Build on top of the GPU codegen hack

spcl · Dec 11, 2024 · 62bc08c · 62bc08c
1 parent 2408ad0
commit 62bc08c
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 4 deletions.
diff --git a/dace/codegen/targets/cuda.py b/dace/codegen/targets/cuda.py
@@ -1502,8 +1502,8 @@ def generate_scope(self, sdfg: SDFG, cfg: ControlFlowRegion, dfg_scope: StateSub
         const_params = _get_const_params(dfg_scope)
         # make dynamic map inputs constant
         # TODO move this into _get_const_params(dfg_scope)
-        # Do not add src as const if the size is being red (src_conn is _read_size)
-        const_params |= set((str(e.src)) for e in dace.sdfg.dynamic_map_inputs(state, scope_entry) if e.src_conn is None or (e.src_conn is not None and e.src_conn == "_read_size"))
+        # Do not add src as const if the size is being read (src_conn is _read_size)
+        const_params |= set((str(e.src)) for e in dace.sdfg.dynamic_map_inputs(state, scope_entry) if e.src_conn is None or not (e.src_conn is not None and e.src_conn == "_read_size"))
 
         # Store init/exit code streams
         old_entry_stream = self.scope_entry_stream

diff --git a/dace/codegen/targets/framecode.py b/dace/codegen/targets/framecode.py
@@ -964,7 +964,6 @@ def generate_code(self,
         # Only allocate arrays that really require deferred allocation (symbol has __dace_defer)
         # Reshaping these arrays are not allowed
         size_arrays = sdfg.size_arrays()
-        callsite_stream.write(f'//Declare size arrays\n', sdfg)
         for size_desc_name in size_arrays:
             size_nodedesc = sdfg.arrays[size_desc_name]
             assert ("__return" not in size_desc_name)
@@ -979,7 +978,7 @@ def generate_code(self,
                     assert len(size_nodedesc.shape) == 1
                     alloc_str = f'{ctypedef} {size_desc_name}[{size_nodedesc.shape[0]}]{{{size_str}}};\n'
                     callsite_stream.write(alloc_str)
-                    self.dispatcher.defined_vars.add(size_desc_name, disp.DefinedType.Pointer, ctypedef)
+                    self.dispatcher.defined_vars.add(size_desc_name, disp.DefinedType.Pointer, ctypedef, allow_shadowing=True)
 
         #######################################################################
         # Generate actual program body