diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
index af2c8a73221..23cbb1b6527 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
@@ -35,7 +35,11 @@ layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
  * output at a single output location.
  */
 void main() {
-  const ivec3 pos = idx_to_ipos_x_wise(gl_GlobalInvocationID.x, out_limits.x, out_limits.y);
+  const uint div_by_x = gl_GlobalInvocationID.x / out_limits.x;
+  const ivec3 pos = ivec3(
+    gl_GlobalInvocationID.x % out_limits.x,
+    div_by_x % out_limits.y,
+    div_by_x / out_limits.y);
 
   if (any(greaterThanEqual(pos, out_limits))) {
     return;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.glsl
index f49bd7bbf19..11b9992a5ff 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.glsl
@@ -47,7 +47,11 @@ void main() {
   // since work size is calculated by x * ((y + B_Y - 1) / B_Y) * z
   const ivec2 out_limits_xy_scaled = (out_limits.xy + ivec2(BATCH_SIZE_X, BATCH_SIZE_Y) - 1) / ivec2(BATCH_SIZE_X, BATCH_SIZE_Y);
 
-  ivec3 pos = idx_to_ipos_x_wise(gl_GlobalInvocationID.x, out_limits_xy_scaled.x, out_limits_xy_scaled.y);
+  const uint div_by_x = gl_GlobalInvocationID.x / out_limits_xy_scaled.x;
+  ivec3 pos = ivec3(
+    gl_GlobalInvocationID.x % out_limits_xy_scaled.x,
+    div_by_x % out_limits_xy_scaled.y,
+    div_by_x / out_limits_xy_scaled.y);
 
   // scale pos.xy by batch sizes, because that's the top pixel to be processed
   pos.x *= BATCH_SIZE_X;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
index caffcdbbdc0..b50a892cad5 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
@@ -44,7 +44,11 @@ void main() {
   const ivec2 out_limits_scaled = (out_limits.xy + TILE_SIZE - 1) / TILE_SIZE;
   const uint shared_mem_stride = gl_WorkGroupSize.x * gl_WorkGroupSize.y * gl_WorkGroupSize.z;
 
-  const ivec3 gpos = idx_to_ipos_x_wise(gl_GlobalInvocationID.x, out_limits_scaled.x, out_limits_scaled.y);
+  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);
 
   // Output position for TILE_SIZE = 2
   // +--------+--------+
diff --git a/backends/vulkan/runtime/graph/ops/glsl/indexing_utils.h b/backends/vulkan/runtime/graph/ops/glsl/indexing_utils.h
index 1d3a60cb293..0b372ab70a4 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/indexing_utils.h
+++ b/backends/vulkan/runtime/graph/ops/glsl/indexing_utils.h
@@ -223,11 +223,6 @@ ivec3 lpos_to_pos(const ivec3 lpos, const ivec4 axis_map) {
   return pos;
 }
 
-ivec3 idx_to_ipos_x_wise(uint idx, int size_x, int size_y) {
-  const uint div_by_x = idx / size_x;
-  return ivec3(idx % size_x, div_by_x % size_y, div_by_x / size_y);
-}
-
 #ifdef USING_BUFFER
 #define load_texel(buf, idx) buf[idx]
 #elif defined(USING_TEXTURE2D)