Merge branch 'develop' into ck_host_lib

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_bf16_n4096_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  true, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_bf16_n4096_tp_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1,  256, 8,  true,  true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1,  256, 4,  true,  true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 2, 1, 1024, 2,  true,  true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t,  1, 4, 1, 1024, 1,  true,  true, true>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_bf16_n512_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 8,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  4,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  8,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_bf16_n64_n128_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn    pd     x      3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 1,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  1,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  2,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_bf16_n768_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  3,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1,  6,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::bf16_t, 1, 12,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n1024_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                                rm  rn  tm  tn  vn  pd      x     3p
+#if 0
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 8,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  8,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1, 16,  4,  64, 1,  true , true, false>>(const S&, A);
+
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  1,  1, 256, 4,  true , true, false>>(const S&, A);
+#endif
+
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 1, 2,  128, 8,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 2,  128, 4,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 2,  128, 2,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 1,  true,  true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n1536_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 4,   64, 8,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 2,  128, 4,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  256, 2,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 6, 1,  256, 1,  true,  true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n2048_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 1, 1,  256, 8,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1,  256, 4,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 2,  true,  true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 8, 1,  256, 1,  true,  true, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n256_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  1,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n3072_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  128, 8,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1,  256, 4,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 6, 1,  256, 2,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 3, 1, 1024, 1,  true, true, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n4096_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1,  256, 8,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 4,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1, 1024, 2,  true, true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1, 1024, 1,  true, true, false>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n4096_tp_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1,  256, 8,  true, true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1,  256, 4,  true, true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 2, 1, 1024, 2,  true, true, true>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t,  1, 4, 1, 1024, 1,  true, true, true>>(const S&, A);
+
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n512_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  1,  4,  64, 8,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  4,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  8,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n64_n128_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  1,  4,  64, 1,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  1,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  2,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_fp16_n768_instance.cpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "add_rmsnorm2d_rdquant_fwd_instance_common.hpp"
+
+// clang-format off
+//                                                               rm  rn  tm  tn  vn     pd    x     3p
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  3,  4,  64, 4,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1,  6,  4,  64, 2,  true , true, false>>(const S&, A);
+template float add_rmsnorm2d_rdquant_fwd_<trait_<ck_tile::fp16_t, 1, 12,  4,  64, 1,  true , true, false>>(const S&, A);
+// clang-format on

example/ck_tile/11_add_rmsnorm2d_rdquant/instances/add_rmsnorm2d_rdquant_fwd_instance_common.hpp

+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <ck_tile/core.hpp>
+#include "add_rmsnorm2d_rdquant_fwd.hpp"
+#include <iostream>
+
+#pragma once
+
+using S = ck_tile::stream_config;
+using A = add_rmsnorm2d_rdquant_fwd_args;
+
+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_ = add_rmsnorm2d_rdquant_fwd_traits_<DataType_,
+                                                 Repeat_M_,
+                                                 Repeat_N_,
+                                                 ThreadPerBlock_M_,
+                                                 ThreadPerBlock_N_,
+                                                 Vector_N_,
+                                                 kPadN_,
+                                                 kSaveInvRms_,
+                                                 kTwoPass_>;
+
+template <typename Traits_>
+float add_rmsnorm2d_rdquant_fwd_(const S& s, A a)
+{
+    using DataType = typename Traits_::DataType;
+
+    using PipelineProblem = ck_tile::AddRmsnorm2dRdquantFwdPipelineProblem<
+        typename AddRmsnormRdquantTypeConfig<DataType>::ADataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::BDataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::GammaDataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::ComputeDataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::XDataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::YScaleDataType,
+        typename AddRmsnormRdquantTypeConfig<DataType>::QYDataType,
+        typename Traits_::Shape,
+        Traits_::kPadN,
+        Traits_::kSaveX,
+        Traits_::kThreePass>;
+
+    using OnePassPipeline   = ck_tile::AddRmsnorm2dRdquantFwdPipelineOnePass<PipelineProblem>;
+    using ThreePassPipeline = ck_tile::AddRmsnorm2dRdquantFwdPipelineThreePass<PipelineProblem>;
+    using Pipeline = std::conditional_t<Traits_::kThreePass, ThreePassPipeline, OnePassPipeline>;
+
+    using Kernel = ck_tile::AddRmsnorm2dRdquantFwd<Pipeline>;
+
+    const dim3 grids                       = Kernel::GridSize(a);
+    constexpr dim3 blocks                  = Kernel::BlockSize();
+    constexpr ck_tile::index_t kBlockPerCu = 1;
+
+    auto kargs = Kernel::MakeKargs(a);
+    if(s.log_level_ > 0)
+        std::cout << ", " << Kernel::GetName() << std::flush;
+
+    return ck_tile::launch_kernel(
+        s, ck_tile::make_kernel<blocks.x, kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
+}

