update layernorm (#1570)

* port layernorm * change warp_welford.hpp * Update warpshuffle * 1. Add save mean and save std back 2. Move construction of tensor_view and tile_window to operator() * refine welford max count calculation * unify layernorm api * Rename file * Remove save mean and inv std * Revert "refine welford max count calculation" This reverts commit 0223658. * Fix order of parameter * refine welford max count calculation again * Remove fp32 instances * Fix bug of padding * refactor api * Support bf16 * Extract common function * Refine arg of operator() * Add kMThreadPerBlock to template parameter * clang format * Refine variable name * Refine file name * remove redundant line * refactor layernorm2d pipeline and add block-per-block utility * fix name * rename more * add more block-per-tile instance * remove duplicated define * update instance for 2048, 1024 case * support up to 2048 now * opt loading * add n1536 * Add two pass pipeline * format * Fix incorrect type * parallel compilation * Use smaller N * fix 2p pass * Support Repeat_M in distribution * Refine nameing * Add reduce example --------- Co-authored-by: letaoqin <[email protected]> Co-authored-by: aska-0096 <[email protected]> Co-authored-by: rocking <[email protected]> Co-authored-by: carlushuang <[email protected]>
ROCm · Oct 22, 2024 · 0394f8a · 0394f8a
1 parent 3f71093
commit 0394f8a
Show file tree

