Merge branch 'develop' into codegen-enable-hiprtc

example/ck_tile/09_topk_softmax/topk_softmax_api.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+#include "ck_tile/core.hpp"
+#include "ck_tile/host.hpp"
+#include "ck_tile/ops/topk_softmax.hpp"
+#include <string>
+
+struct topk_softmax_trait
+{
+    std::string input_type;
+    std::string weight_type; // currently always float
+    int experts;
+};
+
+struct topk_softmax_kargs : public ck_tile::TopkSoftmaxHostArgs
+{
+};
+
+float topk_softmax(topk_softmax_trait t, topk_softmax_kargs a, ck_tile::stream_config s);

example/ck_tile/10_rmsnorm2d/CMakeLists.txt

+set(TILE_RMSNORM2D_FWD "tile_rmsnorm2d_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"
+message("adding ${TILE_RMSNORM2D_FWD}")
+file(GLOB INSTANCE_SRCS instances/*.cpp)
+add_executable(${TILE_RMSNORM2D_FWD} EXCLUDE_FROM_ALL rmsnorm2d_fwd.cpp)
+target_include_directories(${TILE_RMSNORM2D_FWD} PRIVATE ${CMAKE_CURRENT_LIST_DIR})
+target_sources(${TILE_RMSNORM2D_FWD} PRIVATE ${INSTANCE_SRCS})
+
+set(TILE_RMSNORM2D_FWD_COMPILE_OPTIONS)
+
+# NOTE: we turn off undefined-func-template to let source compile without explicit declare function specializations
+list(APPEND TILE_RMSNORM2D_FWD_COMPILE_OPTIONS -Wno-undefined-func-template -Wno-float-equal)
+
+target_compile_options(${TILE_RMSNORM2D_FWD} PRIVATE ${TILE_RMSNORM2D_FWD_COMPILE_OPTIONS})
+
+set(EXAMPLE_RMSNORM2D_FWD "tile_example_rmsnorm2d_fwd")
+add_executable(${EXAMPLE_RMSNORM2D_FWD} EXCLUDE_FROM_ALL example_rmsnorm2d_fwd.cpp)
+target_compile_options(${EXAMPLE_RMSNORM2D_FWD} PRIVATE ${TILE_RMSNORM2D_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)

example/ck_tile/10_rmsnorm2d/README.md

+# Rmsnorm2D forward
+
+This folder contains example for Rmsnorm2D forward using ck_tile tile-programming implementation.
+
+## build
+```
+# in the root of ck_tile
+mkdir build && cd build
+sh ../script/cmake-ck-dev.sh  ../ <arch>  # you can replace this <arch> to gfx90a, gfx942...
+make tile_rmsnorm2d_fwd -j
+```
+This will result in an executable `build/bin/tile_rmsnorm2d_fwd`
+
+## cmdline
+```
+args:
+          -m    m dimension (default:3328)
+          -n    m dimension (default:4096)
+          -e    epsilon (default:1e-5)
+          -v    cpu validation or not (default:1)
+       -prec    precision (default:fp16)
+```

