diff --git a/lib/nncase/v1/include/nncase/kernels/convolution.h b/lib/nncase/v1/include/nncase/kernels/convolution.h
index 7dd9474..7c16558 100644
--- a/lib/nncase/v1/include/nncase/kernels/convolution.h
+++ b/lib/nncase/v1/include/nncase/kernels/convolution.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/convolution.h b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/convolution.h
index 85b887d..673c07d 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/convolution.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/convolution.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/runtime_types.h b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/runtime_types.h
index 3ae0982..8aff824 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/runtime_types.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/runtime_types.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -35,17 +35,6 @@
     }                                 \
     }
 
-#define TYPE_IMPL_SELECT(type, IMPL)          \
-    switch (runtime::get_bytes(type))         \
-    {                                         \
-        IMPL(1, uint8_t);                     \
-        IMPL(2, uint16_t);                    \
-        IMPL(4, uint32_t);                    \
-        IMPL(8, uint64_t);                    \
-    default:                                  \
-        return err(std::errc::not_supported); \
-    }
-
 enum copy_impl_select
 {
     all_contiguous,
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/tensor_compute.h b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/tensor_compute.h
index e0b2723..51e1a0b 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/optimized/tensor_compute.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/optimized/tensor_compute.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,14 +20,33 @@ BEGIN_NS_NNCASE_KERNELS_CPU_OPT
 
 NNCASE_API result<void> concat(datatype_t type, gsl::span<const gsl::byte *const> inputs, gsl::byte *output, const runtime_shape_t &out_shape,
     gsl::span<const runtime_shape_t> in_strides, const runtime_shape_t &out_strides, size_t axis, const runtime_shape_t &concat_dims,
-    kernel_context &context) noexcept;
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> resize_bilinear(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> resize_nearest_neighbor(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> copy(datatype_t type, const gsl::byte *src, gsl::byte *dest,
+    const runtime_shape_t &shape, const runtime_shape_t &src_strides, const runtime_shape_t &dest_strides,
+    int dims_offset, copy_impl_select impl_select, kernel_context &context) noexcept;
+
+NNCASE_API result<void> gather(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, size_t axis,
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> gather_nd(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, size_t batch_dims,
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> onehot(datatype_t type, const int32_t *indices, gsl::byte *output, const runtime_shape_t &indices_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &out_strides, gsl::byte *depth, gsl::byte *off_value, gsl::byte *on_value, size_t axis, onehot_mode_t mode, kernel_context &context) noexcept;
 
 NNCASE_API result<void> slice(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const runtime_shape_t &begins, const runtime_shape_t &ends, const runtime_axis_t &strides,
-    kernel_context &context) noexcept;
-
-NNCASE_API result<void> copy(datatype_t type, const gsl::byte *src, gsl::byte *dest,
-    const runtime_shape_t &shape, const runtime_shape_t &src_strides, const runtime_shape_t &dest_strides, 
-    int dims_offset, copy_impl_select impl_select, kernel_context &context) noexcept;
+    kernel_context &context = default_kernel_context()) noexcept;
 
 END_NS_NNCASE_KERNELS_CPU_OPT
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/reference/convolution.h b/lib/nncase/v1/include/nncase/kernels/cpu/reference/convolution.h
index 6f38b4f..080d441 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/reference/convolution.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/reference/convolution.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/reference/nnil.h b/lib/nncase/v1/include/nncase/kernels/cpu/reference/nnil.h
index 6157865..8302be4 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/reference/nnil.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/reference/nnil.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/reference/reduce_window.h b/lib/nncase/v1/include/nncase/kernels/cpu/reference/reduce_window.h
index ac59809..9258f09 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/reference/reduce_window.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/reference/reduce_window.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/reference/runtime_types.h b/lib/nncase/v1/include/nncase/kernels/cpu/reference/runtime_types.h
index b90e146..0a068c3 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/reference/runtime_types.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/reference/runtime_types.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/cpu/reference/tensor_compute.h b/lib/nncase/v1/include/nncase/kernels/cpu/reference/tensor_compute.h
index 4025d42..22a8b46 100644
--- a/lib/nncase/v1/include/nncase/kernels/cpu/reference/tensor_compute.h
+++ b/lib/nncase/v1/include/nncase/kernels/cpu/reference/tensor_compute.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -49,6 +49,9 @@ NNCASE_API result<void> dequantize(datatype_t in_type, datatype_t out_type, cons
 NNCASE_API result<void> lut1d(datatype_t type, const gsl::byte *input, const gsl::byte *table, gsl::byte *output, const runtime_shape_t &shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const scalar &min, const scalar &max) noexcept;
 
+NNCASE_API result<void> onehot(datatype_t type, const int32_t *indices, gsl::byte *output, const runtime_shape_t &indices_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &out_strides, gsl::byte *depth, gsl::byte *off_value, gsl::byte *on_value, size_t axis, onehot_mode_t mode, kernel_context &context) noexcept;
+
 NNCASE_API result<void> pad(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const runtime_paddings_t &paddings, pad_mode_t mode, const scalar &pad_value,
     kernel_context &context) noexcept;
@@ -63,8 +66,22 @@ NNCASE_API result<void> unary(unary_op_t op, const float *input, float *output,
 NNCASE_API result<void> reduce(reduce_op_t op, float init_value, const float *input, float *output, const runtime_shape_t &in_shape, const runtime_shape_t &axis,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, bool keep_dims, kernel_context &context) noexcept;
 
+NNCASE_API result<void> resize_bilinear(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context) noexcept;
+
+NNCASE_API result<void> resize_nearest_neighbor(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context) noexcept;
+
 NNCASE_API result<void> slice(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const runtime_shape_t &begins, const runtime_shape_t &ends, const runtime_axis_t &strides,
     kernel_context &context) noexcept;
 
+NNCASE_API result<void> gather(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, size_t axis, kernel_context &context) noexcept;
+
+NNCASE_API result<void> gather_nd(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, size_t batch_dims, kernel_context &context) noexcept;
+
 END_NS_NNCASE_KERNELS_CPU_REF
diff --git a/lib/nncase/v1/include/nncase/kernels/k210/k210_kernels.h b/lib/nncase/v1/include/nncase/kernels/k210/k210_kernels.h
index 145a99b..5e448da 100644
--- a/lib/nncase/v1/include/nncase/kernels/k210/k210_kernels.h
+++ b/lib/nncase/v1/include/nncase/kernels/k210/k210_kernels.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/kernel_context.h b/lib/nncase/v1/include/nncase/kernels/kernel_context.h
index 2d61391..bd71438 100644
--- a/lib/nncase/v1/include/nncase/kernels/kernel_context.h
+++ b/lib/nncase/v1/include/nncase/kernels/kernel_context.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/kernel_utils.h b/lib/nncase/v1/include/nncase/kernels/kernel_utils.h
index 181ed20..77e81b4 100644
--- a/lib/nncase/v1/include/nncase/kernels/kernel_utils.h
+++ b/lib/nncase/v1/include/nncase/kernels/kernel_utils.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -15,6 +15,7 @@
 #pragma once
 #include <algorithm>
 #include <cassert>
+#include <cmath>
 #include <cstddef>
 #include <nncase/runtime/datatypes.h>
 #include <numeric>
@@ -27,6 +28,17 @@
 #define CXX_RESTRICT
 #endif
 
