accelerate-backend-kit

accelerate-backend-kit copied to clipboard

The length of each payload array should be the _product_ of it array's dimensions, not the sum.

projanen

Count ETupProject projlen differently / replace with simpler tuple indexing

1

We are using surface tuples throughout accelerate-backend kit, including decoding the `EltRepr` to recover the surface structure (which doesn't help us with shapes, alas). However, there's an inconsistency here, because...

rrnewton

p90a,b,c fails in ExplicitShapes (doBinds) explicitShapes: doBinds got unexpected =TTuple [TArray 1 TInt,TArray 1 TInt,TArray 1 TInt] p92 and p92a sends backend-kit into infinite loop: !! victory is CLOSE --...

svenssonjoel

What is going on here? https://github.com/AccelerateHS/accelerate-backend-kit/blob/master/backend-kit/Data/Array/Accelerate/BackendKit/IRs/Internal/AccClone.hs#L96 Which expression is the head and which is the tail? Moreover, the types make no sense. I feel like this is the equivalent of...

tmcdonell

ToAccClone.hs: TODO: handle Intersect

2

tmcdonell

Running the crystal demo from accelerate-examples

tmcdonell

Delete .genC files except in debug mode

1

rrnewton

This is one possible solution. We could just do a global normalizing pass before getting to the backend where DynamicAcc is used. Currently, it looks like we partially support N-array...

rrnewton

[Related to #19, which is the same thing for clang.] This should be pretty easy, as Intel's changes were into trunk [a year ago](https://software.intel.com/en-us/articles/cilk-plus-array-notation-for-c-accepted-into-gcc-mainline) and [released in GCC 4.9](https://gcc.gnu.org/gcc-4.9/changes.html). I'm...

rrnewton

[Related to #20, which is the same thing for GCC.] If the compiler backends in this repo become full featured enough to be usable, then it will eventually be important...

rrnewton