Commit
·
cae2c48
0
Parent(s):
Add deformable_detr
Browse files- .gitattributes +1 -0
- README.md +6 -0
- build.toml +19 -0
- deformable_detr/ms_deform_attn_cuda.cu +158 -0
- deformable_detr/ms_deform_attn_cuda.cuh +1467 -0
- deformable_detr/ms_deform_attn_cuda.h +46 -0
- deformable_detr/ms_deform_im2col_cuda.cuh +1327 -0
- flake.nix +14 -0
- torch-ext/deformable_detr/__init__.py +45 -0
- torch-ext/deformable_detr/layers.py +81 -0
- torch-ext/ms_deform_attn_cpu.cpp +40 -0
- torch-ext/ms_deform_attn_cpu.h +32 -0
- torch-ext/torch_binding.cpp +19 -0
- torch-ext/torch_binding.h +16 -0
.gitattributes
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
*.so filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- kernel
|
| 5 |
+
---
|
| 6 |
+
|
build.toml
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[general]
|
| 2 |
+
name = "deformable_detr"
|
| 3 |
+
|
| 4 |
+
[torch]
|
| 5 |
+
src = [
|
| 6 |
+
"torch-ext/torch_binding.cpp",
|
| 7 |
+
"torch-ext/torch_binding.h"
|
| 8 |
+
]
|
| 9 |
+
|
| 10 |
+
[kernel.activation]
|
| 11 |
+
cuda-capabilities = [ "7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0" ]
|
| 12 |
+
src = [
|
| 13 |
+
"deformable_detr/ms_deform_attn_cuda.cu",
|
| 14 |
+
"deformable_detr/ms_deform_im2col_cuda.cuh",
|
| 15 |
+
"deformable_detr/ms_deform_attn_cuda.cuh",
|
| 16 |
+
"deformable_detr/ms_deform_attn_cuda.h",
|
| 17 |
+
]
|
| 18 |
+
include = ["."]
|
| 19 |
+
depends = [ "torch" ]
|
deformable_detr/ms_deform_attn_cuda.cu
ADDED
|
@@ -0,0 +1,158 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************************************
|
| 7 |
+
* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
|
| 8 |
+
**************************************************************************************************
|
| 9 |
+
*/
|
| 10 |
+
|
| 11 |
+
#include <vector>
|
| 12 |
+
#include "deformable_detr/ms_deform_im2col_cuda.cuh"
|
| 13 |
+
|
| 14 |
+
#include <ATen/ATen.h>
|
| 15 |
+
#include <ATen/cuda/CUDAContext.h>
|
| 16 |
+
#include <cuda.h>
|
| 17 |
+
#include <cuda_runtime.h>
|
| 18 |
+
|
| 19 |
+
#include <torch/all.h>
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
at::Tensor ms_deform_attn_cuda_forward(
|
| 23 |
+
const at::Tensor &value,
|
| 24 |
+
const at::Tensor &spatial_shapes,
|
| 25 |
+
const at::Tensor &level_start_index,
|
| 26 |
+
const at::Tensor &sampling_loc,
|
| 27 |
+
const at::Tensor &attn_weight,
|
| 28 |
+
const int64_t im2col_step)
|
| 29 |
+
{
|
| 30 |
+
at::DeviceGuard guard(value.device());
|
| 31 |
+
|
| 32 |
+
AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
|
| 33 |
+
AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
|
| 34 |
+
AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
|
| 35 |
+
AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
|
| 36 |
+
AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
|
| 37 |
+
|
| 38 |
+
AT_ASSERTM(value.is_cuda(), "value must be a CUDA tensor");
|
| 39 |
+
AT_ASSERTM(spatial_shapes.is_cuda(), "spatial_shapes must be a CUDA tensor");
|
| 40 |
+
AT_ASSERTM(level_start_index.is_cuda(), "level_start_index must be a CUDA tensor");
|
| 41 |
+
AT_ASSERTM(sampling_loc.is_cuda(), "sampling_loc must be a CUDA tensor");
|
| 42 |
+
AT_ASSERTM(attn_weight.is_cuda(), "attn_weight must be a CUDA tensor");
|
| 43 |
+
|
| 44 |
+
const int batch = value.size(0);
|
| 45 |
+
const int spatial_size = value.size(1);
|
| 46 |
+
const int num_heads = value.size(2);
|
| 47 |
+
const int channels = value.size(3);
|
| 48 |
+
|
| 49 |
+
const int num_levels = spatial_shapes.size(0);
|
| 50 |
+
|
| 51 |
+
const int num_query = sampling_loc.size(1);
|
| 52 |
+
const int num_point = sampling_loc.size(4);
|
| 53 |
+
|
| 54 |
+
const int im2col_step_ = std::min(batch, static_cast<int>(im2col_step));
|
| 55 |
+
|
| 56 |
+
AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
|
| 57 |
+
|
| 58 |
+
auto output = at::zeros({batch, num_query, num_heads, channels}, value.options());
|
| 59 |
+
|
| 60 |
+
const int batch_n = im2col_step_;
|
| 61 |
+
auto output_n = output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
|
| 62 |
+
auto per_value_size = spatial_size * num_heads * channels;
|
| 63 |
+
auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
|
| 64 |
+
auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
|
| 65 |
+
for (int n = 0; n < batch/im2col_step_; ++n)
|
| 66 |
+
{
|
| 67 |
+
auto columns = output_n.select(0, n);
|
| 68 |
+
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16, value.scalar_type(), "ms_deform_attn_forward_cuda", ([&] {
|
| 69 |
+
ms_deformable_im2col_cuda(at::cuda::getCurrentCUDAStream(),
|
| 70 |
+
value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 71 |
+
spatial_shapes.data_ptr<int64_t>(),
|
| 72 |
+
level_start_index.data_ptr<int64_t>(),
|
| 73 |
+
sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 74 |
+
attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
|
| 75 |
+
batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
|
| 76 |
+
columns.data_ptr<scalar_t>());
|
| 77 |
+
|
| 78 |
+
}));
|
| 79 |
+
}
|
| 80 |
+
|
| 81 |
+
output = output.view({batch, num_query, num_heads*channels});
|
| 82 |
+
|
| 83 |
+
return output;
|
| 84 |
+
}
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
std::vector<at::Tensor> ms_deform_attn_cuda_backward(
|
| 88 |
+
const at::Tensor &value,
|
| 89 |
+
const at::Tensor &spatial_shapes,
|
| 90 |
+
const at::Tensor &level_start_index,
|
| 91 |
+
const at::Tensor &sampling_loc,
|
| 92 |
+
const at::Tensor &attn_weight,
|
| 93 |
+
const at::Tensor &grad_output,
|
| 94 |
+
const int64_t im2col_step)
|
| 95 |
+
{
|
| 96 |
+
at::DeviceGuard guard(value.device());
|
| 97 |
+
|
| 98 |
+
AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
|
| 99 |
+
AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
|
| 100 |
+
AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
|
| 101 |
+
AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
|
| 102 |
+
AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
|
| 103 |
+
AT_ASSERTM(grad_output.is_contiguous(), "grad_output tensor has to be contiguous");
|
| 104 |
+
|
| 105 |
+
AT_ASSERTM(value.is_cuda(), "value must be a CUDA tensor");
|
| 106 |
+
AT_ASSERTM(spatial_shapes.is_cuda(), "spatial_shapes must be a CUDA tensor");
|
| 107 |
+
AT_ASSERTM(level_start_index.is_cuda(), "level_start_index must be a CUDA tensor");
|
| 108 |
+
AT_ASSERTM(sampling_loc.is_cuda(), "sampling_loc must be a CUDA tensor");
|
| 109 |
+
AT_ASSERTM(attn_weight.is_cuda(), "attn_weight must be a CUDA tensor");
|
| 110 |
+
AT_ASSERTM(grad_output.is_cuda(), "grad_output must be a CUDA tensor");
|
| 111 |
+
|
| 112 |
+
const int batch = value.size(0);
|
| 113 |
+
const int spatial_size = value.size(1);
|
| 114 |
+
const int num_heads = value.size(2);
|
| 115 |
+
const int channels = value.size(3);
|
| 116 |
+
|
| 117 |
+
const int num_levels = spatial_shapes.size(0);
|
| 118 |
+
|
| 119 |
+
const int num_query = sampling_loc.size(1);
|
| 120 |
+
const int num_point = sampling_loc.size(4);
|
| 121 |
+
|
| 122 |
+
const int im2col_step_ = std::min(batch, static_cast<int>(im2col_step));
|
| 123 |
+
|
| 124 |
+
AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
|
| 125 |
+
|
| 126 |
+
auto grad_value = at::zeros_like(value);
|
| 127 |
+
auto grad_sampling_loc = at::zeros_like(sampling_loc);
|
| 128 |
+
auto grad_attn_weight = at::zeros_like(attn_weight);
|
| 129 |
+
|
| 130 |
+
const int batch_n = im2col_step_;
|
| 131 |
+
auto per_value_size = spatial_size * num_heads * channels;
|
| 132 |
+
auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
|
| 133 |
+
auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
|
| 134 |
+
auto grad_output_n = grad_output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
|
| 135 |
+
|
| 136 |
+
for (int n = 0; n < batch/im2col_step_; ++n)
|
| 137 |
+
{
|
| 138 |
+
auto grad_output_g = grad_output_n.select(0, n);
|
| 139 |
+
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16, value.scalar_type(), "ms_deform_attn_backward_cuda", ([&] {
|
| 140 |
+
ms_deformable_col2im_cuda(at::cuda::getCurrentCUDAStream(),
|
| 141 |
+
grad_output_g.data_ptr<scalar_t>(),
|
| 142 |
+
value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 143 |
+
spatial_shapes.data_ptr<int64_t>(),
|
| 144 |
+
level_start_index.data_ptr<int64_t>(),
|
| 145 |
+
sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 146 |
+
attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
|
| 147 |
+
batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
|
| 148 |
+
grad_value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 149 |
+
grad_sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 150 |
+
grad_attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size);
|
| 151 |
+
|
| 152 |
+
}));
|
| 153 |
+
}
|
| 154 |
+
|
| 155 |
+
return {
|
| 156 |
+
grad_value, grad_sampling_loc, grad_attn_weight
|
| 157 |
+
};
|
| 158 |
+
}
|
deformable_detr/ms_deform_attn_cuda.cuh
ADDED
|
@@ -0,0 +1,1467 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************************************
|
| 7 |
+
* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
|
| 8 |
+
**************************************************************************************************
|
| 9 |
+
*/
|
| 10 |
+
|
| 11 |
+
#include <vector>
|
| 12 |
+
|
| 13 |
+
#include <cuda.h>
|
| 14 |
+
#include <cuda_runtime.h>
|
| 15 |
+
|
| 16 |
+
#include <cstdio>
|
| 17 |
+
#include <algorithm>
|
| 18 |
+
#include <cstring>
|
| 19 |
+
|
| 20 |
+
#include <ATen/ATen.h>
|
| 21 |
+
#include <ATen/cuda/CUDAContext.h>
|
| 22 |
+
|
| 23 |
+
#include <THC/THCAtomics.cuh>
|
| 24 |
+
|
| 25 |
+
#define CUDA_KERNEL_LOOP(i, n) \
|
| 26 |
+
for (int i = blockIdx.x * blockDim.x + threadIdx.x; \
|
| 27 |
+
i < (n); \
|
| 28 |
+
i += blockDim.x * gridDim.x)
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
at::Tensor ms_deform_attn_cuda_forward(
|
| 32 |
+
const at::Tensor &value,
|
| 33 |
+
const at::Tensor &spatial_shapes,
|
| 34 |
+
const at::Tensor &level_start_index,
|
| 35 |
+
const at::Tensor &sampling_loc,
|
| 36 |
+
const at::Tensor &attn_weight,
|
| 37 |
+
const int im2col_step)
|
| 38 |
+
{
|
| 39 |
+
AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
|
| 40 |
+
AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
|
| 41 |
+
AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
|
| 42 |
+
AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
|
| 43 |
+
AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
|
| 44 |
+
|
| 45 |
+
AT_ASSERTM(value.is_cuda(), "value must be a CUDA tensor");
|
| 46 |
+
AT_ASSERTM(spatial_shapes.is_cuda(), "spatial_shapes must be a CUDA tensor");
|
| 47 |
+
AT_ASSERTM(level_start_index.is_cuda(), "level_start_index must be a CUDA tensor");
|
| 48 |
+
AT_ASSERTM(sampling_loc.is_cuda(), "sampling_loc must be a CUDA tensor");
|
| 49 |
+
AT_ASSERTM(attn_weight.is_cuda(), "attn_weight must be a CUDA tensor");
|
| 50 |
+
|
| 51 |
+
const int batch = value.size(0);
|
| 52 |
+
const int spatial_size = value.size(1);
|
| 53 |
+
const int num_heads = value.size(2);
|
| 54 |
+
const int channels = value.size(3);
|
| 55 |
+
|
| 56 |
+
const int num_levels = spatial_shapes.size(0);
|
| 57 |
+
|
| 58 |
+
const int num_query = sampling_loc.size(1);
|
| 59 |
+
const int num_point = sampling_loc.size(4);
|
| 60 |
+
|
| 61 |
+
const int im2col_step_ = std::min(batch, im2col_step);
|
| 62 |
+
|
| 63 |
+
AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
|
| 64 |
+
|
| 65 |
+
auto output = at::zeros({batch, num_query, num_heads, channels}, value.options());
|
| 66 |
+
|
| 67 |
+
const int batch_n = im2col_step_;
|
| 68 |
+
auto output_n = output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
|
| 69 |
+
auto per_value_size = spatial_size * num_heads * channels;
|
| 70 |
+
auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
|
| 71 |
+
auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
|
| 72 |
+
for (int n = 0; n < batch/im2col_step_; ++n)
|
| 73 |
+
{
|
| 74 |
+
auto columns = output_n.select(0, n);
|
| 75 |
+
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16, value.scalar_type(), "ms_deform_attn_forward_cuda", ([&] {
|
| 76 |
+
ms_deformable_im2col_cuda(at::cuda::getCurrentCUDAStream(),
|
| 77 |
+
value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 78 |
+
spatial_shapes.data_ptr<int64_t>(),
|
| 79 |
+
level_start_index.data_ptr<int64_t>(),
|
| 80 |
+
sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 81 |
+
attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
|
| 82 |
+
batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
|
| 83 |
+
columns.data_ptr<scalar_t>());
|
| 84 |
+
|
| 85 |
+
}));
|
| 86 |
+
}
|
| 87 |
+
|
| 88 |
+
output = output.view({batch, num_query, num_heads*channels});
|
| 89 |
+
|
| 90 |
+
return output;
|
| 91 |
+
}
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
std::vector<at::Tensor> ms_deform_attn_cuda_backward(
|
| 95 |
+
const at::Tensor &value,
|
| 96 |
+
const at::Tensor &spatial_shapes,
|
| 97 |
+
const at::Tensor &level_start_index,
|
| 98 |
+
const at::Tensor &sampling_loc,
|
| 99 |
+
const at::Tensor &attn_weight,
|
| 100 |
+
const at::Tensor &grad_output,
|
| 101 |
+
const int im2col_step)
|
| 102 |
+
{
|
| 103 |
+
|
| 104 |
+
AT_ASSERTM(value.is_contiguous(), "value tensor has to be contiguous");
|
| 105 |
+
AT_ASSERTM(spatial_shapes.is_contiguous(), "spatial_shapes tensor has to be contiguous");
|
| 106 |
+
AT_ASSERTM(level_start_index.is_contiguous(), "level_start_index tensor has to be contiguous");
|
| 107 |
+
AT_ASSERTM(sampling_loc.is_contiguous(), "sampling_loc tensor has to be contiguous");
|
| 108 |
+
AT_ASSERTM(attn_weight.is_contiguous(), "attn_weight tensor has to be contiguous");
|
| 109 |
+
AT_ASSERTM(grad_output.is_contiguous(), "grad_output tensor has to be contiguous");
|
| 110 |
+
|
| 111 |
+
AT_ASSERTM(value.is_cuda(), "value must be a CUDA tensor");
|
| 112 |
+
AT_ASSERTM(spatial_shapes.is_cuda(), "spatial_shapes must be a CUDA tensor");
|
| 113 |
+
AT_ASSERTM(level_start_index.is_cuda(), "level_start_index must be a CUDA tensor");
|
| 114 |
+
AT_ASSERTM(sampling_loc.is_cuda(), "sampling_loc must be a CUDA tensor");
|
| 115 |
+
