From 4ca50775840814b72e96da139d5dccd5691d02f0 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E4=BD=B0=E9=98=85?= <43716063+Baiyuetribe@users.noreply.github.com> Date: Wed, 15 Jan 2025 21:14:06 +0800 Subject: [PATCH 1/3] init --- docs/developer-guide/operators.md | 2307 +++++++++-------- tools/pnnx/src/CMakeLists.txt | 1 + .../pnnx/src/pass_ncnn/torch_index_select.cpp | 58 + 3 files changed, 1271 insertions(+), 1095 deletions(-) create mode 100644 tools/pnnx/src/pass_ncnn/torch_index_select.cpp diff --git a/docs/developer-guide/operators.md b/docs/developer-guide/operators.md index 10fe1f03f0f6..1c915cf14fdb 100644 --- a/docs/developer-guide/operators.md +++ b/docs/developer-guide/operators.md @@ -1,168 +1,177 @@ - -* [AbsVal](#absval) -* [ArgMax](#argmax) -* [BatchNorm](#batchnorm) -* [Bias](#bias) -* [BinaryOp](#binaryop) -* [BNLL](#bnll) -* [Cast](#cast) -* [CELU](#celu) -* [Clip](#clip) -* [Concat](#concat) -* [Convolution](#convolution) -* [Convolution1D](#convolution1d) -* [Convolution3D](#convolution3d) -* [ConvolutionDepthWise](#convolutiondepthwise) -* [ConvolutionDepthWise1D](#convolutiondepthwise1d) -* [ConvolutionDepthWise3D](#convolutiondepthwise3d) -* [CopyTo](#copyto) -* [Crop](#crop) -* [CumulativeSum](#cumulativesum) -* [Deconvolution](#deconvolution) -* [Deconvolution1D](#deconvolution1d) -* [Deconvolution3D](#deconvolution3d) -* [DeconvolutionDepthWise](#deconvolutiondepthwise) -* [DeconvolutionDepthWise1D](#deconvolutiondepthwise1d) -* [DeconvolutionDepthWise3D](#deconvolutiondepthwise3d) -* [DeformableConv2D](#deformableconv2d) -* [Dequantize](#dequantize) -* [Diag](#diag) -* [Dropout](#dropout) -* [Eltwise](#eltwise) -* [ELU](#elu) -* [Embed](#embed) -* [Exp](#exp) -* [Flatten](#flatten) -* [Fold](#fold) -* [GELU](#gelu) -* [GLU](#glu) -* [Gemm](#gemm) -* [GridSample](#gridsample) -* [GroupNorm](#groupnorm) -* [GRU](#gru) -* [HardSigmoid](#hardsigmoid) -* [HardSwish](#hardswish) -* [InnerProduct](#innerproduct) -* [Input](#input) -* [InstanceNorm](#instancenorm) -* [Interp](#interp) -* [InverseSpectrogram](#inversespectrogram) -* [LayerNorm](#layernorm) -* [Log](#log) -* [LRN](#lrn) -* [LSTM](#lstm) -* [MemoryData](#memorydata) -* [Mish](#mish) -* [MultiHeadAttention](#multiheadattention) -* [MVN](#mvn) -* [Noop](#noop) -* [Normalize](#normalize) -* [Packing](#packing) -* [Padding](#padding) -* [Permute](#permute) -* [PixelShuffle](#pixelshuffle) -* [Pooling](#pooling) -* [Pooling1D](#pooling1d) -* [Pooling3D](#pooling3d) -* [Power](#power) -* [PReLU](#prelu) -* [Quantize](#quantize) -* [Reduction](#reduction) -* [ReLU](#relu) -* [Reorg](#reorg) -* [Requantize](#requantize) -* [Reshape](#reshape) -* [RMSNorm](#rmsnorm) -* [RNN](#rnn) -* [Scale](#scale) -* [SELU](#selu) -* [Shrink](#shrink) -* [ShuffleChannel](#shufflechannel) -* [Sigmoid](#sigmoid) -* [Slice](#slice) -* [Softmax](#softmax) -* [Softplus](#softplus) -* [Spectrogram](#spectrogram) -* [Split](#split) -* [Swish](#swish) -* [TanH](#tanh) -* [Threshold](#threshold) -* [Tile](#tile) -* [UnaryOp](#unaryop) -* [Unfold](#unfold) +- [AbsVal](#absval) +- [ArgMax](#argmax) +- [BatchNorm](#batchnorm) +- [Bias](#bias) +- [BinaryOp](#binaryop) +- [BNLL](#bnll) +- [Cast](#cast) +- [CELU](#celu) +- [Clip](#clip) +- [Concat](#concat) +- [Convolution](#convolution) +- [Convolution1D](#convolution1d) +- [Convolution3D](#convolution3d) +- [ConvolutionDepthWise](#convolutiondepthwise) +- [ConvolutionDepthWise1D](#convolutiondepthwise1d) +- [ConvolutionDepthWise3D](#convolutiondepthwise3d) +- [CopyTo](#copyto) +- [Crop](#crop) +- [CumulativeSum](#cumulativesum) +- [Deconvolution](#deconvolution) +- [Deconvolution1D](#deconvolution1d) +- [Deconvolution3D](#deconvolution3d) +- [DeconvolutionDepthWise](#deconvolutiondepthwise) +- [DeconvolutionDepthWise1D](#deconvolutiondepthwise1d) +- [DeconvolutionDepthWise3D](#deconvolutiondepthwise3d) +- [DeformableConv2D](#deformableconv2d) +- [Dequantize](#dequantize) +- [Diag](#diag) +- [Dropout](#dropout) +- [Eltwise](#eltwise) +- [ELU](#elu) +- [Embed](#embed) +- [Exp](#exp) +- [Flatten](#flatten) +- [Fold](#fold) +- [GELU](#gelu) +- [GLU](#glu) +- [Gemm](#gemm) +- [GridSample](#gridsample) +- [GroupNorm](#groupnorm) +- [GRU](#gru) +- [HardSigmoid](#hardsigmoid) +- [HardSwish](#hardswish) +- [IndexSelect](#indexselect) +- [InnerProduct](#innerproduct) +- [Input](#input) +- [InstanceNorm](#instancenorm) +- [Interp](#interp) +- [InverseSpectrogram](#inversespectrogram) +- [LayerNorm](#layernorm) +- [Log](#log) +- [LRN](#lrn) +- [LSTM](#lstm) +- [MemoryData](#memorydata) +- [Mish](#mish) +- [MultiHeadAttention](#multiheadattention) +- [MVN](#mvn) +- [Noop](#noop) +- [Normalize](#normalize) +- [Packing](#packing) +- [Padding](#padding) +- [Permute](#permute) +- [PixelShuffle](#pixelshuffle) +- [Pooling](#pooling) +- [Pooling1D](#pooling1d) +- [Pooling3D](#pooling3d) +- [Power](#power) +- [PReLU](#prelu) +- [Quantize](#quantize) +- [Reduction](#reduction) +- [ReLU](#relu) +- [Reorg](#reorg) +- [Requantize](#requantize) +- [Reshape](#reshape) +- [RMSNorm](#rmsnorm) +- [RNN](#rnn) +- [Scale](#scale) +- [SELU](#selu) +- [Shrink](#shrink) +- [ShuffleChannel](#shufflechannel) +- [Sigmoid](#sigmoid) +- [Slice](#slice) +- [Softmax](#softmax) +- [Softplus](#softplus) +- [Spectrogram](#spectrogram) +- [Split](#split) +- [Swish](#swish) +- [TanH](#tanh) +- [Threshold](#threshold) +- [Tile](#tile) +- [UnaryOp](#unaryop) +- [Unfold](#unfold) # AbsVal + ``` y = abs(x) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # ArgMax + ``` y = argmax(x, out_max_val, topk) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | out_max_val | int | 0 | | -| 1 | topk | int | 1 | | +| param id | name | type | default | description | +| -------- | ----------- | ---- | ------- | ----------- | +| 0 | out_max_val | int | 0 | | +| 1 | topk | int | 1 | | # BatchNorm + ``` y = (x - mean) / sqrt(var + eps) * slope + bias ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | channels | int | 0 | | -| 1 | eps | float | 0.f | | +| param id | name | type | default | description | +| -------- | -------- | ----- | ------- | ----------- | +| 0 | channels | int | 0 | | +| 1 | eps | float | 0.f | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| slope_data | float | [channels] | -| mean_data | float | [channels] | -| var_data | float | [channels] | -| bias_data | float | [channels] | +| weight | type | shape | +| ---------- | ----- | ---------- | +| slope_data | float | [channels] | +| mean_data | float | [channels] | +| var_data | float | [channels] | +| bias_data | float | [channels] | # Bias + ``` y = x + bias ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | bias_data_size| int | 0 | | +| param id | name | type | default | description | +| -------- | -------------- | ---- | ------- | ----------- | +| 0 | bias_data_size | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| bias_data | float | [channels] | +| weight | type | shape | +| --------- | ----- | ---------- | +| bias_data | float | [channels] | # BinaryOp - This operation is used for binary computation, and the calculation rule depends on the [broadcasting rule](https://github.com/Tencent/ncnn/wiki/binaryop-broadcasting). + +This operation is used for binary computation, and the calculation rule depends on the [broadcasting rule](https://github.com/Tencent/ncnn/wiki/binaryop-broadcasting). + ``` C = binaryop(A, B) ``` + if with_scalar = 1: + - one_blob_only - support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | op_type | int | 0 | Operation type as follows | -| 1 | with_scalar | int | 0 | with_scalar=0 B is a matrix, with_scalar=1 B is a scalar | -| 2 | b | float | 0.f | When B is a scalar, B = b | +| param id | name | type | default | description | +| -------- | ----------- | ----- | ------- | -------------------------------------------------------- | +| 0 | op_type | int | 0 | Operation type as follows | +| 1 | with_scalar | int | 0 | with_scalar=0 B is a matrix, with_scalar=1 B is a scalar | +| 2 | b | float | 0.f | When B is a scalar, B = b | Operation type: + - 0 = ADD - 1 = SUB - 2 = MUL @@ -177,28 +186,31 @@ Operation type: - 11 = RATAN2 # BNLL + ``` y = log(1 + e^(-x)) , x > 0 y = log(1 + e^x), x < 0 ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # Cast + ``` y = cast(x) ``` -* one_blob_only -* support_packing +- one_blob_only +- support_packing -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | type_from | int | 0 | | -| 1 | type_to | int | 0 | | +| param id | name | type | default | description | +| -------- | --------- | ---- | ------- | ----------- | +| 0 | type_from | int | 0 | | +| 1 | type_to | int | 0 | | Element type: + - 0 = auto - 1 = float32 - 2 = float16 @@ -206,702 +218,730 @@ Element type: - 4 = bfloat16 # CELU + ``` if x < 0 y = (exp(x / alpha) - 1.f) * alpha else y = x ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 1.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | alpha | float | 1.f | | # Clip + ``` y = clamp(x, min, max) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | min | float | -FLT_MAX | | -| 1 | max | float | FLT_MAX | | +| param id | name | type | default | description | +| -------- | ---- | ----- | -------- | ----------- | +| 0 | min | float | -FLT_MAX | | +| 1 | max | float | FLT_MAX | | # Concat + ``` y = concat(x0, x1, x2, ...) by axis ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | axis | int | 0 | | +| param id | name | type | default | description | +| -------- | ---- | ---- | ------- | ----------- | +| 0 | axis | int | 0 | | # Convolution + ``` x2 = pad(x, pads, pad_value) x3 = conv(x2, weight, kernel, stride, dilation) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 8 | int8_scale_term| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 18 | pad_value | float | 0.f | | -| 19 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input, num_output] | -| bias_data | float | [num_output] | -| weight_data_int8_scales| float | [num_output] | -| bottom_blob_int8_scales| float | [1] | -| top_blob_int8_scales| float | [1] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 8 | int8_scale_term | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 18 | pad_value | float | 0.f | | +| 19 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------------------- | --------------- | ------------------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input, num_output] | +| bias_data | float | [num_output] | +| weight_data_int8_scales | float | [num_output] | +| bottom_blob_int8_scales | float | [1] | +| top_blob_int8_scales | float | [1] | # Convolution1D + ``` x2 = pad(x, pads, pad_value) x3 = conv1d(x2, weight, kernel, stride, dilation) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 15 | pad_right | int | pad_left | | -| 18 | pad_value | float | 0.f | | -| 19 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, num_input, num_output] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | -------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 15 | pad_right | int | pad_left | | +| 18 | pad_value | float | 0.f | | +| 19 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | --------------- | --------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, num_input, num_output] | +| bias_data | float | [num_output] | # Convolution3D + ``` x2 = pad(x, pads, pad_value) x3 = conv3d(x2, weight, kernel, stride, dilation) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 17 | pad_behind | int | pad_front | | -| 18 | pad_value | float | 0.f | | -| 21 | kernel_d | int | kernel_w | | -| 22 | dilation_d | int | dilation_w | | -| 23 | stride_d | int | stride_w | | -| 24 | pad_front | int | pad_left | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, kernel_h, kernel_d, num_input, num_output] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 17 | pad_behind | int | pad_front | | +| 18 | pad_value | float | 0.f | | +| 21 | kernel_d | int | kernel_w | | +| 22 | dilation_d | int | dilation_w | | +| 23 | stride_d | int | stride_w | | +| 24 | pad_front | int | pad_left | | + +| weight | type | shape | +| ----------- | --------------- | ----------------------------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, kernel_h, kernel_d, num_input, num_output] | +| bias_data | float | [num_output] | # ConvolutionDepthWise + ``` x2 = pad(x, pads, pad_value) x3 = conv(x2, weight, kernel, stride, dilation, group) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 8 | int8_scale_term| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 18 | pad_value | float | 0.f | | -| 19 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | -| weight_data_int8_scales| float | [group] | -| bottom_blob_int8_scales| float | [1] | -| top_blob_int8_scales| float | [1] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 8 | int8_scale_term | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 18 | pad_value | float | 0.f | | +| 19 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------------------- | --------------- | ------------------------------------------------------------------ | +| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | +| weight_data_int8_scales | float | [group] | +| bottom_blob_int8_scales | float | [1] | +| top_blob_int8_scales | float | [1] | # ConvolutionDepthWise1D + ``` x2 = pad(x, pads, pad_value) x3 = conv1d(x2, weight, kernel, stride, dilation, group) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 15 | pad_right | int | pad_left | | -| 18 | pad_value | float | 0.f | | -| 19 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | -------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 15 | pad_right | int | pad_left | | +| 18 | pad_value | float | 0.f | | +| 19 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | --------------- | -------------------------------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | # ConvolutionDepthWise3D + ``` x2 = pad(x, pads, pad_value) x3 = conv3d(x2, weight, kernel, stride, dilation, group) + bias y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 17 | pad_behind | int | pad_front | | -| 18 | pad_value | float | 0.f | | -| 21 | kernel_d | int | kernel_w | | -| 22 | dilation_d | int | dilation_w | | -| 23 | stride_d | int | stride_w | | -| 24 | pad_front | int | pad_left | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, kernel_h, kernel_d, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 17 | pad_behind | int | pad_front | | +| 18 | pad_value | float | 0.f | | +| 21 | kernel_d | int | kernel_w | | +| 22 | dilation_d | int | dilation_w | | +| 23 | stride_d | int | stride_w | | +| 24 | pad_front | int | pad_left | | + +| weight | type | shape | +| ----------- | --------------- | ---------------------------------------------------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, kernel_h, kernel_d, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | # CopyTo + ``` self[offset] = src ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | woffset | int | 0 | | -| 1 | hoffset | int | 0 | | -| 13 | doffset | int | 0 | | -| 2 | coffset | int | 0 | | -| 9 | starts | array | [ ] | | -| 11 | axes | array | [ ] | | +| param id | name | type | default | description | +| -------- | ------- | ----- | ------- | ----------- | +| 0 | woffset | int | 0 | | +| 1 | hoffset | int | 0 | | +| 13 | doffset | int | 0 | | +| 2 | coffset | int | 0 | | +| 9 | starts | array | [ ] | | +| 11 | axes | array | [ ] | | # Crop + ``` y = crop(x) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | woffset | int | 0 | | -| 1 | hoffset | int | 0 | | -| 13 | doffset | int | 0 | | -| 2 | coffset | int | 0 | | -| 3 | outw | int | 0 | | -| 4 | outh | int | 0 | | -| 14 | outd | int | 0 | | -| 5 | outc | int | 0 | | -| 6 | woffset2 | int | 0 | | -| 7 | hoffset2 | int | 0 | | -| 15 | doffset2 | int | 0 | | -| 8 | coffset2 | int | 0 | | -| 9 | starts | array | [ ] | | -| 10 | ends | array | [ ] | | -| 11 | axes | array | [ ] | | +- one_blob_only + +| param id | name | type | default | description | +| -------- | -------- | ----- | ------- | ----------- | +| 0 | woffset | int | 0 | | +| 1 | hoffset | int | 0 | | +| 13 | doffset | int | 0 | | +| 2 | coffset | int | 0 | | +| 3 | outw | int | 0 | | +| 4 | outh | int | 0 | | +| 14 | outd | int | 0 | | +| 5 | outc | int | 0 | | +| 6 | woffset2 | int | 0 | | +| 7 | hoffset2 | int | 0 | | +| 15 | doffset2 | int | 0 | | +| 8 | coffset2 | int | 0 | | +| 9 | starts | array | [ ] | | +| 10 | ends | array | [ ] | | +| 11 | axes | array | [ ] | | # CumulativeSum If axis < 0, we use axis = x.dims + axis -It implements https://pytorch.org/docs/stable/generated/torch.cumsum.html +It implements -* one_blob_only -* support_inplace - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | axis | int | 0 | | +- one_blob_only +- support_inplace +| param id | name | type | default | description | +| -------- | ---- | ---- | ------- | ----------- | +| 0 | axis | int | 0 | | # Deconvolution + ``` x2 = deconv(x, weight, kernel, stride, dilation) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 18 | output_pad_right| int | 0 | | -| 19 | output_pad_bottom| int | output_pad_right | | -| 20 | output_w | int | 0 | | -| 21 | output_h | int | output_w | | -| 28 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, kernel_h, num_input, num_output] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 18 | output_pad_right | int | 0 | | +| 19 | output_pad_bottom | int | output_pad_right | | +| 20 | output_w | int | 0 | | +| 21 | output_h | int | output_w | | +| 28 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | ---------- | ------------------------------------------- | +| weight_data | float/fp16 | [kernel_w, kernel_h, num_input, num_output] | +| bias_data | float | [num_output] | # Deconvolution1D + ``` x2 = deconv1d(x, weight, kernel, stride, dilation) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 15 | pad_right | int | pad_left | | -| 18 | output_pad_right| int | 0 | | -| 20 | output_w | int | 0 | | -| 28 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, num_input, num_output] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | -------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 15 | pad_right | int | pad_left | | +| 18 | output_pad_right | int | 0 | | +| 20 | output_w | int | 0 | | +| 28 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | ---------- | --------------------------------- | +| weight_data | float/fp16 | [kernel_w, num_input, num_output] | +| bias_data | float | [num_output] | # Deconvolution3D + ``` x2 = deconv3d(x, weight, kernel, stride, dilation) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 17 | pad_behind | int | pad_front | | -| 18 | output_pad_right| int | 0 | | -| 19 | output_pad_bottom| int | output_pad_right | | -| 20 | output_pad_behind| int | output_pad_right | | -| 21 | kernel_d | int | kernel_w | | -| 22 | dilation_d | int | dilation_w | | -| 23 | stride_d | int | stride_w | | -| 24 | pad_front | int | pad_left | | -| 25 | output_w | int | 0 | | -| 26 | output_h | int | output_w | | -| 27 | output_d | int | output_w | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, kernel_h, kernel_d, num_input, num_output] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 17 | pad_behind | int | pad_front | | +| 18 | output_pad_right | int | 0 | | +| 19 | output_pad_bottom | int | output_pad_right | | +| 20 | output_pad_behind | int | output_pad_right | | +| 21 | kernel_d | int | kernel_w | | +| 22 | dilation_d | int | dilation_w | | +| 23 | stride_d | int | stride_w | | +| 24 | pad_front | int | pad_left | | +| 25 | output_w | int | 0 | | +| 26 | output_h | int | output_w | | +| 27 | output_d | int | output_w | | + +| weight | type | shape | +| ----------- | ---------- | ----------------------------------------------------- | +| weight_data | float/fp16 | [kernel_w, kernel_h, kernel_d, num_input, num_output] | +| bias_data | float | [num_output] | # DeconvolutionDepthWise + ``` x2 = deconv(x, weight, kernel, stride, dilation, group) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 18 | output_pad_right| int | 0 | | -| 19 | output_pad_bottom| int | output_pad_right | | -| 20 | output_w | int | 0 | | -| 21 | output_h | int | output_w | | -| 28 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, kernel_h, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 18 | output_pad_right | int | 0 | | +| 19 | output_pad_bottom | int | output_pad_right | | +| 20 | output_w | int | 0 | | +| 21 | output_h | int | output_w | | +| 28 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | ---------- | ------------------------------------------------------------------ | +| weight_data | float/fp16 | [kernel_w, kernel_h, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | # DeconvolutionDepthWise1D + ``` x2 = deconv1d(x, weight, kernel, stride, dilation, group) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 15 | pad_right | int | pad_left | | -| 18 | output_pad_right| int | 0 | | -| 20 | output_w | int | 0 | | -| 28 | dynamic_weight| int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | -------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 15 | pad_right | int | pad_left | | +| 18 | output_pad_right | int | 0 | | +| 20 | output_w | int | 0 | | +| 28 | dynamic_weight | int | 0 | | + +| weight | type | shape | +| ----------- | ---------- | -------------------------------------------------------- | +| weight_data | float/fp16 | [kernel_w, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | # DeconvolutionDepthWise3D + ``` x2 = deconv3d(x, weight, kernel, stride, dilation, group) + bias x3 = depad(x2, pads, pad_value) y = activation(x3, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 7 | group | int | 1 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 17 | pad_behind | int | pad_front | | -| 18 | output_pad_right| int | 0 | | -| 19 | output_pad_bottom| int | output_pad_right | | -| 20 | output_pad_behind| int | output_pad_right | | -| 21 | kernel_d | int | kernel_w | | -| 22 | dilation_d | int | dilation_w | | -| 23 | stride_d | int | stride_w | | -| 24 | pad_front | int | pad_left | | -| 25 | output_w | int | 0 | | -| 26 | output_h | int | output_w | | -| 27 | output_d | int | output_w | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16 | [kernel_w, kernel_h, kernel_d, num_input / group, num_output / group, group] | -| bias_data | float | [num_output] | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 7 | group | int | 1 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 17 | pad_behind | int | pad_front | | +| 18 | output_pad_right | int | 0 | | +| 19 | output_pad_bottom | int | output_pad_right | | +| 20 | output_pad_behind | int | output_pad_right | | +| 21 | kernel_d | int | kernel_w | | +| 22 | dilation_d | int | dilation_w | | +| 23 | stride_d | int | stride_w | | +| 24 | pad_front | int | pad_left | | +| 25 | output_w | int | 0 | | +| 26 | output_h | int | output_w | | +| 27 | output_d | int | output_w | | + +| weight | type | shape | +| ----------- | ---------- | ---------------------------------------------------------------------------- | +| weight_data | float/fp16 | [kernel_w, kernel_h, kernel_d, num_input / group, num_output / group, group] | +| bias_data | float | [num_output] | # DeformableConv2D + ``` x2 = deformableconv2d(x, offset, mask, weight, kernel, stride, dilation) + bias y = activation(x2, act_type, act_params) ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 5 | bias_term | int | 0 | | -| 6 | weight_data_size| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input, num_output] | -| bias_data | float | [num_output] | +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 5 | bias_term | int | 0 | | +| 6 | weight_data_size | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | + +| weight | type | shape | +| ----------- | --------------- | ------------------------------------------- | +| weight_data | float/fp16/int8 | [kernel_w, kernel_h, num_input, num_output] | +| bias_data | float | [num_output] | # Dequantize + ``` y = x * scale + bias ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | scale_data_size| int | 1 | | -| 1 | bias_data_size| int | 0 | | +| param id | name | type | default | description | +| -------- | --------------- | ---- | ------- | ----------- | +| 0 | scale_data_size | int | 1 | | +| 1 | bias_data_size | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| scale_data | float | [scale_data_size] | -| bias_data | float | [bias_data_size] | +| weight | type | shape | +| ---------- | ----- | ----------------- | +| scale_data | float | [scale_data_size] | +| bias_data | float | [bias_data_size] | # Diag + ``` y = diag(x, diagonal) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | diagonal | int | 0 | | +| param id | name | type | default | description | +| -------- | -------- | ---- | ------- | ----------- | +| 0 | diagonal | int | 0 | | # Dropout + ``` y = x * scale ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | scale | float | 1.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | scale | float | 1.f | | # Eltwise + ``` y = elementwise_op(x0, x1, ...) ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | op_type | int | 0 | | -| 1 | coeffs | array | [ ] | | +| param id | name | type | default | description | +| -------- | ------- | ----- | ------- | ----------- | +| 0 | op_type | int | 0 | | +| 1 | coeffs | array | [ ] | | Operation type: + - 0 = PROD - 1 = SUM - 2 = MAX # ELU + ``` if x < 0 y = (exp(x) - 1) * alpha else y = x ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 0.1f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | alpha | float | 0.1f | | # Embed + ``` y = embedding(x) ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | input_dim | int | 0 | | -| 2 | bias_term | int | 0 | | -| 3 | weight_data_size | int | 0 | | -| 18 | int8_scale_term| int | 0 | | +| param id | name | type | default | description | +| -------- | ---------------- | ---- | ------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | input_dim | int | 0 | | +| 2 | bias_term | int | 0 | | +| 3 | weight_data_size | int | 0 | | +| 18 | int8_scale_term | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float | [weight_data_size] | -| bias_term | float | [num_output] | -| weight_data_int8_scales| float | [1] | +| weight | type | shape | +| ----------------------- | ----- | ------------------ | +| weight_data | float | [weight_data_size] | +| bias_term | float | [num_output] | +| weight_data_int8_scales | float | [1] | # Exp + ``` if base == -1 y = exp(shift + x * scale) else y = pow(base, (shift + x * scale)) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | base | float | -1.f | | -| 1 | scale | float | 1.f | | -| 2 | shift | float | 0.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | base | float | -1.f | | +| 1 | scale | float | 1.f | | +| 2 | shift | float | 0.f | | # Flatten + Reshape blob to 1 dimension -* one_blob_only +- one_blob_only # Fold + ``` y = fold(x) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | -| 20 | output_w | int | 0 | | -| 21 | output_h | int | output_w | | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ---------- | ---- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | +| 20 | output_w | int | 0 | | +| 21 | output_h | int | output_w | | # GELU + ``` if fast_gelu == 1 y = 0.5 * x * (1 + tanh(0.79788452 * (x + 0.044715 * x * x * x))); else y = 0.5 * x * erfc(-0.70710678 * x) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | fast_gelu | int | 0 | use approximation | +| param id | name | type | default | description | +| -------- | --------- | ---- | ------- | ----------------- | +| 0 | fast_gelu | int | 0 | use approximation | # GLU @@ -913,13 +953,14 @@ where a is the first half of the input matrix and b is the second half. axis specifies the dimension to split the input -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | axis | int | 0 | | +| param id | name | type | default | description | +| -------- | ---- | ---- | ------- | ----------- | +| 0 | axis | int | 0 | | # Gemm + ``` a = transA ? transpose(x0) : x0 b = transb ? transpose(x1) : x1 @@ -927,88 +968,91 @@ c = x2 y = (gemm(a, b) + c * beta) * alpha ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 1.f | | -| 1 | beta | float | 1.f | | -| 2 | transA | int | 0 | | -| 3 | transb | int | 0 | | -| 4 | constantA | int | 0 | | -| 5 | constantB | int | 0 | | -| 6 | constantC | int | 0 | | -| 7 | constantM | int | 0 | | -| 8 | constantN | int | 0 | | -| 9 | constantK | int | 0 | | -| 10 | constant_broadcast_type_C | int | 0 | | -| 11 | output_N1M | int | 0 | | -| 12 | output_elempack | int | 0 | | -| 13 | output_elemtype | int | 0 | | -| 14 | output_transpose | int| 0 | | -| 18 | int8_scale_term | int | 0 | | -| 20 | constant_TILE_M | int | 0 | | -| 21 | constant_TILE_N | int | 0 | | -| 22 | constant_TILE_K | int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| A_data | float/fp16/int8 | [M, K] or [K, M] | -| B_data | float/fp16/int8 | [N, K] or [K, N] | -| C_data | float | [1], [M] or [N] or [1, M] or [N,1] or [N, M] | -| A_data_int8_scales| float | [M] | -| B_data_int8_scales| float | [1] | +| param id | name | type | default | description | +| -------- | ------------------------- | ----- | ------- | ----------- | +| 0 | alpha | float | 1.f | | +| 1 | beta | float | 1.f | | +| 2 | transA | int | 0 | | +| 3 | transb | int | 0 | | +| 4 | constantA | int | 0 | | +| 5 | constantB | int | 0 | | +| 6 | constantC | int | 0 | | +| 7 | constantM | int | 0 | | +| 8 | constantN | int | 0 | | +| 9 | constantK | int | 0 | | +| 10 | constant_broadcast_type_C | int | 0 | | +| 11 | output_N1M | int | 0 | | +| 12 | output_elempack | int | 0 | | +| 13 | output_elemtype | int | 0 | | +| 14 | output_transpose | int | 0 | | +| 18 | int8_scale_term | int | 0 | | +| 20 | constant_TILE_M | int | 0 | | +| 21 | constant_TILE_N | int | 0 | | +| 22 | constant_TILE_K | int | 0 | | + +| weight | type | shape | +| ------------------ | --------------- | -------------------------------------------- | +| A_data | float/fp16/int8 | [M, K] or [K, M] | +| B_data | float/fp16/int8 | [N, K] or [K, N] | +| C_data | float | [1], [M] or [N] or [1, M] or [N,1] or [N, M] | +| A_data_int8_scales | float | [M] | +| B_data_int8_scales | float | [1] | # GridSample + ``` Given an input and a flow-field grid, computes the output using input values and pixel locations from grid. -For each output location output[:, h2, w2], the size-2 vector grid[h2, w2, 2] specifies input pixel[:, h1, w1] locations x and y, +For each output location output[:, h2, w2], the size-2 vector grid[h2, w2, 2] specifies input pixel[:, h1, w1] locations x and y, which are used to interpolate the output value output[:, h2, w2] This function is often used in conjunction with affine_grid() to build Spatial Transformer Networks . ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | sample_type | int | 1 | | -| 1 | padding_mode | int | 1 | | -| 2 | align_corner | int | 0 | | -| 3 | permute_fusion| int | 0 | fuse with permute | - +| param id | name | type | default | description | +| -------- | -------------- | ---- | ------- | ----------------- | +| 0 | sample_type | int | 1 | | +| 1 | padding_mode | int | 1 | | +| 2 | align_corner | int | 0 | | +| 3 | permute_fusion | int | 0 | fuse with permute | Sample type: + - 1 = Nearest - 2 = Bilinear - 3 = Bicubic Padding mode: + - 1 = zeros - 2 = border - 3 = reflection - # GroupNorm + ``` split x along channel axis into group x0, x1 ... l2 normalize for each group x0, x1 ... y = x * gamma + beta ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | group | int | 1 | | -| 1 | channels | int | 0 | | -| 2 | eps | float | 0.001f | x = x / sqrt(var + eps) | -| 3 | affine | int | 1 | | +| param id | name | type | default | description | +| -------- | -------- | ----- | ------- | ----------------------- | +| 0 | group | int | 1 | | +| 1 | channels | int | 0 | | +| 2 | eps | float | 0.001f | x = x / sqrt(var + eps) | +| 3 | affine | int | 1 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| gamma_data | float | [channels] | -| beta_data | float | [channels] | +| weight | type | shape | +| ---------- | ----- | ---------- | +| gamma_data | float | [channels] | +| beta_data | float | [channels] | # GRU + Apply a single-layer GRU to a feature sequence of `T` timesteps. The input blob shape is `[w=input_size, h=T]` and the output blob shape is `[w=num_output, h=T]`. ``` @@ -1016,134 +1060,149 @@ y = gru(x) y0, hidden y1 = gru(x0, hidden x1) ``` -* one_blob_only if bidirectional +- one_blob_only if bidirectional -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | hidden size of output | -| 1 | weight_data_size| int | 0 | total size of weight matrix | -| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | +| param id | name | type | default | description | +| -------- | ---------------- | ---- | ------- | ------------------------------------- | +| 0 | num_output | int | 0 | hidden size of output | +| 1 | weight_data_size | int | 0 | total size of weight matrix | +| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_xc_data| float/fp16/int8 | [input_size, num_output * 3, num_directions] | -| bias_c_data | float/fp16/int8 | [num_output, 4, num_directions] | -| weight_hc_data| float/fp16/int8 | [num_output, num_output * 3, num_directions] | +| weight | type | shape | +| -------------- | --------------- | -------------------------------------------- | +| weight_xc_data | float/fp16/int8 | [input_size, num_output * 3, num_directions] | +| bias_c_data | float/fp16/int8 | [num_output, 4, num_directions] | +| weight_hc_data | float/fp16/int8 | [num_output, num_output * 3, num_directions] | Direction flag: + - 0 = forward only - 1 = reverse only - 2 = bidirectional # HardSigmoid + ``` y = clamp(x * alpha + beta, 0, 1) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 0.2f | | -| 1 | beta | float | 0.5f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | alpha | float | 0.2f | | +| 1 | beta | float | 0.5f | | # HardSwish + ``` y = x * clamp(x * alpha + beta, 0, 1) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace + +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | alpha | float | 0.2f | | +| 1 | beta | float | 0.5f | | + +# IndexSelect -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 0.2f | | -| 1 | beta | float | 0.5f | | +| param id | name | type | default | description | +| -------- | ----- | ---- | ------- | ----------- | +| 0 | alpha | int | 0 | | # InnerProduct + ``` x2 = innerproduct(x, weight) + bias y = activation(x2, act_type, act_params) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | bias_term | int | 0 | | -| 2 | weight_data_size| int | 0 | | -| 8 | int8_scale_term| int | 0 | | -| 9 | activation_type| int | 0 | | -| 10 | activation_params| array | [ ] | | +- one_blob_only -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_data | float/fp16/int8 | [num_input, num_output] | -| bias_data | float | [num_output] | -| weight_data_int8_scales| float | [num_output] | -| bottom_blob_int8_scales| float | [1] | +| param id | name | type | default | description | +| -------- | ----------------- | ----- | ------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | bias_term | int | 0 | | +| 2 | weight_data_size | int | 0 | | +| 8 | int8_scale_term | int | 0 | | +| 9 | activation_type | int | 0 | | +| 10 | activation_params | array | [ ] | | + +| weight | type | shape | +| ----------------------- | --------------- | ----------------------- | +| weight_data | float/fp16/int8 | [num_input, num_output] | +| bias_data | float | [num_output] | +| weight_data_int8_scales | float | [num_output] | +| bottom_blob_int8_scales | float | [1] | # Input + ``` y = input ``` -* support_inplace +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | w | int | 0 | | -| 1 | h | int | 0 | | -| 11 | d | int | 0 | | -| 2 | c | int | 0 | | +| param id | name | type | default | description | +| -------- | ---- | ---- | ------- | ----------- | +| 0 | w | int | 0 | | +| 1 | h | int | 0 | | +| 11 | d | int | 0 | | +| 2 | c | int | 0 | | # InstanceNorm + ``` split x along channel axis into instance x0, x1 ... l2 normalize for each channel instance x0, x1 ... y = x * gamma + beta ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | channels | int | 0 | | -| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | -| 2 | affine | int | 1 | | +| param id | name | type | default | description | +| -------- | -------- | ----- | ------- | ----------------------- | +| 0 | channels | int | 0 | | +| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | +| 2 | affine | int | 1 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| gamma_data | float | [channels] | -| beta_data | float | [channels] | +| weight | type | shape | +| ---------- | ----- | ---------- | +| gamma_data | float | [channels] | +| beta_data | float | [channels] | # Interp + ``` if dynamic_target_size == 0 y = resize(x) by fixed size or scale else y = resize(x0, size(x1)) ``` -* one_blob_only if dynamic_target_size == 0 +- one_blob_only if dynamic_target_size == 0 -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | resize_type | int | 0 | | -| 1 | height_scale | float | 1.f | | -| 2 | width_scale | float | 1.f | | -| 3 | output_height | int | 0 | | -| 4 | output_width | int | 0 | | -| 5 | dynamic_target_size| int | 0 | | -| 6 | align_corner | int | 0 | | +| param id | name | type | default | description | +| -------- | ------------------- | ----- | ------- | ----------- | +| 0 | resize_type | int | 0 | | +| 1 | height_scale | float | 1.f | | +| 2 | width_scale | float | 1.f | | +| 3 | output_height | int | 0 | | +| 4 | output_width | int | 0 | | +| 5 | dynamic_target_size | int | 0 | | +| 6 | align_corner | int | 0 | | Resize type: + - 1 = Nearest - 2 = Bilinear - 3 = Bicubic # InverseSpectrogram + ``` x1 = x as complex x1 = x1 * sqrt(norm) if normalized @@ -1155,77 +1214,82 @@ if returns == 1 return y1 real if returns == 2 return y1 imag ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | n_fft | int | 0 | | -| 1 | returns | int | 1 | | -| 2 | hoplen | int | n_fft / 4 | | -| 3 | winlen | int | n_fft | | -| 4 | window_type | int | 0 | 0=ones 1=hann 2=hamming | -| 5 | center | int | 1 | | -| 7 | normalized | int | 0 | 0=no 1=n_fft 2=window-l2-energy | +| param id | name | type | default | description | +| -------- | ----------- | ---- | --------- | ------------------------------- | +| 0 | n_fft | int | 0 | | +| 1 | returns | int | 1 | | +| 2 | hoplen | int | n_fft / 4 | | +| 3 | winlen | int | n_fft | | +| 4 | window_type | int | 0 | 0=ones 1=hann 2=hamming | +| 5 | center | int | 1 | | +| 7 | normalized | int | 0 | 0=no 1=n_fft 2=window-l2-energy | # LayerNorm + ``` split x along outmost axis into part x0, x1 ... l2 normalize for each part x0, x1 ... y = x * gamma + beta by elementwise ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | affine_size | int | 0 | | -| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | -| 2 | affine | int | 1 | | +| param id | name | type | default | description | +| -------- | ----------- | ----- | ------- | ----------------------- | +| 0 | affine_size | int | 0 | | +| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | +| 2 | affine | int | 1 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| gamma_data | float | [affine_size] | -| beta_data | float | [affine_size] | +| weight | type | shape | +| ---------- | ----- | ------------- | +| gamma_data | float | [affine_size] | +| beta_data | float | [affine_size] | # Log + ``` if base == -1 y = log(shift + x * scale) else y = log(shift + x * scale) / log(base) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | base | float | -1.f | | -| 1 | scale | float | 1.f | | -| 2 | shift | float | 0.