diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
index 5d7c69ab65..dd4bcb89a8 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
@@ -14,7 +14,7 @@
 
 #define op(X, A, B) ${OPERATOR}
 
-#include "indexing_utils_u16.h"
+#include "indexing_utils.h"
 
 layout(std430) buffer;
 
@@ -35,7 +35,7 @@ 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_u16pos_x_wise(gl_GlobalInvocationID.x, out_limits.x, out_limits.y);
+  const ivec3 pos = idx_to_ipos_x_wise(gl_GlobalInvocationID.x, out_limits.x, out_limits.y);
 
   if (any(greaterThanEqual(pos, out_limits))) {
     return;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
index ad5d4adb13..91a067c690 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
@@ -16,7 +16,7 @@
 
 #define op(X, A, B) ${OPERATOR}
 
-#include "indexing_utils_u16.h"
+#include "indexing_utils.h"
 
 layout(std430) buffer;
 
@@ -32,10 +32,8 @@ ${layout_declare_ubo(8, "float", "out_min", "float", "out_max")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
-#extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
-
 // shared memory to hold calculated positions, this would reduce register usage thus improving performance.
-shared u16vec2 pos_shared[gl_WorkGroupSize.x * gl_WorkGroupSize.y * gl_WorkGroupSize.z * TILE_SIZE * TILE_SIZE];
+shared ivec2 pos_shared[gl_WorkGroupSize.x * gl_WorkGroupSize.y * gl_WorkGroupSize.z * TILE_SIZE * TILE_SIZE];
 
 /*
  * Computes a 2D pointwise convolution of an NxN output tile. Calculating an
@@ -46,7 +44,7 @@ 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 u16vec3 gpos = idx_to_u16pos_x_wise(gl_GlobalInvocationID.x, out_limits_scaled.x, out_limits_scaled.y);
+  const ivec3 gpos = idx_to_ipos_x_wise(gl_GlobalInvocationID.x, out_limits_scaled.x, out_limits_scaled.y);
 
   // Output position for TILE_SIZE = 2
   // +--------+--------+
@@ -54,10 +52,10 @@ void main() {
   // +--------+--------+
   // | pos[2] | pos[3] |
   // +--------+--------+
-  u16vec2 pos[TILE_SIZE * TILE_SIZE];
+  ivec2 pos[TILE_SIZE * TILE_SIZE];
   for (int y = 0, i = 0; y < TILE_SIZE; ++y) {
     for (int x = 0; x < TILE_SIZE; ++x) {
-      pos[i] = u16vec2(
+      pos[i] = ivec2(
           gpos.x * TILE_SIZE + x, gpos.y * TILE_SIZE + y);
       pos_shared[(shared_mem_stride * i) + gl_LocalInvocationIndex] = pos[i];
       i++;
@@ -66,38 +64,38 @@ void main() {
 
   // If the top left position is out of bounds, then this invocation will have
   // no work to do.
-  if (any(greaterThanEqual(u16vec3(pos[0], gpos.z), out_limits))) {
+  if (any(greaterThanEqual(ivec3(pos[0], gpos.z), out_limits))) {
     return;
   }
 
   // Compute the index of the input texture that needs to be loaded for each
   // output position. Note that negative indices can be produced indicating that
   // the top-left element is in a region added by padding.
-  u16vec2 ipos[TILE_SIZE * TILE_SIZE];
+  ivec2 ipos[TILE_SIZE * TILE_SIZE];
   for (int i = 0; i < TILE_SIZE * TILE_SIZE; ++i) {
-    ipos[i] = pos[i] * u16vec2(stride) - u16vec2(padding);
+    ipos[i] = pos[i] * stride - padding;
   }
 
   vec4 sum[TILE_SIZE * TILE_SIZE];
-  sum[0] = texelFetch(t_bias, u16vec2(gpos.z, 0), 0);
+  sum[0] = texelFetch(t_bias, ivec2(gpos.z, 0), 0);
   for (int i = 1; i < TILE_SIZE * TILE_SIZE; ++i) {
     sum[i] = sum[0];
   }
 
   int z4 = 0;
   // Since the kernel is 1x1, we only have to loop over the depth dimension.
-  for (uint16_t z = uint16_t(0); z < uint16_t(in_group_size); z += uint16_t(4), ++z4) {
+  for (int z = 0; z < in_group_size; z += 4, ++z4) {
     // During prepacking, the weight tensor has been permuted so that the
     // channel (IC) dim is along the x-axis, and the batch (OC) dim is along
     // the z-axis.
-    const vec4 ktex_0 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(0, 0));
-    const vec4 ktex_1 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(1, 0));
-    const vec4 ktex_2 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(2, 0));
-    const vec4 ktex_3 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(3, 0));
+    const vec4 ktex_0 = texelFetchOffset(t_kernel, ivec2(z, gpos.z), 0, ivec2(0, 0));
+    const vec4 ktex_1 = texelFetchOffset(t_kernel, ivec2(z, gpos.z), 0, ivec2(1, 0));
+    const vec4 ktex_2 = texelFetchOffset(t_kernel, ivec2(z, gpos.z), 0, ivec2(2, 0));
+    const vec4 ktex_3 = texelFetchOffset(t_kernel, ivec2(z, gpos.z), 0, ivec2(3, 0));
 
 #pragma unroll
     for (int i = 0; i < TILE_SIZE * TILE_SIZE; ++i) {
-      const vec4 in_tex = texelFetch(t_in, u16vec3(ipos[i], z4), 0);
+      const vec4 in_tex = texelFetch(t_in, ivec3(ipos[i], z4), 0);
       // For 2x2 tile size algorithm works as follows.
       // To explain the calculations below, the contents of one in_tex and the
       // group of 4 texels loaded from t_kernel are shown:
@@ -139,9 +137,9 @@ void main() {
   }
 
   for (int i = 0; i < TILE_SIZE * TILE_SIZE; ++i) {
-    const u16vec2 pos = pos_shared[(shared_mem_stride * i) + gl_LocalInvocationIndex];
-    if (all(lessThan(u16vec3(pos, gpos.z), out_limits))) {
-      imageStore(t_out, u16vec3(pos, gpos.z), op(sum[i], out_min, out_max));
+    const ivec2 pos = pos_shared[(shared_mem_stride * i) + gl_LocalInvocationIndex];
+    if (all(lessThan(ivec3(pos, gpos.z), out_limits))) {
+      imageStore(t_out, ivec3(pos, gpos.z), op(sum[i], out_min, out_max));
     }
   }
 }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/indexing_utils_u16.h b/backends/vulkan/runtime/graph/ops/glsl/indexing_utils_u16.h
deleted file mode 100644
index 6dc59b6303..0000000000
--- a/backends/vulkan/runtime/graph/ops/glsl/indexing_utils_u16.h
+++ /dev/null
@@ -1,19 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- * All rights reserved.
- *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#ifndef INDEXING_UTILS_U16_H
-#define INDEXING_UTILS_U16_H
-
-#extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
-
-u16vec3 idx_to_u16pos_x_wise(uint idx, int size_x, int size_y) {
-  const uint div_by_x = idx / size_x;
-  return u16vec3(idx % size_x, div_by_x % size_y, div_by_x / size_y);
-}
-
-#endif // INDEXING_UTILS_U16_H