merge develop to attn-train-develop-qloop

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o

example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o

example/32_batched_gemm_scale_softmax_gemm/grouped_multihead_attention_train_v1.cpp

@@ -71,13 +71,13 @@ using Scale       = ck::tensor_operation::element_wise::Scale;
 using QKVElementOp = PassThrough;
 using YElementOp   = PassThrough;
 
-using InputDataType    = BF16;
-using OutputDataType   = F32;
-using GemmDataType     = BF16;
+using InputDataType    = F16;
+using OutputDataType   = F16;
+using GemmDataType     = F16;
 using AccDataType      = F32;
 using ShuffleDataType  = F32;
 using LSEDataType      = F32;
-using ZDataType        = INT32; // U16
+using ZDataType        = U16; // INT32
 using Acc0BiasDataType = ck::Tuple<>;
 using Acc1BiasDataType = ck::Tuple<>;
 
@@ -88,7 +88,7 @@ static constexpr ck::index_t NumDimK = 1;
 static constexpr ck::index_t NumDimO = 1;
 // When OutputDataType == F32,      bwd CShuffleBlockTransferScalarPerVector_NPerBlock = 4
 // When OutputDataType == F16/BF16, bwd CShuffleBlockTransferScalarPerVector_NPerBlock = 8
-static constexpr ck::index_t CShuffleBlockTransferScalarPerVector_NPerBlock = 4;
+static constexpr ck::index_t CShuffleBlockTransferScalarPerVector_NPerBlock = 8;
 
 static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKOPadding;
 #if USING_MASK
@@ -103,7 +103,7 @@ static constexpr auto TensorSpecQ   = ck::tensor_operation::device::TensorSpecia
 static constexpr auto TensorSpecK   = ck::tensor_operation::device::TensorSpecialization::Default;
 static constexpr auto TensorSpecV   = ck::tensor_operation::device::TensorSpecialization::Default;
 static constexpr auto TensorSpecY   = ck::tensor_operation::device::TensorSpecialization::Default;
-static constexpr bool Deterministic = true;
+static constexpr bool Deterministic = false;
 
 // DIM should be a multiple of 8.
 // If      DIM <= 32 , ues prototype1 1st template.

example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {

example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {

example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {

example/33_multiple_reduce/dual_reduce_common.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
@@ -217,8 +217,8 @@ int mean_meansquare_dual_reduce_test(size_t n,
     size_t invariant_total_length = n;
     size_t reduce_total_length    = h * w * c;
 
-    const AccDataType alpha = ck::type_convert<AccDataType>(1.0f);
-    const AccDataType beta  = ck::type_convert<AccDataType>(0.0f);
+    const double alpha = 1.0f;
+    const double beta  = 0.0f;
 
     std::size_t num_thread = 1;
 
@@ -267,8 +267,8 @@ int mean_meansquare_dual_reduce_test(size_t n,
         i_outLengths,
         {i_outStrides, i_outStrides},
         reduceDims,
-        {&alpha, &alpha},
-        {&beta, &beta},
+        {alpha, alpha},
+        {beta, beta},
         in_dev.GetDeviceBuffer(),
         {mean_dev.GetDeviceBuffer(), meansquare_dev.GetDeviceBuffer()},
         ck::make_tuple(InElementwiseOperation_Mean{}, InElementwiseOperation_Meansquare{}),

example/33_multiple_reduce/dual_reduce_multiblock.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>

example/33_multiple_reduce/dual_reduce_threadwise.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>

example/34_batchnorm/CMakeLists.txt

 add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
+add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
 add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
 add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp)

example/34_batchnorm/batchnorm_backward_nhwc.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>

example/34_batchnorm/batchnorm_common.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once