From 15daccec5423a634cb656eff52e09f0dbc7fdee8 Mon Sep 17 00:00:00 2001
From: Vladan Rakic <vlarakic@amd.com>
Date: Thu, 25 Sep 2025 07:36:35 -0400
Subject: [PATCH] Fix ROCm 7.0 build Remove use of warpSize for setting c-style
 array size. Since ROCm 7.0 warpSize can no longer be used as compile time
 constant.

---
 exllamav2/exllamav2_ext/cuda/layer_norm.cu | 9 ++++++---
 exllamav2/exllamav2_ext/cuda/rms_norm.cu   | 8 ++++++--
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/exllamav2/exllamav2_ext/cuda/layer_norm.cu b/exllamav2/exllamav2_ext/cuda/layer_norm.cu
index b286492e..a2ffa637 100644
--- a/exllamav2/exllamav2_ext/cuda/layer_norm.cu
+++ b/exllamav2/exllamav2_ext/cuda/layer_norm.cu
@@ -3,8 +3,12 @@
 #include "compat.cuh"
 
 #if defined(USE_ROCM)
-#define NUM_WARPS (1024 / warpSize)
-#define WARP_SIZE (warpSize)
+#if defined(__GFX8__) || defined(__GFX9__)
+  #define WARP_SIZE 64
+#else
+  #define WARP_SIZE 32
+#endif
+#define NUM_WARPS (1024 / WARP_SIZE)
 #else
 #define NUM_WARPS 32
 #define WARP_SIZE 32
@@ -230,4 +234,3 @@ void layer_norm_cuda_update_y
 {
     graph->update_param_ptr(label, 0, 3, y);
 }
-
diff --git a/exllamav2/exllamav2_ext/cuda/rms_norm.cu b/exllamav2/exllamav2_ext/cuda/rms_norm.cu
index 94155ade..e6aa6b5d 100644
--- a/exllamav2/exllamav2_ext/cuda/rms_norm.cu
+++ b/exllamav2/exllamav2_ext/cuda/rms_norm.cu
@@ -3,8 +3,12 @@
 #include "compat.cuh"
 
 #if defined(USE_ROCM)
-#define NUM_WARPS (1024 / warpSize)
-#define WARP_SIZE (warpSize)
+#if defined(__GFX8__) || defined(__GFX9__)
+  #define WARP_SIZE 64
+#else
+  #define WARP_SIZE 32
+#endif
+#define NUM_WARPS (1024 / WARP_SIZE)
 #else
 #define NUM_WARPS 32
 #define WARP_SIZE 32