Code

Optimize memory access#67

Code

Draft

sfiligoi wants to merge 63 commits intobiocore:mainbiocore/unifrac-binaries:mainfrom sfiligoi:embinv_250210sfiligoi/unifrac-binaries:embinv_250210Copy head branch name to clipboard

+1,526-783

Commits

Commits on Jan 23, 2025

• Add su::biom::load_n_samples and use in su.cpp

  sfiligoi

  committed
  3c1b044

  Copy full SHA for 3c1b044

• Remove unnecessary header includes

  sfiligoi

  committed
  6838090

  Copy full SHA for 6838090

• Expose the partial functions in libssu

  sfiligoi

  committed
  1efade6

  Copy full SHA for 1efade6

Commits on Jan 24, 2025

• Make libssu trully thread-safe

  sfiligoi

  committed
  897ce3e

  Copy full SHA for 897ce3e

• Make ssu and faithpd use the shared library

  sfiligoi

  committed
  817b7d7

  Copy full SHA for 817b7d7

• Add missing header include

  sfiligoi

  committed
  518c319

  Copy full SHA for 518c319

• Use explicit object list in rapi_test dependency

  sfiligoi

  committed
  b45d9dc

  Copy full SHA for b45d9dc

• Export dl_read_partial and properly clean between rebuilds

  sfiligoi

  committed
  1f65ef6

  Copy full SHA for 1f65ef6

• Use shared library in test_api. Deprecate merge_partial, since not use anywhere anymore

  sfiligoi

  committed
  ef32989

  Copy full SHA for ef32989

• test_ska does not use the unifrac object files

  sfiligoi

  committed
  f9a2aaf

  Copy full SHA for f9a2aaf

• Create an intermediate archive file to avoid dependencies for functions we do not use

  sfiligoi

  committed
  2d106e4

  Copy full SHA for 2d106e4

• Add API-based unifrac tests in test_su

  sfiligoi

  committed
  a1a2c5b

  Copy full SHA for a1a2c5b

• Add API-based faithpd tests in test_su

  sfiligoi

  committed
  b45a9c7

  Copy full SHA for b45a9c7

• Add test_su_api to exercise the optimiza code paths (test_su exercize the cpu-only internals)

  sfiligoi

  committed
  ddceeee

  Copy full SHA for ddceeee

• Handle NOGPU at top-level Makefile, so nv not required

  sfiligoi

  committed
  e53236e

  Copy full SHA for e53236e

• Add support for parallel make

  sfiligoi

  committed
  ba82d84

  Copy full SHA for ba82d84

• Add clean_install make option

  sfiligoi

  committed
  267d158

  Copy full SHA for 267d158

Commits on Jan 25, 2025

• Fix test_su_api checks

  sfiligoi

  committed
  7c9c6ac

  Copy full SHA for 7c9c6ac

• Add cpu-only builds

  sfiligoi

  committed
  a74e67c

  Copy full SHA for a74e67c

Commits on Jan 27, 2025

• Explicitly create all variants of unifrac_task_noclass

  sfiligoi

  committed
  13f7026

  Copy full SHA for 13f7026

• Use internally a C API layer for the accelerated code

  sfiligoi

  committed
  b1b1f69

  Copy full SHA for b1b1f69

• Fix Makefile for acc

  sfiligoi

  committed
  fc1c73d

  Copy full SHA for fc1c73d

• Explicity use h5c++ for everything but acc code. Use CXX for acc code

  sfiligoi

  committed
  df97366

  Copy full SHA for df97366

• Modularize ld loading in src/ssu_ld.c

  sfiligoi

  committed
  aac7d13

  Copy full SHA for aac7d13

• Move handling of UNIFRAC_CPU_INFO into ssu_ld.c

  sfiligoi

  committed
  bec652e

  Copy full SHA for bec652e

Commits on Jan 28, 2025

• Put gpu code into its own shared library. No need for pgi h5c++ wrapper anymore

  sfiligoi

  committed
  042ad3f

  Copy full SHA for 042ad3f

• Put gpu code into its own shared library. No need for pgi h5c++ wrapper anymore - Fix typo