+#define TYPE_IMPL_SELECT(type, IMPL)          \
+    switch (runtime::get_bytes(type))         \
+    {                                         \
+        IMPL(1, uint8_t);                     \
+        IMPL(2, uint16_t);                    \
+        IMPL(4, uint32_t);                    \
+        IMPL(8, uint64_t);                    \
+    default:                                  \
+        return err(std::errc::not_supported); \
+    }
+
 BEGIN_NS_NNCASE_KERNELS
 
 template <class offset_type, class S, class It>
@@ -236,5 +248,46 @@ constexpr T quantize(float value, const quant_param_t &param) noexcept
 {
     return (T)clamp((int32_t)lrintf(value / param.scale + param.zero_point), (int32_t)std::numeric_limits<T>::lowest(), (int32_t)std::numeric_limits<T>::max());
 }
+
+inline std::pair<float, float> get_resize_scales(const runtime_shape_t &in_shape, int32_t out_h, int32_t out_w, bool align_corners)
+{
+    auto height_scale = (float)in_shape[2] / out_h;
+    auto width_scale = (float)in_shape[3] / out_w;
+    if (align_corners && out_h > 1)
+        height_scale = (float)(in_shape[2] - 1) / (out_h - 1);
+    if (align_corners && out_w > 1)
+        width_scale = (float)(in_shape[3] - 1) / (out_w - 1);
+    return { height_scale, width_scale };
+}
+
+inline void set_resize_bilinear(size_t value, float scale, bool half_pixel_centers, size_t shape_size, float &scaled_value, int32_t &v0, int32_t &v1)
+{
+    if (half_pixel_centers)
+    {
+        scaled_value = (value + 0.5f) * scale - 0.5f;
+    }
+    else
+    {
+        scaled_value = value * scale;
+    }
+    float scaled_value_floor = std::floor(scaled_value);
+    v0 = std::max(static_cast<int32_t>(scaled_value_floor), 0);
+    v1 = std::min(static_cast<int32_t>(std::ceil(scaled_value)), static_cast<int32_t>(shape_size - 1));
+}
+
+template <class T>
+inline size_t get_nearest_neighbor(T input_value, size_t shape_size, float scale, bool align_corners, bool half_pixel_centers)
+{
+    const auto offset = half_pixel_centers ? 0.5f : 0.0f;
+    const auto after_scale = (static_cast<float>(input_value) + offset) * scale;
+    const auto align_corners_val = align_corners ? roundf(after_scale) : std::floor(after_scale);
+    int32_t output_value = std::min(static_cast<int32_t>(align_corners_val), static_cast<int32_t>(shape_size - 1));
+    if (half_pixel_centers)
+    {
+        output_value = std::max(0, output_value);
+    }
+    return output_value;
+}
+
 }
 END_NS_NNCASE_KERNELS
diff --git a/lib/nncase/v1/include/nncase/kernels/neutral/neutral_kernels.h b/lib/nncase/v1/include/nncase/kernels/neutral/neutral_kernels.h
index e8b9c81..7c369e4 100644
--- a/lib/nncase/v1/include/nncase/kernels/neutral/neutral_kernels.h
+++ b/lib/nncase/v1/include/nncase/kernels/neutral/neutral_kernels.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/nnil.h b/lib/nncase/v1/include/nncase/kernels/nnil.h
index 3c53881..08851fa 100644
--- a/lib/nncase/v1/include/nncase/kernels/nnil.h
+++ b/lib/nncase/v1/include/nncase/kernels/nnil.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -13,10 +13,10 @@
  * limitations under the License.
  */
 #pragma once
+#include <nncase/kernels/kernel_context.h>
 #include <nncase/runtime/datatypes.h>
 #include <nncase/runtime/error.h>
 #include <nncase/runtime/result.h>
-#include <nncase/kernels/kernel_context.h>
 
 BEGIN_NS_NNCASE_KERNELS
 
diff --git a/lib/nncase/v1/include/nncase/kernels/reduce_window.h b/lib/nncase/v1/include/nncase/kernels/reduce_window.h
index 35c4aae..4b56d86 100644
--- a/lib/nncase/v1/include/nncase/kernels/reduce_window.h
+++ b/lib/nncase/v1/include/nncase/kernels/reduce_window.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/riscv/neutral_kernels.h b/lib/nncase/v1/include/nncase/kernels/riscv/neutral_kernels.h
index cdf224d..127ad0a 100644
--- a/lib/nncase/v1/include/nncase/kernels/riscv/neutral_kernels.h
+++ b/lib/nncase/v1/include/nncase/kernels/riscv/neutral_kernels.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/kernels/tensor_compute.h b/lib/nncase/v1/include/nncase/kernels/tensor_compute.h
index 9d27616..f2ecc9f 100644
--- a/lib/nncase/v1/include/nncase/kernels/tensor_compute.h
+++ b/lib/nncase/v1/include/nncase/kernels/tensor_compute.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -13,15 +13,15 @@
  * limitations under the License.
  */
 #pragma once
+#include <nncase/kernels/kernel_context.h>
 #include <nncase/runtime/datatypes.h>
 #include <nncase/runtime/error.h>
 #include <nncase/runtime/result.h>
-#include <nncase/kernels/kernel_context.h>
 
 BEGIN_NS_NNCASE_KERNELS
 
 NNCASE_API result<void> batch_to_space(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
-    const runtime_shape_t &block_shape, const runtime_paddings_t &crops, const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, 
+    const runtime_shape_t &block_shape, const runtime_paddings_t &crops, const runtime_shape_t &in_strides, const runtime_shape_t &out_strides,
     kernel_context &context = default_kernel_context()) noexcept;
 
 NNCASE_API result<void> broadcast(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
@@ -51,6 +51,10 @@ NNCASE_API result<void> dequantize(datatype_t in_type, datatype_t out_type, cons
 NNCASE_API result<void> lut1d(datatype_t type, const gsl::byte *input, const gsl::byte *table, gsl::byte *output, const runtime_shape_t &shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const scalar &min, const scalar &max) noexcept;
 
+NNCASE_API result<void> onehot(datatype_t type, const int32_t *indices, gsl::byte *output, const runtime_shape_t &indices_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &out_strides, gsl::byte *depth, gsl::byte *off_value, gsl::byte *on_value, size_t axis, onehot_mode_t mode,
+    kernel_context &context = default_kernel_context()) noexcept;
+
 NNCASE_API result<void> pad(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const runtime_paddings_t &paddings, pad_mode_t mode, const scalar &pad_value,
     kernel_context &context = default_kernel_context()) noexcept;
@@ -65,8 +69,21 @@ NNCASE_API result<void> unary(unary_op_t op, const float *input, float *output,
 NNCASE_API result<void> reduce(reduce_op_t op, float init_value, const float *input, float *output, const runtime_shape_t &in_shape, const runtime_shape_t &axis,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, bool keep_dims, kernel_context &context = default_kernel_context()) noexcept;
 
+NNCASE_API result<void> resize_bilinear(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> resize_nearest_neighbor(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, int32_t out_h, int32_t out_w, bool align_corners, bool half_pixel_centers,
+    kernel_context &context = default_kernel_context()) noexcept;
+
 NNCASE_API result<void> slice(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape,
     const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const runtime_shape_t &begins, const runtime_shape_t &ends, const runtime_axis_t &strides,
     kernel_context &context = default_kernel_context()) noexcept;
 
+NNCASE_API result<void> gather(datatype_t in_type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, int32_t axis, kernel_context &context = default_kernel_context()) noexcept;
+
+NNCASE_API result<void> gather_nd(datatype_t type, const gsl::byte *input, gsl::byte *output, const runtime_shape_t &in_shape, const runtime_shape_t &out_shape,
+    const runtime_shape_t &in_strides, const runtime_shape_t &out_strides, const int32_t *indices, const runtime_shape_t &indices_shape, int32_t batch_dims, kernel_context &context = default_kernel_context()) noexcept;
 END_NS_NNCASE_KERNELS
diff --git a/lib/nncase/v1/include/nncase/runtime/allocator.h b/lib/nncase/v1/include/nncase/runtime/allocator.h
index 23fe469..fa995ad 100644
--- a/lib/nncase/v1/include/nncase/runtime/allocator.h
+++ b/lib/nncase/v1/include/nncase/runtime/allocator.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/bfloat16.h b/lib/nncase/v1/include/nncase/runtime/bfloat16.h
index b4373fe..a121cad 100644
--- a/lib/nncase/v1/include/nncase/runtime/bfloat16.h
+++ b/lib/nncase/v1/include/nncase/runtime/bfloat16.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/bitio.h b/lib/nncase/v1/include/nncase/runtime/bitio.h
index 2642406..1303860 100644
--- a/lib/nncase/v1/include/nncase/runtime/bitio.h
+++ b/lib/nncase/v1/include/nncase/runtime/bitio.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/compiler_defs.h b/lib/nncase/v1/include/nncase/runtime/compiler_defs.h
index a1efbe3..4f412fc 100644
--- a/lib/nncase/v1/include/nncase/runtime/compiler_defs.h
+++ b/lib/nncase/v1/include/nncase/runtime/compiler_defs.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/datatypes.h b/lib/nncase/v1/include/nncase/runtime/datatypes.h
index 9146728..af1d543 100644
--- a/lib/nncase/v1/include/nncase/runtime/datatypes.h
+++ b/lib/nncase/v1/include/nncase/runtime/datatypes.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -136,6 +136,22 @@ typedef enum _reduce_op
     reduce_sum
 } reduce_op_t;
 
