c3sr/comm_scope

Issues

• `hipMemcpyHostToDevice` -> `hipMemcpyAsyncDeviceToDevice`

  #47 opened a year ago by cwpearson
  0

• Let's remove `-Wfatal-errors` from the flags

  #45 opened a year ago by cwpearson
  1

• Remove `-Werror`

  #46 opened a year ago by cwpearson
  1

• Add a CUDA 10 docker image

  #12 opened a year ago by cwpearson
  1

• `nvcc fatal: Unknown option '-pthread' in CMake 3.17.2

  #31 opened a year ago by cwpearson
  1

• PCIe A100 can deliver incorrect results when GPU is idle due to low clocks

  #43 opened 4 years ago by cwpearson
  0

• ThetaGPU: error in src/cudaMemcpyPeerAsync_Duplex_GPUGPUPeer: a PTX JIT compilation failed

  #42 opened 4 years ago by cwpearson
  0

• NUMA node boundary benchmarks?

  #41 opened 4 years ago by rlerdorf
  2

• use cudaEvent to measure empty kernel time

  #40 opened 4 years ago by cwpearson
  0

• Allow comm_scope targets to build without being in scope

  #33 opened 4 years ago by cwpearson
  1

• add NVSHMEM benchmarks

  #39 opened 4 years ago by cwpearson
  0

• Break benchmarks out into latency and bandwidth

  #38 opened 4 years ago by cwpearson
  0

• How to fix the number of iterations in a given micro-benchmark?

  #37 opened 4 years ago by Palwisha-18
  5

• Undefined when `USE_NUMA != 1`

  #36 opened 4 years ago by cwpearson
  0

• Undefined when `USE_NUMA != 1`

  #35 opened 4 years ago by cwpearson
  0

• Create a rai_build.yml file for reproducibility.

  #19 opened 4 years ago by cwpearson
  1

• stack-smashing error during do_after_inits

  #14 opened 4 years ago by cwpearson
  1

• Make sure data is random to foil compression

  #34 opened 4 years ago by cwpearson
  0

• Unknown CMake command "sugar_include".

  #32 opened 5 years ago by Yiltan
  4

• How to give size as an argument for comm_scope?

  #30 opened 5 years ago by Palwisha-18
  2

• Any lessons to be learned from EasyPerf?

  #29 opened 5 years ago by cwpearson
  1

• Investigate using cudaLaunchHostFunc for getting wall time when a stream operation ends

  #28 opened 6 years ago by cwpearson
  0

• mfence should clobber memory

  #26 opened 6 years ago by cwpearson
  1

• sync should clobber memory

  #27 opened 6 years ago by cwpearson
  1

• dcbf should clobber memory

  #11 opened 6 years ago by cwpearson
  1

• rename UM_Coherence to UM_Demand

  #20 opened 6 years ago by cwpearson
  1

• Error if numa is not found

  #18 opened 6 years ago by cwpearson
  1

• prefetch-duplex CPU/GPU should probably occur in two different streams

  #23 opened 6 years ago by cwpearson
  1

• create zero-copy H2D

  #24 opened 6 years ago by cwpearson
  1

• use cudaEventWaitStream for multi-device duplex transfers

  #25 opened 6 years ago by cwpearson
  3

• prefetch-duplex GPU/GPU may be able to associate both streams with a single device

  #22 opened 6 years ago by cwpearson
  0

• Anything we should learn from NVIDIA/multi-gpu-programming-models

  #21 opened 6 years ago by cwpearson
  0

• nvcc 7.5 has incompatibility with spdlog 0.16.3-p1

  #17 opened 6 years ago by cwpearson
  0

• add multi-threaded explicit transfer benchmarks

  #16 opened 6 years ago by cwpearson
  1

• Flush caches in unified memory host-to-gpu

  #15 opened 6 years ago by cwpearson
  0

• programatically register benchmarks based on available CUDA and NUMA devices

  #6 opened 6 years ago by cwpearson
  0

• Use clflush with "+m" output operand?

  #10 opened 6 years ago by cwpearson
  0

• zero-copy GPU/GPU should enable bidirectional peer access and free memory

  #13 opened 6 years ago by cwpearson
  0

• This numa/wr.cpp argument no longer exists

  #9 opened 6 years ago by cwpearson
  0

• Use aligned_alloc in numa-memcpy/host_to_pinned

  #7 opened 6 years ago by cwpearson
  0

• Use aligned_alloc in numa-memcpy/pinned_to_gpu

  #8 opened 6 years ago by cwpearson
  0

• Clean up memcpy/host_to_gpu

  #3 opened 6 years ago by cwpearson
  1

• Clean up memcpy/pinned_to_gpu

  #4 opened 6 years ago by cwpearson
  1

• Clean up memcpy/gpu_to_gpu_nopeer

  #5 opened 6 years ago by cwpearson
  1

• Causes scope to crash during benchmark registration on non-gpu systems

  #2 opened 6 years ago by cwpearson
  1

• Graceful handling on non-NUMA systems

  #1 opened 6 years ago by cwpearson
  1