Addng Origami backend - Work in progress

Author: umechand-amd

torch/_inductor/template_heuristics/triton.py

@@ -1638,13 +1638,13 @@ def _get_template_configs_impl(
                                     'EVEN_K': True, 'USE_FAST_ACCUM': False,
                                     'ACC_TYPE': 'tl.float32',
                                     'num_stages': 2,
-                                    'num_warps': 1,
+                                    'num_warps': 8,
                                     'BLOCK_M':OrigamiGemmSelector.config[0],
                                     'BLOCK_N':OrigamiGemmSelector.config[1],
                                     'BLOCK_K':OrigamiGemmSelector.config[2],
                                     'GROUP_M':OrigamiGemmSelector.config[3],
                                     'matrix_instr_nonkdim': 16,
-                                    'waves_per_eu': 8,
+                                    'waves_per_eu': 0,
                                     'kpack': 2,
                                     }
             yield origami_config_kwargs

torch/_inductor/utils.py

@@ -2257,6 +2257,8 @@ def use_ck_tile_gemm_template(layout: Layout, m: int, n: int, k: int) -> bool:
         and V.graph.sizevars.size_hint(m * n * k, fallback=-1) > 0
     )
 
+def use_origami_gemm_template(layout: Layout, m: int, n: int, k: int) -> bool:
+    return _use_autotune_backend("ROCM_ORIGAMI")
 
 def use_ck_conv_template(layout: Layout) -> bool:
     return _use_conv_autotune_backend("CK") and use_ck_template(layout)