  sfiligoi

  committed
  5a0ce0c

  Copy full SHA for 5a0ce0c

• Add symlink unifrac_task.cpp to make R happy

  sfiligoi

  committed
  0fab9aa

  Copy full SHA for 0fab9aa

• Do not check for /proc/driver/nvidia/gpus, not needed

  sfiligoi

  committed
  c65289b

  Copy full SHA for c65289b

• Make loading of GPU shared library optional during initial test

  sfiligoi

  committed
  2ff25ef

  Copy full SHA for 2ff25ef

Commits on Jan 29, 2025

• Clean make process, rename shlibs to mirror gcc naming

  sfiligoi

  committed
  fa7f4bb

  Copy full SHA for fa7f4bb

• Add explicit mtune

  sfiligoi

  committed
  ee35586

  Copy full SHA for ee35586

• Fix typo

  sfiligoi

  committed
  715f0bc

  Copy full SHA for 715f0bc

• Fix typo

  sfiligoi

  committed
  1fe3d10

  Copy full SHA for 1fe3d10

Commits on Jan 30, 2025

• Use a single acc_nv for both OpenACC and OMPGPU. Preliminary support for acc_amd

  sfiligoi

  committed
  37d271e

  Copy full SHA for 37d271e

Commits on Feb 1, 2025

• Do not explicitly source the nv setup script in the Makefile

  sfiligoi

  committed
  ab393ab

  Copy full SHA for ab393ab

• Switch form BUILD_OFFLOAD to BUILD_NV_OFFLOAD. Only build all_nv if NV_CXX defined

  sfiligoi

  committed
  bccccd1

  Copy full SHA for bccccd1

Commits on Feb 3, 2025

• Always build acc_nv on Linux

  sfiligoi

  committed
  e58031d

  Copy full SHA for e58031d

• Add support for AMD GPU compilation

  sfiligoi

  committed
  c8dedf1

  Copy full SHA for c8dedf1

• Add install_amd_clang.sh and use in CI

  sfiligoi

  committed
  3b571f1

  Copy full SHA for 3b571f1

• Make BUILD_NV_OFFLOAD and BUILD_AMD_OFFLOAD optional

  sfiligoi

  committed
  5e521bd

  Copy full SHA for 5e521bd

• Fix typo

  sfiligoi

  committed
  7e0dfc6

  Copy full SHA for 7e0dfc6

• Clarify which GPU is used

  sfiligoi

  committed
  ba0f00a

  Copy full SHA for ba0f00a

• Properly build test_binaries last on Linux

  sfiligoi

  committed
  e4a1ae1

  Copy full SHA for e4a1ae1

Commits on Feb 4, 2025

• Use curl instead of wget

  sfiligoi

  committed
  e3684ee

  Copy full SHA for e3684ee

• Enable curl link following, and use apt install

  sfiligoi

  committed
  f529184

  Copy full SHA for f529184

• Better handle failure in installing gpu compilers

  sfiligoi

  committed
  408309a

  Copy full SHA for 408309a

• Update README with AMD complier instructions

  sfiligoi

  committed
  adf33aa

  Copy full SHA for adf33aa

• Improve installation instructions in README

  sfiligoi

  committed
  23b542a

  Copy full SHA for 23b542a

• Add uprivileged install fallback in scripts/install_amd_clang.sh

  sfiligoi

  committed
  155853d

  Copy full SHA for 155853d

Commits on Feb 5, 2025

• Add support for UNIFRAC_USE_NVIDIA_GPU and UNIFRAC_USE_AMD_GPU. Document clean_install

  sfiligoi

  committed
  664a38c

  Copy full SHA for 664a38c

• Fix typo

  sfiligoi

  committed
  3cb1977

  Copy full SHA for 3cb1977

Commits on Feb 10, 2025

• Update NV setup script to download the latest version

  sfiligoi

  committed
  25beb33

  Copy full SHA for 25beb33

• Improve build docs

  sfiligoi

  committed
  99dd51b

  Copy full SHA for 99dd51b

• Remove popcnt-based selection

  sfiligoi

  committed
  9389792

  Copy full SHA for 9389792

• Transpose embedded_proportions for non-Vaw and non-binary variants

  sfiligoi

  committed
  0f87955

  Copy full SHA for 0f87955

• Tune emb size for new access pattern

  sfiligoi

  committed
  58c97c2

  Copy full SHA for 58c97c2

Commits on Feb 11, 2025

• Fix preprocesssor check

  sfiligoi

  committed
  f5407d9

  Copy full SHA for f5407d9

• NVCC happier when one does not specify vector_length explicitly

  sfiligoi

  committed
  e84beae

  Copy full SHA for e84beae

• Explicitly request vectorization in WeightedVal1

  sfiligoi

  committed
  e158991

  Copy full SHA for e158991

• Use the (old) straight logic for the GPUs

  sfiligoi

  committed
  432a28e

  Copy full SHA for 432a28e

Commits on Feb 12, 2025

• Tune RECOMMENDED_MAX_EMBS_BOOL for CPUs

  sfiligoi

  committed
  64ff847

  Copy full SHA for 64ff847

• NVCC happier when one does not specify vector_length explicitly

  sfiligoi

  committed
  82c9345

  Copy full SHA for 82c9345