar 30, 2025

making sure all merged bp batches are used & remove sign from mdrdata resize
JieyangChen7
committed
6c504fc

Commits on Apr 4, 2025

Update BPEncoder interface data type
JieyangChen7
committed
d13f38e

Commits on Apr 5, 2025

fixing async execution: level interleaver, lossless verify, reconstruction pipeline, level absmax
JieyangChen7
committed
cfc4ec0
Add Array::hostAlloc
JieyangChen7
committed
5099d92
replace exp with absmax
JieyangChen7
committed
3becdef
Decouple stages in refactor and reconstructor & adjust reconstruct pipeline
JieyangChen7
committed
bf4dda5
Fix pipeline for recon qoi
JieyangChen7
committed
c97f729
update auto tuner for hip
JieyangChen7
committed
672e324

Commits on Apr 6, 2025

round BP encoder input to be multiply of BATCH_SIZE
JieyangChen7
committed
452c90b
optimized v1b bpencoder::decode
JieyangChen7
committed
b494893
done v1b bp encoder
JieyangChen7
committed
1e6417b

Commits on Apr 9, 2025

roughly done v2a bp encoder
JieyangChen7
committed
3256be2

Commits on Apr 10, 2025

add absmin
JieyangChen7
committed
820dd7b

Commits on Apr 12, 2025

improve MDR compile speed
JieyangChen7
committed
83d5a20
fix bp encoder v2a for hip
Jieyang Chen
committed
bb451dc
fix block size of v2a to 32 threads
Jieyang Chen
committed
c5c6105
keep bp v2a block size to 32
JieyangChen7
committed
481572f

Commits on Apr 13, 2025

Add missing header in hip
Jieyang Chen
committed
0502e81
Merge cr estimation in Huffman and RLE
JieyangChen7
committed
20b667e
fix bp encoding for double precision
JieyangChen7
committed
13583ce

Commits on Apr 14, 2025

add log csv & fix RLE decode
JieyangChen7
committed
8c4142b
prep for pipeline testing
JieyangChen7
committed
f768a2d

Commits on Apr 17, 2025

clean code after experiments
JieyangChen7
committed
93581b2

Commits on Apr 19, 2025

Move LevelLinearizer
JieyangChen7
committed
0ac5b6e
decouple quantize with huffman
JieyangChen7
committed
79c6e1f

Commits on May 28, 2025

Finished recompose
MasterVChicken
committed
2f0a6f6

Commits on May 30, 2025

Bug unsolved for additional allocated outliers
MasterVChicken
committed
dab7cc5

Commits on Jun 15, 2025

Add try for HybridQuantization and failed to bound L-INF error
MasterVChicken
committed
40556d8

Commits on Jun 29, 2025

Remove global operation for HybridDataRefactor
MasterVChicken
committed
fe4b3db

Commits on Jul 9, 2025

Temporary save
MasterVChicken
committed
c1e1c95

Commits on Jul 14, 2025

temp stage
MasterVChicken
committed
04a1d42

Commits on Jul 22, 2025

temp stage
MasterVChicken
committed
257d7ad

Commits on Aug 9, 2025

Configured to run, but failed to bound tolerance
MasterVChicken
committed
fb9c4fd

Commits on Aug 12, 2025

fix refactoring bug
MasterVChicken
committed
606471a
Auxiliary save for hybrid testing
MasterVChicken
committed
741a2ab

Commits on Aug 17, 2025

Lossless integrated but padding issue not solved
MasterVChicken
committed
48f04f5

Commits on Dec 9, 2025

Fix InCacheBLock refactoring bug
MasterVChicken
committed
bccd34a
Quantization for InCache Block Design
MasterVChicken
committed
ba65c06

Commits on Dec 11, 2025

Finish lossless integration
MasterVChicken
committed
e36ada3

Commits on Dec 31, 2025

Finish hybrid hierarchy refactor(not tested yet)
MasterVChicken
committed
465c81b

Commits on Jan 2, 2026

Add gloabl refactoring parameter
MasterVChicken
committed
6aa52a6

Commits on Jan 5, 2026

Runable but with bugs
MasterVChicken
committed
020f720

Commits on Jan 9, 2026

Temp save for global decomposition bugs
MasterVChicken
committed
e6eb070

Commits on Jan 11, 2026

Fix hybrid bug
MasterVChicken
committed
764e223

Commits on Jan 15, 2026

Replace with new error allocation strategy. May cause bugs of over-conservative compression
MasterVChicken
committed
5c65ad3