+inline std::string reduce_op_to_string(reduce_op_t op)
+{
+    switch (op)
+    {
+    case reduce_mean:
+        return "reduce_mean";
+    case reduce_min:
+        return "reduce_min";
+    case reduce_max:
+        return "reduce_max";
+    case reduce_sum:
+        return "reduce_sum";
+    }
+    return "unknown";
+}
+
 typedef enum _binary_op
 {
     binary_add,
@@ -256,6 +272,12 @@ typedef enum _image_resize_mode
     image_resize_nearest_neighbor
 } image_resize_mode_t;
 
+typedef enum _onehot_mode
+{
+    onehot_normal,
+    onehot_process_neg
+} onehot_mode_t;
+
 typedef enum _pad_mode
 {
     pad_constant,
diff --git a/lib/nncase/v1/include/nncase/runtime/error.h b/lib/nncase/v1/include/nncase/runtime/error.h
index a4f7daf..c56d6f5 100644
--- a/lib/nncase/v1/include/nncase/runtime/error.h
+++ b/lib/nncase/v1/include/nncase/runtime/error.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/host_runtime_tensor.h b/lib/nncase/v1/include/nncase/runtime/host_runtime_tensor.h
index 5d889d2..cdf90a3 100644
--- a/lib/nncase/v1/include/nncase/runtime/host_runtime_tensor.h
+++ b/lib/nncase/v1/include/nncase/runtime/host_runtime_tensor.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -38,9 +38,9 @@ struct host_memory_block
 
     host_memory_block() = default;
     host_memory_block(const host_memory_block &) = delete;
-    host_memory_block(host_memory_block && other) noexcept;
+    host_memory_block(host_memory_block &&other) noexcept;
     host_memory_block &operator=(const host_memory_block &) = delete;
-    host_memory_block &operator=(host_memory_block && other) noexcept;
+    host_memory_block &operator=(host_memory_block &&other) noexcept;
 
     ~host_memory_block()
     {
@@ -52,6 +52,7 @@ struct host_memory_block
         if (auto d = std::move(deleter))
             d(reinterpret_cast<gsl::byte *>(virtual_address));
         deleter = {};
+        physical_block.free(*this);
     }
 
     gsl::span<gsl::byte> virtual_buffer() const noexcept
diff --git a/lib/nncase/v1/include/nncase/runtime/interpreter.h b/lib/nncase/v1/include/nncase/runtime/interpreter.h
index a068ba1..470cfb5 100644
--- a/lib/nncase/v1/include/nncase/runtime/interpreter.h
+++ b/lib/nncase/v1/include/nncase/runtime/interpreter.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/compiler_defs.h b/lib/nncase/v1/include/nncase/runtime/k210/compiler_defs.h
index a79bff0..a57de2c 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/compiler_defs.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/compiler_defs.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/error.h b/lib/nncase/v1/include/nncase/runtime/k210/error.h
index 218c9d0..43776fe 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/error.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/error.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/kernel_context.h b/lib/nncase/v1/include/nncase/runtime/k210/kernel_context.h
index 160a450..11dd3eb 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/kernel_context.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/kernel_context.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/op_reader.h b/lib/nncase/v1/include/nncase/runtime/k210/op_reader.h
index d5a526a..b2c8ec9 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/op_reader.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/op_reader.h
@@ -1,6 +1,4 @@
-/* This file is generated by tools/stackvm_gen/IsaGen at 2021/2/23 16:24:09 +08:00.
- *
- * Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -36,6 +34,7 @@ public:
     virtual result<void> visit(NNCASE_UNUSED const kpu_download_options &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const kpu_conv2d_options &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const kpu_upload_options &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const copy_options &op) noexcept { return ok(); }
 
 protected:
     bool interrupted_;
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/runtime_module.h b/lib/nncase/v1/include/nncase/runtime/k210/runtime_module.h
index 310a7f4..9298d0e 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/runtime_module.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/runtime_module.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/runtime_op_utility.h b/lib/nncase/v1/include/nncase/runtime/k210/runtime_op_utility.h
index 2ffdee0..083d680 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/runtime_op_utility.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/runtime_op_utility.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/k210/runtime_types.h b/lib/nncase/v1/include/nncase/runtime/k210/runtime_types.h
index ac4015f..2d024f1 100644
--- a/lib/nncase/v1/include/nncase/runtime/k210/runtime_types.h
+++ b/lib/nncase/v1/include/nncase/runtime/k210/runtime_types.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -292,7 +292,8 @@ enum class opcode_t : uint8_t
 {
     kpu_upload,
     kpu_download,
-    kpu_conv2d
+    kpu_conv2d,
+    copy
 };
 
 struct kpu_upload_options
@@ -328,4 +329,16 @@ struct kpu_conv2d_options
     kpu_layer_argument_t layer;
 };
 
+struct copy_options
+{
+    opcode_t opcode = opcode_t::copy;
+    uint8_t reserved0[3];
+
+    memory_range input;
+    memory_range output;
+    kpu_shape_t in_shape;
+    kpu_shape_t in_strides;
+    kpu_shape_t out_strides;
+};
+
 END_NS_NNCASE_RT_K210
diff --git a/lib/nncase/v1/include/nncase/runtime/model.h b/lib/nncase/v1/include/nncase/runtime/model.h
index 1c9286e..72b6630 100644
--- a/lib/nncase/v1/include/nncase/runtime/model.h
+++ b/lib/nncase/v1/include/nncase/runtime/model.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/nnil.h b/lib/nncase/v1/include/nncase/runtime/nnil.h
index cb1c092..bae8ab7 100644
--- a/lib/nncase/v1/include/nncase/runtime/nnil.h
+++ b/lib/nncase/v1/include/nncase/runtime/nnil.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/result.h b/lib/nncase/v1/include/nncase/runtime/result.h
index 43eb099..a1fa153 100644
--- a/lib/nncase/v1/include/nncase/runtime/result.h
+++ b/lib/nncase/v1/include/nncase/runtime/result.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -72,7 +72,7 @@ struct Ok
         : value(value) { }
 
     template <class... Args>
-    constexpr explicit Ok(mpark::in_place_t, Args &&...args)
+    constexpr explicit Ok(mpark::in_place_t, Args &&... args)
         : value(std::forward<Args>(args)...) { }
 
     T value;
@@ -101,7 +101,7 @@ inline constexpr Ok<void> ok()
 }
 
 template <class T, class... Args>
-constexpr Ok<T> ok(Args &&...args)
+constexpr Ok<T> ok(Args &&... args)
 {
     return Ok<T>(mpark::in_place, std::forward<Args>(args)...);
 }
diff --git a/lib/nncase/v1/include/nncase/runtime/runtime_loader.h b/lib/nncase/v1/include/nncase/runtime/runtime_loader.h
index 44a61d8..dd873d1 100644
--- a/lib/nncase/v1/include/nncase/runtime/runtime_loader.h
+++ b/lib/nncase/v1/include/nncase/runtime/runtime_loader.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/runtime_module.h b/lib/nncase/v1/include/nncase/runtime/runtime_module.h
index ee06e2c..009affb 100644
--- a/lib/nncase/v1/include/nncase/runtime/runtime_module.h
+++ b/lib/nncase/v1/include/nncase/runtime/runtime_module.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/runtime_op_utility.h b/lib/nncase/v1/include/nncase/runtime/runtime_op_utility.h
index d754cc6..1ebb1a4 100644
--- a/lib/nncase/v1/include/nncase/runtime/runtime_op_utility.h
+++ b/lib/nncase/v1/include/nncase/runtime/runtime_op_utility.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -252,8 +252,10 @@ inline int get_last_not_contiguous_index(const runtime_shape_t &strides, const r
     return -1;
 }
 