AT_ASSERTM(attn_weight.is_cuda(), "attn_weight must be a CUDA tensor");
|
| 116 |
+
AT_ASSERTM(grad_output.is_cuda(), "grad_output must be a CUDA tensor");
|
| 117 |
+
|
| 118 |
+
const int batch = value.size(0);
|
| 119 |
+
const int spatial_size = value.size(1);
|
| 120 |
+
const int num_heads = value.size(2);
|
| 121 |
+
const int channels = value.size(3);
|
| 122 |
+
|
| 123 |
+
const int num_levels = spatial_shapes.size(0);
|
| 124 |
+
|
| 125 |
+
const int num_query = sampling_loc.size(1);
|
| 126 |
+
const int num_point = sampling_loc.size(4);
|
| 127 |
+
|
| 128 |
+
const int im2col_step_ = std::min(batch, im2col_step);
|
| 129 |
+
|
| 130 |
+
AT_ASSERTM(batch % im2col_step_ == 0, "batch(%d) must divide im2col_step(%d)", batch, im2col_step_);
|
| 131 |
+
|
| 132 |
+
auto grad_value = at::zeros_like(value);
|
| 133 |
+
auto grad_sampling_loc = at::zeros_like(sampling_loc);
|
| 134 |
+
auto grad_attn_weight = at::zeros_like(attn_weight);
|
| 135 |
+
|
| 136 |
+
const int batch_n = im2col_step_;
|
| 137 |
+
auto per_value_size = spatial_size * num_heads * channels;
|
| 138 |
+
auto per_sample_loc_size = num_query * num_heads * num_levels * num_point * 2;
|
| 139 |
+
auto per_attn_weight_size = num_query * num_heads * num_levels * num_point;
|
| 140 |
+
auto grad_output_n = grad_output.view({batch/im2col_step_, batch_n, num_query, num_heads, channels});
|
| 141 |
+
|
| 142 |
+
for (int n = 0; n < batch/im2col_step_; ++n)
|
| 143 |
+
{
|
| 144 |
+
auto grad_output_g = grad_output_n.select(0, n);
|
| 145 |
+
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16, value.scalar_type(), "ms_deform_attn_backward_cuda", ([&] {
|
| 146 |
+
ms_deformable_col2im_cuda(at::cuda::getCurrentCUDAStream(),
|
| 147 |
+
grad_output_g.data_ptr<scalar_t>(),
|
| 148 |
+
value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 149 |
+
spatial_shapes.data_ptr<int64_t>(),
|
| 150 |
+
level_start_index.data_ptr<int64_t>(),
|
| 151 |
+
sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 152 |
+
attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size,
|
| 153 |
+
batch_n, spatial_size, num_heads, channels, num_levels, num_query, num_point,
|
| 154 |
+
grad_value.data_ptr<scalar_t>() + n * im2col_step_ * per_value_size,
|
| 155 |
+
grad_sampling_loc.data_ptr<scalar_t>() + n * im2col_step_ * per_sample_loc_size,
|
| 156 |
+
grad_attn_weight.data_ptr<scalar_t>() + n * im2col_step_ * per_attn_weight_size);
|
| 157 |
+
|
| 158 |
+
}));
|
| 159 |
+
}
|
| 160 |
+
|
| 161 |
+
return {
|
| 162 |
+
grad_value, grad_sampling_loc, grad_attn_weight
|
| 163 |
+
};
|
| 164 |
+
}
|
| 165 |
+
|
| 166 |
+
const int CUDA_NUM_THREADS = 1024;
|
| 167 |
+
inline int GET_BLOCKS(const int N, const int num_threads)
|
| 168 |
+
{
|
| 169 |
+
return (N + num_threads - 1) / num_threads;
|
| 170 |
+
}
|
| 171 |
+
|
| 172 |
+
|
| 173 |
+
template <typename scalar_t>
|
| 174 |
+
__device__ scalar_t ms_deform_attn_im2col_bilinear(const scalar_t* &bottom_data,
|
| 175 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 176 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c)
|
| 177 |
+
{
|
| 178 |
+
const int h_low = floor(h);
|
| 179 |
+
const int w_low = floor(w);
|
| 180 |
+
const int h_high = h_low + 1;
|
| 181 |
+
const int w_high = w_low + 1;
|
| 182 |
+
|
| 183 |
+
const scalar_t lh = h - h_low;
|
| 184 |
+
const scalar_t lw = w - w_low;
|
| 185 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 186 |
+
|
| 187 |
+
const int w_stride = nheads * channels;
|
| 188 |
+
const int h_stride = width * w_stride;
|
| 189 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 190 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 191 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 192 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 193 |
+
const int base_ptr = m * channels + c;
|
| 194 |
+
|
| 195 |
+
scalar_t v1 = 0;
|
| 196 |
+
if (h_low >= 0 && w_low >= 0)
|
| 197 |
+
{
|
| 198 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 199 |
+
v1 = bottom_data[ptr1];
|
| 200 |
+
}
|
| 201 |
+
scalar_t v2 = 0;
|
| 202 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 203 |
+
{
|
| 204 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 205 |
+
v2 = bottom_data[ptr2];
|
| 206 |
+
}
|
| 207 |
+
scalar_t v3 = 0;
|
| 208 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 209 |
+
{
|
| 210 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 211 |
+
v3 = bottom_data[ptr3];
|
| 212 |
+
}
|
| 213 |
+
scalar_t v4 = 0;
|
| 214 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 215 |
+
{
|
| 216 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 217 |
+
v4 = bottom_data[ptr4];
|
| 218 |
+
}
|
| 219 |
+
|
| 220 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 221 |
+
|
| 222 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 223 |
+
return val;
|
| 224 |
+
}
|
| 225 |
+
|
| 226 |
+
|
| 227 |
+
template <typename scalar_t>
|
| 228 |
+
__device__ void ms_deform_attn_col2im_bilinear(const scalar_t* &bottom_data,
|
| 229 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 230 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c,
|
| 231 |
+
const scalar_t &top_grad,
|
| 232 |
+
const scalar_t &attn_weight,
|
| 233 |
+
scalar_t* &grad_value,
|
| 234 |
+
scalar_t* grad_sampling_loc,
|
| 235 |
+
scalar_t* grad_attn_weight)
|
| 236 |
+
{
|
| 237 |
+
const int h_low = floor(h);
|
| 238 |
+
const int w_low = floor(w);
|
| 239 |
+
const int h_high = h_low + 1;
|
| 240 |
+
const int w_high = w_low + 1;
|
| 241 |
+
|
| 242 |
+
const scalar_t lh = h - h_low;
|
| 243 |
+
const scalar_t lw = w - w_low;
|
| 244 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 245 |
+
|
| 246 |
+
const int w_stride = nheads * channels;
|
| 247 |
+
const int h_stride = width * w_stride;
|
| 248 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 249 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 250 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 251 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 252 |
+
const int base_ptr = m * channels + c;
|
| 253 |
+
|
| 254 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 255 |
+
const scalar_t top_grad_value = top_grad * attn_weight;
|
| 256 |
+
scalar_t grad_h_weight = 0, grad_w_weight = 0;
|
| 257 |
+
|
| 258 |
+
scalar_t v1 = 0;
|
| 259 |
+
if (h_low >= 0 && w_low >= 0)
|
| 260 |
+
{
|
| 261 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 262 |
+
v1 = bottom_data[ptr1];
|
| 263 |
+
grad_h_weight -= hw * v1;
|
| 264 |
+
grad_w_weight -= hh * v1;
|
| 265 |
+
atomicAdd(grad_value+ptr1, w1*top_grad_value);
|
| 266 |
+
}
|
| 267 |
+
scalar_t v2 = 0;
|
| 268 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 269 |
+
{
|
| 270 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 271 |
+
v2 = bottom_data[ptr2];
|
| 272 |
+
grad_h_weight -= lw * v2;
|
| 273 |
+
grad_w_weight += hh * v2;
|
| 274 |
+
atomicAdd(grad_value+ptr2, w2*top_grad_value);
|
| 275 |
+
}
|
| 276 |
+
scalar_t v3 = 0;
|
| 277 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 278 |
+
{
|
| 279 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 280 |
+
v3 = bottom_data[ptr3];
|
| 281 |
+
grad_h_weight += hw * v3;
|
| 282 |
+
grad_w_weight -= lh * v3;
|
| 283 |
+
atomicAdd(grad_value+ptr3, w3*top_grad_value);
|
| 284 |
+
}
|
| 285 |
+
scalar_t v4 = 0;
|
| 286 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 287 |
+
{
|
| 288 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 289 |
+
v4 = bottom_data[ptr4];
|
| 290 |
+
grad_h_weight += lw * v4;
|
| 291 |
+
grad_w_weight += lh * v4;
|
| 292 |
+
atomicAdd(grad_value+ptr4, w4*top_grad_value);
|
| 293 |
+
}
|
| 294 |
+
|
| 295 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 296 |
+
*grad_attn_weight = top_grad * val;
|
| 297 |
+
*grad_sampling_loc = width * grad_w_weight * top_grad_value;
|
| 298 |
+
*(grad_sampling_loc + 1) = height * grad_h_weight * top_grad_value;
|
| 299 |
+
}
|
| 300 |
+
|
| 301 |
+
|
| 302 |
+
template <typename scalar_t>
|
| 303 |
+
__device__ void ms_deform_attn_col2im_bilinear_gm(const scalar_t* &bottom_data,
|
| 304 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 305 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c,
|
| 306 |
+
const scalar_t &top_grad,
|
| 307 |
+
const scalar_t &attn_weight,
|
| 308 |
+
scalar_t* &grad_value,
|
| 309 |
+
scalar_t* grad_sampling_loc,
|
| 310 |
+
scalar_t* grad_attn_weight)
|
| 311 |
+
{
|
| 312 |
+
const int h_low = floor(h);
|
| 313 |
+
const int w_low = floor(w);
|
| 314 |
+
const int h_high = h_low + 1;
|
| 315 |
+
const int w_high = w_low + 1;
|
| 316 |
+
|
| 317 |
+
const scalar_t lh = h - h_low;
|
| 318 |
+
const scalar_t lw = w - w_low;
|
| 319 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 320 |
+
|
| 321 |
+
const int w_stride = nheads * channels;
|
| 322 |
+
const int h_stride = width * w_stride;
|
| 323 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 324 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 325 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 326 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 327 |
+
const int base_ptr = m * channels + c;
|
| 328 |
+
|
| 329 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 330 |
+
const scalar_t top_grad_value = top_grad * attn_weight;
|
| 331 |
+
scalar_t grad_h_weight = 0, grad_w_weight = 0;
|
| 332 |
+
|
| 333 |
+
scalar_t v1 = 0;
|
| 334 |
+
if (h_low >= 0 && w_low >= 0)
|
| 335 |
+
{
|
| 336 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 337 |
+
v1 = bottom_data[ptr1];
|
| 338 |
+
grad_h_weight -= hw * v1;
|
| 339 |
+
grad_w_weight -= hh * v1;
|
| 340 |
+
atomicAdd(grad_value+ptr1, w1*top_grad_value);
|
| 341 |
+
}
|
| 342 |
+
scalar_t v2 = 0;
|
| 343 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 344 |
+
{
|
| 345 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 346 |
+
v2 = bottom_data[ptr2];
|
| 347 |
+
grad_h_weight -= lw * v2;
|
| 348 |
+
grad_w_weight += hh * v2;
|
| 349 |
+
atomicAdd(grad_value+ptr2, w2*top_grad_value);
|
| 350 |
+
}
|
| 351 |
+
scalar_t v3 = 0;
|
| 352 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 353 |
+
{
|
| 354 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 355 |
+
v3 = bottom_data[ptr3];
|
| 356 |
+
grad_h_weight += hw * v3;
|
| 357 |
+
grad_w_weight -= lh * v3;
|
| 358 |
+
atomicAdd(grad_value+ptr3, w3*top_grad_value);
|
| 359 |
+
}
|
| 360 |
+
scalar_t v4 = 0;
|
| 361 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 362 |
+
{
|
| 363 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 364 |
+
v4 = bottom_data[ptr4];
|
| 365 |
+
grad_h_weight += lw * v4;
|
| 366 |
+
grad_w_weight += lh * v4;
|
| 367 |
+
atomicAdd(grad_value+ptr4, w4*top_grad_value);
|
| 368 |
+
}
|
| 369 |
+
|
| 370 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 371 |
+
atomicAdd(grad_attn_weight, top_grad * val);
|
| 372 |
+
atomicAdd(grad_sampling_loc, width * grad_w_weight * top_grad_value);
|
| 373 |
+
atomicAdd(grad_sampling_loc + 1, height * grad_h_weight * top_grad_value);
|
| 374 |
+
}
|
| 375 |
+
|
| 376 |
+
|
| 377 |
+
template <typename scalar_t>
|
| 378 |
+
__global__ void ms_deformable_im2col_gpu_kernel(const int n,
|
| 379 |
+
const scalar_t *data_value,
|
| 380 |
+
const int64_t *data_spatial_shapes,
|
| 381 |
+
const int64_t *data_level_start_index,
|
| 382 |
+
const scalar_t *data_sampling_loc,
|
| 383 |
+
const scalar_t *data_attn_weight,
|
| 384 |
+
const int batch_size,
|
| 385 |
+
const int spatial_size,
|
| 386 |
+
const int num_heads,
|
| 387 |
+
const int channels,
|
| 388 |
+
const int num_levels,
|
| 389 |
+
const int num_query,
|
| 390 |
+
const int num_point,
|
| 391 |
+
scalar_t *data_col)
|
| 392 |
+
{
|
| 393 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 394 |
+
{
|
| 395 |
+
int _temp = index;
|
| 396 |
+
const int c_col = _temp % channels;
|
| 397 |
+
_temp /= channels;
|
| 398 |
+
const int sampling_index = _temp;
|
| 399 |
+
const int m_col = _temp % num_heads;
|
| 400 |
+
_temp /= num_heads;
|
| 401 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 402 |
+
_temp /= num_query;
|
| 403 |
+
const int b_col = _temp;
|
| 404 |
+
|
| 405 |
+
scalar_t *data_col_ptr = data_col + index;
|
| 406 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 407 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 408 |
+
const int qid_stride = num_heads * channels;
|
| 409 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 410 |
+
scalar_t col = 0;
|
| 411 |
+
|
| 412 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 413 |
+
{
|
| 414 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 415 |
+
const int spatial_h_ptr = l_col << 1;
|
| 416 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 417 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 418 |
+
const scalar_t *data_value_ptr = data_value + (data_value_ptr_init_offset + level_start_id * qid_stride);
|
| 419 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 420 |
+
{
|
| 421 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 422 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 423 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 424 |
+
|
| 425 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 426 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 427 |
+
|
| 428 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 429 |
+
{
|
| 430 |
+
col += ms_deform_attn_im2col_bilinear(data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col) * weight;
|
| 431 |
+
}
|
| 432 |
+
|
| 433 |
+
data_weight_ptr += 1;
|
| 434 |
+
data_loc_w_ptr += 2;
|
| 435 |
+
}
|
| 436 |
+
}
|
| 437 |
+
*data_col_ptr = col;
|
| 438 |
+
}
|
| 439 |
+
}
|
| 440 |
+
|
| 441 |
+
template <typename scalar_t, unsigned int blockSize>
|
| 442 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1(const int n,
|
| 443 |
+
const scalar_t *grad_col,
|
| 444 |
+
const scalar_t *data_value,
|
| 445 |
+
const int64_t *data_spatial_shapes,
|
| 446 |
+
const int64_t *data_level_start_index,
|
| 447 |
+
const scalar_t *data_sampling_loc,
|
| 448 |
+
const scalar_t *data_attn_weight,
|
| 449 |
+
const int batch_size,
|
| 450 |
+
const int spatial_size,
|
| 451 |
+
const int num_heads,
|
| 452 |
+
const int channels,
|
| 453 |
+
const int num_levels,
|
| 454 |
+
