diff --git a/backends/vulkan/runtime/graph/ops/glsl/adamw_step.glsl b/backends/vulkan/runtime/graph/ops/glsl/adamw_step.glsl
index ccb7b34bcb7..860b4f0f1df 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/adamw_step.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/adamw_step.glsl
@@ -36,8 +36,10 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const int i = int(gl_GlobalInvocationID.x);
+  const int i = int(linear_idx_from_gid());
   if (i >= numel) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/arange_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/arange_buffer.glsl
index 906ed91a466..2e9377533c8 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/arange_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/arange_buffer.glsl
@@ -28,8 +28,10 @@ ${layout_declare_ubo(B, "float", "step")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/binary_op_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/binary_op_buffer.glsl
index 4d56d991bdc..ac4751a47e5 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/binary_op_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/binary_op_buffer.glsl
@@ -53,6 +53,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "in_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "other_layout", "CONTIG_LAYOUT_INT")}
@@ -61,7 +63,7 @@ ${layout_declare_spec_const(C, "int", "in_broadcast_packed_dim", "0")}
 ${layout_declare_spec_const(C, "int", "other_broadcast_packed_dim", "0")}
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.glsl
index 1de98128159..9def3666870 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.glsl
@@ -57,11 +57,13 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 $if not IS_COMPARISON_OP:
   #include "binary_op_defs.glslh"
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.glsl
index ac39dd36fc3..ff5554ad6bc 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.glsl
@@ -32,12 +32,14 @@ $else:
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "t_layout", "DEFAULT_LAYOUT")}
 const lowp ivec4 axis_map = unhash_axis_map(t_layout);
 const lowp int packed_dim = unhash_packed_dim(t_layout);
 
 void main() {
-  const int out_buf_idx = int(gl_GlobalInvocationID.x);
+  const int out_buf_idx = int(linear_idx_from_gid());
   // On the CPU, the number of elements is determined based on a buffer of int8
   // elements. However, on the GPU, since the int8 data type is not supported
   // each group of 4 elements is interepreted as 1 int32 element. Thus each
diff --git a/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.glsl
index b0050740e8d..caf1ac80170 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.glsl
@@ -14,8 +14,10 @@ ${layout_declare_ubo(2, "int", "numel")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  int tid = int(gl_GlobalInvocationID.x);
+  int tid = int(linear_idx_from_gid());
   if (tid >= numel) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.glsl b/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.glsl
index 3b487c43887..2e2502f85b2 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.glsl
@@ -19,12 +19,14 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 // This constant is unused in this shader but is kept so that the signature is
 // consistent with image_to_nchw.
 ${layout_declare_spec_const(C, "int", "unused", "0")}
 
 void main() {
-  uint inp_bufi = gl_GlobalInvocationID.x;
+  uint inp_bufi = linear_idx_from_gid();
   if (inp_bufi>= numel(inp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.glsl
index f19cbe7ab40..735660f8866 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.glsl
@@ -37,6 +37,8 @@ ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #define NUM_INPUTS ${NUM_INPUTS}
 
 /*
@@ -45,7 +47,7 @@ layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
  * be inserted along the output's concat_dim starting at concat_offset.
  */
 void main() {
-  const int tid = int(gl_GlobalInvocationID.x);
+  const int tid = int(linear_idx_from_gid());
 
   // The 1-3 input tensors are interpreted as one concatenated tensor ("volume")
   // along the concat_dim for the purposes of tensor indexing. Each thread is
diff --git a/backends/vulkan/runtime/graph/ops/glsl/concat_texture.glsl b/backends/vulkan/runtime/graph/ops/glsl/concat_texture.glsl
index ef6264440cf..7345a61afa0 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/concat_texture.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/concat_texture.glsl
@@ -43,6 +43,8 @@ const int out_packed_dim = get_packed_dim(out_layout);
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #define NUM_INPUTS ${NUM_INPUTS}
 
 /*
@@ -55,7 +57,7 @@ layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
  * input tensor.
  */
 void main() {
-  const int tid = int(gl_GlobalInvocationID.x);
+  const int tid = int(linear_idx_from_gid());
 
