From cf5df82d100c907e2063a24ecf9820632fd64959 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 19 Feb 2025 13:47:39 -0800
Subject: [PATCH] Fix build for gfx950 (#1904)

* fix the gfx950 build issue

* fix typo

[ROCm/composable_kernel commit: 824e2c1737b7dcb883352a7ca9a46f7a77cb97c7]
---
 .../device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp
index 00adc1fa07..f6ea23a1e7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp
@@ -53,8 +53,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const index_t num_k_per_block)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx94__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
     const index_t g_idx = __builtin_amdgcn_readfirstlane(blockIdx.z);
     const index_t k_idx = __builtin_amdgcn_readfirstlane(blockIdx.y * num_k_per_block);
 
@@ -82,7 +81,7 @@ __global__ void
                                         k_idx);
 #else
     ignore = karg;
-#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+#endif // end of if (defined(__gfx9__)
 }
 
 template <typename GridwiseGemm,
@@ -107,8 +106,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const index_t num_k_per_block)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
     // offset base pointer for each work-group
     const index_t g_idx = __builtin_amdgcn_readfirstlane(blockIdx.z);
     const index_t k_idx = __builtin_amdgcn_readfirstlane(blockIdx.y * num_k_per_block);
@@ -142,7 +140,7 @@ __global__ void
                                              k_idx);
 #else
     ignore = karg;
-#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+#endif // end of if (defined(__gfx9__)
 }
 
 // out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]