const int num_query,
|
| 455 |
+
const int num_point,
|
| 456 |
+
scalar_t *grad_value,
|
| 457 |
+
scalar_t *grad_sampling_loc,
|
| 458 |
+
scalar_t *grad_attn_weight)
|
| 459 |
+
{
|
| 460 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 461 |
+
{
|
| 462 |
+
__shared__ scalar_t cache_grad_sampling_loc[blockSize * 2];
|
| 463 |
+
__shared__ scalar_t cache_grad_attn_weight[blockSize];
|
| 464 |
+
unsigned int tid = threadIdx.x;
|
| 465 |
+
int _temp = index;
|
| 466 |
+
const int c_col = _temp % channels;
|
| 467 |
+
_temp /= channels;
|
| 468 |
+
const int sampling_index = _temp;
|
| 469 |
+
const int m_col = _temp % num_heads;
|
| 470 |
+
_temp /= num_heads;
|
| 471 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 472 |
+
_temp /= num_query;
|
| 473 |
+
const int b_col = _temp;
|
| 474 |
+
|
| 475 |
+
const scalar_t top_grad = grad_col[index];
|
| 476 |
+
|
| 477 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 478 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 479 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 480 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 481 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 482 |
+
const int grad_weight_stride = 1;
|
| 483 |
+
const int grad_loc_stride = 2;
|
| 484 |
+
const int qid_stride = num_heads * channels;
|
| 485 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 486 |
+
|
| 487 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 488 |
+
{
|
| 489 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 490 |
+
const int spatial_h_ptr = l_col << 1;
|
| 491 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 492 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 493 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 494 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 495 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 496 |
+
|
| 497 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 498 |
+
{
|
| 499 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 500 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 501 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 502 |
+
|
| 503 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 504 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 505 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 506 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 507 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 508 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 509 |
+
{
|
| 510 |
+
ms_deform_attn_col2im_bilinear(
|
| 511 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 512 |
+
top_grad, weight, grad_value_ptr,
|
| 513 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 514 |
+
}
|
| 515 |
+
|
| 516 |
+
__syncthreads();
|
| 517 |
+
if (tid == 0)
|
| 518 |
+
{
|
| 519 |
+
scalar_t _grad_w=cache_grad_sampling_loc[0], _grad_h=cache_grad_sampling_loc[1], _grad_a=cache_grad_attn_weight[0];
|
| 520 |
+
int sid=2;
|
| 521 |
+
for (unsigned int tid = 1; tid < blockSize; ++tid)
|
| 522 |
+
{
|
| 523 |
+
_grad_w += cache_grad_sampling_loc[sid];
|
| 524 |
+
_grad_h += cache_grad_sampling_loc[sid + 1];
|
| 525 |
+
_grad_a += cache_grad_attn_weight[tid];
|
| 526 |
+
sid += 2;
|
| 527 |
+
}
|
| 528 |
+
|
| 529 |
+
|
| 530 |
+
*grad_sampling_loc = _grad_w;
|
| 531 |
+
*(grad_sampling_loc + 1) = _grad_h;
|
| 532 |
+
*grad_attn_weight = _grad_a;
|
| 533 |
+
}
|
| 534 |
+
__syncthreads();
|
| 535 |
+
|
| 536 |
+
data_weight_ptr += 1;
|
| 537 |
+
data_loc_w_ptr += 2;
|
| 538 |
+
grad_attn_weight += grad_weight_stride;
|
| 539 |
+
grad_sampling_loc += grad_loc_stride;
|
| 540 |
+
}
|
| 541 |
+
}
|
| 542 |
+
}
|
| 543 |
+
}
|
| 544 |
+
|
| 545 |
+
|
| 546 |
+
template <typename scalar_t, unsigned int blockSize>
|
| 547 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2(const int n,
|
| 548 |
+
const scalar_t *grad_col,
|
| 549 |
+
const scalar_t *data_value,
|
| 550 |
+
const int64_t *data_spatial_shapes,
|
| 551 |
+
const int64_t *data_level_start_index,
|
| 552 |
+
const scalar_t *data_sampling_loc,
|
| 553 |
+
const scalar_t *data_attn_weight,
|
| 554 |
+
const int batch_size,
|
| 555 |
+
const int spatial_size,
|
| 556 |
+
const int num_heads,
|
| 557 |
+
const int channels,
|
| 558 |
+
const int num_levels,
|
| 559 |
+
const int num_query,
|
| 560 |
+
const int num_point,
|
| 561 |
+
scalar_t *grad_value,
|
| 562 |
+
scalar_t *grad_sampling_loc,
|
| 563 |
+
scalar_t *grad_attn_weight)
|
| 564 |
+
{
|
| 565 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 566 |
+
{
|
| 567 |
+
__shared__ scalar_t cache_grad_sampling_loc[blockSize * 2];
|
| 568 |
+
__shared__ scalar_t cache_grad_attn_weight[blockSize];
|
| 569 |
+
unsigned int tid = threadIdx.x;
|
| 570 |
+
int _temp = index;
|
| 571 |
+
const int c_col = _temp % channels;
|
| 572 |
+
_temp /= channels;
|
| 573 |
+
const int sampling_index = _temp;
|
| 574 |
+
const int m_col = _temp % num_heads;
|
| 575 |
+
_temp /= num_heads;
|
| 576 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 577 |
+
_temp /= num_query;
|
| 578 |
+
const int b_col = _temp;
|
| 579 |
+
|
| 580 |
+
const scalar_t top_grad = grad_col[index];
|
| 581 |
+
|
| 582 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 583 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 584 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 585 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 586 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 587 |
+
const int grad_weight_stride = 1;
|
| 588 |
+
const int grad_loc_stride = 2;
|
| 589 |
+
const int qid_stride = num_heads * channels;
|
| 590 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 591 |
+
|
| 592 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 593 |
+
{
|
| 594 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 595 |
+
const int spatial_h_ptr = l_col << 1;
|
| 596 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 597 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 598 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 599 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 600 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 601 |
+
|
| 602 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 603 |
+
{
|
| 604 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 605 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 606 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 607 |
+
|
| 608 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 609 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 610 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 611 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 612 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 613 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 614 |
+
{
|
| 615 |
+
ms_deform_attn_col2im_bilinear(
|
| 616 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 617 |
+
top_grad, weight, grad_value_ptr,
|
| 618 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 619 |
+
}
|
| 620 |
+
|
| 621 |
+
__syncthreads();
|
| 622 |
+
|
| 623 |
+
for (unsigned int s=blockSize/2; s>0; s>>=1)
|
| 624 |
+
{
|
| 625 |
+
if (tid < s) {
|
| 626 |
+
const unsigned int xid1 = tid << 1;
|
| 627 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 628 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 629 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 630 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 631 |
+
}
|
| 632 |
+
__syncthreads();
|
| 633 |
+
}
|
| 634 |
+
|
| 635 |
+
if (tid == 0)
|
| 636 |
+
{
|
| 637 |
+
*grad_sampling_loc = cache_grad_sampling_loc[0];
|
| 638 |
+
*(grad_sampling_loc + 1) = cache_grad_sampling_loc[1];
|
| 639 |
+
*grad_attn_weight = cache_grad_attn_weight[0];
|
| 640 |
+
}
|
| 641 |
+
__syncthreads();
|
| 642 |
+
|
| 643 |
+
data_weight_ptr += 1;
|
| 644 |
+
data_loc_w_ptr += 2;
|
| 645 |
+
grad_attn_weight += grad_weight_stride;
|
| 646 |
+
grad_sampling_loc += grad_loc_stride;
|
| 647 |
+
}
|
| 648 |
+
}
|
| 649 |
+
}
|
| 650 |
+
}
|
| 651 |
+
|
| 652 |
+
|
| 653 |
+
template <typename scalar_t>
|
| 654 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v1(const int n,
|
| 655 |
+
const scalar_t *grad_col,
|
| 656 |
+
const scalar_t *data_value,
|
| 657 |
+
const int64_t *data_spatial_shapes,
|
| 658 |
+
const int64_t *data_level_start_index,
|
| 659 |
+
const scalar_t *data_sampling_loc,
|
| 660 |
+
const scalar_t *data_attn_weight,
|
| 661 |
+
const int batch_size,
|
| 662 |
+
const int spatial_size,
|
| 663 |
+
const int num_heads,
|
| 664 |
+
const int channels,
|
| 665 |
+
const int num_levels,
|
| 666 |
+
const int num_query,
|
| 667 |
+
const int num_point,
|
| 668 |
+
scalar_t *grad_value,
|
| 669 |
+
scalar_t *grad_sampling_loc,
|
| 670 |
+
scalar_t *grad_attn_weight)
|
| 671 |
+
{
|
| 672 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 673 |
+
{
|
| 674 |
+
extern __shared__ int _s[];
|
| 675 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 676 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 677 |
+
unsigned int tid = threadIdx.x;
|
| 678 |
+
int _temp = index;
|
| 679 |
+
const int c_col = _temp % channels;
|
| 680 |
+
_temp /= channels;
|
| 681 |
+
const int sampling_index = _temp;
|
| 682 |
+
const int m_col = _temp % num_heads;
|
| 683 |
+
_temp /= num_heads;
|
| 684 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 685 |
+
_temp /= num_query;
|
| 686 |
+
const int b_col = _temp;
|
| 687 |
+
|
| 688 |
+
const scalar_t top_grad = grad_col[index];
|
| 689 |
+
|
| 690 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 691 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 692 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 693 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 694 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 695 |
+
const int grad_weight_stride = 1;
|
| 696 |
+
const int grad_loc_stride = 2;
|
| 697 |
+
const int qid_stride = num_heads * channels;
|
| 698 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 699 |
+
|
| 700 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 701 |
+
{
|
| 702 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 703 |
+
const int spatial_h_ptr = l_col << 1;
|
| 704 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 705 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 706 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 707 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 708 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 709 |
+
|
| 710 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 711 |
+
{
|
| 712 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 713 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 714 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 715 |
+
|
| 716 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 717 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 718 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 719 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 720 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 721 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 722 |
+
{
|
| 723 |
+
ms_deform_attn_col2im_bilinear(
|
| 724 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 725 |
+
top_grad, weight, grad_value_ptr,
|
| 726 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 727 |
+
}
|
| 728 |
+
|
| 729 |
+
__syncthreads();
|
| 730 |
+
if (tid == 0)
|
| 731 |
+
{
|
| 732 |
+
scalar_t _grad_w=cache_grad_sampling_loc[0], _grad_h=cache_grad_sampling_loc[1], _grad_a=cache_grad_attn_weight[0];
|
| 733 |
+
int sid=2;
|
| 734 |
+
for (unsigned int tid = 1; tid < blockDim.x; ++tid)
|
| 735 |
+
{
|
| 736 |
+
_grad_w += cache_grad_sampling_loc[sid];
|
| 737 |
+
_grad_h += cache_grad_sampling_loc[sid + 1];
|
| 738 |
+
_grad_a += cache_grad_attn_weight[tid];
|
| 739 |
+
sid += 2;
|
| 740 |
+
}
|
| 741 |
+
|
| 742 |
+
|
| 743 |
+
*grad_sampling_loc = _grad_w;
|
| 744 |
+
*(grad_sampling_loc + 1) = _grad_h;
|
| 745 |
+
*grad_attn_weight = _grad_a;
|
| 746 |
+
}
|
| 747 |
+
__syncthreads();
|
| 748 |
+
|
| 749 |
+
data_weight_ptr += 1;
|
| 750 |
+
data_loc_w_ptr += 2;
|
| 751 |
+
grad_attn_weight += grad_weight_stride;
|
| 752 |
+
grad_sampling_loc += grad_loc_stride;
|
| 753 |
+
}
|
| 754 |
+
}
|
| 755 |
+
}
|
| 756 |
+
}
|
| 757 |
+
|
| 758 |
+
template <typename scalar_t>
|
| 759 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v2(const int n,
|
| 760 |
+
const scalar_t *grad_col,
|
| 761 |
+
const scalar_t *data_value,
|
| 762 |
+
const int64_t *data_spatial_shapes,
|
| 763 |
+
const int64_t *data_level_start_index,
|
| 764 |
+
const scalar_t *data_sampling_loc,
|
| 765 |
+
const scalar_t *data_attn_weight,
|
| 766 |
+
const int batch_size,
|
| 767 |
+
const int spatial_size,
|
| 768 |
+
const int num_heads,
|
| 769 |
+
const int channels,
|
| 770 |
+
const int num_levels,
|
| 771 |
+
const int num_query,
|
| 772 |
+
const int num_point,
|
| 773 |
+
scalar_t *grad_value,
|
| 774 |
+
scalar_t *grad_sampling_loc,
|
| 775 |
+
scalar_t *grad_attn_weight)
|
| 776 |
+
{
|
| 777 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 778 |
+
{
|
| 779 |
+
extern __shared__ int _s[];
|
| 780 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 781 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 782 |
+
unsigned int tid = threadIdx.x;
|
| 783 |
+
int _temp = index;
|
| 784 |
+
const int c_col = _temp % channels;
|
| 785 |
+
_temp /= channels;
|
| 786 |
+
const int sampling_index = _temp;
|
| 787 |
+
const int m_col = _temp % num_heads;
|
| 788 |
+
_temp /= num_heads;
|
| 789 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 790 |
+
_temp /= num_query;
|
| 791 |
+
const int b_col = _temp;
|
| 792 |
+
|
| 793 |
+
const scalar_t top_grad = grad_col[index];
|
| 794 |
+
|
| 795 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 796 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 797 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 798 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 799 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 800 |
+
const int grad_weight_stride = 1;
|
| 801 |
+
const int grad_loc_stride = 2;
|
| 802 |
+
const int qid_stride = num_heads * channels;
|
