Merge branch 'main' into Arm-backend-Fix-buck2-files-after-moving-TOSA-partitioner

Author: zingo

backends/arm/operator_support/to_copy_support.py

@@ -20,6 +20,8 @@
 
 logger = logging.getLogger(__name__)
 
+SupportedTypeDict = dict[torch.dtype, list[torch.dtype]]
+
 
 @register_tosa_support_check
 class ToCopySupported(SupportedTOSAOperatorCheck):
@@ -33,8 +35,6 @@ class ToCopySupported(SupportedTOSAOperatorCheck):
         TosaSpecification.create_from_string("TOSA-1.0+FP"),
     ]
 
-    SupportedTypeDict = dict[torch.dtype, list[torch.dtype]]
-
     @staticmethod
     def _merge_supported_types(
         # pyre-ignore[11]
@@ -53,11 +53,22 @@ def _merge_supported_types(
         torch.int8: [torch.bool, torch.int16, torch.int32],
         torch.int16: [torch.bool, torch.int8, torch.int32],
         torch.int32: [torch.bool, torch.int8, torch.int16],
+        torch.int64: [torch.bool, torch.int8, torch.int16, torch.int32],
     }
     SUPPORTED_FLOAT_TYPES: SupportedTypeDict = {
         torch.int8: [torch.float16, torch.bfloat16, torch.float32],
         torch.int16: [torch.float16, torch.bfloat16, torch.float32],
         torch.int32: [torch.float16, torch.bfloat16, torch.float32],
+        # INT64 inputs to casts *should* be ok, since they should be rejected by
+        # CheckInt64InputsAndOutputs if the cast can't be done AOT.
+        torch.int64: [
+            torch.int8,
+            torch.int16,
+            torch.int32,
+            torch.float16,
+            torch.bfloat16,
+            torch.float32,
+        ],
         torch.bfloat16: [torch.int8, torch.int16, torch.int32, torch.float32],
         torch.float16: [torch.int8, torch.int16, torch.int32, torch.float32],
         torch.float32: [
@@ -71,22 +82,20 @@ def _merge_supported_types(
     ALL_SUPPORTED_TYPES = _merge_supported_types(
         SUPPORTED_INT_TYPES, SUPPORTED_FLOAT_TYPES
     )
-    POSSIBLE_TYPE_CONVERSIONS = {torch.int64: torch.int32}
 
     def is_node_tosa_supported(
         self, node: fx.Node, tosa_spec: TosaSpecification
     ) -> bool:
-        supported_dtypes = (
-            self.ALL_SUPPORTED_TYPES
-            if tosa_spec.support_float()
-            else self.SUPPORTED_INT_TYPES
-        )
-        # Take into account possible type conversions
-        supported_dtypes.update(
-            (k, supported_dtypes[v])
-            for k, v in self.POSSIBLE_TYPE_CONVERSIONS.items()
-            if v in supported_dtypes
-        )
+
+        supported_dtypes: SupportedTypeDict = {}
+        if tosa_spec.support_integer():
+            supported_dtypes = self._merge_supported_types(
+                self.SUPPORTED_INT_TYPES, supported_dtypes
+            )
+        if tosa_spec.support_float():
+            supported_dtypes = self._merge_supported_types(
+                self.SUPPORTED_FLOAT_TYPES, supported_dtypes
+            )
 
         if len(node.all_input_nodes) != 1:
             self.reporter.report_reject(
@@ -156,7 +165,7 @@ def is_node_tosa_supported(
         if "dim_order" in node.kwargs:
             dim_order = node.kwargs["dim_order"]
             # pyre-ignore[6]
-            if dim_order != list(range(len(dim_order))):  # type: ignore[arg-type]
+            if dim_order is not None and dim_order != list(range(len(dim_order))):  # type: ignore[arg-type]
                 self.reporter.report_reject(
                     node,
                     (

backends/arm/scripts/run_fvp.sh

@@ -92,7 +92,7 @@ if [[ ${target} == *"ethos-u55"*  ]]; then
         -C mps3_board.uart0.shutdown_on_eot=1               \
         -a "${elf_file}"                                    \
         ${data_file}                                        \
-        --timelimit ${timeout} 2>&1 | tee ${log_file} || true # seconds
+        --timelimit ${timeout} 2>&1 | sed 's/\r$//' | tee ${log_file} || true # seconds
     echo "[${BASH_SOURCE[0]}] Simulation complete, $?"
 elif [[ ${target} == *"ethos-u85"*  ]]; then
     ${nobuf} ${fvp_model}                                   \
@@ -104,13 +104,28 @@ elif [[ ${target} == *"ethos-u85"*  ]]; then
         -C mps4_board.uart0.shutdown_on_eot=1               \
         -a "${elf_file}"                                    \
         ${data_file}                                        \
-        --timelimit ${timeout} 2>&1 | tee ${log_file} || true # seconds
+        --timelimit ${timeout} 2>&1 | sed 's/\r$//' | tee ${log_file} || true # seconds
     echo "[${BASH_SOURCE[0]}] Simulation complete, $?"
 else
     echo "Running ${elf_file} for ${target} is not supported"
     exit 1
 fi
 
