issues
search
hikettei
/
cl-waffe2
[Experimental] Graph and Tensor Abstraction for Deep Learning all in Common Lisp
https://hikettei.github.io/cl-waffe2/
MIT License
122
stars
5
forks
source link
[Optimize] do-compiled-loop eliminated Runtime Code Compilation
#125
Closed
hikettei
closed
9 months ago
hikettei
commented
9 months ago
Changes
[Add] New Iterator
do-compiled-loop
: Runtime call-with-view
[Add] cache-id option for define-impl
[Change] Now define-impl with cache-when-compiled=nil produces warnings since not cached operations can lead to a reduction in compilation time.
[Optimize] No runtime compilation when using
!matmul
!where
!compare
!relu
and so on...
[Optimize] call-with-view folds broadcasted axis in order to optimize !sum
[Optimize] defmodel-as: Set freeze-axis=nil and enabled loop collapses
[Add] Row-Major Gemm
[Optimize] Deletes
{GRAD}MOVETENSOR
if the gradient is not permuted
Changes
do-compiled-loop
: Runtime call-with-view!matmul
!where
!compare
!relu
and so on...{GRAD}MOVETENSOR
if the gradient is not permuted