Riyadh Baghdadi

Here are few high level comments: - Can you please add a high level description of the algorithm you are using? (add it in the source code, usually we document...

Thanks. Yes good point. But actually Layer I can introduce false dependences because it supports updates. The user can declare a computation and then update it (i.e., erase the old...

Thanks for reporting this @mikeseven ! Those that are slower have been recently added and/or are not yet optimized so I would expect that they are slower. Those that fail...

For the slow benchmarks, there is another reason. AVX2 is disabled by default in Tiramisu, whereas Halide uses AVX2, so all the Tiramisu benchmarks are expected to be at least...

The distributed Tiramisu backend requires a specific branch of Halide but the other backends will work on up to date Halide without a problem. @jrayzero and @psuriana is there anything...

@mikeseven does the public Halide and LLVM 6.0 work fine for you ? Do all tests succeed ?

Sure. The generated code is just an object file that has the function you generated. You can call it from Python like any other C function. The wrapper file in...

Hi Tiramisu does not have a ready-to-use C backend, you can add one very easily. There are two ways for doing this: - Tiramisu generates Halide IR. You can translate...

Can you follow "Method 3" to install Tiramisu as described in the Readme file

Hi @Msabih , at this stage we still do not support sparse DNNs in this integration. If you are interested in helping with this please let us know. @IHIaadj is...