update

build/torch28-metal-aarch64-darwin/gptoss_kernels/{_gptoss_kernels_4e4a2c4_dirty.abi3.so → _gptoss_kernels_b9ce6e5_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0b828abe067634941d74537de6fe5857e9961ddf17f7525347dd5d7c08a44d5
+oid sha256:b350e0252f34a4b5427cf053b1c45f7df21428a4e1321b0bfd9c6658d775db47
 size 425784

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

@@ -1,9 +1,9 @@
 import torch
-from . import _gptoss_kernels_4e4a2c4_dirty
-ops = torch.ops._gptoss_kernels_4e4a2c4_dirty
+from . import _gptoss_kernels_b9ce6e5_dirty
+ops = torch.ops._gptoss_kernels_b9ce6e5_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_gptoss_kernels_4e4a2c4_dirty::{op_name}"
+    return f"_gptoss_kernels_b9ce6e5_dirty::{op_name}"

build/torch29-metal-aarch64-darwin/gptoss_kernels/{_gptoss_kernels_4e4a2c4_dirty.abi3.so → _gptoss_kernels_b9ce6e5_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcd0ce12caf24430e66480753c2a096377bc824f6a48b6394cf8cd5815015b18
+oid sha256:e7d4ab795104de161cd4d609c639f235b8774c955ffe78ba365a75b97110ba15
 size 410328

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

@@ -1,9 +1,9 @@
 import torch
-from . import _gptoss_kernels_4e4a2c4_dirty
-ops = torch.ops._gptoss_kernels_4e4a2c4_dirty
+from . import _gptoss_kernels_b9ce6e5_dirty
+ops = torch.ops._gptoss_kernels_b9ce6e5_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_gptoss_kernels_4e4a2c4_dirty::{op_name}"
+    return f"_gptoss_kernels_b9ce6e5_dirty::{op_name}"

gptoss_kernels/source/tensor_wrappers.cpp

@@ -169,7 +169,7 @@ void run_metal_kernel(const char* kernel_symbol, EncodeFn&& encode_fn) {
 
         if (borrowed_buffer) {
             TORCH_CHECK(stream != nullptr, "stream must be valid when using borrowed command buffer");
-            stream->synchronize(at::mps::SyncType::COMMIT_AND_WAIT);
+            stream->synchronize(at::mps::SyncType::COMMIT);
         } else {
             TORCH_CHECK(gptoss_metal_command_buffer_commit(&cb) == gptoss_status_success,
                 "commit failed");