   // Compute inp_volume_sizes from output sizes, replacing concat_dim with the
   // sum of all input sizes along that dimension.
diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
index 9089f87d658..ad7074c5e10 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.glsl
@@ -30,6 +30,8 @@ ${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;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "ngroups", "1")}
 
 /*
@@ -37,9 +39,10 @@ ${layout_declare_spec_const(C, "int", "ngroups", "1")}
  * output at a single output location.
  */
 void main() {
-  const uint div_by_x = gl_GlobalInvocationID.x / out_limits.x;
+  const uint linear_idx = linear_idx_from_gid();
+  const uint div_by_x = linear_idx / out_limits.x;
   const ivec3 pos = ivec3(
-    gl_GlobalInvocationID.x % out_limits.x,
+    linear_idx % out_limits.x,
     div_by_x % out_limits.y,
     div_by_x / out_limits.y);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl
index b4196551fa1..a94811ef204 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.glsl
@@ -46,6 +46,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "apply_bias", "1")}
 ${layout_declare_spec_const(C, "int", "conv2d_params_stride_x", "1")}
 ${layout_declare_spec_const(C, "int", "conv2d_params_stride_y", "1")}
@@ -69,7 +71,7 @@ ${layout_declare_spec_const(C, "int", "conv2d_params_groups", "1")}
 #include "conv2d_dw_q8_utils.glslh"
 
 void main() {
-  const int tid = int(gl_GlobalInvocationID.x);
+  const int tid = int(linear_idx_from_gid());
   Conv2dBlockExtents out_block_extents = make_block_extents(output_sizes);
 
   Conv2dBlockIndex out_block_idx = linear_idx_to_block_idx(
diff --git a/backends/vulkan/runtime/graph/ops/glsl/dispatch.glslh b/backends/vulkan/runtime/graph/ops/glsl/dispatch.glslh
new file mode 100644
index 00000000000..a127ff4bb74
--- /dev/null
+++ b/backends/vulkan/runtime/graph/ops/glsl/dispatch.glslh
@@ -0,0 +1,19 @@
+/*
+ * 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 DISPATCH_GLSLH
+#define DISPATCH_GLSLH
+
+uint linear_idx_from_gid() {
+  // Extend this to include Z if two-dimensional linear dispatches become
+  // insufficient.
+  return gl_GlobalInvocationID.x +
+      gl_GlobalInvocationID.y * gl_NumWorkGroups.x * gl_WorkGroupSize.x;
+}
+
+#endif // DISPATCH_GLSLH
diff --git a/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.glsl
index f9a46fda141..5b084a1d5cf 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.glsl
@@ -32,6 +32,8 @@ ${layout_declare_ubo(B, "BufferMetadata", "weight")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 TensorIndex out_tidx_to_indices_tidx(const TensorIndex out_tidx) {
   TensorIndex indices_tidx;
   int d = 0;
@@ -56,7 +58,7 @@ T load_weight_elem(const int embedding_idx, const uint dim_idx) {
 }
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.glsl
index 806c27cfd69..cb3b74d0d87 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.glsl
@@ -27,8 +27,10 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const uint outp_bufi = gl_GlobalInvocationID.x;
+  const uint outp_bufi = linear_idx_from_gid();
   if (outp_bufi >= numel(outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/full_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/full_buffer.glsl
index 729baec324d..8243d13325c 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/full_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/full_buffer.glsl
@@ -26,8 +26,10 @@ ${layout_declare_ubo(B, "float", "fill_value")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.glsl
index eadc3403635..5a390b93480 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.glsl
@@ -32,10 +32,12 @@ ${layout_declare_ubo(B, "BufferMetadata", "index")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "gather_dim", "0")}
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl b/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl
index a60e52fa12b..e5b7d05863e 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.glsl
@@ -60,11 +60,13 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #include "conv2d_int8_output_tile_store.glslh"
 #include "im2col_packed_int8_utils.glslh"
 
 void main() {
-  const int out_buf_idx = int(gl_GlobalInvocationID.x);
+  const int out_buf_idx = int(linear_idx_from_gid());
 