+echo "Checking for a etdump in log"
+! grep "#\[RUN THIS\]" ${log_file} >/dev/null
+if [ $? != 0 ]; then
+    echo "Found ETDump in log!"
+    echo "#!/bin/sh" > etdump_script.sh
+    sed -n '/^#\[RUN THIS\]$/,/^#\[END\]$/p' ${log_file} >> etdump_script.sh
+    # You can run etdump_script.sh if you do
+    # $ chmod a+x etdump_script.sh
+    # $ ./etdump_script.sh
+    # But lets not trust the script as a bad patch would run bad code on your machine
+    grep ">etdump.bin" etdump_script.sh | cut -d\" -f2- | cut -d\" -f1 >etdump.base64
+    base64 -d etdump.base64 >etdump.bin
+    python3 -m devtools.inspector.inspector_cli --etdump_path etdump.bin  --source_time_scale cycles --target_time_scale cycles
+fi
+
 echo "Checking for problems in log:"
 ! grep -E "^(F|E|\\[critical\\]|Hard fault.|Info: Simulation is stopping. Reason: CPU time has been exceeded.).*$" ${log_file}
 if [ $? != 0 ]; then

backends/arm/test/models/stable_diffusion/test_T5EncoderModel.py

@@ -33,8 +33,9 @@ class TestT5EncoderModel(unittest.TestCase):
     # .to_executorch step, i.e. after Arm partitioner.
     ops_after_partitioner = {
         "executorch_exir_dialects_edge__ops_aten__to_copy_default": 2,
+        "executorch_exir_dialects_edge__ops_dim_order_ops__to_dim_order_copy_default": 1,
         "executorch_exir_dialects_edge__ops_aten_view_copy_default": 1,
-        "torch.ops.higher_order.executorch_call_delegate": 2,
+        "torch.ops.higher_order.executorch_call_delegate": 3,
     }
 
     def _prepare_inputs(

backends/arm/test/ops/test_to_copy.py

@@ -70,6 +70,16 @@ def test_copy_tosa_FP(test_data: Tuple):
         aten_op=[],
         exir_op=[],
     )
+    # int to int cast is not supported in TOSA+FP profile
+    if not new_dtype.is_floating_point and not torch.is_floating_point(test_tensor):
+        pipeline.change_args(
+            "check_count.exir",
+            {
+                "torch.ops.higher_order.executorch_call_delegate": 0,
+                "executorch_exir_dialects_edge__ops_dim_order_ops__to_dim_order_copy_default": 1,
+            },
+        )
+        pipeline.pop_stage("run_method_and_compare_outputs")
     pipeline.run()
 
 
@@ -84,6 +94,15 @@ def test_copy_vgf_FP(test_data: Tuple):
         exir_op=[],
         tosa_version="TOSA-1.0+FP",
     )
+    # int to int cast is not supported in TOSA+FP profile
+    if not new_dtype.is_floating_point and not torch.is_floating_point(test_tensor):
+        pipeline.change_args(
+            "check_count.exir",
+            {
+                "torch.ops.higher_order.executorch_call_delegate": 0,
+                "executorch_exir_dialects_edge__ops_dim_order_ops__to_dim_order_copy_default": 1,
+            },
+        )
     pipeline.run()
 
 

backends/arm/test/passes/test_convert_int64_output_ops_to_int32.py

@@ -32,17 +32,8 @@ def forward(self, x: torch.Tensor):
 test_data_suite_convert = {
     "fp32_input": lambda: (torch.rand((1, 2, 3, 4), dtype=torch.float32), torch.int64),
     "fp16_input": lambda: (torch.rand((1, 2, 3, 4), dtype=torch.float16), torch.int64),
-    "int16_input": lambda: (
-        torch.randint(-127, 128, (1, 2, 3, 4), dtype=torch.int16),
-        torch.int64,
-    ),
-    "int8_input": lambda: (
-        torch.randint(-127, 128, (1, 2, 3, 4), dtype=torch.int8),
-        torch.int64,
-    ),
 }
 
