Changes to make CPU backend functions asynchronous calls #1174

9prady9 · 2015-12-17T19:27:49Z

Fixes #832

This fixed `Assign.LinearAssignSeq` unit test in assign unit tests.

This also fixed assign unit test: `ArrayAssign.CPP_ASSIGN_VECTOR_2D`

Fixed lu async function

It was a bug in upstream function cpu::fast

9prady9 · 2015-12-21T23:31:18Z

build arrayfire ci

shehzan10 · 2015-12-22T00:04:08Z

build arrayfire windows ci

pavanky · 2015-12-28T21:06:02Z

@9prady9 Can you make sure cmake detects if the threads submodule is initialized or not (similar to what we are doing with assets).

Right now it fails only when trying to compile

this macro takes care of asynchronous kernel launch and calls sync on the queue when in debug mode.

pavanky · 2015-12-28T22:48:23Z

Code looks good. Things to change require the following:

CMake checks if threads submodule has been initialized
Add getQueue().sync() after each getQueue.enqueue() in debug mode.
Add the ability to make the algorithm synchronous at runtime based on environment variable.

pavanky · 2015-12-28T22:48:37Z

@umar456 What do you think about the last point ?

shehzan10 · 2015-12-28T22:49:42Z

It would be nice to have a compile time option

9prady9 · 2015-12-28T23:49:03Z

build arrayfire ci

umar456 · 2015-12-29T00:03:23Z

What would be the benefit of having both? Couldn't you just export the variable and make it persistent?

pavanky · 2015-12-29T04:05:10Z

@umar456 try setting and unsetting the variable on windows. It is not pretty.

pavanky · 2015-12-30T07:18:57Z

@9prady9 Can you pull from devel and resolve merge conflicts? You will need to fix transform.cpp in CPU backend.

9prady9 · 2015-12-30T15:09:46Z

@pavanky I will start on this after 3.2.2 -> devel PR. I believe that is the last hot fixes before v3.3.0 release, is that right ?

pavanky · 2015-12-30T15:20:10Z

@9prady9 Sure.

This reverts commit 95d9346. * Not necessary with the new queue class (see queue.hpp) * Macros bad

shehzan10 · 2015-12-31T00:10:02Z

build arrayfire ci

Changes to make CPU backend functions asynchronous calls

umar456 and others added 30 commits August 9, 2015 01:18

Setup async project; Setup basic queues;

be0ff22

Convert CPU blas to use async queues

b94c3df

Async CPU approx1 and approx2

3188bdf

Async CPU Assign

f797314

Async CPU Bilateral

c8ecdb9

Async CPU Convolve

759b506

Async CPU diff1 and diff2

c399e75

Avoid sending references to queued lambdas

80903d0

Async CPU Copy, Assign, and Index

96c5602

Async CPU diagonal

46beebc

Async FFT for the CPU backend

b7c83e8

Add eval to copyData

413eea8

Async random on CPU backend

49f0cce

Async where on the CPU backe

fada883

Async CPU reduce and ireduce

1a0802f

Async CPU Transpose. Fix bug in eval

1842bcf

Merge branch 'devel' into async

5d428df

Merge additional changes for async CPU.

4a8e723

async cpu::index function

91f7a1f

Fixed asynchronous indexing & assignment in cpu backend

2796770

converted cpu tile to asychronous call

0aeed42

This fixed `Assign.LinearAssignSeq` unit test in assign unit tests.

converted sort_index cpu function to asynchronous call

330ae1c

This also fixed assign unit test: `ArrayAssign.CPP_ASSIGN_VECTOR_2D`

converted triangle fn in cpu backend to async call

5eea071

converted lu & cholesky decomposition functions to async calls

551433e

svd cpu backend is async now

ed6d26d

converted qr & solve cpu functions to async calls

e0d7c12

Fixed lu async function

Converted wrap & unwrap cpu fns to async calls

d0223f9

converted transform to async call

32a65d8

converted susan fn in cpu backend to asynchronous call

743fb4a

convert sort & sort_by_key cpu fns to async calls

840af46

9prady9 added 2 commits December 20, 2015 11:17

Fixed the bug in cpu ireduce kernel function

1313f98

Fixed orb async cpu fn

b684b06

It was a bug in upstream function cpu::fast

9prady9 added 6 commits December 28, 2015 16:29

cmake fix to check for threads submodule

a0f17b6

moved fft cpu fns implementations to kernel namespace

c539f1d

moved dot cpu implementation to kernel namespace

4831215

moved fftconvolve reorder helper fn to kernel namespace

d1089f8

fixed cpu::kernel::dot fn signature

4f8b3fa

Added ENQUEUE macro in cpu backend

95d9346

this macro takes care of asynchronous kernel launch and calls sync on the queue when in debug mode.

9prady9 added 2 commits December 28, 2015 18:07

Removed obsolete queue sync in cpu::padArray fn

7dad2ef

Fixed cmake condition for threads submodule check

90611a2

Synchronize when AF_SYNCHRONOUS_CALLS is set to 1

1cbffbb

pavanky added this to the 3.3.0 milestone Dec 30, 2015

umar456 and others added 2 commits December 30, 2015 10:41

Revert "Added ENQUEUE macro in cpu backend"

6058dd2

This reverts commit 95d9346. * Not necessary with the new queue class (see queue.hpp) * Macros bad

Merge branch 'devel' into async

f0b9691

pavanky added a commit that referenced this pull request Dec 31, 2015

Merge pull request #1174 from umar456/async

861808e

Changes to make CPU backend functions asynchronous calls

pavanky merged commit 861808e into arrayfire:devel Dec 31, 2015

umar456 deleted the async branch May 17, 2017 22:31

Changes to make CPU backend functions asynchronous calls #1174

Changes to make CPU backend functions asynchronous calls #1174

Uh oh!

Conversation

9prady9 commented Dec 17, 2015

Uh oh!

9prady9 commented Dec 21, 2015

Uh oh!

shehzan10 commented Dec 22, 2015

Uh oh!

pavanky commented Dec 28, 2015

Uh oh!

pavanky commented Dec 28, 2015

Uh oh!

pavanky commented Dec 28, 2015

Uh oh!

shehzan10 commented Dec 28, 2015

Uh oh!

9prady9 commented Dec 28, 2015

Uh oh!

umar456 commented Dec 29, 2015

Uh oh!

pavanky commented Dec 29, 2015

Uh oh!

pavanky commented Dec 30, 2015

Uh oh!

9prady9 commented Dec 30, 2015

Uh oh!

pavanky commented Dec 30, 2015

Uh oh!

shehzan10 commented Dec 31, 2015

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants