[ET-VK] Minor tuning for conv2d pw op to improve performance. (#11185)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: #11113 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/92/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/92/head
Merge bot PR base:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/91/orig
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/trivedivivek/92/orig
@diff-train-skip-merge

---------

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

Author: pytorchbot

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

@@ -49,8 +49,14 @@ void main() {
   const uint div_by_x = gl_GlobalInvocationID.x / out_limits_scaled.x;
   const ivec3 gpos = ivec3(
     gl_GlobalInvocationID.x % out_limits_scaled.x,
-    div_by_x % out_limits_scaled.y,
-    div_by_x / out_limits_scaled.y);
+    div_by_x,
+    gl_GlobalInvocationID.y);
+
+  // If the top left position is out of bounds, then this invocation will have
+  // no work to do.
+  if (gpos.y >= out_limits_scaled.y || gpos.z >= out_limits.z) {
+    return;
+  }
 
   // If the top left position is out of bounds, then this invocation will have
   // no work to do.

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

@@ -398,8 +398,10 @@ void add_conv2d_node(
   utils::uvec3 wg_size = create_conv2d_global_wg_size(
       graph, method, out, weight_data, stride_equals_dilation);
 
-  if (method == Conv2dMethod::Pointwise || method == Conv2dMethod::Depthwise) {
+  if (method == Conv2dMethod::Depthwise) {
     wg_size = {wg_size[0] * wg_size[1] * wg_size[2], 1, 1};
+  } else if (method == Conv2dMethod::Pointwise) {
+    wg_size = {wg_size[0] * wg_size[1], wg_size[2], 1};
   }
 
   vkapi::ParamsBindList param_buffers;