From 118b4f08a80eefc07a1b196786323e0c58b69a53 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?R=C3=A9my=20O?= <remyoudompheng@gmail.com>
Date: Sat, 25 Jan 2025 12:20:35 +0100
Subject: [PATCH] Fix array length mismatches

---
 .../src/ggml-vulkan/vulkan-shaders/types.comp | 32 +++++++------------
 1 file changed, 11 insertions(+), 21 deletions(-)

diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/types.comp b/ggml/src/ggml-vulkan/vulkan-shaders/types.comp
index 9e29ce16e..6b874aad4 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/types.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/types.comp
@@ -383,10 +383,8 @@ shared uvec2 iq2xxs_grid[256];
 void init_iq_shmem()
 {
     // copy the table into shared memory and sync
-    if (gl_LocalInvocationIndex.x < 32) {
-        for (uint i = gl_LocalInvocationIndex.x; i < 512; i += 32) {
-            iq2xxs_grid[i] = iq2xxs_grid_const[i];
-        }
+    for (uint i = gl_LocalInvocationIndex.x; i < iq2xxs_grid.length(); i += gl_WorkGroupSize.x) {
+        iq2xxs_grid[i] = iq2xxs_grid_const[i];
     }
     barrier();
 }
@@ -552,10 +550,8 @@ shared uvec2 iq2xs_grid[512];
 void init_iq_shmem()
 {
     // copy the table into shared memory and sync
-    if (gl_LocalInvocationIndex.x < 32) {
-        for (uint i = gl_LocalInvocationIndex.x; i < 512; i += 32) {
-            iq2xs_grid[i] = iq2xs_grid_const[i];
-        }
+    for (uint i = gl_LocalInvocationIndex.x; i < iq2xs_grid.length(); i += gl_WorkGroupSize.x) {
+        iq2xs_grid[i] = iq2xs_grid_const[i];
     }
     barrier();
 }
@@ -843,10 +839,8 @@ shared uvec2 iq2s_grid[1024];
 void init_iq_shmem()
 {
     // copy the table into shared memory and sync
-    if (gl_LocalInvocationIndex.x < 32) {
-        for (uint i = gl_LocalInvocationIndex.x; i < 1024; i += 32) {
-            iq2s_grid[i] = iq2s_grid_const[i];
-        }
+    for (uint i = gl_LocalInvocationIndex.x; i < iq2s_grid.length(); i += gl_WorkGroupSize.x) {
+        iq2s_grid[i] = iq2s_grid_const[i];
     }
     barrier();
 }
@@ -908,15 +902,13 @@ const uint32_t iq3xxs_grid_const[256] = {
     0x3e1c1c1c, 0x3e1c3404, 0x3e24140c, 0x3e24240c, 0x3e2c0404, 0x3e2c0414, 0x3e2c1424, 0x3e341c04,
 };
 
-shared uint32_t iq3xxs_grid[512];
+shared uint32_t iq3xxs_grid[256];
 
 void init_iq_shmem()
 {
     // copy the table into shared memory and sync
-    if (gl_LocalInvocationIndex.x < 32) {
-        for (uint i = gl_LocalInvocationIndex.x; i < 512; i += 32) {
-            iq3xxs_grid[i] = iq3xxs_grid_const[i];
-        }
+    for (uint i = gl_LocalInvocationIndex.x; i < iq3xxs_grid.length(); i += gl_WorkGroupSize.x) {
+        iq3xxs_grid[i] = iq3xxs_grid_const[i];
     }
     barrier();
 }
@@ -1022,10 +1014,8 @@ shared uint32_t iq3s_grid[512];
 void init_iq_shmem()
 {
     // copy the table into shared memory and sync
-    if (gl_LocalInvocationIndex.x < 32) {
-        for (uint i = gl_LocalInvocationIndex.x; i < 512; i += 32) {
-            iq3s_grid[i] = iq3s_grid_const[i];
-        }
+    for (uint i = gl_LocalInvocationIndex.x; i < iq3s_grid.length(); i += gl_WorkGroupSize.x) {
+        iq3s_grid[i] = iq3s_grid_const[i];
     }
     barrier();
 }