   Conv2dBlockExtents im2col_block_extents = make_block_extents(im2col_sizes);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.glsl b/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.glsl
index c4dcfa688f8..8e5d1f53304 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.glsl
@@ -46,6 +46,8 @@ $if not TO_STAGING:
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "in_layout", "CONTIG_LAYOUT_INT")}
 
 $if not TO_STAGING:
@@ -60,7 +62,7 @@ void main() {
   // the (VRAM-cached) texture reads -- even though each texel is now fetched up
   // to 4 times, once per component. On unified-memory (mobile) GPUs the extra
   // fetches are a net loss, so this variant is gated to discrete GPUs.
-  const int oi = int(gl_GlobalInvocationID.x);
+  const int oi = int(linear_idx_from_gid());
   const int W = inp.sizes.x;
   const int H = inp.sizes.y;
   const int C = inp.sizes.z;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/index_tensor_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/index_tensor_buffer.glsl
index f430a570b92..db61e0859f2 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/index_tensor_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/index_tensor_buffer.glsl
@@ -30,12 +30,14 @@ ${layout_declare_ubo(B, "BufferMetadata", "index")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 // Implements aten.index.Tensor for the case where self is 1D and there is
 // exactly one index tensor. Each output element is:
 //   output[...] = self[index[...]]
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/int8x4_buffer_to_nchw.glsl b/backends/vulkan/runtime/graph/ops/glsl/int8x4_buffer_to_nchw.glsl
index 76e6a6c6238..25fefc1037c 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/int8x4_buffer_to_nchw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/int8x4_buffer_to_nchw.glsl
@@ -26,12 +26,14 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 
 void main() {
   // One thread per output int32 in the NCHW staging buffer.
   // Each output int32 holds 4 consecutive NCHW bytes.
-  const uint out_int32_idx = gl_GlobalInvocationID.x;
+  const uint out_int32_idx = linear_idx_from_gid();
 
   const uint W = inp.sizes[0][0];
   const uint H = inp.sizes[0][1];
diff --git a/backends/vulkan/runtime/graph/ops/glsl/linear_dW.glsl b/backends/vulkan/runtime/graph/ops/glsl/linear_dW.glsl
index 1a8ed94031b..42fde2111e4 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/linear_dW.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/linear_dW.glsl
@@ -28,6 +28,8 @@ ${layout_declare_ubo(B, "ivec4", "x_sizes")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
   // dW[N, K] = sum_m d_out[m, N] * x[m, K]; contraction over the flattened M.
   const int N = dout_sizes.x;
@@ -38,7 +40,7 @@ void main() {
   const int nkt = (K + TILE_K - 1) / TILE_K;
   const int tiles = nnt * nkt;
 
-  const int tile_idx = int(gl_GlobalInvocationID.x);
+  const int tile_idx = int(linear_idx_from_gid());
   if (tile_idx >= tiles) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.glsl b/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.glsl
index 505fc3d0009..0f88dc59702 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.glsl
@@ -57,6 +57,8 @@ ${layout_declare_ubo(B, "ivec4", "input_sizes")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "apply_bias", "0")}
 ${layout_declare_spec_const(C, "int", "K4_per_group", "0")}
 
@@ -73,7 +75,7 @@ shared FPOutTile partial_sums[WGS];
 
 void main() {
   const int lid = int(gl_LocalInvocationID.z);
-  const int n8 = int(gl_GlobalInvocationID.x);
+  const int n8 = int(linear_idx_from_gid());
 
   // The output tensor will have a shape of [n, 1, 1, 1]. Each thread computes
   // 8 output elements, so each thread will write to 8 elements starting at the
diff --git a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.glsl b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.glsl
index 369112bf7fd..2599f3d67f5 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.glsl
@@ -46,6 +46,8 @@ $else:
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 // This header file must be defined after the layout descriptors have been
 // declared because the functions in the header assume some variables have been
 // declared as layout descriptors.
@@ -57,7 +59,7 @@ layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 #endif
 
 void main() {
-  const int out_bufi = int(gl_GlobalInvocationID.x);
+  const int out_bufi = int(linear_idx_from_gid());
   if (out_bufi >= out_numel) {
     return;
   }
@@ -88,9 +90,10 @@ void main() {
 #else // USING_TEXTURE
 