Commits on Jan 16, 2026

Fix over-conservative
MasterVChicken
committed
f90fada

Commits on Jan 19, 2026

Add logic for ROI. API adjustment needed to fully support
MasterVChicken
committed
61d3f93

Commits on Jan 28, 2026

Add support for QOI but fail to lossless compress
MasterVChicken
committed
fc01b88
Add verification module for ROI
MasterVChicken
committed
fd31226

Commits on Mar 19, 2026

Fix switch issue
MasterVChicken
committed
8cf4495

Commits on Apr 20, 2026

Update for AD
MasterVChicken
committed
c3f3bb3

Commits on Jul 26, 2026

fix memcpy for new gcc

Jieyang Chen
authored and
JieyangChen7
committed
103f958
Adjust timing strategies and add throughput calculations

Jieyang Chen
authored and
JieyangChen7
committed
6115f39
Add batched BP encoder interface

Jieyang Chen
authored and
JieyangChen7
committed
801b221
fix huffman for mdr
JieyangChen7
committed
e5fa286
Make pitched/mamaned template variables for Array/SubArray
JieyangChen7
committed
4cf5926
add BPEncoderOptV1 and modify MDR-X lossless interface
JieyangChen7
committed
a87fded
optimized huffman deflate for MDR and MDR low-level pipeline performance optimization
JieyangChen7
committed
5aff842
fix decompression CR output
JieyangChen7
committed
5ba4c57
Add nagabinary to BPEncoderOptV1
JieyangChen7
committed
5dc3b2e
Add run length encoding
JieyangChen7
committed
fd143a0
add time profile for MDR lossless
JieyangChen7
committed
bf3c2d6
Add hybrid compressor for MDR
JieyangChen7
committed
f1fb832
remove reduce memory footprint option (default on now) and add hybrid compressor for MDR
JieyangChen7
committed
4b6dfb1
Fix compiling serial issue with clang

Jieyang Chen
authored and
JieyangChen7
committed
775c239
Fix minor issues with MDR on HIP

Jieyang Chen
authored and
JieyangChen7
committed
3e55c03
Add adaptive lossless for MDR

Jieyang Chen
authored and
JieyangChen7
committed
7ff38ce
Adjust allocation and output estimazation for MDR refactoring

Jieyang Chen
authored and
JieyangChen7
committed
722686b
Adjust allocation and output estimazation for MDR reconstruct

Jieyang Chen
authored and
JieyangChen7
committed
0f3d7a9
Fix domain decomposition for MDR

Jieyang Chen
authored and
JieyangChen7
committed
10a9de5
add QOI pipeline

Jieyang Chen
authored and
JieyangChen7
committed
5971e60
fix lossless signature verify segfault
JieyangChen7
committed
466bc12
clang format
JieyangChen7
committed
b375598
Add mdr-qoi.cpp
JieyangChen7
committed
a7bdf49
Move BP encoder sign reset inside kernel
JieyangChen7
committed
245ecab
Add experimental BP encoders
JieyangChen7
committed
2a19f45
making sure all merged bp batches are used & remove sign from mdrdata resize
JieyangChen7
committed
7e16d63
Update BPEncoder interface data type
JieyangChen7
committed
b32d9fe
fixing async execution: level interleaver, lossless verify, reconstruction pipeline, level absmax
JieyangChen7
committed
c8e0bdd
Add Array::hostAlloc
JieyangChen7
committed
bc24199
replace exp with absmax
JieyangChen7
committed
5f977d2
Decouple stages in refactor and reconstructor & adjust reconstruct pipeline
JieyangChen7
committed
f0e8482
Fix pipeline for recon qoi
JieyangChen7
committed
d2747f5
update auto tuner for hip
JieyangChen7
committed
c54f5e2
round BP encoder input to be multiply of BATCH_SIZE
JieyangChen7
committed
93402e8
optimized v1b bpencoder::decode
JieyangChen7
committed
87ac9d0
done v1b bp encoder
JieyangChen7
committed
5300b3e
roughly done v2a bp encoder
JieyangChen7
committed
7e5007a
add absmin
JieyangChen7
committed
903fd7e
improve MDR compile speed
JieyangChen7
committed
ca1e518
fix bp encoder v2a for hip

