Update on "[ET-VK] Miscellaneous fixes"

Collecting fixes for various models/ops in this diff/PR.

They have all been squashed into this single change to make it easier to cherry pick.

# Fixes

## Wav2Letter

Type: Output correctness failure

This is caused by a bug in swiftshader, and not reproducible on any other platform. Specifically, the issue is in the softmax shader; the exact cause of the issue is unknown, but it is related to using shared memory within shaders. The workaround for this issue is to use separate shared memory arrays for the shared max and shared sum.

## ConvNeXT

Type: Exception during runtime

This is caused by an incompatible memory layout being used for mean2d. More technically, the packed dimension of the tensor cannot be one of the dims being reduced. The current operator registry system did not have a way to select valid tensor representations based on the actual arguments of an op.

To fix, we have to introduce a mechanism for ops to specify valid representations once a node's arguments are known. Once the model is exported with supported memory layout, the model test passes.

## Inception_V3/ViT

Type: Exception during runtime

The root cause of this was an interaction betwen the fuse batch norm pass and how `vulkan_preprocess.py` was applying passes. Essentially, the fuse batch norm pass creates a new param node for the fused weight, but after the pass is applied `_copy_module` is used to copy the transformed graph back into the ExportedProgram. However, it seems that _copy_module lowercases the node names without updating the exported program's graph signature. Therefore, subsequent passes couldn't recognize the weight tensor of convolution tensors as a constant/parameter node.

The solution was to migrate vulkan_preprocess.py to use the _transform() API instead of using _copy_module.

## DenseNet 161 (w/ dynamic shapes)

Type: Output Mismatch

Cause: the native_batch_norm op doesn't support dynamic shapes. However, the backend test runner doesn't set the correct compile option to filter ops without dynamic shape support.

Differential Revision: [D83703496](https://our.internmc.facebook.com/intern/diff/D83703496/)

[ghstack-poisoned]

Author: ssjia

.github/workflows/pull.yml

@@ -976,7 +976,7 @@ jobs:
         done
 
         # For selected vision models, test with dynamic shapes
-        models="mv2 mv3 resnet18 resnet50 ic3 ic4 densenet161"
+        models="mv2 resnet18 resnet50 ic3 densenet161"
         for model in $models; do
           python -m examples.vulkan.export --model_name=$model --test -d
         done

backends/vulkan/runtime/graph/ops/glsl/conv2d.glsl

@@ -60,7 +60,7 @@ void main() {
     int num_steps = ((-ipos.y) + dilation.y - 1) / dilation.y;
     start.y = ipos.y + num_steps * dilation.y;
   }
-  const ivec2 end = min(ipos + overlay_region.xy, ivec2(in_sizes.xy));
+  const ivec2 end = min(ipos + overlay_region.xy, in_sizes.xy);
   // Compute the start of the kernel based on how far we are skipping ahead when
   // reading the input. Note that these are "canonical" indices.
   ivec2 kstart = (start - ipos) / dilation;

backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl

@@ -54,7 +54,7 @@ void main() {
   // Compute the start and end of the input indices to load. Padding is assumed
   // to be constant 0 padding, so reads from the padding region are skipped.
   const ivec2 start = ipos;
-  const ivec2 end = ipos + overlay_region.xy;
+  const ivec2 end = min(ipos + overlay_region.xy, in_sizes.xy);
 
   VEC4_T sum = texelFetch(t_bias, ivec2(pos.z, 0), 0);
   int kx = 0;

backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.glsl

@@ -97,6 +97,10 @@ void main() {
   for (int y = start.y, i = 0; i < TILE_SIZE + BATCH_SIZE_Y - 1; y += dilation.y, i++) {
     for (int x = start.x, j = 0; j < TILE_SIZE + BATCH_SIZE_X - 1; x += dilation.x, j++) {
       in_texels[j] = texelFetch(t_in, ivec3(x, y, pos.z), 0);
+      // Set to zero if reading out of bounds
+      if (any(greaterThanEqual(ivec2(x, y), in_sizes.xy))) {
+        in_texels[j] = VEC4_T(0);
+      }
     }
 
     // from 2nd iteration onwards accumulate dot product in 2nd sum