-template<size_t A, size_t B>
+template <size_t A, size_t B>
 constexpr auto is_not_equal = std::integral_constant<bool, std::not_equal_to<size_t> {}(A, B)> {};
 
-struct DefaultCallable {};
+struct DefaultCallable
+{
+};
 END_NS_NNCASE_RUNTIME
diff --git a/lib/nncase/v1/include/nncase/runtime/runtime_tensor.h b/lib/nncase/v1/include/nncase/runtime/runtime_tensor.h
index eac5e59..396e73d 100644
--- a/lib/nncase/v1/include/nncase/runtime/runtime_tensor.h
+++ b/lib/nncase/v1/include/nncase/runtime/runtime_tensor.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/runtime_tensor_impl.h b/lib/nncase/v1/include/nncase/runtime/runtime_tensor_impl.h
index d53a070..4b01477 100644
--- a/lib/nncase/v1/include/nncase/runtime/runtime_tensor_impl.h
+++ b/lib/nncase/v1/include/nncase/runtime/runtime_tensor_impl.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.h b/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.h
index 01a6f74..73fda54 100644
--- a/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.h
+++ b/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.platform.h b/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.platform.h
index 266d82f..4a2788b 100644
--- a/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.platform.h
+++ b/lib/nncase/v1/include/nncase/runtime/shared_runtime_tensor.platform.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -33,7 +33,7 @@ struct NNCASE_API physical_memory_block
     physical_memory_block &operator=(const physical_memory_block &) = delete;
     physical_memory_block &operator=(physical_memory_block &&other) noexcept;
 
-    result<void> free() noexcept;
+    void free(NNCASE_UNUSED host_memory_block &block) noexcept;
 
     static result<void> acknowledge(host_memory_block &block) noexcept;
     static result<void> allocate(host_memory_block &block) noexcept;
diff --git a/lib/nncase/v1/include/nncase/runtime/small_vector.hpp b/lib/nncase/v1/include/nncase/runtime/small_vector.hpp
index ae82cc1..ebc39a9 100644
--- a/lib/nncase/v1/include/nncase/runtime/small_vector.hpp
+++ b/lib/nncase/v1/include/nncase/runtime/small_vector.hpp
@@ -130,61 +130,69 @@
 //
 #pragma once
 
-#include <type_traits>
 #include <cstddef>
 #include <memory>
+#include <type_traits>
 
-#define ITLIB_SMALL_VECTOR_ERROR_HANDLING_NONE  0
+#define ITLIB_SMALL_VECTOR_ERROR_HANDLING_NONE 0
 #define ITLIB_SMALL_VECTOR_ERROR_HANDLING_THROW 1
 #define ITLIB_SMALL_VECTOR_ERROR_HANDLING_ASSERT 2
 #define ITLIB_SMALL_VECTOR_ERROR_HANDLING_ASSERT_AND_THROW 3
 
 #if !defined(ITLIB_SMALL_VECTOR_ERROR_HANDLING)
-#   define ITLIB_SMALL_VECTOR_ERROR_HANDLING ITLIB_SMALL_VECTOR_ERROR_HANDLING_THROW
+#define ITLIB_SMALL_VECTOR_ERROR_HANDLING ITLIB_SMALL_VECTOR_ERROR_HANDLING_THROW
 #endif
 
-
 #if ITLIB_SMALL_VECTOR_ERROR_HANDLING == ITLIB_SMALL_VECTOR_ERROR_HANDLING_NONE
-#   define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond)
+#define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond)
 #elif ITLIB_SMALL_VECTOR_ERROR_HANDLING == ITLIB_SMALL_VECTOR_ERROR_HANDLING_THROW
-#   include <stdexcept>
-#   define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond) if (cond) throw std::out_of_range("itlib::small_vector out of range")
+#include <stdexcept>
+#define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond) \
+    if (cond)                                      \
+    throw std::out_of_range("itlib::small_vector out of range")
 #elif ITLIB_SMALL_VECTOR_ERROR_HANDLING == ITLIB_SMALL_VECTOR_ERROR_HANDLING_ASSERT
-#   include <cassert>
-#   define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond, rescue_return) assert(!(cond) && "itlib::small_vector out of range")
+#include <cassert>
+#define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond, rescue_return) assert(!(cond) && "itlib::small_vector out of range")
 #elif ITLIB_SMALL_VECTOR_ERROR_HANDLING == ITLIB_SMALL_VECTOR_ERROR_HANDLING_ASSERT_AND_THROW
-#   include <stdexcept>
-#   include <cassert>
-#   define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond, rescue_return) \
-    do { if (cond) { assert(false && "itlib::small_vector out of range"); throw std::out_of_range("itlib::small_vector out of range"); } } while(false)
+#include <cassert>
+#include <stdexcept>
+#define I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(cond, rescue_return)        \
+    do                                                                   \
+    {                                                                    \
+        if (cond)                                                        \
+        {                                                                \
+            assert(false && "itlib::small_vector out of range");         \
+            throw std::out_of_range("itlib::small_vector out of range"); \
+        }                                                                \
+    } while (false)
 #else
 #error "Unknown ITLIB_SMALL_VECTOR_ERRROR_HANDLING"
 #endif
 
-
 #if defined(ITLIB_SMALL_VECTOR_NO_DEBUG_BOUNDS_CHECK)
-#   define I_ITLIB_SMALL_VECTOR_BOUNDS_CHECK(i)
+#define I_ITLIB_SMALL_VECTOR_BOUNDS_CHECK(i)
 #else
-#   include <cassert>
-#   define I_ITLIB_SMALL_VECTOR_BOUNDS_CHECK(i) assert((i) < this->size())
+#include <cassert>
+#define I_ITLIB_SMALL_VECTOR_BOUNDS_CHECK(i) assert((i) < this->size())
 #endif
 
 namespace itlib
 {
 
-template<typename T, size_t StaticCapacity = 16, size_t RevertToStaticSize = 0, class Alloc = std::allocator<T>>
-struct small_vector: Alloc
+template <typename T, size_t StaticCapacity = 16, size_t RevertToStaticSize = 0, class Alloc = std::allocator<T>>
+struct small_vector : Alloc
 {
     static_assert(RevertToStaticSize <= StaticCapacity + 1, "itlib::small_vector: the revert-to-static size shouldn't exceed the static capacity by more than one");
 
     using atraits = std::allocator_traits<Alloc>;
+
 public:
     using allocator_type = Alloc;
     using value_type = typename atraits::value_type;
     using size_type = typename atraits::size_type;
     using difference_type = typename atraits::difference_type;
-    using reference = T&;
-    using const_reference = const T&;
+    using reference = T &;
+    using const_reference = const T &;
     using pointer = typename atraits::pointer;
     using const_pointer = typename atraits::const_pointer;
     using iterator = pointer;
@@ -197,9 +205,10 @@ public:
 
     small_vector()
         : small_vector(Alloc())
-    {}
+    {
+    }
 
-    small_vector(const Alloc& alloc)
+    small_vector(const Alloc &alloc)
         : Alloc(alloc)
         , m_capacity(StaticCapacity)
         , m_dynamic_capacity(0)
@@ -208,36 +217,37 @@ public:
         m_begin = m_end = static_begin_ptr();
     }
 
-    explicit small_vector(size_t count, const Alloc& alloc = Alloc())
+    explicit small_vector(size_t count, const Alloc &alloc = Alloc())
         : small_vector(alloc)
     {
         resize(count);
     }
 
