1. Adding floating point capabilities to reduce block. 2. Fix for gen…

…erate matrix BF16 compressed mode
weiya711 · Feb 29, 2024 · d1ed8f7 · d1ed8f7
1 parent ce5ba84
commit d1ed8f7
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 4 deletions.
diff --git a/sam/onyx/generate_matrices.py b/sam/onyx/generate_matrices.py
@@ -195,10 +195,13 @@ def dump_outputs(self, format=None, tpose=False, dump_shape=True,
                 flat_array.append(val)
             if glb_override:
                 lines = [len(flat_array), *flat_array]
-                self.write_array(lines, name=f"tensor_{self.name}_mode_vals{suffix}", dump_dir=use_dir, dump_hex=print_hex)
+                self.write_array(lines, name=f"tensor_{self.name}_mode_vals{suffix}", 
+                                 dump_dir=use_dir, dump_hex=print_hex, is_val=True)
             else:
-                self.write_array(flat_array, name=f"tensor_{self.name}_mode_vals{suffix}", dump_dir=use_dir,
-                                 dump_hex=print_hex)
+                self.write_array(flat_array, name=f"tensor_{self.name}_mode_vals{suffix}", 
+                                 dump_dir=use_dir, dump_hex=print_hex, is_val=True)
+
+
         elif self.format == "COO":
             crd_dict = dict()
             order = len(self.array.shape)

diff --git a/sam/onyx/hw_nodes/reduce_node.py b/sam/onyx/hw_nodes/reduce_node.py
@@ -112,7 +112,16 @@ def configure(self, attributes):
         # TODO: make this use the metamapper
         instr_type = strip_modifiers(lassen_fc.Py.input_t.field_dict['inst'])
         asm_ = Assembler(instr_type)
-        op = int(asm_.assemble(asm.add()))
+
+        if 'fp' in attributes:
+            is_fp = attributes['fp'].strip('"')
+            if (is_fp == 'true'):
+                op = int(asm_.assemble(asm.fp_add()))
+            else:
+                op = int(asm_.assemble(asm.add()))
+        else:
+            op = int(asm_.assemble(asm.add()))
+
         cfg_kwargs = {
             'stop_lvl': stop_lvl,
             'pe_connected_to_reduce': pe_connected_to_reduce,