 void main() {
+  const uint linear_idx = linear_idx_from_gid();
   const ivec2 out_pos = ivec2(
-    gl_GlobalInvocationID.x % out_limits.x,
-    gl_GlobalInvocationID.x / out_limits.x);
+    linear_idx % out_limits.x,
+    linear_idx / out_limits.x);
 
   if (out_pos.y >= out_limits.y) {
     return;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.glsl b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.glsl
index c1c4f72c600..86c01cfa80e 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.glsl
@@ -46,20 +46,23 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 shared VEC4_T partial_sums[NGROUPS][NWORKERS][TILE_ROWS][TILE_TXCOLS];
 
 void main() {
+  const uint linear_idx = linear_idx_from_gid();
   // txcol stands for "texel column". One txcol corresponds to 4 scalar columns.
   $if TILE_TXCOLS > 1:
     const uint global_wg_x = uint(divup(out_sizes.x, 4 * TILE_TXCOLS));
     const uint out_txcol = uint(
-      (gl_GlobalInvocationID.x % global_wg_x) * TILE_TXCOLS);
+      (linear_idx % global_wg_x) * TILE_TXCOLS);
   $else:
     const uint global_wg_x = uint(divup4(out_sizes.x));
-    const uint out_txcol = uint(gl_GlobalInvocationID.x % global_wg_x);
+    const uint out_txcol = uint(linear_idx % global_wg_x);
 
   const uint out_row = uint(
-    (gl_GlobalInvocationID.x / global_wg_x) * TILE_ROWS);
+    (linear_idx / global_wg_x) * TILE_ROWS);
 
   $if QUANT_NBITS == 4:
     const uint weight_txcol = uint(out_txcol / 2);
diff --git a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.glsl b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.glsl
index 1d4a24bacc3..d8f042f54e8 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.glsl
@@ -42,18 +42,21 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
 
 void main() {
+  const int linear_idx = int(linear_idx_from_gid());
   // txcol stands for "texel column". One txcol corresponds to 4 scalar columns.
   $if TILE_TXCOLS > 1:
     const int global_wg_x = divup(out_sizes.x, 4 * TILE_TXCOLS);
-    const int out_txcol = (int(gl_GlobalInvocationID.x) % global_wg_x) * TILE_TXCOLS;
+    const int out_txcol = (linear_idx % global_wg_x) * TILE_TXCOLS;
   $else:
     const int global_wg_x = divup4(out_sizes.x);
-    const int out_txcol = int(gl_GlobalInvocationID.x) % global_wg_x;
+    const int out_txcol = linear_idx % global_wg_x;
 
-  const int out_row = (int(gl_GlobalInvocationID.x) / global_wg_x) * TILE_ROWS;
+  const int out_row = (linear_idx / global_wg_x) * TILE_ROWS;
 
   $if QUANT_NBITS == 4:
     const int weight_txcol = out_txcol / 2;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.glsl
index a6f9fed4471..ef2511f2d89 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.glsl
@@ -18,13 +18,15 @@ ${layout_declare_ubo(B, "BufferMetadata", "outp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 // This constant is unused in this shader but is kept so that the signature is
 // consistent with nchw_to_image.
 ${layout_declare_spec_const(C, "int", "unused", "0")}
 ${layout_declare_spec_const(C, "int", "transpose_hw", "0")}
 
 void main() {
-  const uint outp_bufi = int(gl_GlobalInvocationID.x);
+  const uint outp_bufi = linear_idx_from_gid();
   if (outp_bufi >= numel(outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/nchw_to_int8x4_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/nchw_to_int8x4_buffer.glsl
index 6b535400554..f911cfe92b7 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/nchw_to_int8x4_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/nchw_to_int8x4_buffer.glsl
@@ -26,10 +26,12 @@ ${layout_declare_ubo(B, "BufferMetadata", "outp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 
 void main() {
-  const uint texel_idx = gl_GlobalInvocationID.x;
+  const uint texel_idx = linear_idx_from_gid();
   const uint num_texels = numel(outp) / 4;
   if (texel_idx >= num_texels) {
     return;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/pad_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/pad_buffer.glsl
index ea5222c74b3..4b8d1558652 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/pad_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/pad_buffer.glsl
@@ -32,10 +32,12 @@ ${layout_declare_ubo(B, "float", "fill_value")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.glsl
index d517003bb8c..b35b0c61488 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.glsl
@@ -33,6 +33,8 @@ ${layout_declare_spec_const(C, "int", "permute_order", "0")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 TensorIndex permute(TensorIndex tidx) {
   TensorIndex new_tidx = tidx;
 