-    explicit small_vector(size_t count, const T& value, const Alloc& alloc = Alloc())
+    explicit small_vector(size_t count, const T &value, const Alloc &alloc = Alloc())
         : small_vector(alloc)
     {
         assign_impl(count, value);
     }
 
     template <class InputIterator, typename = decltype(*std::declval<InputIterator>())>
-    small_vector(InputIterator first, InputIterator last, const Alloc& alloc = Alloc())
+    small_vector(InputIterator first, InputIterator last, const Alloc &alloc = Alloc())
         : small_vector(alloc)
     {
         assign_impl(first, last);
     }
 
-    small_vector(std::initializer_list<T> l, const Alloc& alloc = Alloc())
+    small_vector(std::initializer_list<T> l, const Alloc &alloc = Alloc())
         : small_vector(alloc)
     {
         assign_impl(l);
     }
 
-    small_vector(const small_vector& v)
+    small_vector(const small_vector &v)
         : small_vector(v, atraits::select_on_container_copy_construction(v.get_allocator()))
-    {}
+    {
+    }
 
-    small_vector(const small_vector& v, const Alloc& alloc)
+    small_vector(const small_vector &v, const Alloc &alloc)
         : Alloc(alloc)
         , m_dynamic_capacity(0)
         , m_dynamic_data(nullptr)
@@ -261,7 +271,7 @@ public:
         }
     }
 
-    small_vector(small_vector&& v)
+    small_vector(small_vector &&v)
         : Alloc(std::move(v.get_alloc()))
         , m_capacity(v.m_capacity)
         , m_dynamic_capacity(v.m_dynamic_capacity)
@@ -300,7 +310,7 @@ public:
         }
     }
 
-    small_vector& operator=(const small_vector& v)
+    small_vector &operator=(const small_vector &v)
     {
         if (this == &v)
         {
@@ -323,7 +333,7 @@ public:
         return *this;
     }
 
-    small_vector& operator=(small_vector&& v)
+    small_vector &operator=(small_vector &&v)
     {
         clear();
 
@@ -357,7 +367,7 @@ public:
         return *this;
     }
 
-    void assign(size_type count, const T& value)
+    void assign(size_type count, const T &value)
     {
         clear();
         assign_impl(count, value);
@@ -512,7 +522,8 @@ public:
 
     void reserve(size_type new_cap)
     {
-        if (new_cap <= m_capacity) return;
+        if (new_cap <= m_capacity)
+            return;
 
         auto new_buf = choose_data(new_cap);
 
@@ -520,7 +531,7 @@ public:
         assert(new_buf != static_begin_ptr()); // we should never reserve into static memory
 
         const auto s = size();
-        if(s < RevertToStaticSize)
+        if (s < RevertToStaticSize)
         {
             // we've allocated enough memory for the dynamic buffer but don't move there until we have to
             return;
@@ -558,8 +569,10 @@ public:
     {
         const auto s = size();
 
-        if (s == m_capacity) return;
-        if (m_begin == static_begin_ptr()) return;
+        if (s == m_capacity)
+            return;
+        if (m_begin == static_begin_ptr())
+            return;
 
         auto old_end = m_end;
 
@@ -591,8 +604,10 @@ public:
     void revert_to_static()
     {
         const auto s = size();
-        if (m_begin == static_begin_ptr()) return; //we're already there
-        if (s > StaticCapacity) return; // nothing we can do
+        if (m_begin == static_begin_ptr())
+            return; //we're already there
+        if (s > StaticCapacity)
+            return; // nothing we can do
 
         // revert to static capacity
         auto old_end = m_end;
@@ -625,21 +640,21 @@ public:
         }
     }
 
-    iterator insert(const_iterator position, const value_type& val)
+    iterator insert(const_iterator position, const value_type &val)
     {
         auto pos = grow_at(position, 1);
         atraits::construct(get_alloc(), pos, val);
         return pos;
     }
 
-    iterator insert(const_iterator position, value_type&& val)
+    iterator insert(const_iterator position, value_type &&val)
     {
         auto pos = grow_at(position, 1);
         atraits::construct(get_alloc(), pos, std::move(val));
         return pos;
     }
 
-    iterator insert(const_iterator position, size_type count, const value_type& val)
+    iterator insert(const_iterator position, size_type count, const value_type &val)
     {
         auto pos = grow_at(position, count);
         for (size_type i = 0; i < count; ++i)
@@ -666,7 +681,7 @@ public:
     {
         auto pos = grow_at(position, ilist.size());
         size_type i = 0;
-        for (auto& elem : ilist)
+        for (auto &elem : ilist)
         {
             atraits::construct(get_alloc(), pos + i, elem);
             ++i;
@@ -674,8 +689,8 @@ public:
         return pos;
     }
 
-    template<typename... Args>
-    iterator emplace(const_iterator position, Args&&... args)
+    template <typename... Args>
+    iterator emplace(const_iterator position, Args &&... args)
     {
         auto pos = grow_at(position, 1);
         atraits::construct(get_alloc(), pos, std::forward<Args>(args)...);
@@ -699,14 +714,14 @@ public:
         atraits::construct(get_alloc(), pos, val);
     }
 
-    void push_back(T&& val)
+    void push_back(T &&val)
     {
         auto pos = grow_at(m_end, 1);
         atraits::construct(get_alloc(), pos, std::move(val));
     }
 
-    template<typename... Args>
-    reference emplace_back(Args&&... args)
+    template <typename... Args>
+    reference emplace_back(Args &&... args)
     {
         auto pos = grow_at(m_end, 1);
         atraits::construct(get_alloc(), pos, std::forward<Args>(args)...);
@@ -718,7 +733,7 @@ public:
         shrink_at(m_end - 1, 1);
     }
 
-    void resize(size_type n, const value_type& v)
+    void resize(size_type n, const value_type &v)
     {
         auto new_buf = choose_data(n);
 
@@ -847,7 +862,7 @@ public:
     }
 
 private:
-    T* static_begin_ptr()
+    T *static_begin_ptr()
     {
         return reinterpret_cast<pointer>(m_static_data + 0);
     }
@@ -855,9 +870,9 @@ private:
     // increase the size by splicing the elements in such a way that
     // a hole of uninitialized elements is left at position, with size num
     // returns the (potentially new) address of the hole
-    T* grow_at(const T* cp, size_t num)
+    T *grow_at(const T *cp, size_t num)
     {
-        auto position = const_cast<T*>(cp);
+        auto position = const_cast<T *>(cp);
 
         I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(position < m_begin || position > m_end);
 
@@ -919,9 +934,9 @@ private:
         }
     }
 
-    T* shrink_at(const T* cp, size_t num)
+    T *shrink_at(const T *cp, size_t num)
     {
-        auto position = const_cast<T*>(cp);
+        auto position = const_cast<T *>(cp);
 
         I_ITLIB_SMALL_VECTOR_OUT_OF_RANGE_IF(position < m_begin || position > m_end || position + num > m_end);
 
@@ -985,7 +1000,7 @@ private:
         return ++position;
     }
 
-    void assign_impl(size_type count, const T& value)
+    void assign_impl(size_type count, const T &value)
     {
         assert(m_begin);
         assert(m_begin == m_end);
@@ -1022,7 +1037,7 @@ private:
         assert(m_begin == m_end);
 
         m_begin = m_end = choose_data(ilist.size());
-        for (auto& elem : ilist)
+        for (auto &elem : ilist)
         {
             atraits::construct(get_alloc(), m_end, elem);
             ++m_end;
@@ -1043,7 +1058,7 @@ private:
         }
     }
 
-    T* choose_data(size_t desired_capacity)
+    T *choose_data(size_t desired_capacity)
     {
         if (m_begin == m_dynamic_data)
         {
@@ -1105,8 +1120,8 @@ private:
         }
     }
 