Jieyang Chen
authored and
JieyangChen7
committed
9c72112
fix block size of v2a to 32 threads

Jieyang Chen
authored and
JieyangChen7
committed
982d2ce
keep bp v2a block size to 32
JieyangChen7
committed
4f40578
Add missing header in hip

Jieyang Chen
authored and
JieyangChen7
committed
6fd56b3
Merge cr estimation in Huffman and RLE
JieyangChen7
committed
0901e63
fix bp encoding for double precision
JieyangChen7
committed
cc0d124
add log csv & fix RLE decode
JieyangChen7
committed
0e6184f
prep for pipeline testing
JieyangChen7
committed
7ada35c
clean code after experiments
JieyangChen7
committed
c8bb2f8
Move LevelLinearizer
JieyangChen7
committed
0732f53
decouple quantize with huffman
JieyangChen7
committed
09a88de
Replacing exit function call with C++ error throws
JieyangChen7
committed
b679d16
Adjust space and vol calculation for uniform data
JieyangChen7
committed
6cf91dd
clang format
JieyangChen7
committed
728ab65
Uncompleted QoI

LinusLi037
authored and
JieyangChen7
committed
1fe69cf
Pipelined QoI controlled reconstruction

LinusLi037
authored and
JieyangChen7
committed
97f7e5f
For parallel

Linus-Li-1037
authored and
JieyangChen7
committed
5612391
Portable QoI pipeline

Linus-Li-1037
authored and
JieyangChen7
committed
851a97f
remove hard-code dimension settings

Linus-Li-1037
authored and
JieyangChen7
committed
a19dab5
rebased onto mgard-x
JieyangChen7
committed
8b30032
fix pipeline for qoi
JieyangChen7
committed
ff0d6a5
hip optimize

Linus-Li-1037
authored and
JieyangChen7
committed
d9aa14f
eb guided reconstruction

Linus-Li-1037
authored and
JieyangChen7
committed
27a089b
parallel eb guided reconstruction

Linus-Li-1037
authored and
JieyangChen7
committed
debe45d
True Protable Dimention settings

Linus-Li-1037
authored and
JieyangChen7
committed
4acc56f
fix bitrate measurement

Linus-Li-1037
authored and
JieyangChen7
committed
dc1e453
Options for decrease method & Parallel

Linus-Li-1037
authored and
JieyangChen7
committed
591e63c
Hybrid Decrease

Linus-Li-1037
authored and
JieyangChen7
committed
27ed33b
Hybrid Decrease 2

Linus-Li-1037
authored and
JieyangChen7
committed
b9090e3
Hybrid Decrease 3: fix bitrate measurement

Linus-Li-1037
authored and
JieyangChen7
committed
64521ca
Hybrid Decrease 4

Linus-Li-1037
authored and
JieyangChen7
committed
df3c1c5
CP, MA, MAPE

Xin Liang
authored and
JieyangChen7
committed
4b46f7c
Throughput computation correction

Linus-Li-1037
authored and
JieyangChen7
committed
312eb34
Add run scripts
JieyangChen7
committed
27e26f6
revise for AE
JieyangChen7
committed
07f1f62
Update QoI_NYX.sh
JieyangChen7
committed
6966ac1
HybridLosslessCompressor hyperparameter correction

MasterVChicken
authored and
JieyangChen7
committed
d4bac51
Update QoI_JHTDB.sh

Linus-Li-1037
authored and
JieyangChen7
committed
9b8b0e8
QoI retrieval IO time

Xin Liang
authored and
JieyangChen7
committed
2fe6461
Update QoI_JHTDB.sh

Linus-Li-1037
authored and
JieyangChen7
committed
6b8eea5
Remove dummy files write/read for MDR-X
JieyangChen7
committed
097921c
update run script