-
 test_data_suite_remove = {
     "int32_input": lambda: (
         torch.randint(-127, 128, (1, 2, 3, 4), dtype=torch.int32),
@@ -52,7 +43,7 @@ def forward(self, x: torch.Tensor):
 
 
 @common.parametrize("test_data", test_data_suite_convert)
-def test_convert_or_remove_casting_to_int64_covnert_tosa_FP(test_data: Tuple):
+def test_convert_or_remove_casting_to_int64_convert_tosa_FP(test_data: Tuple):
     test_tensor, target_dtype = test_data()
     module = CastingToInt64Model(target_dtype)
 

examples/arm/executor_runner/arm_executor_runner.cpp

@@ -852,7 +852,7 @@ void write_etdump(RunnerContext& ctx) {
       encoded_buf[encoded_len] = 0x00; // Ensure null termination
       ET_LOG(Info, "Writing etdump.bin [base64]");
       printf(
-          "#---\necho \"%s\" | base64 -d >etdump.bin\npython3 -m devtools.inspector.inspector_cli --etdump_path etdump.bin  --source_time_scale cycles --target_time_scale cycles\n#---\n",
+          "#[RUN THIS]\necho \"%s\" | base64 -d >etdump.bin\npython3 -m devtools.inspector.inspector_cli --etdump_path etdump.bin  --source_time_scale cycles --target_time_scale cycles\n#[END]\n",
           encoded_buf);
     } else {
       ET_LOG(

examples/arm/vgf_minimal_example.ipynb

@@ -23,13 +23,11 @@
     "\n",
     "Before you begin:\n",
     "1. (In a clean virtual environment with a compatible Python version) Install executorch using `./install_executorch.sh`\n",
-    "2. Install MLSDK and Tosa using `examples/arm/setup.sh --disable-ethos-u-deps --enable-mlsdk-deps (For further guidance, refer to https://docs.pytorch.org/executorch/main/tutorial-arm.html)\n",
+    "2. Install MLSDK and Tosa using `examples/arm/setup.sh --disable-ethos-u-deps --enable-mlsdk-deps` (For further guidance, refer to https://docs.pytorch.org/executorch/main/tutorial-arm.html)\n",
     "3. Export vulkan environment variables and add MLSDK components to PATH and LD_LIBRARY_PATH using `examples/arm/ethos-u-scratch/setup_path.sh`\n",
     "\n",
     "With all commands executed from the base `executorch` folder.\n",
     "\n",
-    "\n",
-    "\n",
     "*Some scripts in this notebook produce long output logs: Configuring the 'Customizing Notebook Layout' settings to enable 'Output:scrolling' and setting 'Output:Text Line Limit' makes this more manageable*"
    ]
   },
@@ -165,6 +163,17 @@
     "All of this happens behind the scenes in `to_edge_transform_and_lower`. Printing the graph module shows that what is left in the graph is two quantization nodes for `x` and `y` going into an `executorch_call_delegate` node, followed by a dequantization node."
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%%bash\n",
+    "# Ensure the vulkan environment variables and MLSDK components are available on $PATH\n",
+    "source ethos-u-scratch/setup_path.sh"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -220,10 +229,21 @@
     "\n",
     "\n",
     "After the AOT compilation flow is done, we need to build the executor_runner target. For this example the generic version will be used.\n",
-    "To do this, please ensure the following commands are executed before moving onto the next step.\n",
+    "To do this, please ensure the following commands are executed before moving onto the next step."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%%bash\n",
+    "# Ensure the vulkan environment variables and MLSDK components are available on $PATH\n",
+    "source ethos-u-scratch/setup_path.sh\n",
     "\n",
-    "Clean and configure the CMake build system. Compiled programs will appear in the executorch/cmake-out directory we create here.\n",
-    "```\n",
+    "# Compiled programs will appear in the executorch/cmake-out directory we create here.\n",
+    "# Build example executor runner application to examples/arm/vgf_minimal_example\n",
     "cmake \\\n",
     "  -DCMAKE_INSTALL_PREFIX=cmake-out \\\n",
     "  -DCMAKE_BUILD_TYPE=Debug \\\n",
@@ -237,11 +257,9 @@
     "  -DEXECUTORCH_BUILD_VGF=ON \\\n",
     "  -DEXECUTORCH_ENABLE_LOGGING=ON \\\n",
     "  -DPYTHON_EXECUTABLE=python \\\n",
-    "  -Bcmake-out .\n",
-    "```\n",
+    "  -B../../cmake-out-vkml ../..\n",
     "\n",
-    "Build the executor_runner target\n",
-    "`cmake --build cmake-out --target executor_runner`\n"
+    "cmake --build ../../cmake-out-vkml --target executor_runner"
    ]
   },
   {