From 314d5969bcacd262f28996c3625a82226fe9e521 Mon Sep 17 00:00:00 2001
From: rasbid <104773487+rasbid@users.noreply.github.com>
Date: Mon, 13 Oct 2025 20:00:49 +0300
Subject: [PATCH] Tune AMD GCN warptiles

---
 ggml/src/ggml-vulkan/ggml-vulkan.cpp | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index ebbb412e55f..b42f092aa75 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -2464,8 +2464,18 @@ static void ggml_vk_load_shaders(vk_device& device) {
 
         // chip specific tuning
         if ((device->architecture == AMD_GCN) && (device->driver_id != vk::DriverId::eAmdProprietary)) {
+            l_warptile        = { 256, 128, 128, 16, 16, 64, 2, tm_l, tn_l, tk_l, 16 };
+            l_warptile_mmq    = { 256, 128, 128, 32, 16, 64, 2, tm_l, tn_l, tk_l, 16 };
+            l_warptile_mmq_int = { 256, 128, 128, 32, 16, 64, 2, 4, 4, 1, 16 };
+            l_warptile_mmqid  = { 256, 128, 128, 32, 16, 64, 2, tm_l, tn_l, tk_l, 16 };
+
             m_warptile_mmq = m_warptile_mmq_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
             m_warptile_mmqid = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
+
+            s_warptile        = { 128, 64, 64, 16, 16, 32, 2, tm_s, tn_s, tk_s, 16 };
+            s_warptile_mmq    = { 128, 64, 64, 32, 16, 32, 2, tm_s, tn_s, tk_s, 16 };
+            s_warptile_mmq_int = { 128, 64, 64, 32, 16, 32, 2, 2, 1, 1, 16 };
+            s_warptile_mmqid  = { 128, 64, 64, 32, 16, 32, 2, tm_s, tn_s, tk_s, 16 };
         }
 
         l_mmq_wg_denoms = l_wg_denoms = {128, 128, 1 };