Improve ROCm support. (#780)

- updates to support ROCm 5.2
- workarounds in tests where NV tools were used unconditionally
- implemented `get_num_blocks()` and `add_memfence()` for AMD GPU
- backported from history some atomics
- added bf16 support
- minor warnings cleanup
- added dockerfile to run on a ROCm enabled machine

Co-authored-by: B1tway <andrew.shukshov@gmail.com>
Co-authored-by: Andrey Shukshov <36711069+B1tway@users.noreply.github.com>

lib/ir/dispatch.cc

@@ -47,6 +47,8 @@ ir::type *computation_type(ir::type* a_ty, ir::type* b_ty){
   //     converted to half
   if(a_ty->is_fp16_ty() || b_ty->is_fp16_ty())
     return type::get_fp16_ty(ctx);
+  if(a_ty->is_bf16_ty() || b_ty->is_bf16_ty())
+    return type::get_bf16_ty(ctx);
   if(!a_ty->is_integer_ty() || !b_ty->is_integer_ty())
     throw_unreachable("augment_types");
   // 4 ) both operands are integer and undergo