Linus-Li-1037
authored and
JieyangChen7
committed
c1fbf89
update build script to diable CPU backend when GPU is available
JieyangChen7
committed
75783f0
fix MDR-X lossless baseline tests
JieyangChen7
committed
0bac892
fix BP error after rebase
JieyangChen7
committed
8ca0085
Remove testing scripts
JieyangChen7
committed
a1998d4
fix GreedyBasedSizeInterpreter that causes more bitplanes than necessary being fetched
JieyangChen7
committed
9597370
tmake tests discoverable by CTest. Also, fixing code format
JieyangChen7
committed
9f4daee
Restructured original mgard sourse code by putting it under mgard dir
JieyangChen7
committed
8b3902f
restructed unit tests
JieyangChen7
committed
d010ac7
add test prefix for original mgard unit tests
JieyangChen7
committed
d01f727
add unit tests for mgard-x & fix verify_signature array in Huffman
JieyangChen7
committed
dcc1949
mgard-x: expose orthogonal_projection on low-level (Hybrid)Compressor

JieyangChen7
and
claude
committed
b7aab13
mgard-x/huffman: optimize compress & decompress pipeline (~1.9x compress)

JieyangChen7
and
claude
committed
7e8b351
mgard-x: add BlockDelta lossless backend (cuSZp-style FLE) as a Huffman alternative

JieyangChen7
and
claude
committed
2e995d7
mgard-x: portable warp-cooperative LZ4; drop nvcomp dependency

JieyangChen7
and
claude
committed
186f3c0
mgard-x: add symbol-alphabet rANS lossless backend (beats Huffman CR)

JieyangChen7
and
claude
committed
628b4aa
mgard-x: s=inf skips orthogonal projection (hierarchical L-inf fast path)

JieyangChen7
and
claude
committed
0a1672f
mgard-x: restore split-last-cell coarsening for non-uniform grids

JieyangChen7
and
claude
committed
39ed557
mgard-x: clang-format 14 the tree to fix CI clang-format job

JieyangChen7
and
claude
committed
6568c0a
build_scripts: bump zstd to v1.5.6 and protobuf to v3.21.12

JieyangChen7
and
claude
committed
4869147
mgard-x: throw exceptions instead of exit() on internal errors (#246)

JieyangChen7
and
claude
committed
7c33641
mgard-x: clang-format 14 the #246 exception-handling changes

JieyangChen7
and
claude
committed
321f0fd
Merge tag 'sc26-blockmagrd-v1' into mgard-x

JieyangChen7
and
claude
committed
78d9e7e
mgard-x: zero-initialize the block-local padding region before decompose

JieyangChen7
and
claude
committed
81e4dab
mgard-x: add timing to the hybrid hierarchy pipeline

JieyangChen7
and
claude
committed
c5d0b7b
mgard-x: throw exceptions instead of exit() in the hybrid path

JieyangChen7
and
claude
committed
21dc675
mgard-x: compute ROI local quantizers on device instead of host+H2D per call

authored and
JieyangChen7
committed
56945e5
mgard-x: stop tracking slurm submission scripts

JieyangChen7
and
claude
committed
6f8c5e7
mgard-x: add MI300 HIP support and fix real wavefront-size handling

JieyangChen7
and
claude
committed
c21837a
mgard-x: tune Huffman compress launch configs for MI300

JieyangChen7
and
claude
committed
e3a7c01
mgard-x: guard MI300-tuned Huffman block widths to HIP only

JieyangChen7
and
claude
committed
af7f476

Commits on Jul 27, 2026

mgard-x: apply clang-format to satisfy CI format check

JieyangChen7
and
claude
committed
a14e1ab
mgard-x: clang-format pre-existing files to fix CI format check

JieyangChen7
and
claude
committed
f6bbf38

Commits on Aug 26, 2026

mgard-x: write coarsest level directly to output in local decompose

authored and
Jieyang Chen
committed
786b4c3
mgard-x: fuse quantization into the local decompose kernels

authored and
Jieyang Chen
committed
7e25f65
mgard-x: pad sm_v shared-memory layout to reduce bank conflicts

authored and
Jieyang Chen
committed
2aff64f
mgard-x: fuse dequantization into the local recompose kernels

authored and
Jieyang Chen
committed
eb02765
mgard-x: record hybrid hierarchy parameters in the file header

authored and
Jieyang Chen
committed
6b68383
mgard-x: give every compressor a DequantizeRecompose step

