Code

Flush cache#246

Code

Draft

fjwillemsen wants to merge 30 commits intomasterKernelTuner/kernel_tuner:masterfrom flush_cacheKernelTuner/kernel_tuner:flush_cacheCopy head branch name to clipboard

+253-71

Commits

Commits on Feb 8, 2024

• Added RegisterObserver with common interface among backends

  fjwillemsen

  committed
  81a68a4

• Added test for RegisterObserver, added clause in case of mocktest

  fjwillemsen

  committed
  943b3c4

Commits on Feb 9, 2024

• Added useful error message in case Register Observer is not supported

  fjwillemsen

  committed
  1681730

• Added tests for Register Observer for OpenCL and HIP backends

  fjwillemsen

  committed
  f153945

Commits on Feb 12, 2024

• Added instruction for pointing cache directory elsewhere

  fjwillemsen

  committed
  7bd7c2b

• Non-argument streams are now correctly passed in the CuPy and NVCUDA backends

  fjwillemsen

  committed
  9dea137

Commits on Feb 15, 2024

• Fixed several issues pertaining to the setting of clocks, in particular one where the memory clock would always be seen as not-equal due to a rounding error

  fjwillemsen

  committed
  df54145

• Time spent setting NVML parameters (clock & memory frequency, power) goes to framework time instead of benchmark time

  fjwillemsen

  committed
  4cc4a13

• Time spent setting NVML parameters (clock & memory frequency, power) goes to framework time instead of benchmark time

  fjwillemsen

  committed
  e309bc1

• Removed redundant print statement

  fjwillemsen

  authored
  d6aac8b

Commits on Feb 28, 2024

• Added L2 cache size property to CUDA backends

  fjwillemsen

  committed
  a020791

• Added specification to CUPY compiler options

  fjwillemsen

  committed
  6e6e5fb

• Added L2 cache size property to OpenCL, HIP and mocked PyCUDA backends

  fjwillemsen

  committed
  f15338f

• Added function to check for compute capability validity, improved check on gpu-architecture compiler option, added gpu-architecture auto-adding to CuPy

  fjwillemsen

  committed
  00ac419

• Added a flush kernel to clear the L2 cache between runs

  fjwillemsen

  committed
  55ab074

• Added a flush kernel to clear the L2 cache between runs

  fjwillemsen

  committed
  e106bae

Commits on Feb 29, 2024

• Made function for scaling the compute capability to a valid one, added tests for this function, removed setting --gpu-architecture for CuPy as it is already set internally

  fjwillemsen

  committed
  0cb5e3a

Commits on Mar 1, 2024

• Applied suggestions from comments by @csbnw

  fjwillemsen

  committed
  b682506

• Removed redundant comments / printing

  fjwillemsen

  committed
  da907b1

• Added L2 cache size information to backends

  fjwillemsen

  committed
  2396bdf

• Added L2 flush kernel

  fjwillemsen

  committed
  eced775

• Switched to new attempt for flushing L2 using memset

  fjwillemsen

  committed
  143889f

• Added implementation of allocate numpy array function

  fjwillemsen

  committed
  651eea7

Commits on Mar 2, 2024

• Added new flush L2 cache method using memset

  fjwillemsen

  committed
  7d8d48f

Commits on Mar 4, 2024

• Added a standard method for freeing memory from the GPU

  fjwillemsen

  committed
  9911f4c

• Circumvented an issue where list.remove(val) was not properly implemented by CuPy, and attempt free of previous allocation after checking if flush is possible

  fjwillemsen

  committed
  47c2cca

Commits on Mar 7, 2024

• Added the ability to recopy array arguments with every kernel launch, added interfacing for flushing L2 and recopying arguments

  fjwillemsen

  committed
  157ca41

• Renamed to for clarity, added check

  fjwillemsen

  committed
  98afa60

Commits on Apr 12, 2024

• Improved getting L2 cache size

  fjwillemsen

  committed
  cfecdc5

• Small improvements to flushing arrays

  fjwillemsen

  committed
  108e14c