| 803 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 804 |
+
|
| 805 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 806 |
+
{
|
| 807 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 808 |
+
const int spatial_h_ptr = l_col << 1;
|
| 809 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 810 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 811 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 812 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 813 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 814 |
+
|
| 815 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 816 |
+
{
|
| 817 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 818 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 819 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 820 |
+
|
| 821 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 822 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 823 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 824 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 825 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 826 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 827 |
+
{
|
| 828 |
+
ms_deform_attn_col2im_bilinear(
|
| 829 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 830 |
+
top_grad, weight, grad_value_ptr,
|
| 831 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 832 |
+
}
|
| 833 |
+
|
| 834 |
+
__syncthreads();
|
| 835 |
+
|
| 836 |
+
for (unsigned int s=blockDim.x/2, spre=blockDim.x; s>0; s>>=1, spre>>=1)
|
| 837 |
+
{
|
| 838 |
+
if (tid < s) {
|
| 839 |
+
const unsigned int xid1 = tid << 1;
|
| 840 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 841 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 842 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 843 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 844 |
+
if (tid + (s << 1) < spre)
|
| 845 |
+
{
|
| 846 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + (s << 1)];
|
| 847 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2 + (s << 1)];
|
| 848 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1 + (s << 1)];
|
| 849 |
+
}
|
| 850 |
+
}
|
| 851 |
+
__syncthreads();
|
| 852 |
+
}
|
| 853 |
+
|
| 854 |
+
if (tid == 0)
|
| 855 |
+
{
|
| 856 |
+
*grad_sampling_loc = cache_grad_sampling_loc[0];
|
| 857 |
+
*(grad_sampling_loc + 1) = cache_grad_sampling_loc[1];
|
| 858 |
+
*grad_attn_weight = cache_grad_attn_weight[0];
|
| 859 |
+
}
|
| 860 |
+
__syncthreads();
|
| 861 |
+
|
| 862 |
+
data_weight_ptr += 1;
|
| 863 |
+
data_loc_w_ptr += 2;
|
| 864 |
+
grad_attn_weight += grad_weight_stride;
|
| 865 |
+
grad_sampling_loc += grad_loc_stride;
|
| 866 |
+
}
|
| 867 |
+
}
|
| 868 |
+
}
|
| 869 |
+
}
|
| 870 |
+
|
| 871 |
+
template <typename scalar_t>
|
| 872 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v2_multi_blocks(const int n,
|
| 873 |
+
const scalar_t *grad_col,
|
| 874 |
+
const scalar_t *data_value,
|
| 875 |
+
const int64_t *data_spatial_shapes,
|
| 876 |
+
const int64_t *data_level_start_index,
|
| 877 |
+
const scalar_t *data_sampling_loc,
|
| 878 |
+
const scalar_t *data_attn_weight,
|
| 879 |
+
const int batch_size,
|
| 880 |
+
const int spatial_size,
|
| 881 |
+
const int num_heads,
|
| 882 |
+
const int channels,
|
| 883 |
+
const int num_levels,
|
| 884 |
+
const int num_query,
|
| 885 |
+
const int num_point,
|
| 886 |
+
scalar_t *grad_value,
|
| 887 |
+
scalar_t *grad_sampling_loc,
|
| 888 |
+
scalar_t *grad_attn_weight)
|
| 889 |
+
{
|
| 890 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 891 |
+
{
|
| 892 |
+
extern __shared__ int _s[];
|
| 893 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 894 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 895 |
+
unsigned int tid = threadIdx.x;
|
| 896 |
+
int _temp = index;
|
| 897 |
+
const int c_col = _temp % channels;
|
| 898 |
+
_temp /= channels;
|
| 899 |
+
const int sampling_index = _temp;
|
| 900 |
+
const int m_col = _temp % num_heads;
|
| 901 |
+
_temp /= num_heads;
|
| 902 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 903 |
+
_temp /= num_query;
|
| 904 |
+
const int b_col = _temp;
|
| 905 |
+
|
| 906 |
+
const scalar_t top_grad = grad_col[index];
|
| 907 |
+
|
| 908 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 909 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 910 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 911 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 912 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 913 |
+
const int grad_weight_stride = 1;
|
| 914 |
+
const int grad_loc_stride = 2;
|
| 915 |
+
const int qid_stride = num_heads * channels;
|
| 916 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 917 |
+
|
| 918 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 919 |
+
{
|
| 920 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 921 |
+
const int spatial_h_ptr = l_col << 1;
|
| 922 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 923 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 924 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 925 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 926 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 927 |
+
|
| 928 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 929 |
+
{
|
| 930 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 931 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 932 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 933 |
+
|
| 934 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 935 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 936 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 937 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 938 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 939 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 940 |
+
{
|
| 941 |
+
ms_deform_attn_col2im_bilinear(
|
| 942 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 943 |
+
top_grad, weight, grad_value_ptr,
|
| 944 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 945 |
+
}
|
| 946 |
+
|
| 947 |
+
__syncthreads();
|
| 948 |
+
|
| 949 |
+
for (unsigned int s=blockDim.x/2, spre=blockDim.x; s>0; s>>=1, spre>>=1)
|
| 950 |
+
{
|
| 951 |
+
if (tid < s) {
|
| 952 |
+
const unsigned int xid1 = tid << 1;
|
| 953 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 954 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 955 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 956 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 957 |
+
if (tid + (s << 1) < spre)
|
| 958 |
+
{
|
| 959 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + (s << 1)];
|
| 960 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2 + (s << 1)];
|
| 961 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1 + (s << 1)];
|
| 962 |
+
}
|
| 963 |
+
}
|
| 964 |
+
__syncthreads();
|
| 965 |
+
}
|
| 966 |
+
|
| 967 |
+
if (tid == 0)
|
| 968 |
+
{
|
| 969 |
+
atomicAdd(grad_sampling_loc, cache_grad_sampling_loc[0]);
|
| 970 |
+
atomicAdd(grad_sampling_loc + 1, cache_grad_sampling_loc[1]);
|
| 971 |
+
atomicAdd(grad_attn_weight, cache_grad_attn_weight[0]);
|
| 972 |
+
}
|
| 973 |
+
__syncthreads();
|
| 974 |
+
|
| 975 |
+
data_weight_ptr += 1;
|
| 976 |
+
data_loc_w_ptr += 2;
|
| 977 |
+
grad_attn_weight += grad_weight_stride;
|
| 978 |
+
grad_sampling_loc += grad_loc_stride;
|
| 979 |
+
}
|
| 980 |
+
}
|
| 981 |
+
}
|
| 982 |
+
}
|
| 983 |
+
|
| 984 |
+
|
| 985 |
+
template <typename scalar_t>
|
| 986 |
+
__global__ void ms_deformable_col2im_gpu_kernel_gm(const int n,
|
| 987 |
+
const scalar_t *grad_col,
|
| 988 |
+
const scalar_t *data_value,
|
| 989 |
+
const int64_t *data_spatial_shapes,
|
| 990 |
+
const int64_t *data_level_start_index,
|
| 991 |
+
const scalar_t *data_sampling_loc,
|
| 992 |
+
const scalar_t *data_attn_weight,
|
| 993 |
+
const int batch_size,
|
| 994 |
+
const int spatial_size,
|
| 995 |
+
const int num_heads,
|
| 996 |
+
const int channels,
|
| 997 |
+
const int num_levels,
|
| 998 |
+
const int num_query,
|
| 999 |
+
const int num_point,
|
| 1000 |
+
scalar_t *grad_value,
|
| 1001 |
+
scalar_t *grad_sampling_loc,
|
| 1002 |
+
scalar_t *grad_attn_weight)
|
| 1003 |
+
{
|
| 1004 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 1005 |
+
{
|
| 1006 |
+
int _temp = index;
|
| 1007 |
+
const int c_col = _temp % channels;
|
| 1008 |
+
_temp /= channels;
|
| 1009 |
+
const int sampling_index = _temp;
|
| 1010 |
+
const int m_col = _temp % num_heads;
|
| 1011 |
+
_temp /= num_heads;
|
| 1012 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 1013 |
+
_temp /= num_query;
|
| 1014 |
+
const int b_col = _temp;
|
| 1015 |
+
|
| 1016 |
+
const scalar_t top_grad = grad_col[index];
|
| 1017 |
+
|
| 1018 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 1019 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 1020 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 1021 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 1022 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 1023 |
+
const int grad_weight_stride = 1;
|
| 1024 |
+
const int grad_loc_stride = 2;
|
| 1025 |
+
const int qid_stride = num_heads * channels;
|
| 1026 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 1027 |
+
|
| 1028 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 1029 |
+
{
|
| 1030 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 1031 |
+
const int spatial_h_ptr = l_col << 1;
|
| 1032 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 1033 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 1034 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 1035 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 1036 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 1037 |
+
|
| 1038 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 1039 |
+
{
|
| 1040 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 1041 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 1042 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 1043 |
+
|
| 1044 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 1045 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 1046 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 1047 |
+
{
|
| 1048 |
+
ms_deform_attn_col2im_bilinear_gm(
|
| 1049 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 1050 |
+
top_grad, weight, grad_value_ptr,
|
| 1051 |
+
grad_sampling_loc, grad_attn_weight);
|
| 1052 |
+
}
|
| 1053 |
+
data_weight_ptr += 1;
|
| 1054 |
+
data_loc_w_ptr += 2;
|
| 1055 |
+
grad_attn_weight += grad_weight_stride;
|
| 1056 |
+
grad_sampling_loc += grad_loc_stride;
|
| 1057 |
+
}
|
| 1058 |
+
}
|
| 1059 |
+
}
|
| 1060 |
+
}
|
| 1061 |
+
|
| 1062 |
+
|
| 1063 |
+
template <typename scalar_t>
|
| 1064 |
+
void ms_deformable_im2col_cuda(cudaStream_t stream,
|
| 1065 |
+
const scalar_t* data_value,
|
| 1066 |
+
const int64_t* data_spatial_shapes,
|
| 1067 |
+
const int64_t* data_level_start_index,
|
| 1068 |
+
const scalar_t* data_sampling_loc,
|
| 1069 |
+
const scalar_t* data_attn_weight,
|
| 1070 |
+
const int batch_size,
|
| 1071 |
+
const int spatial_size,
|
| 1072 |
+
const int num_heads,
|
| 1073 |
+
const int channels,
|
| 1074 |
+
const int num_levels,
|
| 1075 |
+
const int num_query,
|
| 1076 |
+
const int num_point,
|
| 1077 |
+
scalar_t* data_col)
|
| 1078 |
+
{
|
| 1079 |
+
const int num_kernels = batch_size * num_query * num_heads * channels;
|
| 1080 |
+
const int num_actual_kernels = batch_size * num_query * num_heads * channels;
|
| 1081 |
+
const int num_threads = CUDA_NUM_THREADS;
|
| 1082 |
+
ms_deformable_im2col_gpu_kernel<scalar_t>
|
| 1083 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1084 |
+
0, stream>>>(
|
| 1085 |
+
num_kernels, data_value, data_spatial_shapes, data_level_start_index, data_sampling_loc, data_attn_weight,
|
| 1086 |
+
batch_size, spatial_size, num_heads, channels, num_levels, num_query, num_point, data_col);
|
| 1087 |
+
|
| 1088 |
+
cudaError_t err = cudaGetLastError();
|
| 1089 |
+
if (err != cudaSuccess)
|
| 1090 |
+
{
|
| 1091 |
+
printf("error in ms_deformable_im2col_cuda: %s\n", cudaGetErrorString(err));
|
| 1092 |
+
}
|
| 1093 |
+
|
| 1094 |
+
}
|
| 1095 |
+
|
| 1096 |
+
template <typename scalar_t>
|
| 1097 |
+
void ms_deformable_col2im_cuda(cudaStream_t stream,
|
| 1098 |
+
const scalar_t* grad_col,
|
| 1099 |
+
const scalar_t* data_value,
|
| 1100 |
+
const int64_t * data_spatial_shapes,
|
| 1101 |
+
const int64_t * data_level_start_index,
|
| 1102 |
+
const scalar_t * data_sampling_loc,
|
| 1103 |
+
const scalar_t * data_attn_weight,
|
| 1104 |
+
const int batch_size,
|
| 1105 |
+
const int spatial_size,
|
| 1106 |
+
const int num_heads,
|
| 1107 |
+
const int channels,
|
| 1108 |
+
const int num_levels,
|
| 1109 |
+
const int num_query,
|
| 1110 |
+
const int num_point,
|
| 1111 |
+
scalar_t* grad_value,
|
| 1112 |
+
scalar_t* grad_sampling_loc,
|
| 1113 |
+
scalar_t* grad_attn_weight)
|
| 1114 |
+
{
|
| 1115 |
+
const int num_threads = (channels > CUDA_NUM_THREADS)?CUDA_NUM_THREADS:channels;
|
| 1116 |
+
const int num_kernels = batch_size * num_query * num_heads * channels;
|
| 1117 |
+
const int num_actual_kernels = batch_size * num_query * num_heads * channels;
|
| 1118 |
+
if (channels > 1024)
|
| 1119 |
+
{
|
| 1120 |
+
if ((channels & 1023) == 0)
|
| 1121 |
+
{
|
| 1122 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v2_multi_blocks<scalar_t>
|
| 1123 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1124 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 1125 |
+
num_kernels,
|
| 1126 |
+
grad_col,
|
| 1127 |
+
data_value,
|
| 1128 |
+
data_spatial_shapes,
|
| 1129 |
+
data_level_start_index,
|
| 1130 |
+
data_sampling_loc,
|
| 1131 |
+
data_attn_weight,
|
| 1132 |
+
batch_size,
|
| 1133 |
+
spatial_size,
|
| 1134 |
+
num_heads,
|
| 1135 |
+
channels,
|
| 1136 |
+
num_levels,
|
| 1137 |
+
num_query,
|
| 1138 |
+
num_point,
|
| 1139 |
+
grad_value,
|
| 1140 |
+
grad_sampling_loc,
|
| 1141 |
+
grad_attn_weight);
|
| 1142 |
+
}
|
| 1143 |
+
else
|
| 1144 |
+
{
|
| 1145 |
+
ms_deformable_col2im_gpu_kernel_gm<scalar_t>
|
| 1146 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1147 |
+
0, stream>>>(
|
| 1148 |
+
num_kernels,
|
| 1149 |
+
grad_col,
|
| 1150 |
