Why using signed integer instead of unsigned for index/size?

[zcbenz]

For example:

mlx/mlx/backend/metal/binary.cpp

Lines 147 to 153 in 54a71f2

	if (large) {
	compute_encoder.set_bytes<int64_t>(out.data_size(), arg_idx++);
	grid_dims = get_2d_grid_dims(out.shape(), out.strides(), work_per_thread);
	} else {
	compute_encoder.set_bytes<int>(out.data_size(), arg_idx++);
	grid_dims = MTL::Size(nthreads, 1, 1);
	}

Why using int/int64_t instead of uint32_t/uint64_t?

[awni]

For the contiguous ops it doesn't matter.. I think they could be unsigned. In some cases (e.g. for the non-contiguous ops) we use signed integers because strides can be negative. I think the use of signed integers in the contiguous ops is just a by product of that as the earlier logic is simpler.

[zcbenz]

Thanks!