From a7614ad5946ea9a958dd69f4482fc132a945c76c Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 12 Mar 2025 07:29:09 -0700
Subject: [PATCH] use old instrinsics with staging compiler (#1970)

[ROCm/composable_kernel commit: 4c97cc511e522eb9cc18786310add7f2a21d203c]
---
 .../device/impl/device_sparse_embeddings_forward_layernorm.hpp  | 2 +-
 include/ck/utility/common_header.hpp                            | 2 +-
 include/ck/utility/dynamic_buffer.hpp                           | 2 +-
 include/ck_tile/core.hpp                                        | 2 +-
 include/ck_tile/core/tensor/buffer_view.hpp                     | 2 +-
 5 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
index d43dab2983..df3c929c2e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
@@ -12,7 +12,7 @@
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#if __clang_major__ >= 20
+#if __clang_major__ == 20
 #include "ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm_builtins.hpp"
 #else
 #include "ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp"
diff --git a/include/ck/utility/common_header.hpp b/include/ck/utility/common_header.hpp
index 69420a6465..c2c3aa002c 100644
--- a/include/ck/utility/common_header.hpp
+++ b/include/ck/utility/common_header.hpp
@@ -33,7 +33,7 @@
 #include "ck/utility/thread_group.hpp"
 #include "ck/utility/debug.hpp"
 
-#if __clang_major__ >= 20
+#if __clang_major__ == 20
 #include "amd_buffer_addressing_builtins.hpp"
 #else
 #include "amd_buffer_addressing.hpp"
diff --git a/include/ck/utility/dynamic_buffer.hpp b/include/ck/utility/dynamic_buffer.hpp
index 6805fba4f9..1a0ea27eab 100644
--- a/include/ck/utility/dynamic_buffer.hpp
+++ b/include/ck/utility/dynamic_buffer.hpp
@@ -7,7 +7,7 @@
 #include "ck/utility/data_type.hpp"
 #include "enable_if.hpp"
 #include "c_style_pointer_cast.hpp"
-#if __clang_major__ >= 20
+#if __clang_major__ == 20
 #include "amd_buffer_addressing_builtins.hpp"
 #else
 #include "amd_buffer_addressing.hpp"
diff --git a/include/ck_tile/core.hpp b/include/ck_tile/core.hpp
index 81b452a53c..94710e584f 100644
--- a/include/ck_tile/core.hpp
+++ b/include/ck_tile/core.hpp
@@ -8,7 +8,7 @@
 #include "ck_tile/core/algorithm/indexing_adaptor.hpp"
 #include "ck_tile/core/algorithm/space_filling_curve.hpp"
 #include "ck_tile/core/algorithm/static_encoding_pattern.hpp"
-#if __clang_major__ >= 20
+#if __clang_major__ == 20
 #include "ck_tile/core/arch/amd_buffer_addressing_builtins.hpp"
 #else
 #include "ck_tile/core/arch/amd_buffer_addressing.hpp"
diff --git a/include/ck_tile/core/tensor/buffer_view.hpp b/include/ck_tile/core/tensor/buffer_view.hpp
index c7e24cbc2b..bdcfbdd920 100644
--- a/include/ck_tile/core/tensor/buffer_view.hpp
+++ b/include/ck_tile/core/tensor/buffer_view.hpp
@@ -5,7 +5,7 @@
 
 #include "ck_tile/core/config.hpp"
 #include "ck_tile/core/arch/arch.hpp"
-#if __clang_major__ >= 20
+#if __clang_major__ == 20
 #include "ck_tile/core/arch/amd_buffer_addressing_builtins.hpp"
 #else
 #include "ck_tile/core/arch/amd_buffer_addressing.hpp"