@@ -50,7 +52,7 @@ TensorIndex permute(TensorIndex tidx) {
 }
 
 void main() {
-  const uint inp_bufi = gl_GlobalInvocationID.x;
+  const uint inp_bufi = linear_idx_from_gid();
   if (inp_bufi >= numel(inp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_buffer.glsl
index 196555a279f..b1b05872538 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_buffer.glsl
@@ -30,6 +30,8 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "upscale_factor", "1")}
@@ -47,7 +49,7 @@ ${layout_declare_spec_const(C, "int", "upscale_factor", "1")}
  * mapping is independent of the tensor's memory layout.
  */
 void main() {
-  const uint outp_bufi = gl_GlobalInvocationID.x;
+  const uint outp_bufi = linear_idx_from_gid();
   if (outp_bufi >= numel(outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q4gsw_backward.glsl b/backends/vulkan/runtime/graph/ops/glsl/q4gsw_backward.glsl
index d89bb83feed..d54fb731956 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q4gsw_backward.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q4gsw_backward.glsl
@@ -30,6 +30,8 @@ ${layout_declare_ubo(B, "ivec4", "dx_sizes")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "group_size", "32")}
 
 // d_x[M, K] = d_out[M, N] @ dequant(W)[N, K], contracting over N.
@@ -44,7 +46,7 @@ void main() {
   const int nkt = (K + TILE_K - 1) / TILE_K;
   const int tiles = nmt * nkt;
 
-  const int tile_idx = int(gl_GlobalInvocationID.x);
+  const int tile_idx = int(linear_idx_from_gid());
   if (tile_idx >= tiles) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_binary.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_binary.glsl
index be93e800436..34fe483db10 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_binary.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_binary.glsl
@@ -44,6 +44,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "in_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "other_layout", "CONTIG_LAYOUT_INT")}
@@ -58,7 +60,7 @@ define_store_int8x4_buffer_fns(t_out)
 
 void main() {
   // Buffer storage: use linear dispatch
-  const uint contig_block_idx = gl_GlobalInvocationID.x;
+  const uint contig_block_idx = linear_idx_from_gid();
   TensorIndex4D tidx = contiguous_block_idx_to_tensor4d_idx_with_block_config(
       out_meta, contig_block_idx, block_config);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_clone.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_clone.glsl
index 0006311e13c..7734b23c32d 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_clone.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_clone.glsl
@@ -28,6 +28,8 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "inp_block_config", "0")}
@@ -47,7 +49,7 @@ void main() {
   TensorIndex4D tidx;
 
   // Buffer storage: use linear dispatch
-  const uint contig_block_idx = gl_GlobalInvocationID.x;
+  const uint contig_block_idx = linear_idx_from_gid();
   tidx = contiguous_block_idx_to_tensor4d_idx_with_block_config(
       inp, contig_block_idx, inp_block_config);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_dequantize.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_dequantize.glsl
index 88089627911..69343ee04d5 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_dequantize.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_dequantize.glsl
@@ -44,6 +44,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "outp_block_config", "0")}
@@ -85,7 +87,7 @@ void main() {
 
 #ifdef USING_BUFFER
   // Buffer storage: use linear dispatch (supports up to 8D)
-  const uint contig_block_idx = gl_GlobalInvocationID.x;
+  const uint contig_block_idx = linear_idx_from_gid();
   tidx = contiguous_block_idx_to_tensor_idx_with_block_config(
       inp, contig_block_idx, inp_block_config);
 #else
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_im2col.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_im2col.glsl
index ed4e124ac45..314788cb857 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_im2col.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_im2col.glsl
@@ -37,8 +37,10 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const int out_buf_idx = int(gl_GlobalInvocationID.x);
+  const int out_buf_idx = int(linear_idx_from_gid());
 
