Layernorm4d (#1022)

* Rename folder * Add layernorm 4d fwd example * Rename original layernorm example * Add layernorm 4d f16 test * Add layernorm4d_fwd client example * Support layernorm4D in ckProfiler * Rename groupnorm to groupnorm fwd in example * Rename layernorm and group fwd in test * Rename normalization to normalization_fwd (instances) * Add fwd to DeviceNormalization * Rename external api header * Rename folder, because we can also add bwd in this folder * Add fwd in layernorm and groupnorm (profiler * Fix compile error --------- Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com> [ROCm/composable_kernel commit: a3d9a2cd42]
2026-05-17 11:30:02 +00:00 · 2023-11-09 08:34:51 +08:00
parent e857da2f34
commit b64f30e733
59 changed files with 1271 additions and 675 deletions
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -139,7 +139,7 @@ add_subdirectory(grouped_convnd_fwd)
 add_subdirectory(grouped_convnd_bwd_weight)
 add_subdirectory(block_to_ctile_map)
 add_subdirectory(softmax)
-add_subdirectory(normalization)
+add_subdirectory(normalization_fwd)
 add_subdirectory(data_type)
 add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
--- a/test/normalization/CMakeLists.txt
+++ b/test/normalization/CMakeLists.txt
@@ -1,21 +0,0 @@
-add_custom_target(test_normalization)
-add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
-if(result EQUAL 0)
-  target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
-  add_dependencies(test_normalization test_layernorm2d_fp32)
-endif()
-add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
-if(result EQUAL 0)
-  target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
-  add_dependencies(test_normalization test_groupnorm_fp32)
-endif()
-add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
-if(result EQUAL 0)
-  target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
-  add_dependencies(test_normalization test_layernorm2d_fp16)
-endif()
-add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
-if(result EQUAL 0)
-  target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
-  add_dependencies(test_normalization test_groupnorm_fp16)
-endif()
--- a/test/normalization_fwd/CMakeLists.txt
+++ b/test/normalization_fwd/CMakeLists.txt
@@ -0,0 +1,30 @@
+add_custom_target(test_normalization_fwd)
+add_gtest_executable(test_layernorm2d_fwd_fp32 test_layernorm2d_fwd_fp32.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_layernorm2d_fwd_fp32 PRIVATE utility device_normalization_fwd_instance)
+  add_dependencies(test_normalization_fwd test_layernorm2d_fwd_fp32)
+endif()
+
+add_gtest_executable(test_groupnorm_fwd_fp32 test_groupnorm_fwd_fp32.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_groupnorm_fwd_fp32 PRIVATE utility device_normalization_fwd_instance)
+  add_dependencies(test_normalization_fwd test_groupnorm_fwd_fp32)
+endif()
+
+add_gtest_executable(test_layernorm2d_fwd_fp16 test_layernorm2d_fwd_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_layernorm2d_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
+  add_dependencies(test_normalization_fwd test_layernorm2d_fwd_fp16)
+endif()
+
+add_gtest_executable(test_layernorm4d_fwd_fp16 test_layernorm4d_fwd_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_layernorm4d_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
+  add_dependencies(test_normalization_fwd test_layernorm4d_fwd_fp16)
+endif()
+
+add_gtest_executable(test_groupnorm_fwd_fp16 test_groupnorm_fwd_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_groupnorm_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
+  add_dependencies(test_normalization_fwd test_groupnorm_fwd_fp16)
+endif()
--- a/test/normalization_fwd/test_groupnorm_fwd_fp16.cpp
+++ b/test/normalization_fwd/test_groupnorm_fwd_fp16.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.

 #include "gtest/gtest.h"
-#include "profiler/profile_groupnorm_impl.hpp"
+#include "profiler/profile_groupnorm_fwd_impl.hpp"

 using F16 = ck::half_t;
 using F32 = float;
--- a/test/normalization_fwd/test_groupnorm_fwd_fp32.cpp
+++ b/test/normalization_fwd/test_groupnorm_fwd_fp32.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.

 #include "gtest/gtest.h"
-#include "profiler/profile_groupnorm_impl.hpp"
+#include "profiler/profile_groupnorm_fwd_impl.hpp"

 using F16 = ck::half_t;
 using F32 = float;
--- a/test/normalization_fwd/test_layernorm2d_fwd_fp16.cpp
+++ b/test/normalization_fwd/test_layernorm2d_fwd_fp16.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.

 #include "gtest/gtest.h"
-#include "profiler/profile_layernorm_impl.hpp"
+#include "profiler/profile_layernorm_fwd_impl.hpp"

 using F16 = ck::half_t;
 using F32 = float;
--- a/test/normalization_fwd/test_layernorm2d_fwd_fp32.cpp
+++ b/test/normalization_fwd/test_layernorm2d_fwd_fp32.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.

 #include "gtest/gtest.h"
-#include "profiler/profile_layernorm_impl.hpp"
+#include "profiler/profile_layernorm_fwd_impl.hpp"

 using F16 = ck::half_t;
 using F32 = float;
--- a/test/normalization_fwd/test_layernorm4d_fwd_fp16.cpp
+++ b/test/normalization_fwd/test_layernorm4d_fwd_fp16.cpp
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_layernorm_fwd_impl.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+using ck::index_t;
+
+template <typename Tuple>
+class TestLayernorm4d : public ::testing::Test
+{
+    protected:
+    using XDataType              = std::tuple_element_t<0, Tuple>;
+    using GammaDataType          = std::tuple_element_t<1, Tuple>;
+    using BetaDataType           = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType        = std::tuple_element_t<3, Tuple>;
+    using YDataType              = std::tuple_element_t<4, Tuple>;
+    using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
+
+    void Run()
+    {
+        // [N, D], reduce D
+        std::vector<std::vector<ck::index_t>> lengths = {
+            {1, 1, 1, 1}, {7, 7, 7, 7}, {256, 16, 16, 8}};
+
+        for(auto length : lengths)
+        {
+            bool success = ck::profiler::profile_layernorm_impl<XDataType,
+                                                                GammaDataType,
+                                                                BetaDataType,
+                                                                ComputeDataType,
+                                                                YDataType,
+                                                                SaveMeanInvStdDataType,
+                                                                true,
+                                                                4>(true, 2, false, false, length);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes = ::testing::Types<
+    // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
+    std::tuple<F16, F16, F16, F32, F16, F32>>;
+
+TYPED_TEST_SUITE(TestLayernorm4d, KernelTypes);
+TYPED_TEST(TestLayernorm4d, Test_FP16) { this->Run(); }