+
data_value,
|
| 1151 |
+
data_spatial_shapes,
|
| 1152 |
+
data_level_start_index,
|
| 1153 |
+
data_sampling_loc,
|
| 1154 |
+
data_attn_weight,
|
| 1155 |
+
batch_size,
|
| 1156 |
+
spatial_size,
|
| 1157 |
+
num_heads,
|
| 1158 |
+
channels,
|
| 1159 |
+
num_levels,
|
| 1160 |
+
num_query,
|
| 1161 |
+
num_point,
|
| 1162 |
+
grad_value,
|
| 1163 |
+
grad_sampling_loc,
|
| 1164 |
+
grad_attn_weight);
|
| 1165 |
+
}
|
| 1166 |
+
}
|
| 1167 |
+
else{
|
| 1168 |
+
switch(channels)
|
| 1169 |
+
{
|
| 1170 |
+
case 1:
|
| 1171 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 1>
|
| 1172 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1173 |
+
0, stream>>>(
|
| 1174 |
+
num_kernels,
|
| 1175 |
+
grad_col,
|
| 1176 |
+
data_value,
|
| 1177 |
+
data_spatial_shapes,
|
| 1178 |
+
data_level_start_index,
|
| 1179 |
+
data_sampling_loc,
|
| 1180 |
+
data_attn_weight,
|
| 1181 |
+
batch_size,
|
| 1182 |
+
spatial_size,
|
| 1183 |
+
num_heads,
|
| 1184 |
+
channels,
|
| 1185 |
+
num_levels,
|
| 1186 |
+
num_query,
|
| 1187 |
+
num_point,
|
| 1188 |
+
grad_value,
|
| 1189 |
+
grad_sampling_loc,
|
| 1190 |
+
grad_attn_weight);
|
| 1191 |
+
break;
|
| 1192 |
+
case 2:
|
| 1193 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 2>
|
| 1194 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1195 |
+
0, stream>>>(
|
| 1196 |
+
num_kernels,
|
| 1197 |
+
grad_col,
|
| 1198 |
+
data_value,
|
| 1199 |
+
data_spatial_shapes,
|
| 1200 |
+
data_level_start_index,
|
| 1201 |
+
data_sampling_loc,
|
| 1202 |
+
data_attn_weight,
|
| 1203 |
+
batch_size,
|
| 1204 |
+
spatial_size,
|
| 1205 |
+
num_heads,
|
| 1206 |
+
channels,
|
| 1207 |
+
num_levels,
|
| 1208 |
+
num_query,
|
| 1209 |
+
num_point,
|
| 1210 |
+
grad_value,
|
| 1211 |
+
grad_sampling_loc,
|
| 1212 |
+
grad_attn_weight);
|
| 1213 |
+
break;
|
| 1214 |
+
case 4:
|
| 1215 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 4>
|
| 1216 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1217 |
+
0, stream>>>(
|
| 1218 |
+
num_kernels,
|
| 1219 |
+
grad_col,
|
| 1220 |
+
data_value,
|
| 1221 |
+
data_spatial_shapes,
|
| 1222 |
+
data_level_start_index,
|
| 1223 |
+
data_sampling_loc,
|
| 1224 |
+
data_attn_weight,
|
| 1225 |
+
batch_size,
|
| 1226 |
+
spatial_size,
|
| 1227 |
+
num_heads,
|
| 1228 |
+
channels,
|
| 1229 |
+
num_levels,
|
| 1230 |
+
num_query,
|
| 1231 |
+
num_point,
|
| 1232 |
+
grad_value,
|
| 1233 |
+
grad_sampling_loc,
|
| 1234 |
+
grad_attn_weight);
|
| 1235 |
+
break;
|
| 1236 |
+
case 8:
|
| 1237 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 8>
|
| 1238 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1239 |
+
0, stream>>>(
|
| 1240 |
+
num_kernels,
|
| 1241 |
+
grad_col,
|
| 1242 |
+
data_value,
|
| 1243 |
+
data_spatial_shapes,
|
| 1244 |
+
data_level_start_index,
|
| 1245 |
+
data_sampling_loc,
|
| 1246 |
+
data_attn_weight,
|
| 1247 |
+
batch_size,
|
| 1248 |
+
spatial_size,
|
| 1249 |
+
num_heads,
|
| 1250 |
+
channels,
|
| 1251 |
+
num_levels,
|
| 1252 |
+
num_query,
|
| 1253 |
+
num_point,
|
| 1254 |
+
grad_value,
|
| 1255 |
+
grad_sampling_loc,
|
| 1256 |
+
grad_attn_weight);
|
| 1257 |
+
break;
|
| 1258 |
+
case 16:
|
| 1259 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 16>
|
| 1260 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1261 |
+
0, stream>>>(
|
| 1262 |
+
num_kernels,
|
| 1263 |
+
grad_col,
|
| 1264 |
+
data_value,
|
| 1265 |
+
data_spatial_shapes,
|
| 1266 |
+
data_level_start_index,
|
| 1267 |
+
data_sampling_loc,
|
| 1268 |
+
data_attn_weight,
|
| 1269 |
+
batch_size,
|
| 1270 |
+
spatial_size,
|
| 1271 |
+
num_heads,
|
| 1272 |
+
channels,
|
| 1273 |
+
num_levels,
|
| 1274 |
+
num_query,
|
| 1275 |
+
num_point,
|
| 1276 |
+
grad_value,
|
| 1277 |
+
grad_sampling_loc,
|
| 1278 |
+
grad_attn_weight);
|
| 1279 |
+
break;
|
| 1280 |
+
case 32:
|
| 1281 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 32>
|
| 1282 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1283 |
+
0, stream>>>(
|
| 1284 |
+
num_kernels,
|
| 1285 |
+
grad_col,
|
| 1286 |
+
data_value,
|
| 1287 |
+
data_spatial_shapes,
|
| 1288 |
+
data_level_start_index,
|
| 1289 |
+
data_sampling_loc,
|
| 1290 |
+
data_attn_weight,
|
| 1291 |
+
batch_size,
|
| 1292 |
+
spatial_size,
|
| 1293 |
+
num_heads,
|
| 1294 |
+
channels,
|
| 1295 |
+
num_levels,
|
| 1296 |
+
num_query,
|
| 1297 |
+
num_point,
|
| 1298 |
+
grad_value,
|
| 1299 |
+
grad_sampling_loc,
|
| 1300 |
+
grad_attn_weight);
|
| 1301 |
+
break;
|
| 1302 |
+
case 64:
|
| 1303 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 64>
|
| 1304 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1305 |
+
0, stream>>>(
|
| 1306 |
+
num_kernels,
|
| 1307 |
+
grad_col,
|
| 1308 |
+
data_value,
|
| 1309 |
+
data_spatial_shapes,
|
| 1310 |
+
data_level_start_index,
|
| 1311 |
+
data_sampling_loc,
|
| 1312 |
+
data_attn_weight,
|
| 1313 |
+
batch_size,
|
| 1314 |
+
spatial_size,
|
| 1315 |
+
num_heads,
|
| 1316 |
+
channels,
|
| 1317 |
+
num_levels,
|
| 1318 |
+
num_query,
|
| 1319 |
+
num_point,
|
| 1320 |
+
grad_value,
|
| 1321 |
+
grad_sampling_loc,
|
| 1322 |
+
grad_attn_weight);
|
| 1323 |
+
break;
|
| 1324 |
+
case 128:
|
| 1325 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 128>
|
| 1326 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1327 |
+
0, stream>>>(
|
| 1328 |
+
num_kernels,
|
| 1329 |
+
grad_col,
|
| 1330 |
+
data_value,
|
| 1331 |
+
data_spatial_shapes,
|
| 1332 |
+
data_level_start_index,
|
| 1333 |
+
data_sampling_loc,
|
| 1334 |
+
data_attn_weight,
|
| 1335 |
+
batch_size,
|
| 1336 |
+
spatial_size,
|
| 1337 |
+
num_heads,
|
| 1338 |
+
channels,
|
| 1339 |
+
num_levels,
|
| 1340 |
+
num_query,
|
| 1341 |
+
num_point,
|
| 1342 |
+
grad_value,
|
| 1343 |
+
grad_sampling_loc,
|
| 1344 |
+
grad_attn_weight);
|
| 1345 |
+
break;
|
| 1346 |
+
case 256:
|
| 1347 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 256>
|
| 1348 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1349 |
+
0, stream>>>(
|
| 1350 |
+
num_kernels,
|
| 1351 |
+
grad_col,
|
| 1352 |
+
data_value,
|
| 1353 |
+
data_spatial_shapes,
|
| 1354 |
+
data_level_start_index,
|
| 1355 |
+
data_sampling_loc,
|
| 1356 |
+
data_attn_weight,
|
| 1357 |
+
batch_size,
|
| 1358 |
+
spatial_size,
|
| 1359 |
+
num_heads,
|
| 1360 |
+
channels,
|
| 1361 |
+
num_levels,
|
| 1362 |
+
num_query,
|
| 1363 |
+
num_point,
|
| 1364 |
+
grad_value,
|
| 1365 |
+
grad_sampling_loc,
|
| 1366 |
+
grad_attn_weight);
|
| 1367 |
+
break;
|
| 1368 |
+
case 512:
|
| 1369 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 512>
|
| 1370 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1371 |
+
0, stream>>>(
|
| 1372 |
+
num_kernels,
|
| 1373 |
+
grad_col,
|
| 1374 |
+
data_value,
|
| 1375 |
+
data_spatial_shapes,
|
| 1376 |
+
data_level_start_index,
|
| 1377 |
+
data_sampling_loc,
|
| 1378 |
+
data_attn_weight,
|
| 1379 |
+
batch_size,
|
| 1380 |
+
spatial_size,
|
| 1381 |
+
num_heads,
|
| 1382 |
+
channels,
|
| 1383 |
+
num_levels,
|
| 1384 |
+
num_query,
|
| 1385 |
+
num_point,
|
| 1386 |
+
grad_value,
|
| 1387 |
+
grad_sampling_loc,
|
| 1388 |
+
grad_attn_weight);
|
| 1389 |
+
break;
|
| 1390 |
+
case 1024:
|
| 1391 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 1024>
|
| 1392 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1393 |
+
0, stream>>>(
|
| 1394 |
+
num_kernels,
|
| 1395 |
+
grad_col,
|
| 1396 |
+
data_value,
|
| 1397 |
+
data_spatial_shapes,
|
| 1398 |
+
data_level_start_index,
|
| 1399 |
+
data_sampling_loc,
|
| 1400 |
+
data_attn_weight,
|
| 1401 |
+
batch_size,
|
| 1402 |
+
spatial_size,
|
| 1403 |
+
num_heads,
|
| 1404 |
+
channels,
|
| 1405 |
+
num_levels,
|
| 1406 |
+
num_query,
|
| 1407 |
+
num_point,
|
| 1408 |
+
grad_value,
|
| 1409 |
+
grad_sampling_loc,
|
| 1410 |
+
grad_attn_weight);
|
| 1411 |
+
break;
|
| 1412 |
+
default:
|
| 1413 |
+
if (channels < 64)
|
| 1414 |
+
{
|
| 1415 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v1<scalar_t>
|
| 1416 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1417 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 1418 |
+
num_kernels,
|
| 1419 |
+
grad_col,
|
| 1420 |
+
data_value,
|
| 1421 |
+
data_spatial_shapes,
|
| 1422 |
+
data_level_start_index,
|
| 1423 |
+
data_sampling_loc,
|
| 1424 |
+
data_attn_weight,
|
| 1425 |
+
batch_size,
|
| 1426 |
+
spatial_size,
|
| 1427 |
+
num_heads,
|
| 1428 |
+
channels,
|
| 1429 |
+
num_levels,
|
| 1430 |
+
num_query,
|
| 1431 |
+
num_point,
|
| 1432 |
+
grad_value,
|
| 1433 |
+
grad_sampling_loc,
|
| 1434 |
+
grad_attn_weight);
|
| 1435 |
+
}
|
| 1436 |
+
else
|
| 1437 |
+
{
|
| 1438 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v2<scalar_t>
|
| 1439 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1440 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 1441 |
+
num_kernels,
|
| 1442 |
+
grad_col,
|
| 1443 |
+
data_value,
|
| 1444 |
+
data_spatial_shapes,
|
| 1445 |
+
data_level_start_index,
|
| 1446 |
+
data_sampling_loc,
|
| 1447 |
+
data_attn_weight,
|
| 1448 |
+
batch_size,
|
| 1449 |
+
spatial_size,
|
| 1450 |
+
num_heads,
|
| 1451 |
+
channels,
|
| 1452 |
+
num_levels,
|
| 1453 |
+
num_query,
|
| 1454 |
+
num_point,
|
| 1455 |
+
grad_value,
|
| 1456 |
+
grad_sampling_loc,
|
| 1457 |
+
grad_attn_weight);
|
| 1458 |
+
}
|
| 1459 |
+
}
|
| 1460 |
+
}
|
| 1461 |
+
cudaError_t err = cudaGetLastError();
|
| 1462 |
+
if (err != cudaSuccess)
|
| 1463 |
+
{
|
| 1464 |
+
printf("error in ms_deformable_col2im_cuda: %s\n", cudaGetErrorString(err));
|
| 1465 |
+
}
|
| 1466 |
+
|
| 1467 |
+
}
|
deformable_detr/ms_deform_attn_cuda.h
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************************************
|
| 7 |
+
* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
|
| 8 |
+
**************************************************************************************************
|
| 9 |
+
*/
|
| 10 |
+
|
| 11 |
+
#pragma once
|
| 12 |
+
#include <torch/torch.h>
|
| 13 |
+
|
| 14 |
+
at::Tensor ms_deform_attn_cuda_forward(
|
| 15 |
+
const at::Tensor &value,
|
| 16 |
+
const at::Tensor &spatial_shapes,
|
| 17 |
+
const at::Tensor &level_start_index,
|
| 18 |
+
const at::Tensor &sampling_loc,
|
| 19 |
+
const at::Tensor &attn_weight,
|
| 20 |
+
const int im2col_step);
|
| 21 |
+
|
| 22 |
+
at::Tensor ms_deform_attn_cuda_forward_bf16(
|
| 23 |
+
const at::Tensor &value,
|
| 24 |
+
const at::Tensor &spatial_shapes,
|
| 25 |
+
const at::Tensor &level_start_index,
|
| 26 |
+
const at::Tensor &sampling_loc,
|
| 27 |
+
const at::Tensor &attn_weight,
|
| 28 |
+
const int im2col_step);
|
| 29 |
+
|
| 30 |
+
std::vector<at::Tensor> ms_deform_attn_cuda_backward(
|
| 31 |
+
const at::Tensor &value,
|
| 32 |
+
const at::Tensor &spatial_shapes,
|
| 33 |
+
const at::Tensor &level_start_index,
|
| 34 |
+
const at::Tensor &sampling_loc,
|
| 35 |
+
const at::Tensor &attn_weight,
|
| 36 |
+
const at::Tensor &grad_output,
|
| 37 |
+
const int im2col_step);
|
| 38 |
+
|
| 39 |
+
std::vector<at::Tensor> ms_deform_attn_cuda_backward_bf16(
|
| 40 |
+
const at::Tensor &value,
|
| 41 |
+
const at::Tensor &spatial_shapes,
|
| 42 |
+
const at::Tensor &level_start_index,
|
| 43 |
+
const at::Tensor &sampling_loc,
|
| 44 |
+
const at::Tensor &attn_weight,
|
| 45 |
+
const at::Tensor &grad_output,
|
| 46 |
+
const int im2col_step);
|
deformable_detr/ms_deform_im2col_cuda.cuh
ADDED
|
@@ -0,0 +1,1327 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************
|
| 7 |
+
* Modified from DCN (https://github.com/msracver/Deformable-ConvNets)
|
| 8 |
+
* Copyright (c) 2018 Microsoft
|
| 9 |
+
**************************************************************************
|
| 10 |
+
*/
|
| 11 |
+
|
| 12 |
+
#include <cstdio>
|
| 13 |
+
#include <algorithm>
|
| 14 |
+
#include <cstring>
|
| 15 |
+
|
| 16 |
+
#include <ATen/ATen.h>
|
| 17 |
+
#include <ATen/cuda/CUDAContext.h>
|
| 18 |
+
|
| 19 |
+
#include <THC/THCAtomics.cuh>
|
| 20 |
+
|
| 21 |
+
#define CUDA_KERNEL_LOOP(i, n) \
|
| 22 |
+
for (int i = blockIdx.x * blockDim.x + threadIdx.x; \
|
| 23 |
+
i < (n); \
|
| 24 |
+
i += blockDim.x * gridDim.x)
|
| 25 |
+
|
| 26 |
+
const int CUDA_NUM_THREADS = 1024;
|
| 27 |
+
inline int GET_BLOCKS(const int N, const int num_threads)
|
| 28 |
+
{
|
| 29 |
+
return (N + num_threads - 1) / num_threads;
|
| 30 |
+
}
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
template <typename scalar_t>
|
| 34 |
+
__device__ scalar_t ms_deform_attn_im2col_bilinear(const scalar_t* &bottom_data,
|
| 35 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 36 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c)
|
| 37 |
+
{
|
| 38 |
+
const int h_low = floor(h);
|
| 39 |
+
const int w_low = floor(w);
|
| 40 |
+
const int h_high = h_low + 1;
|
| 41 |
+
const int w_high = w_low + 1;
|
| 42 |
+
|
| 43 |
+
const scalar_t lh = h - h_low;
|
| 44 |
+
const scalar_t lw = w - w_low;
|
| 45 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 46 |
+
|
| 47 |
+
const int w_stride = nheads * channels;
|
| 48 |
+
const int h_stride = width * w_stride;
|
| 49 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 50 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 51 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 52 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 53 |
+
const int base_ptr = m * channels + c;
|
| 54 |
+
|
| 55 |
+
scalar_t v1 = 0;
|
| 56 |
+
if (h_low >= 0 && w_low >= 0)
|
| 57 |
+
{
|
| 58 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 59 |
+
v1 = bottom_data[ptr1];
|
| 60 |
+
}
|
| 61 |
+
scalar_t v2 = 0;
|
| 62 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 63 |
+
{
|
| 64 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 65 |
+
v2 = bottom_data[ptr2];
|
| 66 |
+
}
|
| 67 |
+
scalar_t v3 = 0;
|
| 68 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 69 |
+
{
|
| 70 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 71 |
+
v3 = bottom_data[ptr3];
|
| 72 |
+
}
|
| 73 |
+
scalar_t v4 = 0;
|
| 74 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 75 |
+
{
|
| 76 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 77 |
+
v4 = bottom_data[ptr4];
|
| 78 |
+
}
|
| 79 |
+
|
| 80 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 81 |
+
|
