[ET-VK] Tuning conv 2d pw op tile size to improve perf. (#11184)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: #11112 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/91/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/91/head
Merge bot PR base:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/90/orig
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/91/orig
@diff-train-skip-merge

---------

Co-authored-by: Vivek Trivedi <5340687+trivedivivek@users.noreply.github.com>

Author: pytorchbot

backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.yaml

@@ -9,8 +9,8 @@ conv2d_pw:
     OPERATOR: X
     NDIM: 3
     DTYPE: float
-    TILE_SIZE_X: 2
-    TILE_SIZE_Y: 2
+    TILE_SIZE_X: 1
+    TILE_SIZE_Y: 4
   generate_variant_forall:
     DTYPE:
       - VALUE: half

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

@@ -305,8 +305,8 @@ utils::uvec3 create_conv2d_global_wg_size(
   if (method == Conv2dMethod::Pointwise) {
     const utils::uvec3 image_extents = graph.logical_limits_of(out);
     return {
-        utils::div_up(image_extents[0u], 2u),
-        utils::div_up(image_extents[1u], 2u),
+        utils::div_up(image_extents[0u], 1u),
+        utils::div_up(image_extents[1u], 4u),
         image_extents[2u]};
   } else if (method == Conv2dMethod::Depthwise && stride_equals_dilation) {
     const utils::uvec3 image_extents = graph.create_global_wg_size(out);