fix linter and force allocation on hostbuf

accel/llvm/ops_llvm.py

@@ -195,7 +195,7 @@ class LLVMBuffer(ExplicitExecAST):
     k.process()
 
     if DEBUG >= 2:
-      print(ast)
+      print(k.ast)
       print("old:", k.shapes)
       print("old:", k.strides)
     
@@ -342,7 +342,7 @@ class LLVMBuffer(ExplicitExecAST):
         phi.add_incoming(reduce_result, loop_exit[store_loop+1+i]._block)
 
     # do the late ast
-    result = ast_parse(loop_exit[store_loop], ast, store_loop, reduce_result=reduce_result)
+    result = ast_parse(loop_exit[store_loop], k.ast, store_loop, reduce_result=reduce_result)
 
     # store result
     builder = loop_exit[store_loop]

tinygrad/llops/ops_gpu.py

@@ -198,6 +198,7 @@ class GPUBuffer(ExplicitExecAST):
 
   def __init__(self, shape:Union[ShapeTracker, Tuple[int, ...]], hostbuf:Optional[GPUBuffer]=None, backing:Optional[np.ndarray]=None):
     super().__init__(shape, hostbuf)
+    if hostbuf is not None: hostbuf.cl  # does the allocation, pylint: disable=W0104
     self._buf : Optional[CLBuffer] = hostbuf._buf if hostbuf is not None else None
     self._base_shape : Tuple[int, ...] = hostbuf._base_shape if hostbuf is not None else self.shape
     self._backing : Optional[np.ndarray] = hostbuf._backing if hostbuf is not None else backing

tinygrad/ops.py

@@ -118,7 +118,6 @@ class ASTKernel:
 
     # create the buffer we are returning (as the same type as the input buffers) and add it as the first buffer
     self.ret = type(self.bufs[0])(output_shape if output_shape else self.info.shape)
-    self.ret.cl   # does the allocation
     self.bufs = [type(self.ret)(self.info.shape, hostbuf=self.ret)] + self.bufs
 
     # check valid AST kernel