   // Extract sizes from BufferMetadata
   const ivec4 im2col_sizes = ivec4(im2col_outp.sizes[0]);
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear_gemv.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear_gemv.glsl
index becff2ab9ab..9153cb0e294 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear_gemv.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear_gemv.glsl
@@ -59,6 +59,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #include "common.glslh"
 #include "linear_int8_weight_tile_load.glslh"
 #include "linear_fp_output_tile_int8_int8_compute.glslh"
@@ -75,7 +77,7 @@ shared ivec4 partial_accums[WGS][TILE_N4];
 
 void main() {
   const int lid = int(gl_LocalInvocationID.z);
-  const int n4 = int(gl_GlobalInvocationID.x) * TILE_N4;
+  const int n4 = int(linear_idx_from_gid()) * TILE_N4;
 
   const int n = mul_4(n4);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_pixel_shuffle.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_pixel_shuffle.glsl
index 2de47e1452e..69ce99ec839 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_pixel_shuffle.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_pixel_shuffle.glsl
@@ -39,6 +39,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 
@@ -90,7 +92,7 @@ void main() {
   // (along the packed dim) at one (n, oh, ow) spatial position.
   const int C_words = div_up_4(C_out);
   const int total_words = N * C_words * H_out * W_out;
-  const int thread_idx = int(gl_GlobalInvocationID.x);
+  const int thread_idx = int(linear_idx_from_gid());
   if (thread_idx >= total_words) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_quantize.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_quantize.glsl
index 2f458054a32..739ca327d4f 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_quantize.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_quantize.glsl
@@ -44,6 +44,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "inp_block_config", "0")}
@@ -83,7 +85,7 @@ void main() {
 
 #ifdef USING_BUFFER
   // Buffer storage: use linear dispatch (supports up to 8D)
-  const uint contig_block_idx = gl_GlobalInvocationID.x;
+  const uint contig_block_idx = linear_idx_from_gid();
   tidx = contiguous_block_idx_to_tensor_idx_with_block_config(
       inp, contig_block_idx, inp_block_config);
 #else
diff --git a/backends/vulkan/runtime/graph/ops/glsl/q8ta_unary.glsl b/backends/vulkan/runtime/graph/ops/glsl/q8ta_unary.glsl
index e97d6d47877..5949f3fab52 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/q8ta_unary.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/q8ta_unary.glsl
@@ -40,6 +40,8 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "in_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "block_config", "0")}
@@ -52,7 +54,7 @@ define_store_int8x4_buffer_fns(t_out)
 
 void main() {
   // Buffer storage: use linear dispatch
-  const uint contig_block_idx = gl_GlobalInvocationID.x;
+  const uint contig_block_idx = linear_idx_from_gid();
   TensorIndex4D tidx = contiguous_block_idx_to_tensor4d_idx_with_block_config(
       out_meta, contig_block_idx, block_config);
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/repeat_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/repeat_buffer.glsl
index be2d87a168f..a7a6de4d8a1 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/repeat_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/repeat_buffer.glsl
@@ -28,8 +28,10 @@ ${layout_declare_ubo(B, "BufferMetadata", "in_meta")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, out_meta)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.glsl b/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.glsl
index b8b73646e8e..372c233ea7c 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.glsl
@@ -27,10 +27,12 @@ ${layout_declare_ubo(B, buffer_scalar_type(SCALAR_VALUE_TYPE), "scalar_value")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 #ifdef USING_BUFFER
 
 void main() {
-  const int i = int(gl_GlobalInvocationID.x);
+  const int i = int(linear_idx_from_gid());
 
   if (i > 0) {
     return;
diff --git a/backends/vulkan/runtime/graph/ops/glsl/split_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/split_buffer.glsl
index 932617ecb36..b1e69bc1215 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/split_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/split_buffer.glsl
@@ -29,12 +29,14 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "split_dim", "0")}
 ${layout_declare_spec_const(C, "int", "split_idx", "0")}
 ${layout_declare_spec_const(C, "int", "split_offset", "0")}
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/tan.glsl b/backends/vulkan/runtime/graph/ops/glsl/tan.glsl
index 4e6dcab8d35..7ce2dd86644 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/tan.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/tan.glsl
@@ -30,12 +30,13 @@ $else:
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
 #include "activations.h"
 