example/ck_tile/10_rmsnorm2d/example_rmsnorm2d_fwd.cpp

+#include "ck_tile/host.hpp"
+#include "ck_tile/core.hpp"
+#include "ck_tile/host/kernel_launch.hpp"
+#include "ck_tile/ops/rmsnorm2d.hpp"
+#include <cstring>
+
+auto create_args(int argc, char* argv[])
+{
+    ck_tile::ArgParser arg_parser;
+    arg_parser.insert("m", "3328", "m dimension")
+        .insert("n", "4096", "n dimension")
+        .insert("stride", "-1", "stride per row, if -1 then equal to n")
+        .insert("e", "1e-5", "epsilon")
+        .insert("v", "1", "cpu validation or not")
+        .insert("prec", "fp16", "precision")
+        .insert("warmup", "0", "cold iter")
+        .insert("repeat", "1", "hot iter");
+
+    bool result = arg_parser.parse(argc, argv);
+    return std::make_tuple(result, arg_parser);
+}
+
+template <typename DataType>
+bool run(const ck_tile::ArgParser& arg_parser)
+{
+    ck_tile::index_t m      = arg_parser.get_int("m");
+    ck_tile::index_t n      = arg_parser.get_int("n");
+    ck_tile::index_t stride = arg_parser.get_int("stride");
+    if(stride < 0)
+        stride = n;
+    float epsilon         = arg_parser.get_float("e");
+    std::string data_type = arg_parser.get_str("prec");
+    int do_validation     = arg_parser.get_int("v");
+    int warmup            = arg_parser.get_int("warmup");
+    int repeat            = arg_parser.get_int("repeat");
+
+    assert(stride >= n);
+
+    using XDataType      = DataType;
+    using YDataType      = DataType;
+    using GammaDataType  = DataType;
+    using InvRmsDataType = ck_tile::null_type;
+
+    using ComputeDataType = float;
+
+    // host verify
+    ck_tile::HostTensor<XDataType> x_host({m, n}, {stride, 1});
+    ck_tile::HostTensor<GammaDataType> gamma_host({n});
+
+    ck_tile::HostTensor<YDataType> y_host_ref({m, n}, {stride, 1});
+    ck_tile::HostTensor<YDataType> y_host_dev({m, n}, {stride, 1});
+
+    ck_tile::HostTensor<InvRmsDataType> invRms_host_ref({m});
+
+    ck_tile::FillUniformDistribution<XDataType>{-.5f, .5f}(x_host);
+    ck_tile::FillUniformDistribution<GammaDataType>{-.5f, .5f}(gamma_host);
+
+    ck_tile::DeviceMem x_buf(x_host.get_element_space_size_in_bytes());
+    ck_tile::DeviceMem gamma_buf(gamma_host.get_element_space_size_in_bytes());
+    ck_tile::DeviceMem y_buf(y_host_dev.get_element_space_size_in_bytes());
+
+    x_buf.ToDevice(x_host.data());
+    gamma_buf.ToDevice(gamma_host.data());
+
+    constexpr bool kTwoPass = true;
+
+    using BlockWarps = ck_tile::sequence<2, 2>;
+    using BlockTile  = ck_tile::sequence<2, 128>;
+    using WarpTile   = ck_tile::sequence<1, 64>;
+    using Vector     = ck_tile::sequence<1, 1>;
+
+    using Shape   = ck_tile::Rmsnorm2dShape<BlockTile, BlockWarps, WarpTile, Vector>;
+    using Problem = ck_tile::Rmsnorm2dFwdPipelineProblem<XDataType,
+                                                         GammaDataType,
+                                                         ComputeDataType,
+                                                         YDataType,
+                                                         InvRmsDataType,
+                                                         Shape,
+                                                         true,  // kPadN
+                                                         false, // kSaveInvRms
+                                                         kTwoPass>;
+
+    using OnePassPipeline = ck_tile::Rmsnorm2dFwdPipelineOnePass<Problem>;
+    using TwoPassPipeline = ck_tile::Rmsnorm2dFwdPipelineTwoPass<Problem>;
+    using Pipeline        = std::conditional_t<kTwoPass, TwoPassPipeline, OnePassPipeline>;
+    using Kernel          = ck_tile::Rmsnorm2dFwd<Pipeline>;
+
+    ck_tile::Rmsnorm2dFwdHostArgs args{x_buf.GetDeviceBuffer(),
+                                       gamma_buf.GetDeviceBuffer(),
+                                       y_buf.GetDeviceBuffer(),
+                                       nullptr,
+                                       epsilon,
+                                       m,
+                                       n,
+                                       stride};
+
+    auto kargs = Kernel::MakeKargs(args);
+
+    const dim3 grids                       = Kernel::GridSize(args);
+    constexpr dim3 blocks                  = Kernel::BlockSize();
+    constexpr ck_tile::index_t kBlockPerCu = 1;
+    auto s = ck_tile::stream_config{nullptr, true, 0, warmup, repeat};
+
+    ck_tile::launch_kernel(
+        s, ck_tile::make_kernel<blocks.x, kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
+
+    bool pass = true;
+
+    if(do_validation)
+    {
+        // reference
+        ck_tile::reference_rmsnorm2d_fwd<XDataType,
+                                         GammaDataType,
+                                         ComputeDataType,
+                                         YDataType,
+                                         InvRmsDataType>(
+            x_host, gamma_host, y_host_ref, invRms_host_ref, epsilon);
+
+        y_buf.FromDevice(y_host_dev.data());
+
+        auto [rtol, atol] = ck_tile::make_tuple(1e-3, 1e-3);
+        if(stride == n)
+        {
+            pass = ck_tile::check_err(
+                y_host_dev, y_host_ref, std::string("OUT Error: Incorrect results!"), rtol, atol);
+        }
+        else
+        {
+            for(int i_r = 0; i_r < m; i_r++)
+            {
+                std::vector<YDataType> y_host_dev_row(y_host_dev.begin() + i_r * stride,
+                                                      y_host_dev.begin() + i_r * stride + n);
+                std::vector<YDataType> y_host_ref_row(y_host_ref.begin() + i_r * stride,
+                                                      y_host_ref.begin() + i_r * stride + n);
+                pass &= ck_tile::check_err(y_host_dev_row,
+                                           y_host_ref_row,
+                                           std::string("OUT[") + std::to_string(i_r) +
+                                               std::string("] Error: Incorrect results!"),
+                                           rtol,
+                                           atol);
+            }
+        }
+
+        std::cout << "[" << data_type << "]"
+                  << " m:" << m << ", n:" << n << ", stride:" << stride
+                  << ", valid:" << (pass ? "y" : "n") << std::flush << std::endl;
+    }
+
+    return pass;
+}
+
+int main(int argc, char* argv[])
+{
+    auto [result, arg_parser] = create_args(argc, argv);
+    if(!result)
+        return -1;
+
+    const std::string data_type = arg_parser.get_str("prec");
+    if(data_type == "fp16")
+    {
+        return run<ck_tile::half_t>(arg_parser) ? 0 : -2;
+    }
+
+    return -3;
+}

