WIP: Cuda autotune

This PR introduces two one changes:

• rotate (32,1,1) depending on field strides to fastest dimension. So (1,1,32) for c-layout and (32,1,1) for fortran layout. So pystencils will be fast also for c-layout (this will always be performed)
• auto-tune the block dimensions to whatevers is fastest for a specific kernel on localhost. On first kernel call different layouts are tried and the kernel will be called henceforth with the fastest configuration (disk_cached). This could be intersting for OpenCL where we don't know which launch config is the fastest (on OpenCL the runtime can alternatively give a hint on that).

One drawback: the test calls are only correct if input and output fields do not overlap (so no in-place kernels).

pystencils/gpucuda/indexing.py

@@ -290,6 +290,8 @@ class LineIndexing(AbstractIndexing):
     This indexing scheme supports up to 4 spatial dimensions, where the innermost dimensions is not larger than the
     maximum amount of threads allowed in a CUDA block (which depends on device).
     """
+    AUTOTUNE_BLOCK_SIZES = ((16, 1, 1), (32, 1, 1), (64, 1, 1), (96, 1, 1), (128, 1, 1), (160, 1, 1), (192, 1, 1),)
+    AUTOTUNE_NUM_CALLS = 10
 
     def __init__(self, field, iteration_slice):
         available_indices = [THREAD_IDX[0]] + BLOCK_IDX