Update on "[ET-VK] Add symint infrastructure to VulkanBackend and ComputeGraph"

Extend the Vulkan backend runtime infrastructure to better support symbolic
integer (symint) arguments. This is a prerequisite for operators that need
to handle dynamic shapes via symint values.

Changes:
- VulkanBackend.cpp: Compute output offset from end of args instead of
  assuming outputs follow inputs directly. Add scalar-to-tensor input
  handling so that Int/Bool EValues can populate tensor inputs. Support
  symint inputs provided as raw Int EValues (not just scalar tensors).
  Add symint output handling to write values back as tensor or Int EValue.
- ComputeGraph.h: Add SymInt case to extract_scalar<T>() so operators
  can transparently read symint values as scalars.
- ComputeGraph.cpp: Add Int fallback in read_symint() so values stored
  as plain Int (rather than SymInt objects) can be read uniformly.

Differential Revision: [D95970167](https://our.internmc.facebook.com/intern/diff/D95970167/)

cc manuelcandales digantdesai cbilgin

[ghstack-poisoned]

Author: ssjia

backends/arm/test/models/test_w2l_arm.py

@@ -24,6 +24,11 @@
 
 input_t = Tuple[torch.Tensor]  # Input x
 
+quant_test_data = {
+    "per_channel_quantization=true": True,
+    "per_channel_quantization=false": False,
+}
+
 
 def get_test_inputs(batch_size, num_features, input_frames):
     return (torch.randn(batch_size, num_features, input_frames),)
@@ -99,14 +104,15 @@ def test_w2l_u55_INT():
 
 @pytest.mark.slow
 @common.XfailIfNoCorstone320
-@pytest.mark.skip(reason="Intermittent timeout issue: MLETORCH-856")
-def test_w2l_u85_INT():
+@common.parametrize("per_channel_quantization", quant_test_data)
+def test_w2l_u85_INT(per_channel_quantization):
     pipeline = EthosU85PipelineINT[input_t](
-        TestW2L.create_model(),
+        TestW2L.create_model("power_spectrum"),
         TestW2L.model_example_inputs,
         aten_ops=[],
         exir_ops=[],
         use_to_edge_transform_and_lower=True,
+        per_channel_quantization=per_channel_quantization,
     )
     pipeline.run()
 

backends/arm/test/ops/test_ceil.py

@@ -87,6 +87,7 @@ def test_ceil_tosa_INT(test_data: input_t1):
         module.exir_op,
         atol=0.06,
         rtol=0.01,
+        frobenius_threshold=0.2,
     )
     pipeline.run()
 

backends/arm/test/pytest.ini

@@ -2,4 +2,5 @@
 addopts = --strict-markers
 markers =
     slow: Tests that take long time
+    flaky: Tests that are known to be flaky/intermittent
     tosa_ref_model: Tests that use TOSA reference model # Temporary!

backends/vulkan/runtime/vk_api/memory/Allocator.cpp

@@ -11,7 +11,8 @@
 namespace vkcompute {
 namespace vkapi {
 
-bool test_host_cached_available(VkPhysicalDevice physical_device) {
+VmaAllocationCreateFlags test_host_cached_available(
+    VkPhysicalDevice physical_device) {
   VkPhysicalDeviceMemoryProperties mem_props;
   vkGetPhysicalDeviceMemoryProperties(physical_device, &mem_props);