 #ifdef USING_BUFFER
 
 void main() {
-  const int i = int(gl_GlobalInvocationID.x);
+  const int i = int(linear_idx_from_gid());
   if (i >= numel) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.glsl
index 79b2cd69790..33ec64034cb 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.glsl
@@ -49,10 +49,11 @@ layout(push_constant) uniform restrict Block {
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
 #include "${OP_NAME}.glslh"
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/unary_op.glsl b/backends/vulkan/runtime/graph/ops/glsl/unary_op.glsl
index d3390229ac1..45aa3ed7133 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/unary_op.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/unary_op.glsl
@@ -37,12 +37,13 @@ float maximum;
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
 #include "activations.h"
 
 #ifdef USING_BUFFER
 
 void main() {
-  const int i = int(gl_GlobalInvocationID.x);
+  const int i = int(linear_idx_from_gid());
   if (i >= numel) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/unfold_copy_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/unfold_copy_buffer.glsl
index 81c4b6f1940..646c2dec2af 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/unfold_copy_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/unfold_copy_buffer.glsl
@@ -30,8 +30,10 @@ ${layout_declare_spec_const(C, "int", "step", "1")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, out_meta)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.glsl b/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.glsl
index 44cb30e96c5..def742cefe1 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.glsl
@@ -29,9 +29,11 @@ ${layout_declare_spec_const(C, "int", "concat_dim", "0")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 void main() {
   // Only one thread needs to update the offset
-  if (gl_GlobalInvocationID.x != 0) {
+  if (linear_idx_from_gid() != 0) {
     return;
   }
 
diff --git a/backends/vulkan/runtime/graph/ops/glsl/view_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/view_buffer.glsl
index b5c9f71144e..7470a5ceba5 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/view_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/view_buffer.glsl
@@ -18,6 +18,8 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "outp_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
 
@@ -26,7 +28,7 @@ ${layout_declare_spec_const(C, "int", "inp_layout", "CONTIG_LAYOUT_INT")}
  * tensor element in the input and output tensors are the same.
  */
 void main() {
-  const uint outp_bufi = gl_GlobalInvocationID.x;
+  const uint outp_bufi = linear_idx_from_gid();
   if (outp_bufi >= numel(outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.glsl b/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.glsl
index f947419fa74..e7622db7572 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.glsl
@@ -20,6 +20,8 @@ ${layout_declare_ubo(B, "BufferMetadata", "inp")}
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "outp_layout", "0")}
 ${layout_declare_spec_const(C, "int", "inp_layout", "0")}
 
@@ -29,7 +31,7 @@ ${layout_declare_spec_const(C, "int", "inp_layout", "0")}
  * may be different and need conversion.
  */
 void main() {
-  const uint outp_bufi = gl_GlobalInvocationID.x;
+  const uint outp_bufi = linear_idx_from_gid();
   if (outp_bufi >= numel(outp)) {
     return;
   }
diff --git a/backends/vulkan/runtime/graph/ops/glsl/where.glsl b/backends/vulkan/runtime/graph/ops/glsl/where.glsl
index 6982d41dd3a..4fae9d94437 100644
--- a/backends/vulkan/runtime/graph/ops/glsl/where.glsl
+++ b/backends/vulkan/runtime/graph/ops/glsl/where.glsl
@@ -41,6 +41,8 @@ $else:
 
 layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
 
+#include "dispatch.glslh"
+
 ${layout_declare_spec_const(C, "int", "out_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "cond_layout", "CONTIG_LAYOUT_INT")}
 ${layout_declare_spec_const(C, "int", "self_layout", "CONTIG_LAYOUT_INT")}
@@ -49,7 +51,7 @@ ${layout_declare_spec_const(C, "int", "other_layout", "CONTIG_LAYOUT_INT")}
 #ifdef USING_BUFFER
 
 void main() {
-  const uint out_bufi = gl_GlobalInvocationID.x;
+  const uint out_bufi = linear_idx_from_gid();
   if (out_of_bounds(out_bufi, outp)) {
     return;
   }