-    allocator_type& get_alloc() { return static_cast<allocator_type&>(*this); }
-    const allocator_type& get_alloc() const { return static_cast<const allocator_type&>(*this); }
+    allocator_type &get_alloc() { return static_cast<allocator_type &>(*this); }
+    const allocator_type &get_alloc() const { return static_cast<const allocator_type &>(*this); }
 
     pointer m_begin;
     pointer m_end;
@@ -1118,9 +1133,9 @@ private:
     pointer m_dynamic_data;
 };
 
-template<typename T, size_t StaticCapacity, size_t RevertToStaticSize, class Alloc>
-bool operator==(const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc>& a,
-    const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc>& b)
+template <typename T, size_t StaticCapacity, size_t RevertToStaticSize, class Alloc>
+bool operator==(const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc> &a,
+    const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc> &b)
 {
     if (a.size() != b.size())
     {
@@ -1136,9 +1151,9 @@ bool operator==(const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc>
     return true;
 }
 
-template<typename T, size_t StaticCapacity, size_t RevertToStaticSize, class Alloc>
-bool operator!=(const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc>& a,
-    const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc>& b)
+template <typename T, size_t StaticCapacity, size_t RevertToStaticSize, class Alloc>
+bool operator!=(const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc> &a,
+    const small_vector<T, StaticCapacity, RevertToStaticSize, Alloc> &b)
 {
     if (a.size() != b.size())
     {
diff --git a/lib/nncase/v1/include/nncase/runtime/span_reader.h b/lib/nncase/v1/include/nncase/runtime/span_reader.h
index e3bb0f8..5f28097 100644
--- a/lib/nncase/v1/include/nncase/runtime/span_reader.h
+++ b/lib/nncase/v1/include/nncase/runtime/span_reader.h
@@ -1,4 +1,4 @@
-/* Copyright 2019-2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/runtime/stackvm/op_reader.h b/lib/nncase/v1/include/nncase/runtime/stackvm/op_reader.h
index 08277d1..252528c 100644
--- a/lib/nncase/v1/include/nncase/runtime/stackvm/op_reader.h
+++ b/lib/nncase/v1/include/nncase/runtime/stackvm/op_reader.h
@@ -1,6 +1,6 @@
-/* This file is generated by tools/stackvm_gen/IsaGen at 2021/5/28 11:52:08 +08:00.
+/* This file is generated by tools/stackvm_gen/IsaGen at 2021/7/14 19:17:48 +08:00.
  *
- * Copyright 2020 Canaan Inc.
+ * Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -1148,24 +1148,6 @@ struct op_reader<tensor_call_op_t>
     }
 };
 
-template <>
-struct op_reader<tensor_concat_op_t>
-{
-    tensor_concat_op_t operator()(span_reader &reader) const
-    {
-        tensor_concat_op_t op(default_init);
-        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
-        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
-        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
-        op.num_src = reader.read_unaligned<uint8_t>();
-        op.axis = reader.read_unaligned<uint8_t>();
-        op.rshape_dims = reader.read_unaligned<uint8_t>();
-        op.rshape_dest = reader.read_unaligned<uint8_t>();
-        op.rstride_dest = reader.read_unaligned<uint8_t>();
-        return op;
-    }
-};
-
 template <>
 struct op_reader<tensor_conv2d_op_t>
 {
@@ -1192,6 +1174,22 @@ struct op_reader<tensor_conv2d_op_t>
     }
 };
 
+template <>
+struct op_reader<tensor_copy_op_t>
+{
+    tensor_copy_op_t operator()(span_reader &reader) const
+    {
+        tensor_copy_op_t op(default_init);
+        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
+        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
+        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
+        op.rshape = reader.read_unaligned<uint8_t>();
+        op.rstride_src = reader.read_unaligned<uint8_t>();
+        op.rstride_dest = reader.read_unaligned<uint8_t>();
+        return op;
+    }
+};
+
 template <>
 struct op_reader<tensor_convert_op_t>
 {
@@ -1226,6 +1224,44 @@ struct op_reader<tensor_dequantize_op_t>
     }
 };
 
+template <>
+struct op_reader<tensor_gather_op_t>
+{
+    tensor_gather_op_t operator()(span_reader &reader) const
+    {
+        tensor_gather_op_t op(default_init);
+        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
+        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
+        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
+        op.rshape_src = reader.read_unaligned<uint8_t>();
+        op.rshape_dest = reader.read_unaligned<uint8_t>();
+        op.rstride_src = reader.read_unaligned<uint8_t>();
+        op.rstride_dest = reader.read_unaligned<uint8_t>();
+        op.rshape_indices = reader.read_unaligned<uint8_t>();
+        op.axis = reader.read_unaligned<uint8_t>();
+        return op;
+    }
+};
+
+template <>
+struct op_reader<tensor_gather_nd_op_t>
+{
+    tensor_gather_nd_op_t operator()(span_reader &reader) const
+    {
+        tensor_gather_nd_op_t op(default_init);
+        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
+        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
+        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
+        op.rshape_src = reader.read_unaligned<uint8_t>();
+        op.rshape_dest = reader.read_unaligned<uint8_t>();
+        op.rstride_src = reader.read_unaligned<uint8_t>();
+        op.rstride_dest = reader.read_unaligned<uint8_t>();
+        op.rshape_indices = reader.read_unaligned<uint8_t>();
+        op.batch_dims = reader.read_unaligned<uint8_t>();
+        return op;
+    }
+};
+
 template <>
 struct op_reader<tensor_lut1d_op_t>
 {
@@ -1243,6 +1279,24 @@ struct op_reader<tensor_lut1d_op_t>
     }
 };
 
+template <>
+struct op_reader<tensor_onehot_op_t>
+{
+    tensor_onehot_op_t operator()(span_reader &reader) const
+    {
+        tensor_onehot_op_t op(default_init);
+        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
+        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
+        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
+        op.rshape_indices = reader.read_unaligned<uint8_t>();
+        op.rshape_dest = reader.read_unaligned<uint8_t>();
+        op.rstride_dest = reader.read_unaligned<uint8_t>();
+        op.axis = reader.read_unaligned<uint8_t>();
+        op.onehot_mode = static_cast<onehot_mode_t>(reader.read_unaligned<uint8_t>());
+        return op;
+    }
+};
+
 template <>
 struct op_reader<tensor_pad_op_t>
 {
@@ -1322,6 +1376,25 @@ struct op_reader<tensor_reduce_window2d_op_t>
     }
 };
 
