jrevels/mixedmodebroadcastad.jl issues and pull requests

#40 - Adapt to broadcast changes for 0.7

Pull Request - State: closed - Opened by vchuravy over 6 years ago - 5 comments

#39 - add ability to mark differentiable parameters to broadcast_gradients! (fix #36)

Pull Request - State: closed - Opened by jrevels over 6 years ago - 5 comments

#38 - Update dependencies

Pull Request - State: closed - Opened by maleadt over 6 years ago

#37 - Fast intrinsics and fastmath

Pull Request - State: closed - Opened by maleadt over 6 years ago - 4 comments

#36 - Don't differentiate boundary states in Julia code

Issue - State: closed - Opened by jrevels over 6 years ago - 2 comments

#35 - Initial data collection and processing infrastructure

Pull Request - State: closed - Opened by maleadt over 6 years ago - 7 comments

#34 - data/plots we want to gather/create

Issue - State: open - Opened by jrevels almost 7 years ago - 2 comments

#33 - add benchmarks to demonstrate scaling with arity

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#32 - [WIP] add TF-style julia benchmarks, re-organize test/perf code, and other stuff

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 2 comments

#31 - disable fast-math for XLA

Pull Request - State: closed - Opened by vchuravy almost 7 years ago - 1 comment

#30 - handroll GPU-friendly dual broadcast method

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#29 - WIP/RFC: fuse forwards pass/backwards pass and remove taping infrastructure

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#28 - add tensorflow hmlstm benchmark python script

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 2 comments

#27 - NFC: couple of improvements

Pull Request - State: closed - Opened by maleadt almost 7 years ago

#26 - get rid of precompute/no-precompute distinction

Pull Request - State: closed - Opened by jrevels almost 7 years ago

#25 - DO NOT MERGE: linear broadcast

Pull Request - State: open - Opened by maleadt almost 7 years ago - 3 comments

#24 - Code simplifications

Pull Request - State: closed - Opened by maleadt almost 7 years ago

#23 - Fused unrolled backwards pass specialized on downstream counter

Pull Request - State: closed - Opened by maleadt almost 7 years ago - 2 comments

#22 - add naive downstream counter to variables, enabling load/add elision for certain backwards pass operations

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 4 comments

#21 - WIP: Manually fuse the backwards pass.

Pull Request - State: closed - Opened by maleadt almost 7 years ago - 5 comments

#20 - unroll dual-based broadcast backwards propagation

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#19 - cache memory for dual buffer; fixes #13

Pull Request - State: closed - Opened by jrevels almost 7 years ago

#18 - WIP: const wrapper for __ldg

Pull Request - State: closed - Opened by vchuravy almost 7 years ago - 3 comments

#17 - dependencies script forgets to actually grab the reference before trying to check it out

Issue - State: closed - Opened by jrevels almost 7 years ago

#16 - replace old benchmarks with new HM-LSTM benchmarks

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#15 - benchmarks round 2

Issue - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#14 - GPU-aware automatic SoA transformation

Pull Request - State: closed - Opened by vchuravy almost 7 years ago - 6 comments

#13 - Cache more stuff

Issue - State: closed - Opened by jrevels almost 7 years ago

#12 - Partial fusion doesn't reduce allocation size

Issue - State: closed - Opened by maleadt almost 7 years ago - 2 comments

#11 - add dependency management script

Pull Request - State: closed - Opened by vchuravy almost 7 years ago - 1 comment

#10 - go back to array-of-structs

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 5 comments

#9 - go full struct-of-arrays

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 6 comments

#8 - use CUDAapi to automatically build kernels.so

Pull Request - State: closed - Opened by vchuravy almost 7 years ago

#7 - build script

Issue - State: closed - Opened by jrevels almost 7 years ago - 2 comments

#6 - add partially fused kernels, remove unnecessary tag parameter, fix tests, simplify benchmark code

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 3 comments

#5 - refactor to get rid of DiffResults and handle dual numbers directly

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 1 comment

#4 - get rid of unnecessary propagate macro

Pull Request - State: closed - Opened by jrevels almost 7 years ago

#3 - add required machinery for AD of unfused benchmark

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 2 comments

#2 - benchmarks

Issue - State: closed - Opened by jrevels almost 7 years ago - 5 comments

#1 - rewrite for lstm benchmark kernel

Pull Request - State: closed - Opened by jrevels almost 7 years ago - 5 comments

GitHub / jrevels/mixedmodebroadcastad.jl issues and pull requests