Add receipt 4 option to codegen (#1875)

* Add receipt 4 option to codegen

* Remove repeated code

* Review comments

example/ck_tile/01_fmha/codegen/ops/fmha_bwd.py

@@ -506,6 +506,14 @@ def get_bwd_dq_dk_dv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                     cond &= deterministic == "f"
                     if not cond:
                         continue
+            if receipt == 4:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= bias in ['no', 'bias']
+                    cond &= dropout in ['no', 'dropout_wg32',  'dropout_wg16']
+                    cond &= dpad == dvpad
+                    cond &= deterministic == "f"
+                    if not cond:
+                        continue
             api_pool.register_dq_dk_dv_traits(k.api_trait())
             gen.append(k)
 

example/ck_tile/01_fmha/codegen/ops/fmha_fwd.py

@@ -487,13 +487,20 @@ def get_fwd_blobs(kernel_filter : Optional[str], receipt, mask_impl) -> Tuple[Fm
                 if kernel_filter != None:
                     if not fnmatch.fnmatch(k.name, kernel_filter):
                         continue
-                if receipt == 2:
+                if receipt in (2, 3):
                     cond = dtype in ['fp16', 'bf16']
                     cond &= pipeline.F_vlayout == 'row'
                     cond &= pipeline.F_bias in ['no', 'alibi']
                     cond &= pipeline.F_squant == 'f'
                     if not cond:
                         continue
+                if receipt == 4:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= pipeline.F_vlayout == 'row'
+                    cond &= pipeline.F_bias in ['no', 'bias']
+                    cond &= pipeline.F_squant == 'f'
+                    if not cond:
+                        continue
                 api_pool.register_traits(k.api_trait())
                 gen.append(k)
 

example/ck_tile/01_fmha/generate.py

@@ -103,7 +103,8 @@ if __name__ == "__main__":
         required=False,
         help="codegen receipt. 0: generate only 8xhdim coverage\n"  + \
              "  1: generate more instance to cover all hdim\n"  + \
-             "  2: Only generate instance for Flash attention integration"
+             "  2: Only generate instance for Flash attention integration\n"  + \
+             "  4: Only generate instance for PyTorch integration"
     )
 
     args = parser.parse_args()