From 6667a9021713f794a2c9aee4696c19f6cf376235 Mon Sep 17 00:00:00 2001
From: Emily Martins <65371150+ecamartins@users.noreply.github.com>
Date: Tue, 14 Jul 2026 08:23:43 +0000
Subject: [PATCH] [rocm-libraries] ROCm/rocm-libraries#8595 (commit 7fbebbb)

[CK] Clean up Stream-K remnants in old CK and fix
 static_assert in CK Tile (#8595)

## Motivation

This PR is cleans up some Stream-K documentation (e.g., typos, unneeded
docs) and fixes an error in a Stream-K static assert. This PR adds no
new functionality.

## Technical Details
Exact details about the clean-up changes are as follows:
- Removes Stream-K documentation from old CK that were missed in PR
#8141.
- Fixes `static_assert` syntax in Stream-K kernel by adding `constexpr`
to if-else condition and correcting empty message parameter.
- Cleans up some typos in some Doxygen comments in the CK Tile Stream-K
kernel file.

## Test Plan

Ran existing tests locally; no new functionality was added, so no new
tests were added.

## Test Result

Tests passed.

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
---
 client_example/01_gemm/README.md                  |  2 --
 .../kernel/streamk_gemm/streamk_gemm_kernel.hpp   | 15 ++++++++-------
 2 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/client_example/01_gemm/README.md b/client_example/01_gemm/README.md
index 6ff4958cee..fd457265f8 100644
--- a/client_example/01_gemm/README.md
+++ b/client_example/01_gemm/README.md
@@ -139,8 +139,6 @@ Table of supported cases by instance factory with XDL instruction for Row/Row/Ro
 * **DeviceGemmMultipleDLayernorm** - GEMM fused with layernorm.
 * **DeviceGemmMultipleDMultipleR** - GEMM fused with reductions and custom global reductions operators.
 * **DeviceGemmReduce** - GEMM fused with reduction.
-* **DeviceGemm_Streamk_V2** - GEMM stream K implementation. Implementation allows to use reduction instead of AtomicAdd.
-* **DeviceGemmStreamK** - GEMM stream K implementation using AtomicAdd.
 
 ## How to Run
 
diff --git a/include/ck_tile/ops/gemm/kernel/streamk_gemm/streamk_gemm_kernel.hpp b/include/ck_tile/ops/gemm/kernel/streamk_gemm/streamk_gemm_kernel.hpp
index 6f359e1bab..2c006ecfe7 100644
--- a/include/ck_tile/ops/gemm/kernel/streamk_gemm/streamk_gemm_kernel.hpp
+++ b/include/ck_tile/ops/gemm/kernel/streamk_gemm/streamk_gemm_kernel.hpp
@@ -325,11 +325,11 @@ struct StreamKKernel
     }
 
     /**
-     * @brief Runs the main Stream - K algorithm.
-     * @param kargs Stream - K kernel arguments.
-     * @param cta_idx The current Stream - K workgroup's index.
+     * @brief Runs the main Stream-K algorithm.
+     * @param kargs Stream-K kernel arguments.
+     * @param cta_idx The current Stream-K workgroup's index.
      * @param smem_ptr_0 Pointer to LDS.
-     * @note It is assumed that the first Stream - K workgroup has a `cta_idx` of zero. If a
+     * @note It is assumed that the first Stream-K workgroup has a `cta_idx` of zero. If a
      * non-persistent data-parallel (DP) section is used, then a Stream-K workgroup's `cta_idx`
      * *should be something like `blockIdx.x` minus number of DP workgroups.
      */
@@ -374,8 +374,9 @@ struct StreamKKernel
             {
                 BaseGemm(kargs, tile_idx, num_loop_sk, i_k_a, i_k_b, k_size, smem_ptr_0);
             }
-            else if(TilePartitioner::ReductionStrategy == StreamKReductionStrategy::Linear ||
-                    TilePartitioner::ReductionStrategy == StreamKReductionStrategy::Tree)
+            else if constexpr(TilePartitioner::ReductionStrategy ==
+                                  StreamKReductionStrategy::Linear ||
+                              TilePartitioner::ReductionStrategy == StreamKReductionStrategy::Tree)
             {
                 const auto c_macro_tile_idx =
                     kargs.tile_partitioner.get_output_tile_index(tile_idx);
@@ -517,7 +518,7 @@ struct StreamKKernel
             else
             {
                 static_assert(
-                    "An implementation does not exist for the chosen reduction strategy.");
+                    false, "An implementation does not exist for the chosen reduction strategy.");
             }
 
             // Prepare for next Stream-K loop iteration.