update

build/torch28-metal-aarch64-darwin/gptoss_kernels/{_gptoss_kernels_c93d5eb_dirty.abi3.so → _gptoss_kernels_7f87253_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fd3e7b13129a1cff2ac95be086c7062563ff68001f2df5d42d51eba2b6326a4
-size 425560
+oid sha256:e018e38aa94dfae2f05d57e4f666d5a3f26add864cd977ca5052f4ffa6aceb38
+size 425432

build/torch28-metal-aarch64-darwin/gptoss_kernels/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _gptoss_kernels_c93d5eb_dirty
-ops = torch.ops._gptoss_kernels_c93d5eb_dirty
+from . import _gptoss_kernels_7f87253_dirty
+ops = torch.ops._gptoss_kernels_7f87253_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_gptoss_kernels_c93d5eb_dirty::{op_name}"
+    return f"_gptoss_kernels_7f87253_dirty::{op_name}"

build/torch29-metal-aarch64-darwin/gptoss_kernels/{_gptoss_kernels_c93d5eb_dirty.abi3.so → _gptoss_kernels_7f87253_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af7daaaedab979a266a8ad8326309f318ec89cb31e8e9434d482c4dff46599c4
-size 410088
+oid sha256:1446e5adae835d61201c5d66dd701bc8dbb996c771474eb4ab7420cf03b32042
+size 409960

build/torch29-metal-aarch64-darwin/gptoss_kernels/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _gptoss_kernels_c93d5eb_dirty
-ops = torch.ops._gptoss_kernels_c93d5eb_dirty
+from . import _gptoss_kernels_7f87253_dirty
+ops = torch.ops._gptoss_kernels_7f87253_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_gptoss_kernels_c93d5eb_dirty::{op_name}"
+    return f"_gptoss_kernels_7f87253_dirty::{op_name}"

gptoss_kernels/source/tensor_wrappers.cpp

@@ -133,7 +133,6 @@ void run_metal_kernel(const char* kernel_symbol, EncodeFn&& encode_fn) {
             "function_create failed");
         stream = at::mps::getCurrentMPSStream();
         if (stream != nullptr) {
-            stream->synchronize(at::mps::SyncType::COMMIT_AND_WAIT);
             void* command_queue_obj = stream->commandQueue();
             if (command_queue_obj != nullptr) {
                 TORCH_CHECK(