Merge branch 'develop' into lwpck-471

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank3_reduce1_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank3_reduce2_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank3_reduce3_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank4_reduce1_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank4_reduce2_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank4_reduce3_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_f32_f32_rank4_reduce4_instances(
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <index_t Rank, index_t Reduce>
+using device_softmax_f32_f32_instances = std::tuple<
+    // clang-format off
+    //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,               1>, // fallback kernel
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,                8,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,                8,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               16,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               32,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,                8,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               16,              1,               4,               4>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               32,              1,               4,               4>,
+    // Reduction on middle dimensions
+    // InSrcVectorDim is 0 since we want to coalesce reads on M dimension
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                8,                4,              0,               1,               1>,
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                8,                4,              0,               4,               4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank3_instances(
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+void add_device_softmax_i8_i8_rank4_instances(
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank3_reduce1_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank3_reduce2_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank3_reduce3_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank4_reduce1_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank4_reduce2_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank4_reduce3_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_softmax_i8_i8_rank4_reduce4_instances(
+    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <index_t Rank, index_t Reduce>
+using device_softmax_i8_i8_instances = std::tuple<
+    // clang-format off
+    //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
+    // fallback kernel
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,               1,              1>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,               16,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               16,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               32,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               64,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               16,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               32,              1,              16,             16>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               64,              1,              16,             16>,
+    // Reduction on middle dimensions
+    // InSrcVectorDim is 0 since we want to coalesce reads on M dimension
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                8,                8,              0,               1,              1>,
+    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                 32,                  8,               32,                8,              0,              16,              8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp"

library/include/ck/library/utility/algorithm.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <algorithm>
+#include <iterator>
+#include <type_traits>
+#include <utility>
+
+namespace ck {
+namespace ranges {
+template <typename InputRange, typename OutputIterator>
+auto copy(InputRange&& range, OutputIterator iter)
+    -> decltype(std::copy(std::begin(std::forward<InputRange>(range)),
+                          std::end(std::forward<InputRange>(range)),
+                          iter))
+{
+    return std::copy(std::begin(std::forward<InputRange>(range)),
+                     std::end(std::forward<InputRange>(range)),
+                     iter);
+}
+
+template <typename T, typename OutputRange>
+auto fill(OutputRange&& range, const T& init)
+    -> std::void_t<decltype(std::fill(std::begin(std::forward<OutputRange>(range)),
+                                      std::end(std::forward<OutputRange>(range)),
+                                      init))>
+{
+    std::fill(std::begin(std::forward<OutputRange>(range)),
+              std::end(std::forward<OutputRange>(range)),
+              init);
+}
+
+template <typename InputRange, typename OutputIterator, typename UnaryOperation>
+auto transform(InputRange&& range, OutputIterator iter, UnaryOperation unary_op)
+    -> decltype(std::transform(std::begin(range), std::end(range), iter, unary_op))
+{
+    return std::transform(std::begin(range), std::end(range), iter, unary_op);
+}
+
+} // namespace ranges
+} // namespace ck

library/include/ck/library/utility/check_err.hpp

@@ -15,18 +15,22 @@
 
 #include "ck/ck.hpp"
 #include "ck/utility/data_type.hpp"
-#include "ck/utility/span.hpp"
 #include "ck/utility/type.hpp"
 #include "ck/host_utility/io.hpp"
 
+#include "ck/library/utility/ranges.hpp"
+
 namespace ck {
 namespace utils {
 
-template <typename T>
-typename std::enable_if<std::is_floating_point<T>::value && !std::is_same<T, half_t>::value,
-                        bool>::type
-check_err(const std::vector<T>& out,
-          const std::vector<T>& ref,
+template <typename Range, typename RefRange>
+typename std::enable_if<
+    std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+        std::is_floating_point_v<ranges::range_value_t<Range>> &&
+        !std::is_same_v<ranges::range_value_t<Range>, half_t>,
+    bool>::type
+check_err(const Range& out,
+          const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-5,
           double atol            = 3e-6)
@@ -44,15 +48,17 @@ check_err(const std::vector<T>& out,
     double max_err = std::numeric_limits<double>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
-        err = std::abs(out[i] - ref[i]);
-        if(err > atol + rtol * std::abs(ref[i]) || !std::isfinite(out[i]) || !std::isfinite(ref[i]))
+        const double o = *std::next(std::begin(out), i);
+        const double r = *std::next(std::begin(ref), i);
+        err            = std::abs(o - r);
+        if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
             if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
-                          << "] != ref[" << i << "]: " << out[i] << " != " << ref[i] << std::endl;
+                          << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
             }
             res = false;
         }
@@ -64,10 +70,13 @@ check_err(const std::vector<T>& out,
     return res;
 }
 
