Code

Various CUDA Optimizations#1

Code

Open

lilinitsy wants to merge 51 commits intoZwickyTransientFacility:masterZwickyTransientFacility/periodfind:masterfrom lilinitsy:masterlilinitsy/periodfind:masterCopy head branch name to clipboard

+75,505-55,065

Commits

Commits on Jun 17, 2024

• Remove old cuda architectures incompatible with version 12

  lilinitsy

  committed
  168217d

  Copy full SHA for 168217d

Commits on Jun 30, 2024

• buggy (wrong results) of LS kernel batched

  lilinitsy

  committed
  e8ee3e6

  Copy full SHA for e8ee3e6

• some updates???

  lilinitsy

  committed
  0a649bb

  Copy full SHA for 0a649bb

• index properly in the curve_bytes calculation...

  lilinitsy

  committed
  b8b9e15

  Copy full SHA for b8b9e15

Commits on Jul 1, 2024

• Multiple lightcurves batched, but not faster; need some improvements

  lilinitsy

  committed
  ace28ad

  Copy full SHA for ace28ad

Commits on Jul 13, 2024

• backup

  lilinitsy

  committed
  52e54b2

  Copy full SHA for 52e54b2

Commits on Jul 14, 2024

• Batch updates

  lilinitsy

  committed
  be405e3

  Copy full SHA for be405e3

• Minor updates

  lilinitsy

  committed
  a16e378

  Copy full SHA for a16e378

• ffastmath on and some GPU intrinsics

  lilinitsy

  committed
  7c5c7c8

  Copy full SHA for 7c5c7c8

• Use two asynchronous cuda streams, which leads to a large performance increase

  lilinitsy

  committed
  2f36a81

  Copy full SHA for 2f36a81

Commits on Jul 16, 2024

• Updates to improve accuracy; runtime for testing 100k periods decreases from 170s to 148s

  lilinitsy

  committed
  a681cd3

  Copy full SHA for a681cd3

• Get multiple streams working, minor improvement

  lilinitsy

  committed
  7fe18a0

  Copy full SHA for 7fe18a0

Commits on Jul 17, 2024

• clang-format

  lilinitsy

  committed
  ac75555

  Copy full SHA for ac75555

• Move cuda stream synchronization outside loop

  lilinitsy

  committed
  310de6a

  Copy full SHA for 310de6a

• Remove extraneous chronos usage

  lilinitsy

  committed
  04b2955

  Copy full SHA for 04b2955

Commits on Jul 27, 2024

• Synchronize streams before destroying

  lilinitsy

  committed
  8419b2f

  Copy full SHA for 8419b2f

• Increase the number of batched curves

  lilinitsy

  committed
  0b60864

  Copy full SHA for 0b60864

Commits on Jul 28, 2024

• Modify inner loop condition

  lilinitsy

  committed
  9f6eb21

  Copy full SHA for 9f6eb21

• Move synchronize

  lilinitsy

  committed
  18500e0

  Copy full SHA for 18500e0

• restrict conditional entropy pointers

  lilinitsy

  committed
  b6cc7e2

  Copy full SHA for b6cc7e2

• Pinned memory for CE

  lilinitsy

  committed
  b03f020

  Copy full SHA for b03f020

• 3 cuda streams for CE

  lilinitsy

  committed
  036159c

  Copy full SHA for 036159c

Commits on Aug 7, 2024

• Use registers instead of shared memory in conditional entropy kernel

  lilinitsy

  committed
  8521b23

  Copy full SHA for 8521b23

• Copy local hists to local memory

  lilinitsy

  committed
  5aa4f3e

  Copy full SHA for 5aa4f3e

• Cleanup on CE

  lilinitsy

  committed
  1a19575

  Copy full SHA for 1a19575

• Cleanup on CE

  lilinitsy

  committed
  45031c8

  Copy full SHA for 45031c8

Commits on Aug 13, 2024

• Change block size

  

  Ville Cantory

  committed
  21d122d

  Copy full SHA for 21d122d

• Pyx speedups using list comprehension

  lilinitsy

  committed
  a37cef5

  Copy full SHA for a37cef5

Commits on Aug 14, 2024

• Tune threadcount and batch size

  lilinitsy

  committed
  34b8769

  Copy full SHA for 34b8769

• More tuning for v100

  lilinitsy

  committed
  b60e830

  Copy full SHA for b60e830

• More tuning for running on 1 v100

  lilinitsy

  committed
  a5e347b

  Copy full SHA for a5e347b

• List comprehension when returning stats in pyx

  lilinitsy

  committed
  5083b32

  Copy full SHA for 5083b32

• Update pyx to reserve memory beforehand

  lilinitsy

  committed
  6c19ff4

  Copy full SHA for 6c19ff4

• update pyx

  lilinitsy

  committed
  39b5617

  Copy full SHA for 39b5617

• ls.pyx updates

  lilinitsy

  committed
  5edbd7e

  Copy full SHA for 5edbd7e

• pyx update, set streams to 4

  lilinitsy

  committed
  a16331a

  Copy full SHA for a16331a

• Add OpenMP to Cmakelists

  lilinitsy

  committed
  8feafdf

  Copy full SHA for 8feafdf

Commits on Aug 15, 2024

• Pyx updates for CE

  lilinitsy

  committed
  0593b7a

  Copy full SHA for 0593b7a

• Adjust thread count

  lilinitsy

  committed
  ba4a9e9

  Copy full SHA for ba4a9e9

• Change threads for ce

  lilinitsy

  committed
  40854f9

  Copy full SHA for 40854f9

Commits on Aug 16, 2024

• Testingw ithout pinned on v100

  lilinitsy

  committed
  f135693

  Copy full SHA for f135693

• Use shared memory (Faster on V100, slower on 1080)

  lilinitsy

  committed
  dfeadbb

  Copy full SHA for dfeadbb

• ce update

  lilinitsy

  committed
  6adb53a

  Copy full SHA for 6adb53a

Commits on Aug 17, 2024

• updates to threadcounts per kernel

  lilinitsy

  committed
  49d69c3

  Copy full SHA for 49d69c3

• Pinned memory for aov

  lilinitsy

  committed
  4b416c0

  Copy full SHA for 4b416c0

Commits on Aug 18, 2024

• Threadcount adjustments

  lilinitsy

  committed
  13fed4c

  Copy full SHA for 13fed4c

• streams for AOV

  lilinitsy

  committed
  9593fbc

  Copy full SHA for 9593fbc

Commits on Aug 19, 2024

• Move atomic adds in FoldBinKernel

  lilinitsy

  committed
  c4c2bf5

  Copy full SHA for c4c2bf5

• Remove numphasebinoverlap loop from FoldBinKernel

  lilinitsy

  committed
  54785b6

  Copy full SHA for 54785b6

Commits on Feb 16, 2025

• Finally ran clang-format

  lilinitsy

  committed
  7e59a60

  Copy full SHA for 7e59a60

• Remove extraneous files, clang-format, etc, ready for merge

  lilinitsy

  committed
  dd4c9ea

  Copy full SHA for dd4c9ea