[Bugfix][ROCm][GPT-OSS] Use old triton_kernels implementation on ROCm if the new API is not available (#34153)

Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>

vllm/model_executor/layers/fused_moe/gpt_oss_triton_kernels_moe.py

@@ -19,11 +19,14 @@ from vllm.model_executor.layers.fused_moe.utils import _resize_cache
 from vllm.model_executor.layers.quantization.utils.quant_utils import (
     QuantKey,
 )
+from vllm.platforms import current_platform
 from vllm.triton_utils import tl, triton
 from vllm.utils.import_utils import has_triton_kernels
 
 logger = init_logger(__name__)
 
+use_legacy_triton_kernels = False
+
 if has_triton_kernels():
     try:
         import triton_kernels.swiglu
@@ -38,10 +41,20 @@ if has_triton_kernels():
         from triton_kernels.tensor import (
             BIT,
             Bitmatrix,
+        )
+        from triton_kernels.topk import topk
+
+        try:
+            from triton_kernels.tensor import (
                 SparseMatrix,
                 make_ragged_tensor_metadata,
             )
-        from triton_kernels.topk import topk
+        except ImportError:
+            if current_platform.is_rocm():
+                logger.warning_once("Using legacy triton_kernels on ROCm")
+                use_legacy_triton_kernels = True
+            else:
+                raise
     except (AttributeError, ImportError) as e:
         logger.error(
             "Failed to import Triton kernels. Please make sure your triton "
@@ -101,6 +114,12 @@ def legacy_routing_from_bitmatrix(
     Replacement for the removed triton_kernels.routing.routing_from_bitmatrix.
     Creates routing data from a bitmatrix representation.
     """
+    if use_legacy_triton_kernels:
+        from triton_kernels.routing import routing_from_bitmatrix
+
+        return routing_from_bitmatrix(
+            bitmatrix, expt_scal, expt_indx, n_expts_tot, n_expts_act
+        )
     sparse_logits = SparseMatrix(indx=expt_indx, vals=expt_scal, mask=bitmatrix)
     dispatch_indx = sparse_logits.mask_metadata.row_sorted_indx
     combine_indx = sparse_logits.mask_metadata.col_sorted_indx
@@ -130,6 +149,10 @@ def legacy_routing(
     Replacement for the removed triton_kernels.routing.routing function.
     Computes routing data from gating logits.
     """
+    if use_legacy_triton_kernels:
+        from triton_kernels.routing import routing
+
+        return routing(logits, n_expts_act, sm_first=sm_first)
     if sm_first:
         logits = torch.softmax(logits, dim=-1)
     sparse_logits = topk(logits, n_expts_act, apply_softmax=not sm_first)
@@ -231,12 +254,23 @@ def triton_kernel_fused_experts(
     )
     output_tensor = _resize_cache(output_tensor, (batch_dim, M, K))
 
-    act = FusedActivation(
+    act = (
+        FusedActivation(
             FnSpecs(
-            "swiglu", triton_kernels.swiglu.swiglu_fn, ("alpha", "limit"), reduction_n=2
+                "swiglu",
+                triton_kernels.swiglu.swiglu_fn,
+                ("alpha", "limit"),
+                reduction_n=2,
             ),
             (swiglu_alpha, swiglu_limit),
         )
+        if not use_legacy_triton_kernels
+        else FusedActivation(
+            FnSpecs("swiglu", triton_kernels.swiglu.swiglu_fn, ("alpha", "limit")),
+            (swiglu_alpha, swiglu_limit),
+            2,
+        )
+    )
     gammas = routing_data.gate_scal if routing_data else None
 
     matmul_ogs(
@@ -296,9 +330,18 @@ def make_routing_data(
 
     bitmatrix_shape = [n_rows, bm_cols * 32]
     bitmatrix_shape_max = [n_rows, None]
-    bitmatrix = Bitmatrix(
+    bitmatrix = (
+        Bitmatrix(
             bitmatrix, dtype=BIT, shape=bitmatrix_shape, shape_max=bitmatrix_shape_max
         )
+        if not use_legacy_triton_kernels
+        else Bitmatrix(
+            bitmatrix,
+            shape=bitmatrix_shape,
+            shape_max=bitmatrix_shape_max,
+            scratchpad=None,
+        )
+    )
 
     # matmul_ogs expects invalid topk_weights to be -1s
     topk_weights = torch.where(topk_ids == -1, -1.0, topk_weights)