-template <typename T>
-typename std::enable_if<std::is_same<T, bhalf_t>::value, bool>::type
-check_err(const std::vector<T>& out,
-          const std::vector<T>& ref,
+template <typename Range, typename RefRange>
+typename std::enable_if<
+    std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+        std::is_same_v<ranges::range_value_t<Range>, bhalf_t>,
+    bool>::type
+check_err(const Range& out,
+          const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-3,
           double atol            = 1e-3)
@@ -86,9 +95,9 @@ check_err(const std::vector<T>& out,
     double max_err = std::numeric_limits<float>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
-        double o = type_convert<float>(out[i]);
-        double r = type_convert<float>(ref[i]);
-        err      = std::abs(o - r);
+        const double o = type_convert<float>(*std::next(std::begin(out), i));
+        const double r = type_convert<float>(*std::next(std::begin(ref), i));
+        err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
@@ -108,10 +117,13 @@ check_err(const std::vector<T>& out,
     return res;
 }
 
-template <typename T>
-typename std::enable_if<std::is_same_v<T, half_t>, bool>::type
-check_err(span<const T> out,
-          span<const T> ref,
+template <typename Range, typename RefRange>
+typename std::enable_if<
+    std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+        std::is_same_v<ranges::range_value_t<Range>, half_t>,
+    bool>::type
+check_err(const Range& out,
+          const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double rtol            = 1e-3,
           double atol            = 1e-3)
@@ -126,12 +138,12 @@ check_err(span<const T> out,
     bool res{true};
     int err_count  = 0;
     double err     = 0;
-    double max_err = std::numeric_limits<T>::min();
+    double max_err = std::numeric_limits<ranges::range_value_t<Range>>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
-        double o = type_convert<float>(out[i]);
-        double r = type_convert<float>(ref[i]);
-        err      = std::abs(o - r);
+        const double o = type_convert<float>(*std::next(std::begin(out), i));
+        const double r = type_convert<float>(*std::next(std::begin(ref), i));
+        err            = std::abs(o - r);
         if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
         {
             max_err = err > max_err ? err : max_err;
@@ -151,26 +163,17 @@ check_err(span<const T> out,
     return res;
 }
 
-template <typename T>
-typename std::enable_if<std::is_same<T, half_t>::value, bool>::type
-check_err(const std::vector<T>& out,
-          const std::vector<T>& ref,
-          const std::string& msg = "Error: Incorrect results!",
-          double rtol            = 1e-3,
-          double atol            = 1e-3)
-{
-    return check_err(span<const T>{out}, span<const T>{ref}, msg, rtol, atol);
-}
-
-template <typename T>
-std::enable_if_t<(std::is_integral_v<T> && !std::is_same_v<T, bhalf_t>)
+template <typename Range, typename RefRange>
+std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+                  std::is_integral_v<ranges::range_value_t<Range>> &&
+                  !std::is_same_v<ranges::range_value_t<Range>, bhalf_t>)
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
-                     || std::is_same_v<T, int4_t>
+                     || std::is_same_v<ranges::range_value_t<Range>, int4_t>
 #endif
                  ,
                  bool>
-check_err(const std::vector<T>& out,
-          const std::vector<T>& ref,
+check_err(const Range& out,
+          const RefRange& ref,
           const std::string& msg = "Error: Incorrect results!",
           double                 = 0,
           double atol            = 0)
@@ -188,9 +191,9 @@ check_err(const std::vector<T>& out,
     int64_t max_err = std::numeric_limits<int64_t>::min();
     for(std::size_t i = 0; i < ref.size(); ++i)
     {
-        int64_t o = out[i];
-        int64_t r = ref[i];
-        err       = std::abs(o - r);
+        const int64_t o = *std::next(std::begin(out), i);
+        const int64_t r = *std::next(std::begin(ref), i);
+        err             = std::abs(o - r);
 
         if(err > atol)
         {