example/ck_tile/11_add_rmsnorm2d_rdquant/script/perf_test.sh

+#!/bin/sh
+EXE="$(find . -name tile_add_rmsnorm2d_rdquant_fwd -type f | head -n 1)"
+
+$EXE -m=1 -n=1 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=80 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=128 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=144 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=168 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=184 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=256 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=288 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=344 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=376 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=448 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=512 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=924 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=1024 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=1078 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=1996 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+$EXE -m=700 -n=4080 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
+
+$EXE -m=700 -n=80 -e=1e-12 -v=1  -prec=fp16 -repeat=1000
+$EXE -m=700 -n=128 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=144 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=168 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=184 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=256 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=288 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=344 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=376 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=448 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=512 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=924 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=1024 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=1078 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=1996 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
+$EXE -m=700 -n=4080 -e=1e-12 -v=1 -prec=fp16 -repeat=1000
\ No newline at end of file

example/ck_tile/11_add_rmsnorm2d_rdquant/script/smoke_test.sh

+#!/bin/sh
+EXE="$(find . -name tile_add_rmsnorm2d_rdquant_fwd -type f | head -n 1)"
+
+for pr_i in "fp16" "bf16" ; do
+$EXE -prec=$pr_i -m=99  -n=13
+$EXE -prec=$pr_i -m=17  -n=16
+$EXE -prec=$pr_i -m=1   -n=100
+$EXE -prec=$pr_i -m=4   -n=128
+$EXE -prec=$pr_i -m=80  -n=127
+$EXE -prec=$pr_i -m=22  -n=255 -stride=256
+$EXE -prec=$pr_i -m=7   -n=599
+$EXE -prec=$pr_i -m=19  -n=512
+$EXE -prec=$pr_i -m=33  -n=313 -stride=1000
+$EXE -prec=$pr_i -m=11  -n=510
+$EXE -prec=$pr_i -m=171 -n=676 -stride=818
+$EXE -prec=$pr_i -m=91  -n=636
+$EXE -prec=$pr_i -m=12  -n=768 -stride=800
+$EXE -prec=$pr_i -m=100 -n=766 -stride=812
+$EXE -prec=$pr_i -m=31  -n=1024
+$EXE -prec=$pr_i -m=64  -n=1000 -stride=1004
+$EXE -prec=$pr_i -m=8   -n=1501
+$EXE -prec=$pr_i -m=3   -n=1826
+$EXE -prec=$pr_i -m=5   -n=2040
+$EXE -prec=$pr_i -m=7   -n=2734
+$EXE -prec=$pr_i -m=1   -n=3182
+$EXE -prec=$pr_i -m=9   -n=4096
+$EXE -prec=$pr_i -m=3   -n=8192
+$EXE -prec=$pr_i -m=1   -n=10547
+$EXE -prec=$pr_i -m=3   -n=17134
+done

example/ck_tile/12_smoothquant/CMakeLists.txt

+function (add_smoothquant_example TARGET_NAME MAIN_SRC)
+    message("adding ${TARGET_NAME}")
+    # not using add_example_executable() to add target, since we don't want this to have
+    # to be included in "make all/install/check"
+    add_executable(${TARGET_NAME} EXCLUDE_FROM_ALL ${MAIN_SRC})
+    target_include_directories(${TARGET_NAME} PRIVATE ${CMAKE_CURRENT_LIST_DIR})
+
+    foreach(source IN LISTS ARGN)
+        list(APPEND INSTANCE_SRCS ${source})
+    endforeach()
+
+    target_sources(${TARGET_NAME} PRIVATE ${INSTANCE_SRCS})
+
+    set(COMPILE_OPTIONS)
+    # NOTE: we turn off undefined-func-template to let source compile without explicit declare function specializations
+    list(APPEND COMPILE_OPTIONS -Wno-undefined-func-template -Wno-float-equal)
+
+    target_compile_options(${TARGET_NAME} PRIVATE ${COMPILE_OPTIONS})
+endfunction(add_smoothquant_example TARGET_NAME MAIN_SRC)
+
+file(GLOB INSTANCE_SRCS instances/*.cpp)
+
+add_smoothquant_example(tile_smoothquant smoothquant.cpp ${INSTANCE_SRCS})
+add_smoothquant_example(tile_example_smoothquant example_smoothquant.cpp)

example/ck_tile/12_smoothquant/README.md

+# smoothquant
+
+This folder contains example for smoothquant 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_smoothquant -j
+```
+This will result in an executable `build/bin/tile_smoothquant`
+
+## cmdline
+```
+args:
+          -m    m dimension (default:3328)
+          -n    m dimension (default:4096)
+          -v    cpu validation or not (default:1)
+       -prec    precision (default:fp16)
+```