+template <>
+struct op_reader<tensor_resize_image_op_t>
+{
+    tensor_resize_image_op_t operator()(span_reader &reader) const
+    {
+        tensor_resize_image_op_t op(default_init);
+        op.opcode = static_cast<opcode_t>(reader.read_unaligned<uint8_t>());
+        op.funct = static_cast<tensor_function_t>(reader.read_unaligned<uint16_t>());
+        op.datatype = static_cast<datatype_t>(reader.read_unaligned<uint8_t>());
+        op.rshape_src = reader.read_unaligned<uint8_t>();
+        op.rstride_src = reader.read_unaligned<uint8_t>();
+        op.rstride_dest = reader.read_unaligned<uint8_t>();
+        op.align_corners = reader.read_unaligned<bool>();
+        op.half_pixel_centers = reader.read_unaligned<bool>();
+        op.image_resize_mode = static_cast<image_resize_mode_t>(reader.read_unaligned<uint8_t>());
+        return op;
+    }
+};
+
 template <>
 struct op_reader<tensor_slice_op_t>
 {
@@ -1485,15 +1558,19 @@ public:
     virtual result<void> visit(NNCASE_UNUSED const tensor_broadcast_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_binary_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_call_op_t &op) noexcept { return ok(); }
-    virtual result<void> visit(NNCASE_UNUSED const tensor_concat_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_conv2d_op_t &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const tensor_copy_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_convert_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_dequantize_op_t &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const tensor_gather_op_t &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const tensor_gather_nd_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_lut1d_op_t &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const tensor_onehot_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_pad_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_quantize_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_reduce_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_reduce_window2d_op_t &op) noexcept { return ok(); }
+    virtual result<void> visit(NNCASE_UNUSED const tensor_resize_image_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_slice_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_unary_op_t &op) noexcept { return ok(); }
     virtual result<void> visit(NNCASE_UNUSED const tensor_transpose_op_t &op) noexcept { return ok(); }
diff --git a/lib/nncase/v1/include/nncase/runtime/stackvm/opcode.h b/lib/nncase/v1/include/nncase/runtime/stackvm/opcode.h
index 13f5ad3..122a4d2 100644
--- a/lib/nncase/v1/include/nncase/runtime/stackvm/opcode.h
+++ b/lib/nncase/v1/include/nncase/runtime/stackvm/opcode.h
@@ -1,6 +1,6 @@
-/* This file is generated by tools/stackvm_gen/IsaGen at 2021/5/28 11:52:08 +08:00.
+/* This file is generated by tools/stackvm_gen/IsaGen at 2021/7/14 19:17:48 +08:00.
  *
- * Copyright 2020 Canaan Inc.
+ * Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -127,25 +127,28 @@ enum class tensor_function_t
     BROADCAST = 0x0002,
     CALL = 0x0003,
     CLAMP = 0x0004,
-    CONCAT = 0x0005,
-    CONV2D = 0x0006,
-    CONV2D_TRANSPOSE = 0x0007,
-    CONVERT = 0x0008,
+    CONV2D = 0x0005,
+    CONV2D_TRANSPOSE = 0x0006,
+    CONVERT = 0x0007,
+    COPY = 0x0008,
     DEQUANTIZE = 0x0009,
-    LOGISTIC = 0x000A,
-    LUT1D = 0x000B,
-    MATMUL = 0x000C,
-    PAD = 0x000D,
-    QUANTIZE = 0x000E,
-    REDUCE = 0x000F,
-    REDUCE_WINDOW2D = 0x0010,
-    RESIZE_IMAGE = 0x0011,
-    SLICE = 0x0012,
-    SOFTMAX = 0x0013,
-    SPACE_TO_BATCH = 0x0014,
-    TAKE = 0x0015,
-    TRANSPOSE = 0x0016,
-    UNARY = 0x0017,
+    GATHER = 0x000A,
+    GATHER_ND = 0x000B,
+    LOGISTIC = 0x000C,
+    LUT1D = 0x000D,
+    MATMUL = 0x000E,
+    ONEHOT = 0x000F,
+    PAD = 0x0010,
+    QUANTIZE = 0x0011,
+    REDUCE = 0x0012,
+    REDUCE_WINDOW2D = 0x0013,
+    RESIZE_IMAGE = 0x0014,
+    SLICE = 0x0015,
+    SOFTMAX = 0x0016,
+    SPACE_TO_BATCH = 0x0017,
+    TAKE = 0x0018,
+    TRANSPOSE = 0x0019,
+    UNARY = 0x001A,
 };
 
 // Instructions
@@ -1273,24 +1276,6 @@ struct tensor_call_op_t
     }
 };
 
-struct tensor_concat_op_t
-{
-    opcode_t opcode;
-    tensor_function_t funct;
-    datatype_t datatype;
-    uint8_t num_src;
-    uint8_t axis;
-    uint8_t rshape_dims;
-    uint8_t rshape_dest;
-    uint8_t rstride_dest;
-
-    tensor_concat_op_t(default_init_t) noexcept { }
-    explicit tensor_concat_op_t(datatype_t datatype, uint8_t num_src, uint8_t axis, uint8_t rshape_dims, uint8_t rshape_dest, uint8_t rstride_dest) noexcept
-        : opcode(opcode_t::TENSOR), funct(tensor_function_t::CONCAT), datatype(datatype), num_src(num_src), axis(axis), rshape_dims(rshape_dims), rshape_dest(rshape_dest), rstride_dest(rstride_dest)
-    {
-    }
-};
-
 struct tensor_conv2d_op_t
 {
     opcode_t opcode;
@@ -1317,6 +1302,22 @@ struct tensor_conv2d_op_t
     }
 };
 
+struct tensor_copy_op_t
+{
+    opcode_t opcode;
+    tensor_function_t funct;
+    datatype_t datatype;
+    uint8_t rshape;
+    uint8_t rstride_src;
+    uint8_t rstride_dest;
+
+    tensor_copy_op_t(default_init_t) noexcept { }
+    explicit tensor_copy_op_t(datatype_t datatype, uint8_t rshape, uint8_t rstride_src, uint8_t rstride_dest) noexcept
+        : opcode(opcode_t::TENSOR), funct(tensor_function_t::COPY), datatype(datatype), rshape(rshape), rstride_src(rstride_src), rstride_dest(rstride_dest)
+    {
+    }
+};
+
 struct tensor_convert_op_t
 {
     opcode_t opcode;
@@ -1351,6 +1352,44 @@ struct tensor_dequantize_op_t
     }
 };
 
+struct tensor_gather_op_t
+{
+    opcode_t opcode;
+    tensor_function_t funct;
+    datatype_t datatype;
+    uint8_t rshape_src;
+    uint8_t rshape_dest;
+    uint8_t rstride_src;
+    uint8_t rstride_dest;
+    uint8_t rshape_indices;
+    uint8_t axis;
+
+    tensor_gather_op_t(default_init_t) noexcept { }
+    explicit tensor_gather_op_t(datatype_t datatype, uint8_t rshape_src, uint8_t rshape_dest, uint8_t rstride_src, uint8_t rstride_dest, uint8_t rshape_indices, uint8_t axis) noexcept
+        : opcode(opcode_t::TENSOR), funct(tensor_function_t::GATHER), datatype(datatype), rshape_src(rshape_src), rshape_dest(rshape_dest), rstride_src(rstride_src), rstride_dest(rstride_dest), rshape_indices(rshape_indices), axis(axis)
+    {
+    }
+};
+
+struct tensor_gather_nd_op_t
+{
+    opcode_t opcode;
+    tensor_function_t funct;
+    datatype_t datatype;
+    uint8_t rshape_src;
+    uint8_t rshape_dest;
+    uint8_t rstride_src;
+    uint8_t rstride_dest;
+    uint8_t rshape_indices;
+    uint8_t batch_dims;
+
+    tensor_gather_nd_op_t(default_init_t) noexcept { }
+    explicit tensor_gather_nd_op_t(datatype_t datatype, uint8_t rshape_src, uint8_t rshape_dest, uint8_t rstride_src, uint8_t rstride_dest, uint8_t rshape_indices, uint8_t batch_dims) noexcept
+        : opcode(opcode_t::TENSOR), funct(tensor_function_t::GATHER_ND), datatype(datatype), rshape_src(rshape_src), rshape_dest(rshape_dest), rstride_src(rstride_src), rstride_dest(rstride_dest), rshape_indices(rshape_indices), batch_dims(batch_dims)
+    {
+    }
+};
+
 struct tensor_lut1d_op_t
 {
     opcode_t opcode;
@@ -1368,6 +1407,24 @@ struct tensor_lut1d_op_t
     }
 };
 
+struct tensor_onehot_op_t
+{
+    opcode_t opcode;
+    tensor_function_t funct;
+    datatype_t datatype;
+    uint8_t rshape_indices;
+    uint8_t rshape_dest;
+    uint8_t rstride_dest;
+    uint8_t axis;
+    onehot_mode_t onehot_mode;
+
+    tensor_onehot_op_t(default_init_t) noexcept { }
+    explicit tensor_onehot_op_t(datatype_t datatype, uint8_t rshape_indices, uint8_t rshape_dest, uint8_t rstride_dest, uint8_t axis, onehot_mode_t onehot_mode) noexcept
+        : opcode(opcode_t::TENSOR), funct(tensor_function_t::ONEHOT), datatype(datatype), rshape_indices(rshape_indices), rshape_dest(rshape_dest), rstride_dest(rstride_dest), axis(axis), onehot_mode(onehot_mode)
+    {
+    }
+};
+
 struct tensor_pad_op_t
 {
     opcode_t opcode;
@@ -1447,6 +1504,25 @@ struct tensor_reduce_window2d_op_t
     }
 };
 
