Merge remote-tracking branch 'origin/develop' into myamlak/cgemm

bb1f8082 · root · 97ac5007 · 82d7d993 · bb1f8082 · bb1f8082
Commit bb1f8082 authored May 26, 2022 by root
20 changed files
--- a/library/include/ck/library/host_tensor/host_reduce_util.hpp
+++ b/library/include/ck/library/host_tensor/host_reduce_util.hpp
--- a/library/include/ck/library/host_tensor/host_reduction.hpp
+++ b/library/include/ck/library/host_tensor/host_reduction.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16.hpp
 #ifndef DEVICE_REDUCE_INSTANCE_BLOCKWISE_B16_F32_B16_HPP
 #define DEVICE_REDUCE_INSTANCE_BLOCKWISE_B16_F32_B16_HPP

-#include "reduction_enums.hpp"
-#include "reduction_operator_mapping.hpp"
+#include "data_type.hpp"
 #include "device_reduce_instance_blockwise.hpp"

 namespace ck {

--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f16_f16_f16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f16_f16_f16.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_b16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_b16.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_f16.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_f16.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_f32.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f32_f32_f32.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f64_f64_f32.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f64_f64_f32.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f64_f64_f64.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_f64_f64_f64.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_i32_i32_i8.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_second_call_i32_i32_i8.hpp