Infinoid

The input files I received are in `.mkv` format and I was hoping to read from them as-is. FWIW, it looks like scikit-video works around this by running `ffprobe -count_frames`.

> I believe the C++ library does currently have a code caching mechanism that should automatically reuse any kernel that's been previously compiled. Neat. I wrote an example program to...

Oh ok, thanks. If I hadn't reused the same input buffers, it would have worked fine.

Sure. ```c // Generated by the Tensor Algebra Compiler (tensor-compiler.org) __global__ void computeDeviceKernel0(taco_tensor_t * __restrict__ A, taco_tensor_t * __restrict__ B, taco_tensor_t * __restrict__ C){ double* __restrict__ A_vals = (double*)(A->vals); int*...

Hi, Taco does support tensor contraction. But there are issues with sparse outputs at the moment. See issues #184, #297. I think that's what's happening here.

I've been seeing missing local variables with `precompute` too. I ran into it with MTTKRP, but managed to boil it down to this: ```sh taco 'A(i,j) = B(i,j) * 5'...

The missing variable thing is fixed. The original "Not lowerable" bug described by @roastduck is still present. I tried to minimize that reproducer, here's the result: ``` $ bin/taco "C(i0)=A(i0,p0)*B(p0)"...

Taco generates code in a temp directory and compiles it from there. Debug builds of taco leave the temp directory around after the application ends, so you can take a...

Hi, I wrote a program based on your description. One problem I noticed is that `time()` has a granularity of 1 second. I replaced it with a call to `gettimeofday()`...

For a general algorithm, I don't think it would be as simple as that. Taco is good at regular operations over sparse data. But [there is no simple algorithm to...