From 6164d076de34c69ed158cd16c2dba3666e132cbb Mon Sep 17 00:00:00 2001
From: "assistant-librarian[bot]"
 <assistant-librarian[bot]@users.noreply.github.com>
Date: Mon, 15 Dec 2025 07:16:25 +0000
Subject: [PATCH] Merge commit '3143a5a480e4fcf216670012fe491b44324f03b6' into
 develop

---
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp | 39 ++++++++++++++++++-
 1 file changed, 38 insertions(+), 1 deletion(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index ec48beb789..1db9fd45b8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -620,7 +620,44 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         bool isWave64  = get_warp_size() == 64;
         for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
         {
-            const auto& a        = arg.gemm_kernel_args_[i].karg_;
+            const auto& a = arg.gemm_kernel_args_[i].karg_;
+
+            // Validate stride requirements for SplitK (k_batch > 1)
+            // TODO: Enable splitK
+            if(a.k_batch > 1)
+            {
+                if constexpr(std::is_same_v<ELayout, tensor_layout::gemm::RowMajor>)
+                {
+                    if(a.StrideC != a.N)
+                    {
+                        if(ck::EnvIsEnabled(CK_ENV(CK_LOGGING)))
+                        {
+                            std::cout << "[" << __func__ << "] group id: " << i
+                                      << " SplitK (k_batch=" << a.k_batch
+                                      << ") requires contiguous output stride."
+                                      << " For RowMajor layout: StrideC must equal N."
+                                      << " Got StrideC=" << a.StrideC << ", N=" << a.N << std::endl;
+                        }
+                        return false;
+                    }
+                }
+                else if constexpr(std::is_same_v<ELayout, tensor_layout::gemm::ColumnMajor>)
+                {
+                    if(a.StrideC != a.M)
+                    {
+                        if(ck::EnvIsEnabled(CK_ENV(CK_LOGGING)))
+                        {
+                            std::cout << "[" << __func__ << "] group id: " << i
+                                      << " SplitK (k_batch=" << a.k_batch
+                                      << ") requires contiguous output stride."
+                                      << " For ColumnMajor layout: StrideC must equal M."
+                                      << " Got StrideC=" << a.StrideC << ", M=" << a.M << std::endl;
+                        }
+                        return false;
+                    }
+                }
+            }
+
             bool group_arg_valid = false;
             if(isWave64)
             {