Merge commit '6635d1bb888e3f51ec1125ff7d6f54a2ec054a10' into develop

docs/sphinx/requirements.in

@@ -1,2 +1,2 @@
-rocm-docs-core[api_reference]==1.20.0
+rocm-docs-core[api_reference]==1.20.1
 sphinxcontrib-bibtex==2.6.3

docs/sphinx/requirements.txt

@@ -237,7 +237,7 @@ requests==2.32.3
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core[api-reference]==1.20.0
+rocm-docs-core[api-reference]==1.20.1
     # via -r requirements.in
 rpds-py==0.24.0
     # via

include/ck/utility/get_id.hpp

@@ -9,8 +9,11 @@ namespace ck {
 
 __host__ __device__ constexpr index_t get_warp_size()
 {
-    // warpSize is defined by HIP
-    return warpSize;
+#if defined(__GFX9__) || !defined(__HIP_DEVICE_COMPILE__)
+    return 64;
+#else
+    return 32;
+#endif
 }
 
 __device__ index_t get_thread_local_1d_id() { return threadIdx.x; }

include/ck_tile/core/arch/arch.hpp

@@ -50,8 +50,11 @@ enum struct memory_operation_enum : std::uint16_t
 
 CK_TILE_HOST_DEVICE constexpr index_t get_warp_size()
 {
-    // warpSize is defined by HIP
-    return warpSize;
+#if defined(__GFX9__) || !defined(__HIP_DEVICE_COMPILE__)
+    return 64;
+#else
+    return 32;
+#endif
 }
 
 CK_TILE_DEVICE index_t get_grid_size() { return gridDim.x; }

library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp

@@ -5,6 +5,7 @@
 
 #include <vector>
 #include <type_traits>
+#include <memory>
 
 #include "ck/utility/functional2.hpp"
 

test/scatter_gather/scatter_gather.cpp

@@ -1,13 +1,9 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <vector>
-#include <iostream>
-#include <numeric>
 #include <cassert>
 #include <cstdlib>
-#include <iostream>
-#include <time.h>
+#include <ctime>
 #include <unordered_set>
 
 #include "ck_tile/core.hpp"