Hide file tree

Showing 59 changed files with 2,916 additions and 1,041 deletions.
diff --git a/example/66_complex_contraction_bilinear/run_complex_contraction_bilinear_example.inc b/example/66_complex_contraction_bilinear/run_complex_contraction_bilinear_example.inc
diff --git a/example/ck_tile/02_layernorm2d/CMakeLists.txt b/example/ck_tile/02_layernorm2d/CMakeLists.txt
@@ -1,4 +1,21 @@
+set(EXAMPLE_LAYERNORM2D_FWD "tile_example_layernorm2d_fwd")
 # not using add_example_executable() to add this target, since we don't want this to have
 # to be included in "make all/install/check"
-add_executable(tile_example_layernorm2d_fwd EXCLUDE_FROM_ALL layernorm2d_fwd.cpp)
-target_compile_options(tile_example_layernorm2d_fwd PRIVATE -DSAVE_MEAN_INV_STD)
+message("adding example ${EXAMPLE_LAYERNORM2D_FWD}")
+file(GLOB INSTANCE_SRCS instances/*.cpp)
+add_executable(${EXAMPLE_LAYERNORM2D_FWD} EXCLUDE_FROM_ALL layernorm2d_fwd.cpp)
+target_include_directories(${EXAMPLE_LAYERNORM2D_FWD} PRIVATE ${CMAKE_CURRENT_LIST_DIR})
+target_sources(${EXAMPLE_LAYERNORM2D_FWD} PRIVATE ${INSTANCE_SRCS})
+
+set(EXAMPLE_LAYERNORM2D_FWD_COMPILE_OPTIONS)
+
+# NOTE: we turn off undefined-func-template to let source compile without explicit declare function specializations
+list(APPEND EXAMPLE_LAYERNORM2D_FWD_COMPILE_OPTIONS -Wno-undefined-func-template -Wno-float-equal)
+
+target_compile_options(${EXAMPLE_LAYERNORM2D_FWD} PRIVATE ${EXAMPLE_LAYERNORM2D_FWD_COMPILE_OPTIONS})
+
+# TODO: we have to turn off this global prop, otherwise the progress bar generated
+# by cmake will print too many files, execvp: /bin/sh: Argument list too long
+# however, this property may affect global
+# TODO: consider codegen a makefile by us
+set_property(GLOBAL PROPERTY RULE_MESSAGES OFF)
diff --git a/example/ck_tile/02_layernorm2d/README.md b/example/ck_tile/02_layernorm2d/README.md
@@ -6,8 +6,7 @@ This folder contains example for Layernorm2D forward using ck_tile tile-programm
 ```
 # in the root of ck_tile
 mkdir build && cd build
-# you can replace <arch> with the appropriate architecture (for example gfx90a or gfx942) or leave it blank
-sh ../script/cmake-ck-dev.sh  ../ <arch>
+sh ../script/cmake-ck-dev.sh  ../ <arch>  # you can replace this <arch> to gfx90a, gfx942...
 make tile_example_layernorm2d_fwd -j
 ```
 This will result in an executable `build/bin/tile_example_layernorm2d_fwd`
@@ -20,4 +19,4 @@ args:
           -e    epsilon (default:1e-5)
           -v    cpu validation or not (default:1)
        -prec    precision (default:fp16)
-```
+```
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_api.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_api.cpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <ck_tile/core.hpp>
+#include "layernorm2d_fwd.hpp"
+
+template <typename DataType_,
+          ck_tile::index_t Repeat_M_,         // each thread repeat along M
+          ck_tile::index_t Repeat_N_,         // each thread repeat along N
+          ck_tile::index_t ThreadPerBlock_M_, // num threads along M
+          ck_tile::index_t ThreadPerBlock_N_, // num threads along N
+          ck_tile::index_t Vector_N_,         // vector size along N
+          bool kPadN_,
+          bool kSaveMeanInvStd_,
+          bool kTwoPass_>
+using trait_ = layernorm2d_fwd_traits_<DataType_,
+                                       Repeat_M_,
+                                       Repeat_N_,
+                                       ThreadPerBlock_M_,
+                                       ThreadPerBlock_N_,
+                                       Vector_N_,
+                                       kPadN_,
+                                       kSaveMeanInvStd_,
+                                       kTwoPass_>;
+
+template <typename data_type>
+float layernorm2d_fwd_b16_(layernorm2d_fwd_traits /*t*/,
+                           layernorm2d_fwd_args a,
+                           const ck_tile::stream_config& s)
+{
+#if 1
+    float r = -1;
+    // clang-format off
+    //                                            rm  rn  tm   tn  vn  pd     mv     2p
+    if(a.n <= 64) {
+            r = layernorm2d_fwd_<trait_<data_type, 1,  1,  4,  64, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 128) {
+        if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type, 1,  1,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type, 1,  2,  4,  64, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 256) {
+        if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 1,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4,  4,  64, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 512) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 1,  4,  64, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 8,  4,  64, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 768) {
+        if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 6,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1,12,  4,  64, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 1024) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 1, 2,  128, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 2,  128, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 2,  128, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 1536) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 4,   64, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 2,  128, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 6, 1,  256, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 2048) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 1, 1,  256, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 8, 1,  256, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 3072) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 1,  128, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 6, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 3, 1, 1024, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n <= 4096) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 1, 1024, 2,  true,  false, false>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1, 1024, 1,  true,  false, false>>(s, a);
+    }
+    else if(a.n > 4096) {
+        if (a.n % 8 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 8,  true,  false, true>>(s, a);
+        else if (a.n % 4 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 4,  true,  false, true>>(s, a);
+        else if (a.n % 2 == 0)
+            r = layernorm2d_fwd_<trait_<data_type,  1, 2, 1, 1024, 2,  true,  false, true>>(s, a);
+        else
+            r = layernorm2d_fwd_<trait_<data_type,  1, 4, 1, 1024, 1,  true,  false, true>>(s, a);
+    }
+    return r;
+#else
+    return layernorm2d_fwd_<trait_<data_type,  1, 1,  1,  256, 4,  true,  false, false>>(s, a);
+#endif
+    // clang-format on
+}
+
+float layernorm2d_fwd(layernorm2d_fwd_traits t,
+                      layernorm2d_fwd_args a,
+                      const ck_tile::stream_config& s)
+{
+
+    float r = -1;
+    if(t.data_type.compare("fp16") == 0)
+    {
+        return layernorm2d_fwd_b16_<ck_tile::fp16_t>(t, a, s);
+    }
+    else if(t.data_type.compare("bf16") == 0)
+    {
+        return layernorm2d_fwd_b16_<ck_tile::bf16_t>(t, a, s);
+    }
+    if(r < 0)
+        throw std::runtime_error("Without supported instances!");
+
+    return r;
+}
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n1024_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n1024_instance.cpp
@@ -0,0 +1,22 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+#if 0
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 8,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 4,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  8,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1, 16,  4,  64, 1,  true , false, false>>(const S&, A);
+
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  1, 256, 4,  true , false, false>>(const S&, A);
+#endif
+
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 1, 2,  128, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 2,  128, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n1536_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n1536_instance.cpp
@@ -0,0 +1,13 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 4,   64, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 6, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n2048_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n2048_instance.cpp
@@ -0,0 +1,14 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 1, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 8, 1,  256, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n256_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n256_instance.cpp
@@ -0,0 +1,12 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 4,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n3072_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n3072_instance.cpp
@@ -0,0 +1,14 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  128, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 6, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1, 1024, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n4096_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n4096_instance.cpp
@@ -0,0 +1,14 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n4096_tp_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n4096_tp_instance.cpp
@@ -0,0 +1,14 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  false, true>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  false, true>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  false, true>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  false, true>>(const S&, A);
+
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n512_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n512_instance.cpp
@@ -0,0 +1,13 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 8,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 4,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  8,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n64_n128_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n64_n128_instance.cpp
@@ -0,0 +1,12 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 1,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n768_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_bf16_n768_instance.cpp
@@ -0,0 +1,12 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  3,  4,  64, 4,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  6,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::bf16_t, 1, 12,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n1024_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n1024_instance.cpp
@@ -0,0 +1,22 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+#if 0
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 8,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 4,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  8,  4,  64, 2,  true , false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t, 1, 16,  4,  64, 1,  true , false, false>>(const S&, A);
+
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  1,  1, 256, 4,  true , false, false>>(const S&, A);
+#endif
+
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 1, 2,  128, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 2, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 2,  128, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n1536_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n1536_instance.cpp
@@ -0,0 +1,13 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 4,   64, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 6, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on
diff --git a/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n2048_instance.cpp b/example/ck_tile/02_layernorm2d/instances/layernorm2d_fwd_fp16_n2048_instance.cpp
@@ -0,0 +1,14 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "layernorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd     mv     2p
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 1, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float layernorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 8, 1,  256, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on