| 82 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 83 |
+
return val;
|
| 84 |
+
}
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
template <typename scalar_t>
|
| 88 |
+
__device__ void ms_deform_attn_col2im_bilinear(const scalar_t* &bottom_data,
|
| 89 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 90 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c,
|
| 91 |
+
const scalar_t &top_grad,
|
| 92 |
+
const scalar_t &attn_weight,
|
| 93 |
+
scalar_t* &grad_value,
|
| 94 |
+
scalar_t* grad_sampling_loc,
|
| 95 |
+
scalar_t* grad_attn_weight)
|
| 96 |
+
{
|
| 97 |
+
const int h_low = floor(h);
|
| 98 |
+
const int w_low = floor(w);
|
| 99 |
+
const int h_high = h_low + 1;
|
| 100 |
+
const int w_high = w_low + 1;
|
| 101 |
+
|
| 102 |
+
const scalar_t lh = h - h_low;
|
| 103 |
+
const scalar_t lw = w - w_low;
|
| 104 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 105 |
+
|
| 106 |
+
const int w_stride = nheads * channels;
|
| 107 |
+
const int h_stride = width * w_stride;
|
| 108 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 109 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 110 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 111 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 112 |
+
const int base_ptr = m * channels + c;
|
| 113 |
+
|
| 114 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 115 |
+
const scalar_t top_grad_value = top_grad * attn_weight;
|
| 116 |
+
scalar_t grad_h_weight = 0, grad_w_weight = 0;
|
| 117 |
+
|
| 118 |
+
scalar_t v1 = 0;
|
| 119 |
+
if (h_low >= 0 && w_low >= 0)
|
| 120 |
+
{
|
| 121 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 122 |
+
v1 = bottom_data[ptr1];
|
| 123 |
+
grad_h_weight -= hw * v1;
|
| 124 |
+
grad_w_weight -= hh * v1;
|
| 125 |
+
atomicAdd(grad_value+ptr1, w1*top_grad_value);
|
| 126 |
+
}
|
| 127 |
+
scalar_t v2 = 0;
|
| 128 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 129 |
+
{
|
| 130 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 131 |
+
v2 = bottom_data[ptr2];
|
| 132 |
+
grad_h_weight -= lw * v2;
|
| 133 |
+
grad_w_weight += hh * v2;
|
| 134 |
+
atomicAdd(grad_value+ptr2, w2*top_grad_value);
|
| 135 |
+
}
|
| 136 |
+
scalar_t v3 = 0;
|
| 137 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 138 |
+
{
|
| 139 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 140 |
+
v3 = bottom_data[ptr3];
|
| 141 |
+
grad_h_weight += hw * v3;
|
| 142 |
+
grad_w_weight -= lh * v3;
|
| 143 |
+
atomicAdd(grad_value+ptr3, w3*top_grad_value);
|
| 144 |
+
}
|
| 145 |
+
scalar_t v4 = 0;
|
| 146 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 147 |
+
{
|
| 148 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 149 |
+
v4 = bottom_data[ptr4];
|
| 150 |
+
grad_h_weight += lw * v4;
|
| 151 |
+
grad_w_weight += lh * v4;
|
| 152 |
+
atomicAdd(grad_value+ptr4, w4*top_grad_value);
|
| 153 |
+
}
|
| 154 |
+
|
| 155 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 156 |
+
*grad_attn_weight = top_grad * val;
|
| 157 |
+
*grad_sampling_loc = width * grad_w_weight * top_grad_value;
|
| 158 |
+
*(grad_sampling_loc + 1) = height * grad_h_weight * top_grad_value;
|
| 159 |
+
}
|
| 160 |
+
|
| 161 |
+
|
| 162 |
+
template <typename scalar_t>
|
| 163 |
+
__device__ void ms_deform_attn_col2im_bilinear_gm(const scalar_t* &bottom_data,
|
| 164 |
+
const int &height, const int &width, const int &nheads, const int &channels,
|
| 165 |
+
const scalar_t &h, const scalar_t &w, const int &m, const int &c,
|
| 166 |
+
const scalar_t &top_grad,
|
| 167 |
+
const scalar_t &attn_weight,
|
| 168 |
+
scalar_t* &grad_value,
|
| 169 |
+
scalar_t* grad_sampling_loc,
|
| 170 |
+
scalar_t* grad_attn_weight)
|
| 171 |
+
{
|
| 172 |
+
const int h_low = floor(h);
|
| 173 |
+
const int w_low = floor(w);
|
| 174 |
+
const int h_high = h_low + 1;
|
| 175 |
+
const int w_high = w_low + 1;
|
| 176 |
+
|
| 177 |
+
const scalar_t lh = h - h_low;
|
| 178 |
+
const scalar_t lw = w - w_low;
|
| 179 |
+
const scalar_t hh = 1 - lh, hw = 1 - lw;
|
| 180 |
+
|
| 181 |
+
const int w_stride = nheads * channels;
|
| 182 |
+
const int h_stride = width * w_stride;
|
| 183 |
+
const int h_low_ptr_offset = h_low * h_stride;
|
| 184 |
+
const int h_high_ptr_offset = h_low_ptr_offset + h_stride;
|
| 185 |
+
const int w_low_ptr_offset = w_low * w_stride;
|
| 186 |
+
const int w_high_ptr_offset = w_low_ptr_offset + w_stride;
|
| 187 |
+
const int base_ptr = m * channels + c;
|
| 188 |
+
|
| 189 |
+
const scalar_t w1 = hh * hw, w2 = hh * lw, w3 = lh * hw, w4 = lh * lw;
|
| 190 |
+
const scalar_t top_grad_value = top_grad * attn_weight;
|
| 191 |
+
scalar_t grad_h_weight = 0, grad_w_weight = 0;
|
| 192 |
+
|
| 193 |
+
scalar_t v1 = 0;
|
| 194 |
+
if (h_low >= 0 && w_low >= 0)
|
| 195 |
+
{
|
| 196 |
+
const int ptr1 = h_low_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 197 |
+
v1 = bottom_data[ptr1];
|
| 198 |
+
grad_h_weight -= hw * v1;
|
| 199 |
+
grad_w_weight -= hh * v1;
|
| 200 |
+
atomicAdd(grad_value+ptr1, w1*top_grad_value);
|
| 201 |
+
}
|
| 202 |
+
scalar_t v2 = 0;
|
| 203 |
+
if (h_low >= 0 && w_high <= width - 1)
|
| 204 |
+
{
|
| 205 |
+
const int ptr2 = h_low_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 206 |
+
v2 = bottom_data[ptr2];
|
| 207 |
+
grad_h_weight -= lw * v2;
|
| 208 |
+
grad_w_weight += hh * v2;
|
| 209 |
+
atomicAdd(grad_value+ptr2, w2*top_grad_value);
|
| 210 |
+
}
|
| 211 |
+
scalar_t v3 = 0;
|
| 212 |
+
if (h_high <= height - 1 && w_low >= 0)
|
| 213 |
+
{
|
| 214 |
+
const int ptr3 = h_high_ptr_offset + w_low_ptr_offset + base_ptr;
|
| 215 |
+
v3 = bottom_data[ptr3];
|
| 216 |
+
grad_h_weight += hw * v3;
|
| 217 |
+
grad_w_weight -= lh * v3;
|
| 218 |
+
atomicAdd(grad_value+ptr3, w3*top_grad_value);
|
| 219 |
+
}
|
| 220 |
+
scalar_t v4 = 0;
|
| 221 |
+
if (h_high <= height - 1 && w_high <= width - 1)
|
| 222 |
+
{
|
| 223 |
+
const int ptr4 = h_high_ptr_offset + w_high_ptr_offset + base_ptr;
|
| 224 |
+
v4 = bottom_data[ptr4];
|
| 225 |
+
grad_h_weight += lw * v4;
|
| 226 |
+
grad_w_weight += lh * v4;
|
| 227 |
+
atomicAdd(grad_value+ptr4, w4*top_grad_value);
|
| 228 |
+
}
|
| 229 |
+
|
| 230 |
+
const scalar_t val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
|
| 231 |
+
atomicAdd(grad_attn_weight, top_grad * val);
|
| 232 |
+
atomicAdd(grad_sampling_loc, width * grad_w_weight * top_grad_value);
|
| 233 |
+
atomicAdd(grad_sampling_loc + 1, height * grad_h_weight * top_grad_value);
|
| 234 |
+
}
|
| 235 |
+
|
| 236 |
+
|
| 237 |
+
template <typename scalar_t>
|
| 238 |
+
__global__ void ms_deformable_im2col_gpu_kernel(const int n,
|
| 239 |
+
const scalar_t *data_value,
|
| 240 |
+
const int64_t *data_spatial_shapes,
|
| 241 |
+
const int64_t *data_level_start_index,
|
| 242 |
+
const scalar_t *data_sampling_loc,
|
| 243 |
+
const scalar_t *data_attn_weight,
|
| 244 |
+
const int batch_size,
|
| 245 |
+
const int spatial_size,
|
| 246 |
+
const int num_heads,
|
| 247 |
+
const int channels,
|
| 248 |
+
const int num_levels,
|
| 249 |
+
const int num_query,
|
| 250 |
+
const int num_point,
|
| 251 |
+
scalar_t *data_col)
|
| 252 |
+
{
|
| 253 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 254 |
+
{
|
| 255 |
+
int _temp = index;
|
| 256 |
+
const int c_col = _temp % channels;
|
| 257 |
+
_temp /= channels;
|
| 258 |
+
const int sampling_index = _temp;
|
| 259 |
+
const int m_col = _temp % num_heads;
|
| 260 |
+
_temp /= num_heads;
|
| 261 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 262 |
+
_temp /= num_query;
|
| 263 |
+
const int b_col = _temp;
|
| 264 |
+
|
| 265 |
+
scalar_t *data_col_ptr = data_col + index;
|
| 266 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 267 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 268 |
+
const int qid_stride = num_heads * channels;
|
| 269 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 270 |
+
scalar_t col = 0;
|
| 271 |
+
|
| 272 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 273 |
+
{
|
| 274 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 275 |
+
const int spatial_h_ptr = l_col << 1;
|
| 276 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 277 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 278 |
+
const scalar_t *data_value_ptr = data_value + (data_value_ptr_init_offset + level_start_id * qid_stride);
|
| 279 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 280 |
+
{
|
| 281 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 282 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 283 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 284 |
+
|
| 285 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 286 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 287 |
+
|
| 288 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 289 |
+
{
|
| 290 |
+
col += ms_deform_attn_im2col_bilinear(data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col) * weight;
|
| 291 |
+
}
|
| 292 |
+
|
| 293 |
+
data_weight_ptr += 1;
|
| 294 |
+
data_loc_w_ptr += 2;
|
| 295 |
+
}
|
| 296 |
+
}
|
| 297 |
+
*data_col_ptr = col;
|
| 298 |
+
}
|
| 299 |
+
}
|
| 300 |
+
|
| 301 |
+
template <typename scalar_t, unsigned int blockSize>
|
| 302 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1(const int n,
|
| 303 |
+
const scalar_t *grad_col,
|
| 304 |
+
const scalar_t *data_value,
|
| 305 |
+
const int64_t *data_spatial_shapes,
|
| 306 |
+
const int64_t *data_level_start_index,
|
| 307 |
+
const scalar_t *data_sampling_loc,
|
| 308 |
+
const scalar_t *data_attn_weight,
|
| 309 |
+
const int batch_size,
|
| 310 |
+
const int spatial_size,
|
| 311 |
+
const int num_heads,
|
| 312 |
+
const int channels,
|
| 313 |
+
const int num_levels,
|
| 314 |
+
const int num_query,
|
| 315 |
+
const int num_point,
|
| 316 |
+
scalar_t *grad_value,
|
| 317 |
+
scalar_t *grad_sampling_loc,
|
| 318 |
+
scalar_t *grad_attn_weight)
|
| 319 |
+
{
|
| 320 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 321 |
+
{
|
| 322 |
+
__shared__ scalar_t cache_grad_sampling_loc[blockSize * 2];
|
| 323 |
+
__shared__ scalar_t cache_grad_attn_weight[blockSize];
|
| 324 |
+
unsigned int tid = threadIdx.x;
|
| 325 |
+
int _temp = index;
|
| 326 |
+
const int c_col = _temp % channels;
|
| 327 |
+
_temp /= channels;
|
| 328 |
+
const int sampling_index = _temp;
|
| 329 |
+
const int m_col = _temp % num_heads;
|
| 330 |
+
_temp /= num_heads;
|
| 331 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 332 |
+
_temp /= num_query;
|
| 333 |
+
const int b_col = _temp;
|
| 334 |
+
|
| 335 |
+
const scalar_t top_grad = grad_col[index];
|
| 336 |
+
|
| 337 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 338 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 339 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 340 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 341 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 342 |
+
const int grad_weight_stride = 1;
|
| 343 |
+
const int grad_loc_stride = 2;
|
| 344 |
+
const int qid_stride = num_heads * channels;
|
| 345 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 346 |
+
|
| 347 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 348 |
+
{
|
| 349 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 350 |
+
const int spatial_h_ptr = l_col << 1;
|
| 351 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 352 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 353 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 354 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 355 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 356 |
+
|
| 357 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 358 |
+
{
|
| 359 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 360 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 361 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 362 |
+
|
| 363 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 364 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 365 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 366 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 367 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 368 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 369 |
+
{
|
| 370 |
+
ms_deform_attn_col2im_bilinear(
|
| 371 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 372 |
+
top_grad, weight, grad_value_ptr,
|
| 373 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 374 |
+
}
|
| 375 |
+
|
| 376 |
+
__syncthreads();
|
| 377 |
+
if (tid == 0)
|
| 378 |
+
{
|
| 379 |
+
scalar_t _grad_w=cache_grad_sampling_loc[0], _grad_h=cache_grad_sampling_loc[1], _grad_a=cache_grad_attn_weight[0];
|
| 380 |
+
int sid=2;
|
| 381 |
+
for (unsigned int tid = 1; tid < blockSize; ++tid)
|
| 382 |
+
{
|
| 383 |
+
_grad_w += cache_grad_sampling_loc[sid];
|
| 384 |
+
_grad_h += cache_grad_sampling_loc[sid + 1];
|
| 385 |
+
_grad_a += cache_grad_attn_weight[tid];
|
| 386 |
+
sid += 2;
|
| 387 |
+
}
|
| 388 |
+
|
| 389 |
+
|
| 390 |
+
*grad_sampling_loc = _grad_w;
|
| 391 |
+
*(grad_sampling_loc + 1) = _grad_h;
|
| 392 |
+
*grad_attn_weight = _grad_a;
|
| 393 |
+
}
|
| 394 |
+
__syncthreads();
|
| 395 |
+
|
| 396 |
+
data_weight_ptr += 1;
|
| 397 |
+
data_loc_w_ptr += 2;
|
| 398 |
+
grad_attn_weight += grad_weight_stride;
|
| 399 |
+
grad_sampling_loc += grad_loc_stride;
|
| 400 |
+
}
|
| 401 |
+
}
|
| 402 |
+
}
|
| 403 |
+
}
|
| 404 |
+
|
| 405 |
+
|
| 406 |
+
template <typename scalar_t, unsigned int blockSize>
|
| 407 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2(const int n,
|
| 408 |
+
const scalar_t *grad_col,
|
| 409 |
+
const scalar_t *data_value,
|
| 410 |
+
const int64_t *data_spatial_shapes,
|
| 411 |
+
const int64_t *data_level_start_index,
|
| 412 |
+
const scalar_t *data_sampling_loc,
|
| 413 |
+
const scalar_t *data_attn_weight,
|
| 414 |
+
const int batch_size,
|
| 415 |
+
const int spatial_size,
|
| 416 |
+
const int num_heads,
|
| 417 |
+
const int channels,
|
| 418 |
+
const int num_levels,
|
| 419 |
+
const int num_query,
|
| 420 |
+
const int num_point,
|
| 421 |
+
scalar_t *grad_value,
|
| 422 |
+
scalar_t *grad_sampling_loc,
|
| 423 |
+
scalar_t *grad_attn_weight)
|
| 424 |
+
{
|
| 425 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 426 |
+
{
|
| 427 |
+