+struct tensor_resize_image_op_t
+{
+    opcode_t opcode;
+    tensor_function_t funct;
+    datatype_t datatype;
+    uint8_t rshape_src;
+    uint8_t rstride_src;
+    uint8_t rstride_dest;
+    bool align_corners;
+    bool half_pixel_centers;
+    image_resize_mode_t image_resize_mode;
+
+    tensor_resize_image_op_t(default_init_t) noexcept { }
+    explicit tensor_resize_image_op_t(datatype_t datatype, uint8_t rshape_src, uint8_t rstride_src, uint8_t rstride_dest, bool align_corners, bool half_pixel_centers, image_resize_mode_t image_resize_mode) noexcept
+        : opcode(opcode_t::TENSOR), funct(tensor_function_t::RESIZE_IMAGE), datatype(datatype), rshape_src(rshape_src), rstride_src(rstride_src), rstride_dest(rstride_dest), align_corners(align_corners), half_pixel_centers(half_pixel_centers), image_resize_mode(image_resize_mode)
+    {
+    }
+};
+
 struct tensor_slice_op_t
 {
     opcode_t opcode;
diff --git a/lib/nncase/v1/include/nncase/runtime/stackvm/runtime_module.h b/lib/nncase/v1/include/nncase/runtime/stackvm/runtime_module.h
index f7d1cda..3e0a361 100644
--- a/lib/nncase/v1/include/nncase/runtime/stackvm/runtime_module.h
+++ b/lib/nncase/v1/include/nncase/runtime/stackvm/runtime_module.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/lib/nncase/v1/include/nncase/version.h b/lib/nncase/v1/include/nncase/version.h
index 1b67733..6aff662 100644
--- a/lib/nncase/v1/include/nncase/version.h
+++ b/lib/nncase/v1/include/nncase/version.h
@@ -1,4 +1,4 @@
-/* Copyright 2020 Canaan Inc.
+/* Copyright 2019-2021 Canaan Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,4 +14,4 @@
  */
 #pragma once
 #define NNCASE_VERSION "1.0.0"
-#define NNCASE_GIT_DESC "0dc08ae"
+#define NNCASE_VERSION_SUFFIX "-8c384a4"
diff --git a/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeConfig.cmake b/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeConfig.cmake
index b4500a2..f5effc2 100644
--- a/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeConfig.cmake
+++ b/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeConfig.cmake
@@ -1 +1,10 @@
 include(${CMAKE_CURRENT_LIST_DIR}/nncaseruntimeTargets.cmake)
+if(NOT TARGET mpark_variant)
+    find_package(mpark_variant REQUIRED)
+endif()
+
+if(NOT TARGET gsl-lite)
+    find_package(gsl-lite REQUIRED)
+endif()
+
+
diff --git a/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeTargets.cmake b/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeTargets.cmake
index 5d4c6e8..92dcf51 100644
--- a/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeTargets.cmake
+++ b/lib/nncase/v1/lib/cmake/nncaseruntime/nncaseruntimeTargets.cmake
@@ -54,14 +54,14 @@ endif()
 add_library(kernels INTERFACE IMPORTED)
 
 set_target_properties(kernels PROPERTIES
-  INTERFACE_LINK_LIBRARIES "CONAN_PKG::gsl-lite;CONAN_PKG::mpark-variant"
+  INTERFACE_LINK_LIBRARIES "gsl::gsl-lite;mpark_variant::mpark_variant"
 )
 
 # Create imported target runtime
 add_library(runtime INTERFACE IMPORTED)
 
 set_target_properties(runtime PROPERTIES
-  INTERFACE_LINK_LIBRARIES "CONAN_PKG::gsl-lite;CONAN_PKG::mpark-variant;\$<LINK_ONLY:kernels>"
+  INTERFACE_LINK_LIBRARIES "gsl::gsl-lite;mpark_variant::mpark_variant;\$<LINK_ONLY:kernels>"
 )
 
 # Create imported target nncaseruntime
@@ -69,7 +69,7 @@ add_library(nncaseruntime STATIC IMPORTED)
 
 set_target_properties(nncaseruntime PROPERTIES
   INTERFACE_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include"
-  INTERFACE_LINK_LIBRARIES "\$<LINK_ONLY:kernels>;\$<LINK_ONLY:runtime>;\$<LINK_ONLY:runtime_stackvm>;CONAN_PKG::gsl-lite;CONAN_PKG::mpark-variant"
+  INTERFACE_LINK_LIBRARIES "\$<LINK_ONLY:kernels>;\$<LINK_ONLY:runtime>;\$<LINK_ONLY:runtime_stackvm>;gsl::gsl-lite;mpark_variant::mpark_variant"
 )
 
 # Create imported target runtime_stackvm
diff --git a/lib/nncase/v1/lib/libnncase.rt_modules.k210.a b/lib/nncase/v1/lib/libnncase.rt_modules.k210.a
index 9a9293e..e81135e 100644
Binary files a/lib/nncase/v1/lib/libnncase.rt_modules.k210.a and b/lib/nncase/v1/lib/libnncase.rt_modules.k210.a differ
diff --git a/lib/nncase/v1/lib/libnncase.runtime.a b/lib/nncase/v1/lib/libnncase.runtime.a
index c6dd40c..2fdadb6 100644
Binary files a/lib/nncase/v1/lib/libnncase.runtime.a and b/lib/nncase/v1/lib/libnncase.runtime.a differ
diff --git a/lib/nncase/v1/nncase_v1.cpp b/lib/nncase/v1/nncase_v1.cpp
index 8ac32a2..4c54325 100644
--- a/lib/nncase/v1/nncase_v1.cpp
+++ b/lib/nncase/v1/nncase_v1.cpp
@@ -49,7 +49,7 @@ public:
 
     result<void> run_kmodel(const uint8_t *src, dmac_channel_number_t dma_ch)
     {
-#ifdef NNCASE_DEBUG
+#if NNCASE_DEBUG
         auto micro = sysctl_get_time_us();
 #endif
         try_(interp_.options().set("dma_ch", (uint32_t)dma_ch));
@@ -59,7 +59,7 @@ public:
         try_(hrt::sync(input_tensor, hrt::sync_write_back));
         try_(interp_.input_tensor(0, input_tensor));
         try_(interp_.run());
-#ifdef NNCASE_DEBUG
+#if NNCASE_DEBUG
         auto duration = sysctl_get_time_us() - micro;
         printf("run kmodel takes %f ms.\n", duration / 1e3f);
 #endif
diff --git a/third_party/gsl-lite/CMakeLists.txt b/third_party/gsl-lite/CMakeLists.txt
index 7c69a80..0a024e0 100644
--- a/third_party/gsl-lite/CMakeLists.txt
+++ b/third_party/gsl-lite/CMakeLists.txt
@@ -3,3 +3,4 @@
 add_library(gsl-lite INTERFACE)
 target_include_directories(gsl-lite INTERFACE include)
 add_library(CONAN_PKG::gsl-lite ALIAS gsl-lite)
+add_library(gsl::gsl-lite ALIAS gsl-lite)
diff --git a/third_party/mpark-variant/CMakeLists.txt b/third_party/mpark-variant/CMakeLists.txt
index bb278e5..ce134ce 100644
--- a/third_party/mpark-variant/CMakeLists.txt
+++ b/third_party/mpark-variant/CMakeLists.txt
@@ -1,5 +1,6 @@
 cmake_minimum_required (VERSION 3.8)
 
-add_library(mpark-variant INTERFACE)
-target_include_directories(mpark-variant INTERFACE include)
-add_library(CONAN_PKG::mpark-variant ALIAS mpark-variant)
+add_library(mpark_variant INTERFACE)
+target_include_directories(mpark_variant INTERFACE include)
+add_library(mpark_variant::mpark_variant ALIAS mpark_variant)
+add_library(CONAN_PKG::mpark-variant ALIAS mpark_variant)