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_api.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <ck_tile/core.hpp>
+#include "rmsnorm2d_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 kSaveInvRms_,
+          bool kTwoPass_>
+using trait_ = rmsnorm2d_fwd_traits_<DataType_,
+                                     Repeat_M_,
+                                     Repeat_N_,
+                                     ThreadPerBlock_M_,
+                                     ThreadPerBlock_N_,
+                                     Vector_N_,
+                                     kPadN_,
+                                     kSaveInvRms_,
+                                     kTwoPass_>;
+
+template <typename data_type>
+float rmsnorm2d_fwd_b16_(rmsnorm2d_fwd_traits /*t*/,
+                         rmsnorm2d_fwd_args a,
+                         const ck_tile::stream_config& s)
+{
+#if 1
+    float r = -1;
+    // clang-format off
+    //                                            rm  rn  tm   tn  vn  pd    rms     2p
+    if(a.n <= 64) {
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type, 1,  1,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 1,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 1,  4,  64, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 3,  4,  64, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 6,  4,  64, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 1, 2,  128, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 2,  128, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4, 2,  128, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 3, 4,   64, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 3, 2,  128, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 3, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 1, 1,  256, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 3, 1,  128, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 3, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 6, 1,  256, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 8,  true,  false, false>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 4,  true,  false, false>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 1, 1024, 2,  true,  false, false>>(s, a);
+        else
+            r = rmsnorm2d_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 = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 1,  256, 8,  true,  false, true>>(s, a);
+        else if (a.n % 4 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4, 1,  256, 4,  true,  false, true>>(s, a);
+        else if (a.n % 2 == 0)
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 2, 1, 1024, 2,  true,  false, true>>(s, a);
+        else
+            r = rmsnorm2d_fwd_<trait_<data_type,  1, 4, 1, 1024, 1,  true,  false, true>>(s, a);
+    }
+    return r;
+#else
+    return rmsnorm2d_fwd_<trait_<data_type,  1, 1,  1,  256, 4,  true,  false, false>>(s, a);
+#endif
+    // clang-format on
+}
+
+float rmsnorm2d_fwd(rmsnorm2d_fwd_traits t, rmsnorm2d_fwd_args a, const ck_tile::stream_config& s)
+{
+
+    float r = -1;
+    if(t.data_type.compare("fp16") == 0)
+    {
+        return rmsnorm2d_fwd_b16_<ck_tile::fp16_t>(t, a, s);
+    }
+    else if(t.data_type.compare("bf16") == 0)
+    {
+        return rmsnorm2d_fwd_b16_<ck_tile::bf16_t>(t, a, s);
+    }
+    if(r < 0)
+        throw std::runtime_error("Without supported instances!");
+
+    return r;
+}

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n1024_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+#if 0
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 8,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  8,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1, 16,  4,  64, 1,  true , false, false>>(const S&, A);
+
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  1, 256, 4,  true , false, false>>(const S&, A);
+#endif
+
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 1, 2,  128, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 2,  128, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n1536_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 4,   64, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 6, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n2048_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 1, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 8, 1,  256, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n256_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n3072_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  128, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 6, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 3, 1, 1024, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n4096_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n4096_tp_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  false, true>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  false, true>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  false, true>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  false, true>>(const S&, A);
+
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n512_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 8,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  8,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n64_n128_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 1,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_bf16_n768_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  3,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1,  6,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::bf16_t, 1, 12,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_fp16_n1024_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+#if 0
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 8,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  8,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1, 16,  4,  64, 1,  true , false, false>>(const S&, A);
+
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  1,  1, 256, 4,  true , false, false>>(const S&, A);
+#endif
+
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 1, 2,  128, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 2, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 2,  128, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_fp16_n1536_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 4,   64, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 2,  128, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 6, 1,  256, 1,  true,  false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_fp16_n2048_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 1, 1,  256, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 8, 1,  256, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_fp16_n256_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  1,  4,  64, 4,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 2,  true , false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 1,  true , false, false>>(const S&, A);
+// clang-format on

example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_fp16_n3072_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "rmsnorm2d_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                       rm  rn  tm  tn  vn  pd    rms     2p
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  128, 8,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  256, 4,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 6, 1,  256, 2,  true,  false, false>>(const S&, A);
+template float rmsnorm2d_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1, 1024, 1,  true,  false, false>>(const S&, A);
+
+// clang-format on