[ET-VK] Tuning local workgroup size calculation for conv2d pw to improve performance. (pytorch#11188)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: pytorch#11135 by
@trivedivivek
^ Please use this as the source of truth for the PR details, comments,
and reviews
ghstack PR base:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/95/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/95/head
Merge bot PR base:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/94/orig
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/95/orig
@diff-train-skip-merge

---------

Co-authored-by: Vivek Trivedi <[email protected]>

Author: pytorchbot

backends/vulkan/runtime/graph/ops/impl/Convolution.cpp

@@ -404,6 +404,21 @@ void add_conv2d_node(
     wg_size = {wg_size[0] * wg_size[1], wg_size[2], 1};
   }
 
+  utils::uvec3 local_wg_size;
+  if (method == Conv2dMethod::Pointwise) {
+    uint32_t local_wg_size_y = 1;
+    if (wg_size[1] % 8 == 0) {
+      local_wg_size_y = 8;
+    } else if (wg_size[1] % 4 == 0) {
+      local_wg_size_y = 4;
+    } else if (wg_size[1] % 2 == 0) {
+      local_wg_size_y = 2;
+    }
+    local_wg_size = {64 / local_wg_size_y, local_wg_size_y, 1};
+  } else {
+    local_wg_size = graph.create_local_wg_size(wg_size);
+  }
+
   vkapi::ParamsBindList param_buffers;
   std::vector<PushConstantDataInfo> push_constants;
   if (method == Conv2dMethod::Pointwise) {
@@ -464,7 +479,7 @@ void add_conv2d_node(
       graph,
       shader,
       wg_size,
-      graph.create_local_wg_size(wg_size),
+      local_wg_size,
       // Inputs and Outputs
       {{out, vkapi::kWrite}, {{in, arg_weight, arg_bias}, vkapi::kRead}},
       // Shader params buffers