Merge remote-tracking branch 'origin/develop' into wavelet_model

example/12_reduce/reduce_blockwise_two_call.cpp

@@ -14,11 +14,11 @@
 #include "ck/tensor_operation/gpu/device/device_reduce_multiblock.hpp"
 
 #include "ck/library/utility/check_err.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/host_tensor/host_common_util.hpp"
-#include "ck/library/host_tensor/host_reduction.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/utility/host_reduction.hpp"
 
 using namespace ck;
 using namespace ck::tensor_operation::device;
@@ -171,13 +171,13 @@ int main(int argc, char* argv[])
         }
 
         if(beta != 0.0f)
-            for(size_t i = 0; i < out_ref.mDesc.GetElementSpace(); i++)
+            for(size_t i = 0; i < out_ref.mDesc.GetElementSpaceSize(); i++)
                 out.mData[i] = out_ref.mData[i];
     };
 
-    DeviceMem in_1_dev(sizeof(InOutDataType) * in_1.mDesc.GetElementSpace());
-    DeviceMem in_2_dev(sizeof(InOutDataType) * in_2.mDesc.GetElementSpace());
-    DeviceMem out_dev(sizeof(InOutDataType) * out.mDesc.GetElementSpace());
+    DeviceMem in_1_dev(sizeof(InOutDataType) * in_1.mDesc.GetElementSpaceSize());
+    DeviceMem in_2_dev(sizeof(InOutDataType) * in_2.mDesc.GetElementSpaceSize());
+    DeviceMem out_dev(sizeof(InOutDataType) * out.mDesc.GetElementSpaceSize());
 
     in_1_dev.ToDevice(in_1.mData.data());
 

example/12_reduce/reduce_example_common.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+
+template <ck::index_t Rank, ck::index_t NumReduceDim>
+std::vector<int> get_invariant_dims(const std::vector<int>& reduceDims)
+{
+    assert(NumReduceDim == reduceDims.size());
+
+    int reduceFlag = 0;
+
+    // flag the bits for the reduceDims
+    for(int i = 0; i < NumReduceDim; i++)
+    {
+        reduceFlag |= 1 << reduceDims[i];
+    };
+
+    std::vector<int> invariantDims;
+
+    // collect invariant dimensions
+    for(int i = 0; i < Rank; i++)
+        if((reduceFlag & (1 << i)) == 0)
+        {
+            invariantDims.push_back(i);
+        };
+
+    return invariantDims;
+};
+
+template <ck::index_t Rank, ck::index_t NumReduceDim>
+struct ReduceShape
+{
+    static constexpr ck::index_t Rank_         = Rank;
+    static constexpr ck::index_t NumReduceDim_ = NumReduceDim;
+};
+
+using reduce_shape_instances = std::tuple<ReduceShape<3, 1>,
+                                          ReduceShape<3, 2>,
+                                          ReduceShape<4, 1>,
+                                          ReduceShape<4, 2>,
+                                          ReduceShape<4, 3>,
+                                          ReduceShape<5, 1>,
+                                          ReduceShape<5, 2>,
+                                          ReduceShape<5, 3>,
+                                          ReduceShape<5, 4>>;

example/16_gemm_multi_d_multi_reduces/CMakeLists.txt

+add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+
+#exclude GEMM+max exampe from testing, since there is random failure on gfx908
+#https://github.com/ROCmSoftwarePlatform/composable_kernel/issues/358
+#TODO: fix the failure and re-enable this test
+add_example_executable_no_testing(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)

example/16_gemm_reduce/CMakeLists.txt

-add_example_executable(example_gemm_reduce_xdl_max_fp16 gemm_reduce_xdl_max_fp16.cpp)
-add_example_executable(example_gemm_reduce_xdl_mean_squaremean_fp16 gemm_reduce_xdl_mean_squaremean_fp16.cpp)

example/17_convnd_bwd_data/CMakeLists.txt

+add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
+target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)