Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Oct 5, 2016

• Added check_reduction test.

  Show description for 96ed416
  obilaniu

  committed
  96ed416

  Copy full SHA for 96ed416

Commits on Jul 28, 2016

• Add gpu_ prefix and change signature to unsigned ints

  tsirif

  committed
  f49da95

  Copy full SHA for f49da95

Commits on Jul 26, 2016

• Add cuda extension to return cuda device count

  tsirif

  committed
  5fce66f

  Copy full SHA for 5fce66f

Commits on Jun 22, 2016

• Complete tests on collectives and fix CMakeLists

  Show description for 0605a1a
  tsirif

  committed
  0605a1a

  Copy full SHA for 0605a1a

• Fix broadcast root and nonroot synchronization rules

  Show description for 32f049c
  tsirif

  committed
  32f049c

  Copy full SHA for 32f049c

• Change include to <> in header files to be installed

  tsirif

  committed
  ff86df4

  Copy full SHA for ff86df4

• Document `gpuarray/buffer_collectives.h` and nccl implementations

  tsirif

  committed
  65dde33

  Copy full SHA for 65dde33

• Replace `int` with `size_t` in count argument and add restriction for nccl

  tsirif

  committed
  dff32f2

  Copy full SHA for dff32f2

• Add documentation for `gpuarray/collectives.h` and implementation

  tsirif

  committed
  8968327

  Copy full SHA for 8968327

• Fix flag checking in array_collectives

  Show description for 4d3ad15
  tsirif

  committed
  4d3ad15

  Copy full SHA for 4d3ad15

• Add `GpuArray` collectives implementations

  tsirif

  committed
  98b3a2b

  Copy full SHA for 98b3a2b

• Rename `comm_error` to `error_msg` in `gpucontext`

  Show description for 7879d1c
  tsirif

  committed
  7879d1c

  Copy full SHA for 7879d1c

• Fix things according to comments

  Show description for 7021828
  tsirif

  committed
  7021828

  Copy full SHA for 7021828

• Implement generate_clique_id of nccl

  tsirif

  committed
  f36deee

  Copy full SHA for f36deee

• Remove macro for GA_CTX_PROP_START

  tsirif

  committed
  b44496a

  Copy full SHA for b44496a

• Add comm_error impl and remove gpucomm_get_device

  tsirif

  committed
  dcb0fa5

  Copy full SHA for dcb0fa5

• Make gpucomm_free a void function

  tsirif

  committed
  9838eca

  Copy full SHA for 9838eca

• Fix mistake in definition of gpucommCliqueId

  tsirif

  committed
  b301960

  Copy full SHA for b301960

• Add funct to init comm_ops properly

  Show description for 3c99b2b
  tsirif

  committed
  3c99b2b

  Copy full SHA for 3c99b2b

• Add dummy gpucommCliqueId compatible with ncclUniqueId

  tsirif

  committed
  5da2bc1

  Copy full SHA for 5da2bc1

• Complete gpucomm in private_cuda

  Show description for ebe95af
  tsirif

  committed
  ebe95af

  Copy full SHA for ebe95af

• Add GA_COMM_ERROR and string repr

  tsirif

  committed
  a99783a

  Copy full SHA for a99783a

• Complete second buffer level api for gpucomm

  tsirif

  committed
  bd58890

  Copy full SHA for bd58890

• Split collectives to separate files, re-consider API

  tsirif

  committed
  0e81021

  Copy full SHA for 0e81021

• Add API for gpu collectives and communicator

  tsirif

  committed
  60f04b7

  Copy full SHA for 60f04b7

Commits on Jun 14, 2016

• Merge pull request #192 from abergeron/faster_wait

  Show description for 5d60445
  nouiz

  authored
  5d60445

  Copy full SHA for 5d60445

Commits on Jun 10, 2016

• Improve comments.

  abergeron

  committed
  a20132b

  Copy full SHA for a20132b

Commits on Jun 9, 2016

• Add a flags to disable multi-stream for a context.

  abergeron

  committed
  41458eb

  Copy full SHA for 41458eb

Commits on Jun 8, 2016

• Add a way to get a better description of blas errors.

  abergeron

  committed
  4eeac6e

  Copy full SHA for 4eeac6e

Commits on May 31, 2016

• Add facility to convert float16 values in GpuElemwise.

  abergeron

  committed
  4d2b847

  Copy full SHA for 4d2b847

Commits on May 25, 2016

• Add comment about PTX to GA_USE_BINARY.

  abergeron

  committed
  594c670

  Copy full SHA for 594c670

Commits on May 24, 2016

• Make the compile cache per-context and make it cache a bigger portion

  Show description for 3fa284b
  abergeron

  committed
  3fa284b

  Copy full SHA for 3fa284b

• Add some documentation to the gpuelemwise.h header.

  abergeron

  committed
  73b730f

  Copy full SHA for 73b730f

• Move the extcopy functionality out of the backends.

  Show description for 89f4f4f
  abergeron

  committed
  89f4f4f

  Copy full SHA for 89f4f4f

• Make GpuElemwise support an arbitrary number of dimensions.

  Show description for ed56dd3
  abergeron

  committed
  ed56dd3

  Copy full SHA for ed56dd3