Merge remote-tracking branch 'origin/develop' into bwroblew/direct_loads

example/44_elementwise_permute/elementwise_permute_4D_fp16.cpp

@@ -19,13 +19,13 @@ using BDataType = F16;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 using DeviceElementwisePermuteInstance =
-    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ADataType>,
-                                                        ck::Tuple<BDataType>,
-                                                        PassThrough,
-                                                        4,
-                                                        8,
-                                                        ck::Sequence<8>,
-                                                        ck::Sequence<1>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ADataType>, // InDataTypeTuple
+                                                        ck::Tuple<BDataType>, // OutDataTypeTuple
+                                                        PassThrough,          // Elementwise op
+                                                        4,                    // NumDim
+                                                        8,                    // MPerThread
+                                                        ck::Sequence<8>,  // InScalarPerVectorSeq
+                                                        ck::Sequence<1>>; // OutScalarPerVectorSeq
 
 template <typename HostTensorA, typename HostTensorB, typename Functor>
 void host_elementwise4D(HostTensorB& B_nhwc, const HostTensorA& A_nchw, Functor functor)
@@ -99,7 +99,6 @@ int main()
 
     std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
               << std::endl;
-
     bool pass = true;
 
     if(do_verification)

example/53_layernorm_bwd/CMakeLists.txt

+add_example_executable(example_layernorm2d_bwd_fp16 layernorm2d_bwd_fp16.cpp)

example/54_groupnorm_bwd/CMakeLists.txt

+add_example_executable(example_groupnorm_bwd_fp16 groupnorm_bwd_fp16.cpp)

example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp

@@ -34,6 +34,7 @@ using AccDataType      = F32;
 using CShuffleDataType = F32;
 using DDataType        = F16;
 using EDataType        = F16;
+using ComputeDataType  = F16;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -291,6 +292,7 @@ int main(int argc, char* argv[])
                                                                       BDataType,
                                                                       CShuffleDataType,
                                                                       AccDataType,
+                                                                      ComputeDataType,
                                                                       PassThrough,
                                                                       BElementOp>;
 

example/62_conv_fwd_activ/CMakeLists.txt

@@ -30,6 +30,18 @@ foreach(gpu IN LISTS GPU_TARGETS)
       # Elu
       add_example_executable(example_convnd_fwd_xdl_elu_fp16 convnd_fwd_xdl_elu_fp16.cpp)
       add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_elu_fp16)
+      # ScaleAdd on A and B
+      add_example_executable(example_conv_fwd_xdl_scaleadd_ab_fp16 multi_AB/conv_fwd_xdl_scaleadd_ab_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_fp16)
+      add_example_executable(example_conv_fwd_xdl_scaleadd_ab_fp32 multi_AB/conv_fwd_xdl_scaleadd_ab_fp32.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_fp32)
+      add_example_executable(example_conv_fwd_xdl_scaleadd_ab_bf16 multi_AB/conv_fwd_xdl_scaleadd_ab_bf16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_bf16)
+      add_example_executable(example_conv_fwd_xdl_scaleadd_ab_int8 multi_AB/conv_fwd_xdl_scaleadd_ab_int8.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_conv_fwd_xdl_scaleadd_ab_int8)
+      # ScaleAdd ScaleAdd Relu
+      add_example_executable(example_convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16 convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_scaleadd_scaleadd_relu_fp16)
    set(target 1)
  endif()
 endforeach()