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | base | float | -1.f | | +| 1 | scale | float | 1.f | | +| 2 | shift | float | 0.f | | # LRN + ``` if region_type == ACROSS_CHANNELS square_sum = sum of channel window of local_size if region_type == WITHIN_CHANNEL square_sum = sum of spatial window of local_size y = x * pow(bias + alpha * square_sum / (local_size * local_size), -beta) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | region_type | int | 0 | | -| 1 | local_size | int | 5 | | -| 2 | alpha | float | 1.f | | -| 3 | beta | float | 0.75f | | -| 4 | bias | float | 1.f | | +| param id | name | type | default | description | +| -------- | ----------- | ----- | ------- | ----------- | +| 0 | region_type | int | 0 | | +| 1 | local_size | int | 5 | | +| 2 | alpha | float | 1.f | | +| 3 | beta | float | 0.75f | | +| 4 | bias | float | 1.f | | Region type: + - 0 = ACROSS_CHANNELS - 1 = WITHIN_CHANNEL # LSTM + Apply a single-layer LSTM to a feature sequence of `T` timesteps. The input blob shape is `[w=input_size, h=T]` and the output blob shape is `[w=num_output, h=T]`. ``` @@ -1233,53 +1297,57 @@ y = lstm(x) y0, hidden y1, cell y2 = lstm(x0, hidden x1, cell x2) ``` -* one_blob_only if bidirectional +- one_blob_only if bidirectional -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | output size of output | -| 1 | weight_data_size| int | 0 | total size of IFOG weight matrix | -| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | -| 3 | hidden_size | int | num_output| hidden size | +| param id | name | type | default | description | +| -------- | ---------------- | ---- | ---------- | ------------------------------------- | +| 0 | num_output | int | 0 | output size of output | +| 1 | weight_data_size | int | 0 | total size of IFOG weight matrix | +| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | +| 3 | hidden_size | int | num_output | hidden size | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_xc_data| float/fp16/int8 | [input_size, hidden_size * 4, num_directions] | -| bias_c_data | float/fp16/int8 | [hidden_size, 4, num_directions] | -| weight_hc_data| float/fp16/int8 | [num_output, hidden_size * 4, num_directions] | -| weight_hr_data| float/fp16/int8 | [hidden_size, num_output, num_directions] | +| weight | type | shape | +| -------------- | --------------- | --------------------------------------------- | +| weight_xc_data | float/fp16/int8 | [input_size, hidden_size * 4, num_directions] | +| bias_c_data | float/fp16/int8 | [hidden_size, 4, num_directions] | +| weight_hc_data | float/fp16/int8 | [num_output, hidden_size * 4, num_directions] | +| weight_hr_data | float/fp16/int8 | [hidden_size, num_output, num_directions] | Direction flag: + - 0 = forward only - 1 = reverse only - 2 = bidirectional # MemoryData + ``` y = data ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | w | int | 0 | | -| 1 | h | int | 0 | | -| 11 | d | int | 0 | | -| 2 | c | int | 0 | | -| 21 | load_type | int | 1 | 1=fp32 | +| param id | name | type | default | description | +| -------- | --------- | ---- | ------- | ----------- | +| 0 | w | int | 0 | | +| 1 | h | int | 0 | | +| 11 | d | int | 0 | | +| 2 | c | int | 0 | | +| 21 | load_type | int | 1 | 1=fp32 | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| data | float | [w, h, d, c] | +| weight | type | shape | +| ------ | ----- | ------------ | +| data | float | [w, h, d, c] | # Mish + ``` y = x * tanh(log(exp(x) + 1)) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # MultiHeadAttention + ``` split q k v into num_head part q0, k0, v0, q1, k1, v1 ... for each num_head part @@ -1294,33 +1362,34 @@ for each num_head part y = affine(out) ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | embed_dim | int | 0 | | -| 1 | num_heads | int | 1 | | -| 2 | weight_data_size| int | 0 | qdim = weight_data_size / embed_dim | -| 3 | kdim | int | embed_dim | | -| 4 | vdim | int | embed_dim | | -| 5 | attn_mask | int | 0 | | -| 6 | scale | float | 1.f / sqrt(embed_dim / num_heads) | | -| 18 | int8_scale_term | int | 0 | | - -| weight | type | shape | -| ------------- | ----- | --------------------- | -| q_weight_data | float/fp16/int8 | [embed_dim * qdim] | -| q_bias_data | float | [embed_dim] | -| k_weight_data | float/fp16/int8 | [embed_dim * kdim] | -| k_bias_data | float | [embed_dim] | -| v_weight_data | float/fp16/int8 | [embed_dim * vdim] | -| v_bias_data | float | [embed_dim] | -| out_weight_data| float/fp16/int8 | [qdim * embed_dim] | -| out_bias_data | float | [qdim] | -| q_weight_data_int8_scales| float | [embed_dim] | -| k_weight_data_int8_scales| float | [embed_dim] | -| v_weight_data_int8_scales| float | [embed_dim] | -| out_weight_data_int8_scales| float | [1] | +| param id | name | type | default | description | +| -------- | ---------------- | ----- | --------------------------------- | ----------------------------------- | +| 0 | embed_dim | int | 0 | | +| 1 | num_heads | int | 1 | | +| 2 | weight_data_size | int | 0 | qdim = weight_data_size / embed_dim | +| 3 | kdim | int | embed_dim | | +| 4 | vdim | int | embed_dim | | +| 5 | attn_mask | int | 0 | | +| 6 | scale | float | 1.f / sqrt(embed_dim / num_heads) | | +| 18 | int8_scale_term | int | 0 | | + +| weight | type | shape | +| --------------------------- | --------------- | ------------------ | +| q_weight_data | float/fp16/int8 | [embed_dim * qdim] | +| q_bias_data | float | [embed_dim] | +| k_weight_data | float/fp16/int8 | [embed_dim * kdim] | +| k_bias_data | float | [embed_dim] | +| v_weight_data | float/fp16/int8 | [embed_dim * vdim] | +| v_bias_data | float | [embed_dim] | +| out_weight_data | float/fp16/int8 | [qdim * embed_dim] | +| out_bias_data | float | [qdim] | +| q_weight_data_int8_scales | float | [embed_dim] | +| k_weight_data_int8_scales | float | [embed_dim] | +| v_weight_data_int8_scales | float | [embed_dim] | +| out_weight_data_int8_scales | float | [1] | # MVN + ``` if normalize_variance == 1 && across_channels == 1 y = (x - mean) / (sqrt(var) + eps) of whole blob if normalize_variance == 1 && across_channels == 0 y = (x - mean) / (sqrt(var) + eps) of each channel @@ -1328,20 +1397,22 @@ if normalize_variance == 0 && across_channels == 1 y = x - mean of whole bl if normalize_variance == 0 && across_channels == 0 y = x - mean of each channel ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | normalize_variance| int | 0 | | -| 1 | across_channels| int | 0 | | -| 2 | eps | float | 0.0001f | x = x / (sqrt(var) + eps) | +| param id | name | type | default | description | +| -------- | ------------------ | ----- | ------- | ------------------------- | +| 0 | normalize_variance | int | 0 | | +| 1 | across_channels | int | 0 | | +| 2 | eps | float | 0.0001f | x = x / (sqrt(var) + eps) | # Noop + ``` y = x ``` # Normalize + ``` if across_spatial == 1 && across_channel == 1 x2 = normalize(x) of whole blob if across_spatial == 1 && across_channel == 0 x2 = normalize(x) of each channel @@ -1349,79 +1420,85 @@ if across_spatial == 0 && across_channel == 1 x2 = normalize(x) of each pos y = x2 * scale ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | across_spatial| int | 0 | | -| 1 | channel_shared| int | 0 | | -| 2 | eps | float | 0.0001f | see eps mode | -| 3 | scale_data_size| int | 0 | | -| 4 | across_channel| int | 0 | | -| 9 | eps_mode | int | 0 | | +| param id | name | type | default | description | +| -------- | --------------- | ----- | ------- | ------------ | +| 0 | across_spatial | int | 0 | | +| 1 | channel_shared | int | 0 | | +| 2 | eps | float | 0.0001f | see eps mode | +| 3 | scale_data_size | int | 0 | | +| 4 | across_channel | int | 0 | | +| 9 | eps_mode | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| scale_data | float | [scale_data_size] | +| weight | type | shape | +| ---------- | ----- | ----------------- | +| scale_data | float | [scale_data_size] | Eps Mode: -- 0 = caffe/mxnet x = x / sqrt(var + eps) -- 1 = pytorch x = x / max(sqrt(var), eps) -- 2 = tensorflow x = x / sqrt(max(var, eps)) + +- 0 = caffe/mxnet x = x / sqrt(var + eps) +- 1 = pytorch x = x / max(sqrt(var), eps) +- 2 = tensorflow x = x / sqrt(max(var, eps)) # Packing + ``` y = wrap_packing(x) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | out_elempack | int | 1 | | -| 1 | use_padding | int | 0 | | -| 2 | cast_type_from| int | 0 | | -| 3 | cast_type_to | int | 0 | | -| 4 | storage_type_from| int | 0 | | -| 5 | storage_type_to| int | 0 | | +| param id | name | type | default | description | +| -------- | ----------------- | ---- | ------- | ----------- | +| 0 | out_elempack | int | 1 | | +| 1 | use_padding | int | 0 | | +| 2 | cast_type_from | int | 0 | | +| 3 | cast_type_to | int | 0 | | +| 4 | storage_type_from | int | 0 | | +| 5 | storage_type_to | int | 0 | | # Padding + ``` y = pad(x, pads) ``` -| param id | name | type | default | description | -| --------- | ------------- | ---- | --------- | ----------------- | -| 0 | top | int | 0 | | -| 1 | bottom | int | 0 | | -| 2 | left | int | 0 | | -| 3 | right | int | 0 | | -| 4 | type | int | 0 | | -| 5 | value | float | 0 | | -| 6 | per_channel_pad_data_size| int | 0 | | -| 7 | front | int | stride_w | | -| 8 | behind | int | pad_left | | +| param id | name | type | default | description | +| -------- | ------------------------- | ----- | -------- | ----------- | +| 0 | top | int | 0 | | +| 1 | bottom | int | 0 | | +| 2 | left | int | 0 | | +| 3 | right | int | 0 | | +| 4 | type | int | 0 | | +| 5 | value | float | 0 | | +| 6 | per_channel_pad_data_size | int | 0 | | +| 7 | front | int | stride_w | | +| 8 | behind | int | pad_left | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| per_channel_pad_data| float | [per_channel_pad_data_size] | +| weight | type | shape | +| -------------------- | ----- | --------------------------- | +| per_channel_pad_data | float | [per_channel_pad_data_size] | Padding type: + - 0 = CONSTANT - 1 = REPLICATE - 2 = REFLECT # Permute + ``` y = reorder(x) ``` -| param id | name | type | default | description | -| --------- | ------------- | ---- | --------- | ----------------- | -| 0 | order_type | int | 0 | | +| param id | name | type | default | description | +| -------- | ---------- | ---- | ------- | ----------- | +| 0 | order_type | int | 0 | | Order Type: + - 0 = WH WHC WHDC - 1 = HW HWC HWDC - 2 = WCH WDHC @@ -1448,183 +1525,198 @@ Order Type: - 23 = CDHW # PixelShuffle + ``` if mode == 0 y = depth_to_space(x) where x channel order is sw-sh-outc if mode == 1 y = depth_to_space(x) where x channel order is outc-sw-sh ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ---- | --------- | ----------------- | -| 0 | upscale_factor| int | 1 | | -| 1 | mode | int | 0 | | +| param id | name | type | default | description | +| -------- | -------------- | ---- | ------- | ----------- | +| 0 | upscale_factor | int | 1 | | +| 1 | mode | int | 0 | | # Pooling + ``` x2 = pad(x, pads) x3 = pooling(x2, kernel, stride) ``` -| param id | name | type | default | description | -| --------- | --------------| ---- | --------- | ----------------- | -| 0 | pooling_type | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | stride_w | int | 1 | | -| 3 | pad_left | int | 0 | | -| 4 | global_pooling| int | 0 | | -| 5 | pad_mode | int | 0 | | -| 6 | avgpool_count_include_pad| int | 0 | | -| 7 | adaptive_pooling| int | 0 | | -| 8 | out_w | int | 0 | | -| 11 | kernel_h | int | kernel_w | | -| 12 | stride_h | int | stride_w | | -| 13 | pad_top | int | pad_left | | -| 14 | pad_right | int | pad_left | | -| 15 | pad_bottom | int | pad_top | | -| 18 | out_h | int | out_w | | +| param id | name | type | default | description | +| -------- | ------------------------- | ---- | -------- | ----------- | +| 0 | pooling_type | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | stride_w | int | 1 | | +| 3 | pad_left | int | 0 | | +| 4 | global_pooling | int | 0 | | +| 5 | pad_mode | int | 0 | | +| 6 | avgpool_count_include_pad | int | 0 | | +| 7 | adaptive_pooling | int | 0 | | +| 8 | out_w | int | 0 | | +| 11 | kernel_h | int | kernel_w | | +| 12 | stride_h | int | stride_w | | +| 13 | pad_top | int | pad_left | | +| 14 | pad_right | int | pad_left | | +| 15 | pad_bottom | int | pad_top | | +| 18 | out_h | int | out_w | | Pooling type: + - 0 = MAX - 1 = AVG Pad mode: + - 0 = full padding - 1 = valid padding - 2 = tensorflow padding=SAME or onnx padding=SAME_UPPER - 3 = onnx padding=SAME_LOWER # Pooling1D + ``` x2 = pad(x, pads) x3 = pooling1d(x2, kernel, stride) ``` -| param id | name | type | default | description | -| --------- | --------------| ---- | --------- | ----------------- | -| 0 | pooling_type | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | stride_w | int | 1 | | -| 3 | pad_left | int | 0 | | -| 4 | global_pooling| int | 0 | | -| 5 | pad_mode | int | 0 | | -| 6 | avgpool_count_include_pad| int | 0 | | -| 7 | adaptive_pooling| int | 0 | | -| 8 | out_w | int | 0 | | -| 14 | pad_right | int | pad_left | | +| param id | name | type | default | description | +| -------- | ------------------------- | ---- | -------- | ----------- | +| 0 | pooling_type | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | stride_w | int | 1 | | +| 3 | pad_left | int | 0 | | +| 4 | global_pooling | int | 0 | | +| 5 | pad_mode | int | 0 | | +| 6 | avgpool_count_include_pad | int | 0 | | +| 7 | adaptive_pooling | int | 0 | | +| 8 | out_w | int | 0 | | +| 14 | pad_right | int | pad_left | | Pooling type: + - 0 = MAX - 1 = AVG Pad mode: + - 0 = full padding - 1 = valid padding - 2 = tensorflow padding=SAME or onnx padding=SAME_UPPER - 3 = onnx padding=SAME_LOWER # Pooling3D + ``` x2 = pad(x, pads) x3 = pooling3d(x2, kernel, stride) ``` -| param id | name | type | default | description | -| --------- | --------------| ---- | --------- | ----------------- | -| 0 | pooling_type | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | stride_w | int | 1 | | -| 3 | pad_left | int | 0 | | -| 4 | global_pooling| int | 0 | | -| 5 | pad_mode | int | 0 | | -| 6 | avgpool_count_include_pad| int | 0 | | -| 7 | adaptive_pooling| int | 0 | | -| 8 | out_w | int | 0 | | -| 11 | kernel_h | int | kernel_w | | -| 12 | stride_h | int | stride_w | | -| 13 | pad_top | int | pad_left | | -| 14 | pad_right | int | pad_left | | -| 15 | pad_bottom | int | pad_top | | -| 16 | pad_behind | int | pad_front | | -| 18 | out_h | int | out_w | | -| 21 | kernel_d | int | kernel_w | | -| 22 | stride_d | int | stride_w | | -| 23 | pad_front | int | pad_left | | -| 28 | out_d | int | out_w | | +| param id | name | type | default | description | +| -------- | ------------------------- | ---- | --------- | ----------- | +| 0 | pooling_type | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | stride_w | int | 1 | | +| 3 | pad_left | int | 0 | | +| 4 | global_pooling | int | 0 | | +| 5 | pad_mode | int | 0 | | +| 6 | avgpool_count_include_pad | int | 0 | | +| 7 | adaptive_pooling | int | 0 | | +| 8 | out_w | int | 0 | | +| 11 | kernel_h | int | kernel_w | | +| 12 | stride_h | int | stride_w | | +| 13 | pad_top | int | pad_left | | +| 14 | pad_right | int | pad_left | | +| 15 | pad_bottom | int | pad_top | | +| 16 | pad_behind | int | pad_front | | +| 18 | out_h | int | out_w | | +| 21 | kernel_d | int | kernel_w | | +| 22 | stride_d | int | stride_w | | +| 23 | pad_front | int | pad_left | | +| 28 | out_d | int | out_w | | Pooling type: + - 0 = MAX - 1 = AVG Pad mode: + - 0 = full padding - 1 = valid padding - 2 = tensorflow padding=SAME or onnx padding=SAME_UPPER - 3 = onnx padding=SAME_LOWER # Power + ``` y = pow((shift + x * scale), power) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | power | float | 1.f | | -| 1 | scale | float | 1.f | | -| 2 | shift | float | 0.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | power | float | 1.f | | +| 1 | scale | float | 1.f | | +| 2 | shift | float | 0.f | | # PReLU + ``` if x < 0 y = x * slope else y = x ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_slope | int | 0 | | +| param id | name | type | default | description | +| -------- | --------- | ---- | ------- | ----------- | +| 0 | num_slope | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| slope_data | float | [num_slope] | +| weight | type | shape | +| ---------- | ----- | ----------- | +| slope_data | float | [num_slope] | # Quantize + ``` y = float2int8(x * scale) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | scale_data_size| int | 1 | | +| param id | name | type | default | description | +| -------- | --------------- | ---- | ------- | ----------- | +| 0 | scale_data_size | int | 1 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| scale_data | float | [scale_data_size] | +| weight | type | shape | +| ---------- | ----- | ----------------- | +| scale_data | float | [scale_data_size] | # Reduction + ``` y = reduce_op(x * coeff) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | operation | int | 0 | | -| 1 | reduce_all | int | 1 | | -| 2 | coeff | float | 1.f | | -| 3 | axes | array | [ ] | | -| 4 | keepdims | int | 0 | | -| 5 | fixbug0 | int | 0 | hack for bug fix, should be 1 | +| param id | name | type | default | description | +| -------- | ---------- | ----- | ------- | ----------------------------- | +| 0 | operation | int | 0 | | +| 1 | reduce_all | int | 1 | | +| 2 | coeff | float | 1.f | | +| 3 | axes | array | [ ] | | +| 4 | keepdims | int | 0 | | +| 5 | fixbug0 | int | 0 | hack for bug fix, should be 1 | Operation type: + - 0 = SUM - 1 = ASUM - 2 = SUMSQ @@ -1638,96 +1730,103 @@ Operation type: - 10 = LogSumExp # ReLU + ``` if x < 0 y = x * slope else y = x ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | slope | float | 0.f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | slope | float | 0.f | | # Reorg + ``` if mode == 0 y = space_to_depth(x) where x channel order is sw-sh-outc if mode == 1 y = space_to_depth(x) where x channel order is outc-sw-sh ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ---- | --------- | ----------------- | -| 0 | stride | int | 1 | | -| 1 | mode | int | 0 | | +| param id | name | type | default | description | +| -------- | ------ | ---- | ------- | ----------- | +| 0 | stride | int | 1 | | +| 1 | mode | int | 0 | | # Requantize + ``` x2 = x * scale_in + bias x3 = activation(x2) y = float2int8(x3 * scale_out) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | scale_in_data_size| int | 1 | | -| 1 | scale_out_data_size| int | 1 | | -| 2 | bias_data_size| int | 0 | | -| 3 | activation_type| int | 0 | | -| 4 | activation_params| int | [ ] | | +| param id | name | type | default | description | +| -------- | ------------------- | ---- | ------- | ----------- | +| 0 | scale_in_data_size | int | 1 | | +| 1 | scale_out_data_size | int | 1 | | +| 2 | bias_data_size | int | 0 | | +| 3 | activation_type | int | 0 | | +| 4 | activation_params | int | [ ] | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| scale_in_data | float | [scale_in_data_size] | -| scale_out_data| float | [scale_out_data_size] | -| bias_data | float | [bias_data_size] | +| weight | type | shape | +| -------------- | ----- | --------------------- | +| scale_in_data | float | [scale_in_data_size] | +| scale_out_data | float | [scale_out_data_size] | +| bias_data | float | [bias_data_size] | # Reshape + ``` if permute == 1 y = hwc2chw(reshape(chw2hwc(x))) else y = reshape(x) ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | w | int | -233 | | -| 1 | h | int | -233 | | -| 11 | d | int | -233 | | -| 2 | c | int | -233 | | -| 3 | permute | int | 0 | | +| param id | name | type | default | description | +| -------- | ------- | ---- | ------- | ----------- | +| 0 | w | int | -233 | | +| 1 | h | int | -233 | | +| 11 | d | int | -233 | | +| 2 | c | int | -233 | | +| 3 | permute | int | 0 | | Reshape flag: + - 0 = copy from bottom - -1 = remaining - -233 = drop this dim(default) # RMSNorm + ``` split x along outmost axis into part x0, x1 ... root mean square normalize for each part x0, x1 ... y = x * gamma by elementwise ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | affine_size | int | 0 | | -| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | -| 2 | affine | int | 1 | | +| param id | name | type | default | description | +| -------- | ----------- | ----- | ------- | ----------------------- | +| 0 | affine_size | int | 0 | | +| 1 | eps | float | 0.001f | x = x / sqrt(var + eps) | +| 2 | affine | int | 1 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| gamma_data | float | [affine_size] | +| weight | type | shape | +| ---------- | ----- | ------------- | +| gamma_data | float | [affine_size] | # RNN + Apply a single-layer RNN to a feature sequence of `T` timesteps. The input blob shape is `[w=input_size, h=T]` and the output blob shape is `[w=num_output, h=T]`. ``` @@ -1735,127 +1834,137 @@ y = rnn(x) y0, hidden y1 = rnn(x0, hidden x1) ``` -* one_blob_only if bidirectional +- one_blob_only if bidirectional -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | hidden size of output | -| 1 | weight_data_size| int | 0 | total size of weight matrix | -| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | +| param id | name | type | default | description | +| -------- | ---------------- | ---- | ------- | ------------------------------------- | +| 0 | num_output | int | 0 | hidden size of output | +| 1 | weight_data_size | int | 0 | total size of weight matrix | +| 2 | direction | int | 0 | 0=forward, 1=reverse, 2=bidirectional | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| weight_xc_data| float/fp16/int8 | [input_size, num_output, num_directions] | -| bias_c_data | float/fp16/int8 | [num_output, 1, num_directions] | -| weight_hc_data| float/fp16/int8 | [num_output, num_output, num_directions] | +| weight | type | shape | +| -------------- | --------------- | ---------------------------------------- | +| weight_xc_data | float/fp16/int8 | [input_size, num_output, num_directions] | +| bias_c_data | float/fp16/int8 | [num_output, 1, num_directions] | +| weight_hc_data | float/fp16/int8 | [num_output, num_output, num_directions] | Direction flag: + - 0 = forward only - 1 = reverse only - 2 = bidirectional # Scale + ``` if scale_data_size == -233 y = x0 * x1 else y = x * scale + bias ``` -* one_blob_only if scale_data_size != -233 -* support_inplace +- one_blob_only if scale_data_size != -233 +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | scale_data_size| int | 0 | | -| 1 | bias_term | int | 0 | | +| param id | name | type | default | description | +| -------- | --------------- | ---- | ------- | ----------- | +| 0 | scale_data_size | int | 0 | | +| 1 | bias_term | int | 0 | | -| weight | type | shape | -| ------------- | ----- | --------------------- | -| scale_data | float | [scale_data_size] | -| bias_data | float | [scale_data_size] | +| weight | type | shape | +| ---------- | ----- | ----------------- | +| scale_data | float | [scale_data_size] | +| bias_data | float | [scale_data_size] | # SELU + ``` if x < 0 y = (exp(x) - 1.f) * alpha * lambda else y = x * lambda ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | alpha | float | 1.67326324f| | -| 1 | lambda | float | 1.050700987f| | +| param id | name | type | default | description | +| -------- | ------ | ----- | ------------ | ----------- | +| 0 | alpha | float | 1.67326324f | | +| 1 | lambda | float | 1.050700987f | | # Shrink + ``` if x < -lambd y = x + bias if x > lambd y = x - bias else y = x ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | bias | float | 0.0f | | -| 1 | lambd | float | 0.5f | | +| param id | name | type | default | description | +| -------- | ----- | ----- | ------- | ----------- | +| 0 | bias | float | 0.0f | | +| 1 | lambd | float | 0.5f | | # ShuffleChannel + ``` if reverse == 0 y = shufflechannel(x) by group if reverse == 1 y = shufflechannel(x) by channel / group ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ---- | --------- | ----------------- | -| 0 | group | int | 1 | | -| 1 | reverse | int | 0 | | +| param id | name | type | default | description | +| -------- | ------- | ---- | ------- | ----------- | +| 0 | group | int | 1 | | +| 1 | reverse | int | 0 | | # Sigmoid + ``` y = 1 / (1 + exp(-x)) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # Slice + ``` split x along axis into slices, each part slice size is based on slices array ``` -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | slices | array | [ ] | | -| 1 | axis | int | 0 | | -| 2 | indices | array | [ ] | | +| param id | name | type | default | description | +| -------- | ------- | ----- | ------- | ----------- | +| 0 | slices | array | [ ] | | +| 1 | axis | int | 0 | | +| 2 | indices | array | [ ] | | # Softmax + ``` softmax(x, axis) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | axis | int | 0 | | -| 1 | fixbug0 | int | 0 | hack for bug fix, should be 1 | +| param id | name | type | default | description | +| -------- | ------- | ---- | ------- | ----------------------------- | +| 0 | axis | int | 0 | | +| 1 | fixbug0 | int | 0 | hack for bug fix, should be 1 | # Softplus + ``` y = log(exp(x) + 1) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # Spectrogram + ``` x1 = pad(x) if center y = stft(x1) @@ -1866,68 +1975,74 @@ if power == 1 return magnitude if power == 2 return square of magnitude ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | n_fft | int | 0 | | -| 1 | power | int | 0 | | -| 2 | hoplen | int | n_fft / 4 | | -| 3 | winlen | int | n_fft | | -| 4 | window_type | int | 0 | 0=ones 1=hann 2=hamming | -| 5 | center | int | 1 | | -| 6 | pad_type | int | 2 | 0=CONSTANT 1=REPLICATE 2=REFLECT | -| 7 | normalized | int | 0 | 0=no 1=n_fft 2=window-l2-energy | -| 8 | onesided | int | 1 | | +| param id | name | type | default | description | +| -------- | ----------- | ---- | --------- | -------------------------------- | +| 0 | n_fft | int | 0 | | +| 1 | power | int | 0 | | +| 2 | hoplen | int | n_fft / 4 | | +| 3 | winlen | int | n_fft | | +| 4 | window_type | int | 0 | 0=ones 1=hann 2=hamming | +| 5 | center | int | 1 | | +| 6 | pad_type | int | 2 | 0=CONSTANT 1=REPLICATE 2=REFLECT | +| 7 | normalized | int | 0 | 0=no 1=n_fft 2=window-l2-energy | +| 8 | onesided | int | 1 | | # Split + ``` y0, y1 ... = x ``` # Swish + ``` y = x / (1 + exp(-x)) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # TanH + ``` y = tanh(x) ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace # Threshold + ``` if x > threshold y = 1 else y = 0 ``` -* one_blob_only -* support_inplace +- one_blob_only +- support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | threshold | float | 0.f | | +| param id | name | type | default | description | +| -------- | --------- | ----- | ------- | ----------- | +| 0 | threshold | float | 0.f | | # Tile + ``` y = repeat tiles along axis for x ``` -* one_blob_only +- one_blob_only -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | axis | int | 0 | | -| 1 | tiles | int | 1 | | -| 2 | repeats | array | [ ] | | +| param id | name | type | default | description | +| -------- | ------- | ----- | ------- | ----------- | +| 0 | axis | int | 0 | | +| 1 | tiles | int | 1 | | +| 2 | repeats | array | [ ] | | # UnaryOp + ``` y = unaryop(x) ``` @@ -1935,11 +2050,12 @@ y = unaryop(x) - one_blob_only - support_inplace -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | op_type | int | 0 | Operation type as follows | +| param id | name | type | default | description | +| -------- | ------- | ---- | ------- | ------------------------- | +| 0 | op_type | int | 0 | Operation type as follows | Operation type: + - 0 = ABS - 1 = NEG - 2 = FLOOR @@ -1962,22 +2078,23 @@ Operation type: - 19 = TRUNC # Unfold + ``` y = unfold(x) ``` -* one_blob_only - -| param id | name | type | default | description | -| --------- | ------------- | ----- | --------- | ----------------- | -| 0 | num_output | int | 0 | | -| 1 | kernel_w | int | 0 | | -| 2 | dilation_w | int | 1 | | -| 3 | stride_w | int | 1 | | -| 4 | pad_left | int | 0 | | -| 11 | kernel_h | int | kernel_w | | -| 12 | dilation_h | int | dilation_w | | -| 13 | stride_h | int | stride_w | | -| 14 | pad_top | int | pad_left | | -| 15 | pad_right | int | pad_left | | -| 16 | pad_bottom | int | pad_top | | +- one_blob_only + +| param id | name | type | default | description | +| -------- | ---------- | ---- | ---------- | ----------- | +| 0 | num_output | int | 0 | | +| 1 | kernel_w | int | 0 | | +| 2 | dilation_w | int | 1 | | +| 3 | stride_w | int | 1 | | +| 4 | pad_left | int | 0 | | +| 11 | kernel_h | int | kernel_w | | +| 12 | dilation_h | int | dilation_w | | +| 13 | stride_h | int | stride_w | | +| 14 | pad_top | int | pad_left | | +| 15 | pad_right | int | pad_left | | +| 16 | pad_bottom | int | pad_top | | diff --git a/tools/pnnx/src/CMakeLists.txt b/tools/pnnx/src/CMakeLists.txt index b1ac6f5c0245..cbc31e217ef8 100644 --- a/tools/pnnx/src/CMakeLists.txt +++ b/tools/pnnx/src/CMakeLists.txt @@ -575,6 +575,7 @@ set(pnnx_pass_ncnn_SRCS pass_ncnn/torch_cumsum.cpp pass_ncnn/torch_diag.cpp pass_ncnn/torch_flatten.cpp + pass_ncnn/torch_index_select.cpp pass_ncnn/torch_istft.cpp pass_ncnn/torch_logsumexp.cpp pass_ncnn/torch_matmul.cpp diff --git a/tools/pnnx/src/pass_ncnn/torch_index_select.cpp b/tools/pnnx/src/pass_ncnn/torch_index_select.cpp new file mode 100644 index 000000000000..8cb0cef548b6 --- /dev/null +++ b/tools/pnnx/src/pass_ncnn/torch_index_select.cpp @@ -0,0 +1,58 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2024 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// Unless required by applicable law or agreed to in writing, software distributed +// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +// CONDITIONS OF ANY KIND, either express or implied. See the License for the +// specific language governing permissions and limitations under the License. + +#include "pass_ncnn.h" + +namespace pnnx { + +namespace ncnn { + +class torch_index_select : public GraphRewriterPass +{ +public: + const char* match_pattern_graph() const + { + return R"PNNXIR(7767517 +4 4 +pnnx.Input input_0 0 1 input +pnnx.Input input_1 0 1 index +torch.index_select op_0 2 1 input index out dim=%dim +pnnx.Output output 1 0 out +)PNNXIR"; + } + + const char* type_str() const + { + return "IndexSelect"; + } + + const char* name_str() const + { + return "index_select"; + } + + void write(Operator* op, const std::map& captured_params) const + { + int dim = captured_params.at("dim").i; + + // 设置参数 + op->params["0"] = dim; + } +}; + +REGISTER_GLOBAL_PNNX_NCNN_GRAPH_REWRITER_PASS(torch_index_select, 60) + +} // namespace ncnn + +} // namespace pnnx From 4dbfd7193ec3fdcce1017225517dea4ee9c9b702 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E4=BD=B0=E9=98=85?= <43716063+Baiyuetribe@users.noreply.github.com> Date: Thu, 16 Jan 2025 21:10:51 +0800 Subject: [PATCH 2/3] check c++03 and ctest --- .github/workflows/linux-x64-cpu-gcc.yml | 218 +++++++------- src/CMakeLists.txt | 1 + src/layer/indexselect.cpp | 270 ++++++++++++++++++ src/layer/indexselect.h | 37 +++ tests/CMakeLists.txt | 1 + tests/test_indexselect.cpp | 113 ++++++++ tools/pnnx/tests/ncnn/CMakeLists.txt | 1 + .../tests/ncnn/test_torch_index_select.py | 149 ++++++++++ 8 files changed, 681 insertions(+), 109 deletions(-) create mode 100644 src/layer/indexselect.cpp create mode 100644 src/layer/indexselect.h create mode 100644 tests/test_indexselect.cpp create mode 100644 tools/pnnx/tests/ncnn/test_torch_index_select.py diff --git a/.github/workflows/linux-x64-cpu-gcc.yml b/.github/workflows/linux-x64-cpu-gcc.yml index ab2185be3e74..31abbe47c251 100644 --- a/.github/workflows/linux-x64-cpu-gcc.yml +++ b/.github/workflows/linux-x64-cpu-gcc.yml @@ -1,33 +1,33 @@ name: linux-x64-cpu-gcc on: push: - branches: [master] + # branches: [master] paths: - - '.github/workflows/linux-x64-cpu-gcc.yml' - - 'toolchains/host-c.gcc.toolchain.cmake' - - 'CMakeLists.txt' - - 'cmake/**' - - 'src/*' - - 'src/layer/*' - - 'src/layer/x86/**' - - 'tests/**' - - 'tools/**' - - '!tools/pnnx/**' - - 'examples/**' + - ".github/workflows/linux-x64-cpu-gcc.yml" + - "toolchains/host-c.gcc.toolchain.cmake" + - "CMakeLists.txt" + - "cmake/**" + - "src/*" + - "src/layer/*" + - "src/layer/x86/**" + - "tests/**" + - "tools/**" + - "!tools/pnnx/**" + - "examples/**" pull_request: branches: [master] paths: - - '.github/workflows/linux-x64-cpu-gcc.yml' - - 'toolchains/host-c.gcc.toolchain.cmake' - - 'CMakeLists.txt' - - 'cmake/**' - - 'src/*' - - 'src/layer/*' - - 'src/layer/x86/**' - - 'tests/**' - - 'tools/**' - - '!tools/pnnx/**' - - 'examples/**' + - ".github/workflows/linux-x64-cpu-gcc.yml" + - "toolchains/host-c.gcc.toolchain.cmake" + - "CMakeLists.txt" + - "cmake/**" + - "src/*" + - "src/layer/*" + - "src/layer/x86/**" + - "tests/**" + - "tools/**" + - "!tools/pnnx/**" + - "examples/**" concurrency: group: linux-x64-cpu-gcc-${{ github.ref }} cancel-in-progress: true @@ -38,97 +38,97 @@ jobs: linux-gcc: runs-on: ubuntu-20.04 steps: - - uses: actions/checkout@v4 - - name: update - run: sudo apt-get update - - name: protobuf - run: sudo apt-get install libprotobuf-dev protobuf-compiler libopencv-dev - - name: build-sse2 - run: | - mkdir build-sse2 && cd build-sse2 - cmake -DNCNN_AVX=OFF -DNCNN_AVX2=OFF -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-sse2 - run: cd build-sse2 && ctest --output-on-failure -j $(nproc) - - name: build-shared - run: | - mkdir build-shared && cd build-shared - cmake -DNCNN_AVX2=ON -DNCNN_SHARED_LIB=ON .. - cmake --build . -j $(nproc) - - name: build-avx2 - run: | - mkdir build-avx2 && cd build-avx2 - cmake -DNCNN_AVX2=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx2 - run: cd build-avx2 && ctest --output-on-failure -j $(nproc) - - name: build-avx - run: | - mkdir build-avx && cd build-avx - cmake -DNCNN_AVX2=OFF -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx - run: cd build-avx && ctest --output-on-failure -j $(nproc) - - name: build-avx1-2 - run: | - mkdir build-avx1-2 && cd build-avx1-2 - cmake -DNCNN_AVX2=ON -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx1-2 - run: cd build-avx1-2 && ctest --output-on-failure -j $(nproc) - - name: build-noint8 - run: | - mkdir build-noint8 && cd build-noint8 - cmake -DNCNN_INT8=OFF -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-noint8 - run: cd build-noint8 && ctest --output-on-failure -j $(nproc) + - uses: actions/checkout@v4 + - name: update + run: sudo apt-get update + - name: protobuf + run: sudo apt-get install libprotobuf-dev protobuf-compiler libopencv-dev + - name: build-sse2 + run: | + mkdir build-sse2 && cd build-sse2 + cmake -DNCNN_AVX=OFF -DNCNN_AVX2=OFF -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-sse2 + run: cd build-sse2 && ctest --output-on-failure -j $(nproc) + - name: build-shared + run: | + mkdir build-shared && cd build-shared + cmake -DNCNN_AVX2=ON -DNCNN_SHARED_LIB=ON .. + cmake --build . -j $(nproc) + - name: build-avx2 + run: | + mkdir build-avx2 && cd build-avx2 + cmake -DNCNN_AVX2=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx2 + run: cd build-avx2 && ctest --output-on-failure -j $(nproc) + - name: build-avx + run: | + mkdir build-avx && cd build-avx + cmake -DNCNN_AVX2=OFF -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx + run: cd build-avx && ctest --output-on-failure -j $(nproc) + - name: build-avx1-2 + run: | + mkdir build-avx1-2 && cd build-avx1-2 + cmake -DNCNN_AVX2=ON -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx1-2 + run: cd build-avx1-2 && ctest --output-on-failure -j $(nproc) + - name: build-noint8 + run: | + mkdir build-noint8 && cd build-noint8 + cmake -DNCNN_INT8=OFF -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-noint8 + run: cd build-noint8 && ctest --output-on-failure -j $(nproc) linux-gcc-cpp03-nostdio-nostring-simplestl: runs-on: ubuntu-20.04 steps: - - uses: actions/checkout@v4 - - name: build-nostdio - run: | - mkdir build-nostdio && cd build-nostdio - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host.gcc-c++03.toolchain.cmake -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-nostdio - run: cd build-nostdio && ctest --output-on-failure -j $(nproc) - - name: build-nostdio-nostring - run: | - mkdir build-nostdio-nostring && cd build-nostdio-nostring - cmake -DNCNN_STDIO=OFF -DNCNN_STRING=OFF -DNCNN_BUILD_TESTS=OFF -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: build-simplestl - run: | - mkdir build-simplestl && cd build-simplestl - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-simplestl - run: cd build-simplestl && ctest --output-on-failure -j $(nproc) - - name: build-simplestl-simpleomp - run: | - mkdir build-simplestl-simpleomp && cd build-simplestl-simpleomp - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_SIMPLEOMP=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-simplestl-simpleomp - run: cd build-simplestl-simpleomp && ctest --output-on-failure -j $(nproc) + - uses: actions/checkout@v4 + - name: build-nostdio + run: | + mkdir build-nostdio && cd build-nostdio + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host.gcc-c++03.toolchain.cmake -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-nostdio + run: cd build-nostdio && ctest --output-on-failure -j $(nproc) + - name: build-nostdio-nostring + run: | + mkdir build-nostdio-nostring && cd build-nostdio-nostring + cmake -DNCNN_STDIO=OFF -DNCNN_STRING=OFF -DNCNN_BUILD_TESTS=OFF -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: build-simplestl + run: | + mkdir build-simplestl && cd build-simplestl + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-simplestl + run: cd build-simplestl && ctest --output-on-failure -j $(nproc) + - name: build-simplestl-simpleomp + run: | + mkdir build-simplestl-simpleomp && cd build-simplestl-simpleomp + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_SIMPLEOMP=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-simplestl-simpleomp + run: cd build-simplestl-simpleomp && ctest --output-on-failure -j $(nproc) linux-gcc-avx512: runs-on: [self-hosted, linux, t4] steps: - - uses: actions/checkout@v4 - - name: build - env: - CC: gcc - CXX: g++ - LD_LIBRARY_PATH: /data/action/install/lib64 - run: | - mkdir build && cd build - cmake -DNCNN_AVX2=ON -DNCNN_AVX512=ON -DNCNN_AVX512VNNI=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j 4 - - name: test - env: - LD_LIBRARY_PATH: /data/action/install/lib64 - run: cd build && ctest --output-on-failure -j 4 + - uses: actions/checkout@v4 + - name: build + env: + CC: gcc + CXX: g++ + LD_LIBRARY_PATH: /data/action/install/lib64 + run: | + mkdir build && cd build + cmake -DNCNN_AVX2=ON -DNCNN_AVX512=ON -DNCNN_AVX512VNNI=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j 4 + - name: test + env: + LD_LIBRARY_PATH: /data/action/install/lib64 + run: cd build && ctest --output-on-failure -j 4 diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index c97235d97a00..33f9db2a630f 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -169,6 +169,7 @@ ncnn_add_layer(Shrink) ncnn_add_layer(RMSNorm) ncnn_add_layer(Spectrogram) ncnn_add_layer(InverseSpectrogram) +ncnn_add_layer(IndexSelect) if(NCNN_VULKAN) ncnn_add_shader(${CMAKE_CURRENT_SOURCE_DIR}/convert_ycbcr.comp) diff --git a/src/layer/indexselect.cpp b/src/layer/indexselect.cpp new file mode 100644 index 000000000000..45dfc63b791b --- /dev/null +++ b/src/layer/indexselect.cpp @@ -0,0 +1,270 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// Unless required by applicable law or agreed to in writing, software distributed +// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +// CONDITIONS OF ANY KIND, either express or implied. See the License for the +// specific language governing permissions and limitations under the License. + +#include "indexselect.h" + +namespace ncnn { +IndexSelect::IndexSelect() +{ + one_blob_only = false; // 是否单一输入 + support_inplace = false; // 是否支持原地运算 +} + +int IndexSelect::load_param(const ParamDict& pd) +{ + dim = pd.get(0, -1); // dim = [-dim~dim-1] + return 0; +} + +int IndexSelect::forward(const std::vector& bottom_blobs, std::vector& top_blobs, const Option& opt) const +{ + const Mat& bottom_blob = bottom_blobs[0]; + const Mat& index_blob = bottom_blobs[1]; + Mat& top_blob = top_blobs[0]; // 仅1个输出 + int dims = bottom_blob.dims; + int w = bottom_blob.w; + int h = bottom_blob.h; + int d = bottom_blob.d; + int channels = bottom_blob.c; + size_t elemsize = bottom_blob.elemsize; + + int index_len = index_blob.w; // 索引数据 + + int axis = dim < 0 ? dim + dims : dim; + // 检查k值是否有效 + if (index_len < 1 || axis >= dims) + { + return -1; + } + + if (dims == 1) + { + // 创建输出blob + top_blob.create(index_len, elemsize, opt.blob_allocator); + const float* ptr = bottom_blob; + float* outptr = top_blob; + const int* index_ptr = index_blob; + for (int i = 0; i < index_len; i++) + { + outptr[i] = ptr[index_ptr[i]]; + } + } + else if (dims == 2) + { + if (axis == 0) + { + top_blob.create(w, index_len, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int i = 0; i < index_len; i++) + { + int index = index_ptr[i]; + const float* ptr_row = bottom_blob.row(index); + float* outptr_row = top_blob.row(i); + memcpy(outptr_row, ptr_row, w * sizeof(float)); + } + } + else if (axis == 1) + { + top_blob.create(index_len, h, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + for (int i = 0; i < h; i++) + { + const float* ptr_row = bottom_blob.row(i); + float* outptr_row = top_blob.row(i); + + // 对每一行,根据索引选择对应列 + for (int j = 0; j < index_len; j++) + { + int index = index_ptr[j]; + outptr_row[j] = ptr_row[index]; + } + } + } + } + else if (dims == 3) + { + if (axis == 0) // channels维度 + { + top_blob.create(w, h, index_len, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < index_len; q++) + { + int index = index_ptr[q]; + const Mat bottom_channel = bottom_blob.channel(index); + Mat top_channel = top_blob.channel(q); + + for (int i = 0; i < h; i++) + { + const float* ptr = bottom_channel.row(i); + float* outptr = top_channel.row(i); + memcpy(outptr, ptr, w * sizeof(float)); + } + } + } + else if (axis == 1) // h维度 + { + top_blob.create(w, index_len, channels, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < channels; q++) + { + const Mat bottom_channel = bottom_blob.channel(q); + Mat top_channel = top_blob.channel(q); + + for (int i = 0; i < index_len; i++) + { + int index = index_ptr[i]; + const float* ptr = bottom_channel.row(index); + float* outptr = top_channel.row(i); + memcpy(outptr, ptr, w * sizeof(float)); + } + } + } + else if (axis == 2) // w维度 + { + top_blob.create(index_len, h, channels, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < channels; q++) + { + const Mat bottom_channel = bottom_blob.channel(q); + Mat top_channel = top_blob.channel(q); + + for (int i = 0; i < h; i++) + { + const float* ptr = bottom_channel.row(i); + float* outptr = top_channel.row(i); + for (int j = 0; j < index_len; j++) + { + int index = index_ptr[j]; + outptr[j] = ptr[index]; + } + } + } + } + } + + else if (dims == 4) + { + if (axis == 0) // channels维度 + { + top_blob.create(w, h, d, index_len, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < index_len; q++) + { + int index = index_ptr[q]; + const Mat bottom_c = bottom_blob.channel(index); + Mat top_c = top_blob.channel(q); + + for (int z = 0; z < d; z++) + { + const Mat bottom_d = bottom_c.channel(z); + Mat top_d = top_c.channel(z); + for (int i = 0; i < h; i++) + { + const float* ptr = bottom_d.row(i); + float* outptr = top_d.row(i); + memcpy(outptr, ptr, w * sizeof(float)); + } + } + } + } + else if (axis == 1) // d维度 + { + top_blob.create(w, h, index_len, channels, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < channels; q++) + { + const Mat bottom_c = bottom_blob.channel(q); + Mat top_c = top_blob.channel(q); + + for (int z = 0; z < index_len; z++) + { + int index = index_ptr[z]; + const Mat bottom_d = bottom_c.channel(index); + Mat top_d = top_c.channel(z); + for (int i = 0; i < h; i++) + { + const float* ptr = bottom_d.row(i); + float* outptr = top_d.row(i); + memcpy(outptr, ptr, w * sizeof(float)); + } + } + } + } + else if (axis == 2) // h维度 + { + top_blob.create(w, index_len, d, channels, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < channels; q++) + { + const Mat bottom_c = bottom_blob.channel(q); + Mat top_c = top_blob.channel(q); + + for (int z = 0; z < d; z++) + { + const Mat bottom_d = bottom_c.channel(z); + Mat top_d = top_c.channel(z); + for (int i = 0; i < index_len; i++) + { + int index = index_ptr[i]; + const float* ptr = bottom_d.row(index); + float* outptr = top_d.row(i); + memcpy(outptr, ptr, w * sizeof(float)); + } + } + } + } + else if (axis == 3) // w维度 + { + top_blob.create(index_len, h, d, channels, elemsize, opt.blob_allocator); + const int* index_ptr = index_blob; + + for (int q = 0; q < channels; q++) + { + const Mat bottom_c = bottom_blob.channel(q); + Mat top_c = top_blob.channel(q); + + for (int z = 0; z < d; z++) + { + const Mat bottom_d = bottom_c.channel(z); + Mat top_d = top_c.channel(z); + for (int i = 0; i < h; i++) + { + const float* ptr = bottom_d.row(i); + float* outptr = top_d.row(i); + for (int j = 0; j < index_len; j++) + { + int index = index_ptr[j]; + outptr[j] = ptr[index]; + } + } + } + } + } + } + else + { + return -1; + } + + return 0; +} + +} // namespace ncnn \ No newline at end of file diff --git a/src/layer/indexselect.h b/src/layer/indexselect.h new file mode 100644 index 000000000000..1a0da5712082 --- /dev/null +++ b/src/layer/indexselect.h @@ -0,0 +1,37 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// Unless required by applicable law or agreed to in writing, software distributed +// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +// CONDITIONS OF ANY KIND, either express or implied. See the License for the +// specific language governing permissions and limitations under the License. + +#ifndef LAYER_INDEXSELECT_H +#define LAYER_INDEXSELECT_H + +#include "layer.h" + +namespace ncnn { + +class IndexSelect : public Layer +{ +public: + IndexSelect(); + + virtual int load_param(const ParamDict& pd); + + virtual int forward(const std::vector& bottom_blobs, std::vector& top_blobs, const Option& opt) const; + +public: + int dim; +}; + +} // namespace ncnn + +#endif // LAYER_TOPK_H diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt index f55859e736ea..4f44e8c9825b 100644 --- a/tests/CMakeLists.txt +++ b/tests/CMakeLists.txt @@ -115,6 +115,7 @@ ncnn_add_layer_test(GRU) ncnn_add_layer_test(HardSigmoid) ncnn_add_layer_test(HardSwish) ncnn_add_layer_test(InnerProduct) +ncnn_add_layer_test(IndexSelect) ncnn_add_layer_test(InstanceNorm) ncnn_add_layer_test(Interp) ncnn_add_layer_test(InverseSpectrogram) diff --git a/tests/test_indexselect.cpp b/tests/test_indexselect.cpp new file mode 100644 index 000000000000..9577df813469 --- /dev/null +++ b/tests/test_indexselect.cpp @@ -0,0 +1,113 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// Unless required by applicable law or agreed to in writing, software distributed +// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +// CONDITIONS OF ANY KIND, either express or implied. See the License for the +// specific language governing permissions and limitations under the License. + +#include "testutil.h" + +static ncnn::Mat IntArrayMat(int a0) +{ + ncnn::Mat m(1); + int* p = m; + p[0] = a0; + return m; +} + +static ncnn::Mat IntArrayMat(int a0, int a1) +{ + ncnn::Mat m(2); + int* p = m; + p[0] = a0; + p[1] = a1; + return m; +} + +static ncnn::Mat IntArrayMat(int a0, int a1, int a2) +{ + ncnn::Mat m(3); + int* p = m; + p[0] = a0; + p[1] = a1; + p[2] = a2; + return m; +} + +static ncnn::Mat IntArrayMat(int a0, int a1, int a2, int a3) +{ + ncnn::Mat m(4); + int* p = m; + p[0] = a0; + p[1] = a1; + p[2] = a2; + p[3] = a3; + return m; +} + +static int test_index_select(const ncnn::Mat& a, const ncnn::Mat& index, int axis) +{ + ncnn::ParamDict pd; + pd.set(0, axis); + + std::vector weights(0); + + std::vector a0(2); + a0[0] = a; + a0[1] = index; + + int ret = test_layer("IndexSelect", pd, weights, a0); + if (ret != 0) + { + fprintf(stderr, "test_index_select failed a.dims=%d a=(%d %d %d %d) index.w=%d axis=%d\n", a.dims, a.w, a.h, a.d, a.c, index.w, axis); + } + + return ret; +} + +static int test_index_select_0() +{ + return 0 + || test_index_select(RandomMat(3, 4, 5, 6), IntArrayMat(1, 0), 0) + || test_index_select(RandomMat(4, 4, 5, 6), IntArrayMat(1, 0), 1) + || test_index_select(RandomMat(3, 4, 7, 6), IntArrayMat(1, 0), 2) + || test_index_select(RandomMat(7, 2, 5, 6), IntArrayMat(1, 0, 3), 3); +} + +static int test_index_select_1() +{ + return 0 + || test_index_select(RandomMat(2, 3, 5), IntArrayMat(1, 0), -3) + || test_index_select(RandomMat(4, 3, 5), IntArrayMat(0, 1), -2) + || test_index_select(RandomMat(6, 4, 5), IntArrayMat(2, 0), -1); +} + +static int test_index_select_2() +{ + return 0 + || test_index_select(RandomMat(8, 6), IntArrayMat(1, 4, 3), 0) + || test_index_select(RandomMat(8, 7), IntArrayMat(3, 5), 1); +} + +static int test_index_select_3() +{ + return 0 + || test_index_select(RandomMat(18), IntArrayMat(1, 7, 9, 15), -1); +} + +int main() +{ + SRAND(7767517); + return 0 + || test_index_select_0() + || test_index_select_1() + || test_index_select_2() + || test_index_select_3(); +} \ No newline at end of file diff --git a/tools/pnnx/tests/ncnn/CMakeLists.txt b/tools/pnnx/tests/ncnn/CMakeLists.txt index 42c3bed32e05..614f17e80d86 100644 --- a/tools/pnnx/tests/ncnn/CMakeLists.txt +++ b/tools/pnnx/tests/ncnn/CMakeLists.txt @@ -203,6 +203,7 @@ pnnx_ncnn_add_test(torch_square) pnnx_ncnn_add_test(torch_tan) pnnx_ncnn_add_test(torch_tanh) pnnx_ncnn_add_test(torch_trunc) +pnnx_ncnn_add_test(torch_index_select) pnnx_ncnn_add_test(convnext_tiny) pnnx_ncnn_add_test(mobilenet_v2) diff --git a/tools/pnnx/tests/ncnn/test_torch_index_select.py b/tools/pnnx/tests/ncnn/test_torch_index_select.py new file mode 100644 index 000000000000..902159cc856b --- /dev/null +++ b/tools/pnnx/tests/ncnn/test_torch_index_select.py @@ -0,0 +1,149 @@ +# Tencent is pleased to support the open source community by making ncnn available. +# +# Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved. +# +# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +# in compliance with the License. You may obtain a copy of the License at +# +# https://opensource.org/licenses/BSD-3-Clause +# +# Unless required by applicable law or agreed to in writing, software distributed +# under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +# CONDITIONS OF ANY KIND, either express or implied. See the License for the +# specific language governing permissions and limitations under the License. + +import torch +import torch.nn as nn +import torch.nn.functional as F + +# Tencent is pleased to support the open source community by making ncnn available. +# +# Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved. +# +# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +# in compliance with the License. You may obtain a copy of the License at +# +# https://opensource.org/licenses/BSD-3-Clause +# +# Unless required by applicable law or agreed to in writing, software distributed +# under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +# CONDITIONS OF ANY KIND, either express or implied. See the License for the +# specific language governing permissions and limitations under the License. +# Tencent is pleased to support the open source community by making ncnn available. +# +# Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved. +# +# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +# in compliance with the License. You may obtain a copy of the License at +# +# https://opensource.org/licenses/BSD-3-Clause +# +# Unless required by applicable law or agreed to in writing, software distributed +# under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR +# CONDITIONS OF ANY KIND, either express or implied. See the License for the +# specific language governing permissions and limitations under the License. + +import torch +import torch.nn as nn +import torch.nn.functional as F + + +# 由于ncnn的MemoryData层不支持int64，所以先浮点存档，再转回int64 +# class Model(nn.Module): +# def __init__(self): +# super(Model, self).