__shared__ scalar_t cache_grad_sampling_loc[blockSize * 2];
|
| 428 |
+
__shared__ scalar_t cache_grad_attn_weight[blockSize];
|
| 429 |
+
unsigned int tid = threadIdx.x;
|
| 430 |
+
int _temp = index;
|
| 431 |
+
const int c_col = _temp % channels;
|
| 432 |
+
_temp /= channels;
|
| 433 |
+
const int sampling_index = _temp;
|
| 434 |
+
const int m_col = _temp % num_heads;
|
| 435 |
+
_temp /= num_heads;
|
| 436 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 437 |
+
_temp /= num_query;
|
| 438 |
+
const int b_col = _temp;
|
| 439 |
+
|
| 440 |
+
const scalar_t top_grad = grad_col[index];
|
| 441 |
+
|
| 442 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 443 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 444 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 445 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 446 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 447 |
+
const int grad_weight_stride = 1;
|
| 448 |
+
const int grad_loc_stride = 2;
|
| 449 |
+
const int qid_stride = num_heads * channels;
|
| 450 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 451 |
+
|
| 452 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 453 |
+
{
|
| 454 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 455 |
+
const int spatial_h_ptr = l_col << 1;
|
| 456 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 457 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 458 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 459 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 460 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 461 |
+
|
| 462 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 463 |
+
{
|
| 464 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 465 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 466 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 467 |
+
|
| 468 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 469 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 470 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 471 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 472 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 473 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 474 |
+
{
|
| 475 |
+
ms_deform_attn_col2im_bilinear(
|
| 476 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 477 |
+
top_grad, weight, grad_value_ptr,
|
| 478 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 479 |
+
}
|
| 480 |
+
|
| 481 |
+
__syncthreads();
|
| 482 |
+
|
| 483 |
+
for (unsigned int s=blockSize/2; s>0; s>>=1)
|
| 484 |
+
{
|
| 485 |
+
if (tid < s) {
|
| 486 |
+
const unsigned int xid1 = tid << 1;
|
| 487 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 488 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 489 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 490 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 491 |
+
}
|
| 492 |
+
__syncthreads();
|
| 493 |
+
}
|
| 494 |
+
|
| 495 |
+
if (tid == 0)
|
| 496 |
+
{
|
| 497 |
+
*grad_sampling_loc = cache_grad_sampling_loc[0];
|
| 498 |
+
*(grad_sampling_loc + 1) = cache_grad_sampling_loc[1];
|
| 499 |
+
*grad_attn_weight = cache_grad_attn_weight[0];
|
| 500 |
+
}
|
| 501 |
+
__syncthreads();
|
| 502 |
+
|
| 503 |
+
data_weight_ptr += 1;
|
| 504 |
+
data_loc_w_ptr += 2;
|
| 505 |
+
grad_attn_weight += grad_weight_stride;
|
| 506 |
+
grad_sampling_loc += grad_loc_stride;
|
| 507 |
+
}
|
| 508 |
+
}
|
| 509 |
+
}
|
| 510 |
+
}
|
| 511 |
+
|
| 512 |
+
|
| 513 |
+
template <typename scalar_t>
|
| 514 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v1(const int n,
|
| 515 |
+
const scalar_t *grad_col,
|
| 516 |
+
const scalar_t *data_value,
|
| 517 |
+
const int64_t *data_spatial_shapes,
|
| 518 |
+
const int64_t *data_level_start_index,
|
| 519 |
+
const scalar_t *data_sampling_loc,
|
| 520 |
+
const scalar_t *data_attn_weight,
|
| 521 |
+
const int batch_size,
|
| 522 |
+
const int spatial_size,
|
| 523 |
+
const int num_heads,
|
| 524 |
+
const int channels,
|
| 525 |
+
const int num_levels,
|
| 526 |
+
const int num_query,
|
| 527 |
+
const int num_point,
|
| 528 |
+
scalar_t *grad_value,
|
| 529 |
+
scalar_t *grad_sampling_loc,
|
| 530 |
+
scalar_t *grad_attn_weight)
|
| 531 |
+
{
|
| 532 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 533 |
+
{
|
| 534 |
+
extern __shared__ int _s[];
|
| 535 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 536 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 537 |
+
unsigned int tid = threadIdx.x;
|
| 538 |
+
int _temp = index;
|
| 539 |
+
const int c_col = _temp % channels;
|
| 540 |
+
_temp /= channels;
|
| 541 |
+
const int sampling_index = _temp;
|
| 542 |
+
const int m_col = _temp % num_heads;
|
| 543 |
+
_temp /= num_heads;
|
| 544 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 545 |
+
_temp /= num_query;
|
| 546 |
+
const int b_col = _temp;
|
| 547 |
+
|
| 548 |
+
const scalar_t top_grad = grad_col[index];
|
| 549 |
+
|
| 550 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 551 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 552 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 553 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 554 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 555 |
+
const int grad_weight_stride = 1;
|
| 556 |
+
const int grad_loc_stride = 2;
|
| 557 |
+
const int qid_stride = num_heads * channels;
|
| 558 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 559 |
+
|
| 560 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 561 |
+
{
|
| 562 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 563 |
+
const int spatial_h_ptr = l_col << 1;
|
| 564 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 565 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 566 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 567 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 568 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 569 |
+
|
| 570 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 571 |
+
{
|
| 572 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 573 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 574 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 575 |
+
|
| 576 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 577 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 578 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 579 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 580 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 581 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 582 |
+
{
|
| 583 |
+
ms_deform_attn_col2im_bilinear(
|
| 584 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 585 |
+
top_grad, weight, grad_value_ptr,
|
| 586 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 587 |
+
}
|
| 588 |
+
|
| 589 |
+
__syncthreads();
|
| 590 |
+
if (tid == 0)
|
| 591 |
+
{
|
| 592 |
+
scalar_t _grad_w=cache_grad_sampling_loc[0], _grad_h=cache_grad_sampling_loc[1], _grad_a=cache_grad_attn_weight[0];
|
| 593 |
+
int sid=2;
|
| 594 |
+
for (unsigned int tid = 1; tid < blockDim.x; ++tid)
|
| 595 |
+
{
|
| 596 |
+
_grad_w += cache_grad_sampling_loc[sid];
|
| 597 |
+
_grad_h += cache_grad_sampling_loc[sid + 1];
|
| 598 |
+
_grad_a += cache_grad_attn_weight[tid];
|
| 599 |
+
sid += 2;
|
| 600 |
+
}
|
| 601 |
+
|
| 602 |
+
|
| 603 |
+
*grad_sampling_loc = _grad_w;
|
| 604 |
+
*(grad_sampling_loc + 1) = _grad_h;
|
| 605 |
+
*grad_attn_weight = _grad_a;
|
| 606 |
+
}
|
| 607 |
+
__syncthreads();
|
| 608 |
+
|
| 609 |
+
data_weight_ptr += 1;
|
| 610 |
+
data_loc_w_ptr += 2;
|
| 611 |
+
grad_attn_weight += grad_weight_stride;
|
| 612 |
+
grad_sampling_loc += grad_loc_stride;
|
| 613 |
+
}
|
| 614 |
+
}
|
| 615 |
+
}
|
| 616 |
+
}
|
| 617 |
+
|
| 618 |
+
template <typename scalar_t>
|
| 619 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v2(const int n,
|
| 620 |
+
const scalar_t *grad_col,
|
| 621 |
+
const scalar_t *data_value,
|
| 622 |
+
const int64_t *data_spatial_shapes,
|
| 623 |
+
const int64_t *data_level_start_index,
|
| 624 |
+
const scalar_t *data_sampling_loc,
|
| 625 |
+
const scalar_t *data_attn_weight,
|
| 626 |
+
const int batch_size,
|
| 627 |
+
const int spatial_size,
|
| 628 |
+
const int num_heads,
|
| 629 |
+
const int channels,
|
| 630 |
+
const int num_levels,
|
| 631 |
+
const int num_query,
|
| 632 |
+
const int num_point,
|
| 633 |
+
scalar_t *grad_value,
|
| 634 |
+
scalar_t *grad_sampling_loc,
|
| 635 |
+
scalar_t *grad_attn_weight)
|
| 636 |
+
{
|
| 637 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 638 |
+
{
|
| 639 |
+
extern __shared__ int _s[];
|
| 640 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 641 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 642 |
+
unsigned int tid = threadIdx.x;
|
| 643 |
+
int _temp = index;
|
| 644 |
+
const int c_col = _temp % channels;
|
| 645 |
+
_temp /= channels;
|
| 646 |
+
const int sampling_index = _temp;
|
| 647 |
+
const int m_col = _temp % num_heads;
|
| 648 |
+
_temp /= num_heads;
|
| 649 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 650 |
+
_temp /= num_query;
|
| 651 |
+
const int b_col = _temp;
|
| 652 |
+
|
| 653 |
+
const scalar_t top_grad = grad_col[index];
|
| 654 |
+
|
| 655 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 656 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 657 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 658 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 659 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 660 |
+
const int grad_weight_stride = 1;
|
| 661 |
+
const int grad_loc_stride = 2;
|
| 662 |
+
const int qid_stride = num_heads * channels;
|
| 663 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 664 |
+
|
| 665 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 666 |
+
{
|
| 667 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 668 |
+
const int spatial_h_ptr = l_col << 1;
|
| 669 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 670 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 671 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 672 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 673 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 674 |
+
|
| 675 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 676 |
+
{
|
| 677 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 678 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 679 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 680 |
+
|
| 681 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 682 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 683 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 684 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 685 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 686 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 687 |
+
{
|
| 688 |
+
ms_deform_attn_col2im_bilinear(
|
| 689 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 690 |
+
top_grad, weight, grad_value_ptr,
|
| 691 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 692 |
+
}
|
| 693 |
+
|
| 694 |
+
__syncthreads();
|
| 695 |
+
|
| 696 |
+
for (unsigned int s=blockDim.x/2, spre=blockDim.x; s>0; s>>=1, spre>>=1)
|
| 697 |
+
{
|
| 698 |
+
if (tid < s) {
|
| 699 |
+
const unsigned int xid1 = tid << 1;
|
| 700 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 701 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 702 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 703 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 704 |
+
if (tid + (s << 1) < spre)
|
| 705 |
+
{
|
| 706 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + (s << 1)];
|
| 707 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2 + (s << 1)];
|
| 708 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1 + (s << 1)];
|
| 709 |
+
}
|
| 710 |
+
}
|
| 711 |
+
__syncthreads();
|
| 712 |
+
}
|
| 713 |
+
|
| 714 |
+
if (tid == 0)
|
| 715 |
+
{
|
| 716 |
+
*grad_sampling_loc = cache_grad_sampling_loc[0];
|
| 717 |
+
*(grad_sampling_loc + 1) = cache_grad_sampling_loc[1];
|
| 718 |
+
*grad_attn_weight = cache_grad_attn_weight[0];
|
| 719 |
+
}
|
| 720 |
+
__syncthreads();
|
| 721 |
+
|
| 722 |
+
data_weight_ptr += 1;
|
| 723 |
+
data_loc_w_ptr += 2;
|
| 724 |
+
grad_attn_weight += grad_weight_stride;
|
| 725 |
+
grad_sampling_loc += grad_loc_stride;
|
| 726 |
+
}
|
| 727 |
+
}
|
| 728 |
+
}
|
| 729 |
+
}
|
| 730 |
+
|
| 731 |
+
template <typename scalar_t>
|
| 732 |
+
__global__ void ms_deformable_col2im_gpu_kernel_shm_reduce_v2_multi_blocks(const int n,
|
| 733 |
+
const scalar_t *grad_col,
|
| 734 |
+
const scalar_t *data_value,
|
| 735 |
+
const int64_t *data_spatial_shapes,
|
| 736 |
+
const int64_t *data_level_start_index,
|
| 737 |
+
const scalar_t *data_sampling_loc,
|
| 738 |
+
const scalar_t *data_attn_weight,
|
| 739 |
+
const int batch_size,
|
| 740 |
+
const int spatial_size,
|
| 741 |
+
const int num_heads,
|
| 742 |
+
const int channels,
|
| 743 |
+
const int num_levels,
|
| 744 |
+
const int num_query,
|
| 745 |
+
const int num_point,
|
| 746 |
+
scalar_t *grad_value,
|
| 747 |
+
scalar_t *grad_sampling_loc,
|
| 748 |
+
scalar_t *grad_attn_weight)
|
| 749 |
+
{
|
| 750 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 751 |
+
{
|
| 752 |
+
extern __shared__ int _s[];
|
| 753 |
+
scalar_t* cache_grad_sampling_loc = (scalar_t*)_s;
|
| 754 |
+
scalar_t* cache_grad_attn_weight = cache_grad_sampling_loc + 2 * blockDim.x;
|
| 755 |
+
unsigned int tid = threadIdx.x;
|
| 756 |
+
int _temp = index;
|
| 757 |
+
const int c_col = _temp % channels;
|
| 758 |
+
_temp /= channels;
|
| 759 |
+
const int sampling_index = _temp;
|
| 760 |
+
const int m_col = _temp % num_heads;
|
| 761 |
+
_temp /= num_heads;
|
| 762 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 763 |
+
_temp /= num_query;
|
| 764 |
+
const int b_col = _temp;
|
| 765 |
+
|
| 766 |
+
const scalar_t top_grad = grad_col[index];
|
| 767 |
+
|
| 768 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 769 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 770 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 771 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 772 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 773 |
+
