[PYTHON] CUTLASS wrapper for fair benchmarks (#75)

Before this commit, the benchmarking infrastructure used heterogeneous protocols between library (e.g., CUTLASS uses a C++ binary that reports mean TFLOPS; torch and triton use python call and report 10th, 50th and 90th quantiles). For the sake of uniformity and fair benchmark practices, this PR adds a python wrapper for auto-tuned CUTLASS matrix multiplication. Benchmarks have been rewritten to use this wrapper with `triton.testing.do_bench` rather than system calls to CUTLASS profiler. Importantly, this also ensures that all the matmuls are done on the *same* input data which should stabilize clock across providers.

python/src/main.cc

@@ -3,10 +3,14 @@
 void init_superblocking(pybind11::module &m);
 void init_torch_utils(pybind11::module &m);
 void init_triton(pybind11::module &m);
+void init_cutlass(pybind11::module &m);
 
 PYBIND11_MODULE(libtriton, m) {
   m.doc() = "Python bindings to the C++ Triton API";
   init_triton(m);
   init_torch_utils(m);
   init_superblocking(m);
+#ifdef WITH_CUTLASS_BINDINGS
+  init_cutlass(m);
+#endif
 }