__init__() + +# def forward(self, x, y, z, d): +# # 1D +# x0 = torch.index_select(x, 0, torch.tensor([7, 9, 11])) +# # 2D +# y0 = torch.index_select(y, 0, torch.tensor([1, 0])) +# y1 = torch.index_select(y, 1, torch.tensor([1, 3, 2])) +# # 3D +# z0 = torch.index_select(z, -3, torch.tensor([1, 0])) +# z1 = torch.index_select(z, -2, torch.tensor([2, 0])) +# z2 = torch.index_select(z, -1, torch.tensor([1, 2])) +# # 4D +# d0 = torch.index_select(d, 0, torch.tensor([1, 0])) +# d1 = torch.index_select(d, 1, torch.tensor([1, 0])) +# d2 = torch.index_select(d, 2, torch.tensor([1, 0])) +# d3 = torch.index_select(d, 3, torch.tensor([1, 0])) + +# return x0, y0, y1, z0, z1, z2, d0, d1, d2, d3 + + +# 成功 +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + # 注册所有需要的索引缓冲区 + self.register_buffer("idx_x", torch.tensor([7.0, 9.0, 11.0])) + self.register_buffer("idx_y0", torch.tensor([1.0, 0.0])) + self.register_buffer("idx_y1", torch.tensor([1.0, 3.0, 2.0])) + self.register_buffer("idx_z0", torch.tensor([1.0, 0.0])) + self.register_buffer("idx_z1", torch.tensor([2.0, 0.0])) + self.register_buffer("idx_z2", torch.tensor([1.0, 2.0, 0.0])) + self.register_buffer("idx_d0", torch.tensor([1.0, 0.0, 3.0])) + self.register_buffer("idx_d1", torch.tensor([0.0, 1.0])) + self.register_buffer("idx_d2", torch.tensor([4.0, 3.0, 0.0])) + self.register_buffer("idx_d3", torch.tensor([3.0, 6.0, 2.0])) + + def float_to_int(self, idx_float): + mask = torch.ones_like(idx_float) + return (torch.max(idx_float * mask, mask * 0)).int() + + def forward(self, x, y, z, d): + # 使用辅助函数进行转换 + x0 = torch.index_select(x, 0, self.float_to_int(self.idx_x)) + y0 = torch.index_select(y, 0, self.float_to_int(self.idx_y0)) + y1 = torch.index_select(y, 1, self.float_to_int(self.idx_y1)) + z0 = torch.index_select(z, -3, self.float_to_int(self.idx_z0)) + z1 = torch.index_select(z, -2, self.float_to_int(self.idx_z1)) + z2 = torch.index_select(z, -1, self.float_to_int(self.idx_z2)) + d0 = torch.index_select(d, 0, self.float_to_int(self.idx_d0)) + d1 = torch.index_select(d, 1, self.float_to_int(self.idx_d1)) + d2 = torch.index_select(d, 2, self.float_to_int(self.idx_d2)) + d3 = torch.index_select(d, 3, self.float_to_int(self.idx_d3)) + + return x0, y0, y1, z0, z1, z2, d0, d1, d2, d3 + + +def test(): + net = Model() + net.eval() + + torch.manual_seed(0) + x = torch.rand(36) # 1D + y = torch.rand(5, 7) # 2D + z = torch.rand(3, 5, 8) # 3D + d = torch.rand(4, 3, 6, 7) # 4D + + a = net(x, y, z, d) + + # export torchscript + mod = torch.jit.trace(net, (x, y, z, d)) + mod.save("test_torch_index_select.pt") + + # torchscript to pnnx + import os + + os.system( + "../../src/pnnx test_torch_index_select.pt inputshape=[36],[5,7],[3,5,8],[4,3,6,7]" + ) + + # pnnx inference + import test_torch_index_select_ncnn + + b = test_torch_index_select_ncnn.test_inference() + + for a0, b0 in zip(a, b): + if not torch.allclose(a0, b0, 1e-3, 1e-3): + return False + return True + + +if __name__ == "__main__": + if test(): + exit(0) + else: + exit(1) From 8266f4b7f7e43195fe5747a330196c61d6779b3d Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E4=BD=B0=E9=98=85?= <43716063+Baiyuetribe@users.noreply.github.com> Date: Thu, 16 Jan 2025 21:27:44 +0800 Subject: [PATCH 3/3] clean code --- .github/workflows/linux-x64-cpu-gcc.yml | 218 ++++++++++++------------ 1 file changed, 109 insertions(+), 109 deletions(-) diff --git a/.github/workflows/linux-x64-cpu-gcc.yml b/.github/workflows/linux-x64-cpu-gcc.yml index 31abbe47c251..ab2185be3e74 100644 --- a/.github/workflows/linux-x64-cpu-gcc.yml +++ b/.github/workflows/linux-x64-cpu-gcc.yml @@ -1,33 +1,33 @@ name: linux-x64-cpu-gcc on: push: - # branches: [master] + branches: [master] paths: - - ".github/workflows/linux-x64-cpu-gcc.yml" - - "toolchains/host-c.gcc.toolchain.cmake" - - "CMakeLists.txt" - - "cmake/**" - - "src/*" - - "src/layer/*" - - "src/layer/x86/**" - - "tests/**" - - "tools/**" - - "!tools/pnnx/**" - - "examples/**" + - '.github/workflows/linux-x64-cpu-gcc.yml' + - 'toolchains/host-c.gcc.toolchain.cmake' + - 'CMakeLists.txt' + - 'cmake/**' + - 'src/*' + - 'src/layer/*' + - 'src/layer/x86/**' + - 'tests/**' + - 'tools/**' + - '!tools/pnnx/**' + - 'examples/**' pull_request: branches: [master] paths: - - ".github/workflows/linux-x64-cpu-gcc.yml" - - "toolchains/host-c.gcc.toolchain.cmake" - - "CMakeLists.txt" - - "cmake/**" - - "src/*" - - "src/layer/*" - - "src/layer/x86/**" - - "tests/**" - - "tools/**" - - "!tools/pnnx/**" - - "examples/**" + - '.github/workflows/linux-x64-cpu-gcc.yml' + - 'toolchains/host-c.gcc.toolchain.cmake' + - 'CMakeLists.txt' + - 'cmake/**' + - 'src/*' + - 'src/layer/*' + - 'src/layer/x86/**' + - 'tests/**' + - 'tools/**' + - '!tools/pnnx/**' + - 'examples/**' concurrency: group: linux-x64-cpu-gcc-${{ github.ref }} cancel-in-progress: true @@ -38,97 +38,97 @@ jobs: linux-gcc: runs-on: ubuntu-20.04 steps: - - uses: actions/checkout@v4 - - name: update - run: sudo apt-get update - - name: protobuf - run: sudo apt-get install libprotobuf-dev protobuf-compiler libopencv-dev - - name: build-sse2 - run: | - mkdir build-sse2 && cd build-sse2 - cmake -DNCNN_AVX=OFF -DNCNN_AVX2=OFF -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-sse2 - run: cd build-sse2 && ctest --output-on-failure -j $(nproc) - - name: build-shared - run: | - mkdir build-shared && cd build-shared - cmake -DNCNN_AVX2=ON -DNCNN_SHARED_LIB=ON .. - cmake --build . -j $(nproc) - - name: build-avx2 - run: | - mkdir build-avx2 && cd build-avx2 - cmake -DNCNN_AVX2=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx2 - run: cd build-avx2 && ctest --output-on-failure -j $(nproc) - - name: build-avx - run: | - mkdir build-avx && cd build-avx - cmake -DNCNN_AVX2=OFF -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx - run: cd build-avx && ctest --output-on-failure -j $(nproc) - - name: build-avx1-2 - run: | - mkdir build-avx1-2 && cd build-avx1-2 - cmake -DNCNN_AVX2=ON -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-avx1-2 - run: cd build-avx1-2 && ctest --output-on-failure -j $(nproc) - - name: build-noint8 - run: | - mkdir build-noint8 && cd build-noint8 - cmake -DNCNN_INT8=OFF -DNCNN_BUILD_TESTS=ON .. - cmake --build . -j $(nproc) - - name: test-noint8 - run: cd build-noint8 && ctest --output-on-failure -j $(nproc) + - uses: actions/checkout@v4 + - name: update + run: sudo apt-get update + - name: protobuf + run: sudo apt-get install libprotobuf-dev protobuf-compiler libopencv-dev + - name: build-sse2 + run: | + mkdir build-sse2 && cd build-sse2 + cmake -DNCNN_AVX=OFF -DNCNN_AVX2=OFF -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-sse2 + run: cd build-sse2 && ctest --output-on-failure -j $(nproc) + - name: build-shared + run: | + mkdir build-shared && cd build-shared + cmake -DNCNN_AVX2=ON -DNCNN_SHARED_LIB=ON .. + cmake --build . -j $(nproc) + - name: build-avx2 + run: | + mkdir build-avx2 && cd build-avx2 + cmake -DNCNN_AVX2=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx2 + run: cd build-avx2 && ctest --output-on-failure -j $(nproc) + - name: build-avx + run: | + mkdir build-avx && cd build-avx + cmake -DNCNN_AVX2=OFF -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx + run: cd build-avx && ctest --output-on-failure -j $(nproc) + - name: build-avx1-2 + run: | + mkdir build-avx1-2 && cd build-avx1-2 + cmake -DNCNN_AVX2=ON -DNCNN_AVX=ON -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-avx1-2 + run: cd build-avx1-2 && ctest --output-on-failure -j $(nproc) + - name: build-noint8 + run: | + mkdir build-noint8 && cd build-noint8 + cmake -DNCNN_INT8=OFF -DNCNN_BUILD_TESTS=ON .. + cmake --build . -j $(nproc) + - name: test-noint8 + run: cd build-noint8 && ctest --output-on-failure -j $(nproc) linux-gcc-cpp03-nostdio-nostring-simplestl: runs-on: ubuntu-20.04 steps: - - uses: actions/checkout@v4 - - name: build-nostdio - run: | - mkdir build-nostdio && cd build-nostdio - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host.gcc-c++03.toolchain.cmake -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-nostdio - run: cd build-nostdio && ctest --output-on-failure -j $(nproc) - - name: build-nostdio-nostring - run: | - mkdir build-nostdio-nostring && cd build-nostdio-nostring - cmake -DNCNN_STDIO=OFF -DNCNN_STRING=OFF -DNCNN_BUILD_TESTS=OFF -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: build-simplestl - run: | - mkdir build-simplestl && cd build-simplestl - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-simplestl - run: cd build-simplestl && ctest --output-on-failure -j $(nproc) - - name: build-simplestl-simpleomp - run: | - mkdir build-simplestl-simpleomp && cd build-simplestl-simpleomp - cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_SIMPLEOMP=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j $(nproc) - - name: test-simplestl-simpleomp - run: cd build-simplestl-simpleomp && ctest --output-on-failure -j $(nproc) + - uses: actions/checkout@v4 + - name: build-nostdio + run: | + mkdir build-nostdio && cd build-nostdio + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host.gcc-c++03.toolchain.cmake -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-nostdio + run: cd build-nostdio && ctest --output-on-failure -j $(nproc) + - name: build-nostdio-nostring + run: | + mkdir build-nostdio-nostring && cd build-nostdio-nostring + cmake -DNCNN_STDIO=OFF -DNCNN_STRING=OFF -DNCNN_BUILD_TESTS=OFF -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: build-simplestl + run: | + mkdir build-simplestl && cd build-simplestl + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-simplestl + run: cd build-simplestl && ctest --output-on-failure -j $(nproc) + - name: build-simplestl-simpleomp + run: | + mkdir build-simplestl-simpleomp && cd build-simplestl-simpleomp + cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/host-c.gcc.toolchain.cmake -DNCNN_STDIO=ON -DNCNN_STRING=ON -DNCNN_SIMPLESTL=ON -DNCNN_SIMPLEOMP=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_BENCHMARK=OFF -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j $(nproc) + - name: test-simplestl-simpleomp + run: cd build-simplestl-simpleomp && ctest --output-on-failure -j $(nproc) linux-gcc-avx512: runs-on: [self-hosted, linux, t4] steps: - - uses: actions/checkout@v4 - - name: build - env: - CC: gcc - CXX: g++ - LD_LIBRARY_PATH: /data/action/install/lib64 - run: | - mkdir build && cd build - cmake -DNCNN_AVX2=ON -DNCNN_AVX512=ON -DNCNN_AVX512VNNI=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. - cmake --build . -j 4 - - name: test - env: - LD_LIBRARY_PATH: /data/action/install/lib64 - run: cd build && ctest --output-on-failure -j 4 + - uses: actions/checkout@v4 + - name: build + env: + CC: gcc + CXX: g++ + LD_LIBRARY_PATH: /data/action/install/lib64 + run: | + mkdir build && cd build + cmake -DNCNN_AVX2=ON -DNCNN_AVX512=ON -DNCNN_AVX512VNNI=ON -DNCNN_BUILD_TESTS=ON -DNCNN_BUILD_TOOLS=OFF -DNCNN_BUILD_EXAMPLES=OFF .. + cmake --build . -j 4 + - name: test + env: + LD_LIBRARY_PATH: /data/action/install/lib64 + run: cd build && ctest --output-on-failure -j 4