Merge commit '30480288' into conv_quant_int8

include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp → include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp

@@ -13,7 +13,7 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_multiple_r.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
@@ -101,9 +101,9 @@ struct ComputePtrOffsetOfStridedBatch
  * \note Using \p ComputePtrOffsetOfBatch gives us the flexibility that 2 workgroups can compute 2
  * tiles from different matrices. Keep in mind that these 2 matrices can share the same grid
  * descriptor (like in BatchedGEMM), or use their own grid descriptors (in GroupedGemm). \link
- * device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for \link
- * DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the computing of
- * pointer offset into \p ComputePtrOffsetOfStridedBatch.
+ * impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for
+ * \link DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the
+ * computing of pointer offset into \p ComputePtrOffsetOfStridedBatch.
  *
  * \note \p Block2ETileMap allows customized mapping between a workgroup and the C-tile it computes.
  * Together with \p ComputePtrOffsetOfBatch, we can reuse GridwiseGemm (and GridwiseGemm fusion ) to

include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp → include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp

@@ -90,9 +90,9 @@ struct ComputePtrOffsetOfStridedBatch
  * \note Using \p ComputePtrOffsetOfBatch gives us the flexibility that 2 workgroups can compute 2
  * tiles from different matrices. Keep in mind that these 2 matrices can share the same grid
  * descriptor (like in BatchedGEMM), or use their own grid descriptors (in GroupedGemm). \link
- * device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for \link
- * DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the computing of
- * pointer offset into \p ComputePtrOffsetOfStridedBatch.
+ * impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for
+ * \link DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the
+ * computing of pointer offset into \p ComputePtrOffsetOfStridedBatch.
  *
  * \note \p Block2ETileMap allows customized mapping between a workgroup and the C-tile it computes.
  * Together with \p ComputePtrOffsetOfBatch, we can reuse GridwiseGemm (and GridwiseGemm fusion ) to

include/ck/tensor_operation/gpu/device/device_multiple_reduce_multiblock.hpp → include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp

@@ -11,7 +11,7 @@
 
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
 #include "ck/tensor_operation/gpu/device/device_multiple_reduce.hpp"
-#include "ck/tensor_operation/gpu/device/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp"
 

include/ck/tensor_operation/gpu/device/device_multiple_reduce_threadwise.hpp → include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp

@@ -11,7 +11,7 @@
 
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
 #include "ck/tensor_operation/gpu/device/device_multiple_reduce.hpp"
-#include "ck/tensor_operation/gpu/device/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"

include/ck/tensor_operation/gpu/device/device_normalization_impl.hpp → include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp

@@ -9,7 +9,7 @@
 #include "ck/utility/reduction_operator.hpp"
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
 #include "ck/tensor_operation/gpu/device/device_reduce.hpp"
-#include "ck/tensor_operation/gpu/device/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_layernorm_welford_variance.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp"
 #include "ck/host_utility/device_prop.hpp"