authored and
Jieyang Chen
committed
419196a
mgard-x: make the hybrid kernel-fusion switch usable and honest

authored and
Jieyang Chen
committed
5f554f6
mgard-x: make kernel fusion a CLI flag instead of environment variables

authored and
Jieyang Chen
committed
63eb82e

Commits on Aug 27, 2026

mgard-x: require ROCm 7.0 for HIP, drop redundant SERIAL overrides

Jieyang Chen
and
claude
committed
8128970

Commits on Sep 3, 2026

clang format fix
Jieyang Chen
committed
b1e9c68

Commits on Sep 6, 2026

mgard-x: fix stale leading dimensions on Array shrink, and test the paths

JieyangChen7
and
claude
committed
fbf3da9

Commits on Sep 8, 2026

mgard-x: fix CPUPipelines so OpenMP builds compile again

JieyangChen7
and
claude
committed
f40ccf9
mgard-x: project SubArrays onto their own dimensions, not always (0, 1, 2)

JieyangChen7
and
claude
committed
9a6863a
mgard-x: extend the in-cache block refactoring to 1D and 2D

JieyangChen7
and
claude
committed
0f22175
mgard-x: quantize the coarsest layer in ROI mode when there is no global stage

JieyangChen7
and
claude
committed
a9aad02

Commits on Sep 9, 2026

mgard-x: size Huffman's merge-path array from the grid that actually runs

JieyangChen7
and
claude
committed
361bc65
mgard-x: support an L2 (s != inf) error bound in the hybrid pipeline

JieyangChen7
and
claude
committed
26412bd

Commits on Sep 10, 2026

mgard-x: restrict the hybrid s-norm bound to one block-local level

JieyangChen7
and
claude
committed
3018a77

Commits on Sep 11, 2026

update docs
JieyangChen7
committed
e4ef1e7
Fix typos and improve formatting in README.md
JieyangChen7
authored
c199010
Update README for MGARD-Lambda formatting and clarity
JieyangChen7
authored
407c4fb
build scripts: stop building nvcomp

JieyangChen7
and
claude
committed
e479d9e
remove mgard-cuda (legacy CUDA backend)

JieyangChen7
and
claude
committed
6187036
mgard-x: add Blackwell (sm_120) build support

JieyangChen7
and
claude
committed
e873988
update mgard family tree
JieyangChen7
committed
e2ba9c2
mgard-x: remove L2 projection for BlockMGARD L-inf mode
MasterVChicken
committed
8d7d33f

Commits on Sep 12, 2026

Commits on Sep 13, 2026

Commits on Sep 14, 2026

mgard-x: fix broken examples, migrate MDR-X to current API

MasterVChicken
and
claude
committed
e35a282

Commits on Sep 17, 2026

mgard-x: fix stale active-device bug in CUDA/HIP DeviceSpecification and DeviceQueues

JieyangChen7
and
claude
committed
92b686c
docs: update MGARD-X software requirements and dependency versions

JieyangChen7
and
claude
committed
08cc0cb

Commits on Sep 18, 2026

Commits on Sep 19, 2026

mgard-x: fix MaxErrorEstimatorHB missing negabinary error margin

JieyangChen7
and
claude
committed
1c6f604
mgard-x: fix uninitialized padding in ComposedRefactor level buffers

JieyangChen7
and
claude
committed
7c1dd66

Commits on Sep 20, 2026

mgard-x: fix MDR-X sign-buffer heap overflow in bitplane decode

JieyangChen7
and
claude
committed
a73c30a
mgard-x: unify projection mode control between Compressor and hybrid

JieyangChen7
and
claude
committed
3b1f8b0
mgard-x: fix clang-format violations flagged by CI

JieyangChen7
and
claude
committed
84e4a5a

Commits on Sep 21, 2026

mgard-x: bump to 1.7.0, add release notes, remove dead prefetch/reorder config

JieyangChen7
and
claude
committed
cf8417b
mgard-x: fix clang-format after reorder-argument removal

JieyangChen7
and
claude
committed
ef7fd3b
update main readme to include the release month/year and revise the publication list
JieyangChen7
committed
786bde3
Update source file headers to 2026 and personal email; update LICENSE copyright

JieyangChen7
and
claude
committed
6b43196