const int grad_weight_stride = 1;
|
| 774 |
+
const int grad_loc_stride = 2;
|
| 775 |
+
const int qid_stride = num_heads * channels;
|
| 776 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 777 |
+
|
| 778 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 779 |
+
{
|
| 780 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 781 |
+
const int spatial_h_ptr = l_col << 1;
|
| 782 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 783 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 784 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 785 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 786 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 787 |
+
|
| 788 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 789 |
+
{
|
| 790 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 791 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 792 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 793 |
+
|
| 794 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 795 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 796 |
+
*(cache_grad_sampling_loc+(threadIdx.x << 1)) = 0;
|
| 797 |
+
*(cache_grad_sampling_loc+((threadIdx.x << 1) + 1)) = 0;
|
| 798 |
+
*(cache_grad_attn_weight+threadIdx.x)=0;
|
| 799 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 800 |
+
{
|
| 801 |
+
ms_deform_attn_col2im_bilinear(
|
| 802 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 803 |
+
top_grad, weight, grad_value_ptr,
|
| 804 |
+
cache_grad_sampling_loc+(threadIdx.x << 1), cache_grad_attn_weight+threadIdx.x);
|
| 805 |
+
}
|
| 806 |
+
|
| 807 |
+
__syncthreads();
|
| 808 |
+
|
| 809 |
+
for (unsigned int s=blockDim.x/2, spre=blockDim.x; s>0; s>>=1, spre>>=1)
|
| 810 |
+
{
|
| 811 |
+
if (tid < s) {
|
| 812 |
+
const unsigned int xid1 = tid << 1;
|
| 813 |
+
const unsigned int xid2 = (tid + s) << 1;
|
| 814 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + s];
|
| 815 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2];
|
| 816 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1];
|
| 817 |
+
if (tid + (s << 1) < spre)
|
| 818 |
+
{
|
| 819 |
+
cache_grad_attn_weight[tid] += cache_grad_attn_weight[tid + (s << 1)];
|
| 820 |
+
cache_grad_sampling_loc[xid1] += cache_grad_sampling_loc[xid2 + (s << 1)];
|
| 821 |
+
cache_grad_sampling_loc[xid1 + 1] += cache_grad_sampling_loc[xid2 + 1 + (s << 1)];
|
| 822 |
+
}
|
| 823 |
+
}
|
| 824 |
+
__syncthreads();
|
| 825 |
+
}
|
| 826 |
+
|
| 827 |
+
if (tid == 0)
|
| 828 |
+
{
|
| 829 |
+
atomicAdd(grad_sampling_loc, cache_grad_sampling_loc[0]);
|
| 830 |
+
atomicAdd(grad_sampling_loc + 1, cache_grad_sampling_loc[1]);
|
| 831 |
+
atomicAdd(grad_attn_weight, cache_grad_attn_weight[0]);
|
| 832 |
+
}
|
| 833 |
+
__syncthreads();
|
| 834 |
+
|
| 835 |
+
data_weight_ptr += 1;
|
| 836 |
+
data_loc_w_ptr += 2;
|
| 837 |
+
grad_attn_weight += grad_weight_stride;
|
| 838 |
+
grad_sampling_loc += grad_loc_stride;
|
| 839 |
+
}
|
| 840 |
+
}
|
| 841 |
+
}
|
| 842 |
+
}
|
| 843 |
+
|
| 844 |
+
|
| 845 |
+
template <typename scalar_t>
|
| 846 |
+
__global__ void ms_deformable_col2im_gpu_kernel_gm(const int n,
|
| 847 |
+
const scalar_t *grad_col,
|
| 848 |
+
const scalar_t *data_value,
|
| 849 |
+
const int64_t *data_spatial_shapes,
|
| 850 |
+
const int64_t *data_level_start_index,
|
| 851 |
+
const scalar_t *data_sampling_loc,
|
| 852 |
+
const scalar_t *data_attn_weight,
|
| 853 |
+
const int batch_size,
|
| 854 |
+
const int spatial_size,
|
| 855 |
+
const int num_heads,
|
| 856 |
+
const int channels,
|
| 857 |
+
const int num_levels,
|
| 858 |
+
const int num_query,
|
| 859 |
+
const int num_point,
|
| 860 |
+
scalar_t *grad_value,
|
| 861 |
+
scalar_t *grad_sampling_loc,
|
| 862 |
+
scalar_t *grad_attn_weight)
|
| 863 |
+
{
|
| 864 |
+
CUDA_KERNEL_LOOP(index, n)
|
| 865 |
+
{
|
| 866 |
+
int _temp = index;
|
| 867 |
+
const int c_col = _temp % channels;
|
| 868 |
+
_temp /= channels;
|
| 869 |
+
const int sampling_index = _temp;
|
| 870 |
+
const int m_col = _temp % num_heads;
|
| 871 |
+
_temp /= num_heads;
|
| 872 |
+
[[maybe_unused]] const int q_col = _temp % num_query;
|
| 873 |
+
_temp /= num_query;
|
| 874 |
+
const int b_col = _temp;
|
| 875 |
+
|
| 876 |
+
const scalar_t top_grad = grad_col[index];
|
| 877 |
+
|
| 878 |
+
int data_weight_ptr = sampling_index * num_levels * num_point;
|
| 879 |
+
int data_loc_w_ptr = data_weight_ptr << 1;
|
| 880 |
+
const int grad_sampling_ptr = data_weight_ptr;
|
| 881 |
+
grad_sampling_loc += grad_sampling_ptr << 1;
|
| 882 |
+
grad_attn_weight += grad_sampling_ptr;
|
| 883 |
+
const int grad_weight_stride = 1;
|
| 884 |
+
const int grad_loc_stride = 2;
|
| 885 |
+
const int qid_stride = num_heads * channels;
|
| 886 |
+
const int data_value_ptr_init_offset = b_col * spatial_size * qid_stride;
|
| 887 |
+
|
| 888 |
+
for (int l_col=0; l_col < num_levels; ++l_col)
|
| 889 |
+
{
|
| 890 |
+
const int level_start_id = data_level_start_index[l_col];
|
| 891 |
+
const int spatial_h_ptr = l_col << 1;
|
| 892 |
+
const int spatial_h = data_spatial_shapes[spatial_h_ptr];
|
| 893 |
+
const int spatial_w = data_spatial_shapes[spatial_h_ptr + 1];
|
| 894 |
+
const int value_ptr_offset = data_value_ptr_init_offset + level_start_id * qid_stride;
|
| 895 |
+
const scalar_t *data_value_ptr = data_value + value_ptr_offset;
|
| 896 |
+
scalar_t *grad_value_ptr = grad_value + value_ptr_offset;
|
| 897 |
+
|
| 898 |
+
for (int p_col=0; p_col < num_point; ++p_col)
|
| 899 |
+
{
|
| 900 |
+
const scalar_t loc_w = data_sampling_loc[data_loc_w_ptr];
|
| 901 |
+
const scalar_t loc_h = data_sampling_loc[data_loc_w_ptr + 1];
|
| 902 |
+
const scalar_t weight = data_attn_weight[data_weight_ptr];
|
| 903 |
+
|
| 904 |
+
const scalar_t h_im = loc_h * spatial_h - 0.5;
|
| 905 |
+
const scalar_t w_im = loc_w * spatial_w - 0.5;
|
| 906 |
+
if (h_im > -1 && w_im > -1 && h_im < spatial_h && w_im < spatial_w)
|
| 907 |
+
{
|
| 908 |
+
ms_deform_attn_col2im_bilinear_gm(
|
| 909 |
+
data_value_ptr, spatial_h, spatial_w, num_heads, channels, h_im, w_im, m_col, c_col,
|
| 910 |
+
top_grad, weight, grad_value_ptr,
|
| 911 |
+
grad_sampling_loc, grad_attn_weight);
|
| 912 |
+
}
|
| 913 |
+
data_weight_ptr += 1;
|
| 914 |
+
data_loc_w_ptr += 2;
|
| 915 |
+
grad_attn_weight += grad_weight_stride;
|
| 916 |
+
grad_sampling_loc += grad_loc_stride;
|
| 917 |
+
}
|
| 918 |
+
}
|
| 919 |
+
}
|
| 920 |
+
}
|
| 921 |
+
|
| 922 |
+
|
| 923 |
+
template <typename scalar_t>
|
| 924 |
+
void ms_deformable_im2col_cuda(cudaStream_t stream,
|
| 925 |
+
const scalar_t* data_value,
|
| 926 |
+
const int64_t* data_spatial_shapes,
|
| 927 |
+
const int64_t* data_level_start_index,
|
| 928 |
+
const scalar_t* data_sampling_loc,
|
| 929 |
+
const scalar_t* data_attn_weight,
|
| 930 |
+
const int batch_size,
|
| 931 |
+
const int spatial_size,
|
| 932 |
+
const int num_heads,
|
| 933 |
+
const int channels,
|
| 934 |
+
const int num_levels,
|
| 935 |
+
const int num_query,
|
| 936 |
+
const int num_point,
|
| 937 |
+
scalar_t* data_col)
|
| 938 |
+
{
|
| 939 |
+
const int num_kernels = batch_size * num_query * num_heads * channels;
|
| 940 |
+
const int num_actual_kernels = batch_size * num_query * num_heads * channels;
|
| 941 |
+
const int num_threads = CUDA_NUM_THREADS;
|
| 942 |
+
ms_deformable_im2col_gpu_kernel<scalar_t>
|
| 943 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 944 |
+
0, stream>>>(
|
| 945 |
+
num_kernels, data_value, data_spatial_shapes, data_level_start_index, data_sampling_loc, data_attn_weight,
|
| 946 |
+
batch_size, spatial_size, num_heads, channels, num_levels, num_query, num_point, data_col);
|
| 947 |
+
|
| 948 |
+
cudaError_t err = cudaGetLastError();
|
| 949 |
+
if (err != cudaSuccess)
|
| 950 |
+
{
|
| 951 |
+
printf("error in ms_deformable_im2col_cuda: %s\n", cudaGetErrorString(err));
|
| 952 |
+
}
|
| 953 |
+
|
| 954 |
+
}
|
| 955 |
+
|
| 956 |
+
template <typename scalar_t>
|
| 957 |
+
void ms_deformable_col2im_cuda(cudaStream_t stream,
|
| 958 |
+
const scalar_t* grad_col,
|
| 959 |
+
const scalar_t* data_value,
|
| 960 |
+
const int64_t * data_spatial_shapes,
|
| 961 |
+
const int64_t * data_level_start_index,
|
| 962 |
+
const scalar_t * data_sampling_loc,
|
| 963 |
+
const scalar_t * data_attn_weight,
|
| 964 |
+
const int batch_size,
|
| 965 |
+
const int spatial_size,
|
| 966 |
+
const int num_heads,
|
| 967 |
+
const int channels,
|
| 968 |
+
const int num_levels,
|
| 969 |
+
const int num_query,
|
| 970 |
+
const int num_point,
|
| 971 |
+
scalar_t* grad_value,
|
| 972 |
+
scalar_t* grad_sampling_loc,
|
| 973 |
+
scalar_t* grad_attn_weight)
|
| 974 |
+
{
|
| 975 |
+
const int num_threads = (channels > CUDA_NUM_THREADS)?CUDA_NUM_THREADS:channels;
|
| 976 |
+
const int num_kernels = batch_size * num_query * num_heads * channels;
|
| 977 |
+
const int num_actual_kernels = batch_size * num_query * num_heads * channels;
|
| 978 |
+
if (channels > 1024)
|
| 979 |
+
{
|
| 980 |
+
if ((channels & 1023) == 0)
|
| 981 |
+
{
|
| 982 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v2_multi_blocks<scalar_t>
|
| 983 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 984 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 985 |
+
num_kernels,
|
| 986 |
+
grad_col,
|
| 987 |
+
data_value,
|
| 988 |
+
data_spatial_shapes,
|
| 989 |
+
data_level_start_index,
|
| 990 |
+
data_sampling_loc,
|
| 991 |
+
data_attn_weight,
|
| 992 |
+
batch_size,
|
| 993 |
+
spatial_size,
|
| 994 |
+
num_heads,
|
| 995 |
+
channels,
|
| 996 |
+
num_levels,
|
| 997 |
+
num_query,
|
| 998 |
+
num_point,
|
| 999 |
+
grad_value,
|
| 1000 |
+
grad_sampling_loc,
|
| 1001 |
+
grad_attn_weight);
|
| 1002 |
+
}
|
| 1003 |
+
else
|
| 1004 |
+
{
|
| 1005 |
+
ms_deformable_col2im_gpu_kernel_gm<scalar_t>
|
| 1006 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1007 |
+
0, stream>>>(
|
| 1008 |
+
num_kernels,
|
| 1009 |
+
grad_col,
|
| 1010 |
+
data_value,
|
| 1011 |
+
data_spatial_shapes,
|
| 1012 |
+
data_level_start_index,
|
| 1013 |
+
data_sampling_loc,
|
| 1014 |
+
data_attn_weight,
|
| 1015 |
+
batch_size,
|
| 1016 |
+
spatial_size,
|
| 1017 |
+
num_heads,
|
| 1018 |
+
channels,
|
| 1019 |
+
num_levels,
|
| 1020 |
+
num_query,
|
| 1021 |
+
num_point,
|
| 1022 |
+
grad_value,
|
| 1023 |
+
grad_sampling_loc,
|
| 1024 |
+
grad_attn_weight);
|
| 1025 |
+
}
|
| 1026 |
+
}
|
| 1027 |
+
else{
|
| 1028 |
+
switch(channels)
|
| 1029 |
+
{
|
| 1030 |
+
case 1:
|
| 1031 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 1>
|
| 1032 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1033 |
+
0, stream>>>(
|
| 1034 |
+
num_kernels,
|
| 1035 |
+
grad_col,
|
| 1036 |
+
data_value,
|
| 1037 |
+
data_spatial_shapes,
|
| 1038 |
+
data_level_start_index,
|
| 1039 |
+
data_sampling_loc,
|
| 1040 |
+
data_attn_weight,
|
| 1041 |
+
batch_size,
|
| 1042 |
+
spatial_size,
|
| 1043 |
+
num_heads,
|
| 1044 |
+
channels,
|
| 1045 |
+
num_levels,
|
| 1046 |
+
num_query,
|
| 1047 |
+
num_point,
|
| 1048 |
+
grad_value,
|
| 1049 |
+
grad_sampling_loc,
|
| 1050 |
+
grad_attn_weight);
|
| 1051 |
+
break;
|
| 1052 |
+
case 2:
|
| 1053 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 2>
|
| 1054 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1055 |
+
0, stream>>>(
|
| 1056 |
+
num_kernels,
|
| 1057 |
+
grad_col,
|
| 1058 |
+
data_value,
|
| 1059 |
+
data_spatial_shapes,
|
| 1060 |
+
data_level_start_index,
|
| 1061 |
+
data_sampling_loc,
|
| 1062 |
+
data_attn_weight,
|
| 1063 |
+
batch_size,
|
| 1064 |
+
spatial_size,
|
| 1065 |
+
num_heads,
|
| 1066 |
+
channels,
|
| 1067 |
+
num_levels,
|
| 1068 |
+
num_query,
|
| 1069 |
+
num_point,
|
| 1070 |
+
grad_value,
|
| 1071 |
+
grad_sampling_loc,
|
| 1072 |
+
grad_attn_weight);
|
| 1073 |
+
break;
|
| 1074 |
+
case 4:
|
| 1075 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 4>
|
| 1076 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1077 |
+
0, stream>>>(
|
| 1078 |
+
num_kernels,
|
| 1079 |
+
grad_col,
|
| 1080 |
+
data_value,
|
| 1081 |
+
data_spatial_shapes,
|
| 1082 |
+
data_level_start_index,
|
| 1083 |
+
data_sampling_loc,
|
| 1084 |
+
data_attn_weight,
|
| 1085 |
+
batch_size,
|
| 1086 |
+
spatial_size,
|
| 1087 |
+
num_heads,
|
| 1088 |
+
channels,
|
| 1089 |
+
num_levels,
|
| 1090 |
+
num_query,
|
| 1091 |
+
num_point,
|
| 1092 |
+
grad_value,
|
| 1093 |
+
grad_sampling_loc,
|
| 1094 |
+
grad_attn_weight);
|
| 1095 |
+
break;
|
| 1096 |
+
case 8:
|
| 1097 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 8>
|
| 1098 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1099 |
+
0, stream>>>(
|
| 1100 |
+
num_kernels,
|
| 1101 |
+
grad_col,
|
| 1102 |
+
data_value,
|
| 1103 |
+
data_spatial_shapes,
|
| 1104 |
+
data_level_start_index,
|
| 1105 |
+
data_sampling_loc,
|
| 1106 |
+
data_attn_weight,
|
| 1107 |
+
batch_size,
|
| 1108 |
+
spatial_size,
|
| 1109 |
+
num_heads,
|
| 1110 |
+
channels,
|
| 1111 |
+
num_levels,
|
| 1112 |
+
num_query,
|
| 1113 |
+
num_point,
|
| 1114 |
+
grad_value,
|
| 1115 |
+
grad_sampling_loc,
|
| 1116 |
+
grad_attn_weight);
|
| 1117 |
+
break;
|
| 1118 |
+
case 16:
|
| 1119 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 16>
|
| 1120 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1121 |
+
0, stream>>>(
|
| 1122 |
+
num_kernels,
|
| 1123 |
+
grad_col,
|
| 1124 |
+
data_value,
|
| 1125 |
+
data_spatial_shapes,
|
| 1126 |
+
data_level_start_index,
|
| 1127 |
+
data_sampling_loc,
|
| 1128 |
+
data_attn_weight,
|
| 1129 |
+
batch_size,
|
| 1130 |
+
spatial_size,
|
| 1131 |
+
num_heads,
|
| 1132 |
+
channels,
|
| 1133 |
+
num_levels,
|
| 1134 |
+
num_query,
|
| 1135 |
+
num_point,
|
| 1136 |
+
grad_value,
|
| 1137 |
+
grad_sampling_loc,
|
| 1138 |
+
grad_attn_weight);
|
| 1139 |
+
break;
|
| 1140 |
+
case 32:
|
| 1141 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v1<scalar_t, 32>
|
| 1142 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1143 |
+
0, stream>>>(
|
| 1144 |
+
num_kernels,
|
| 1145 |
+
grad_col,
|
| 1146 |
+
data_value,
|
| 1147 |
+
data_spatial_shapes,
|
| 1148 |
+
data_level_start_index,
|
| 1149 |
+
data_sampling_loc,
|
| 1150 |
+
data_attn_weight,
|
| 1151 |
+
batch_size,
|
| 1152 |
+
spatial_size,
|
| 1153 |
+
num_heads,
|
| 1154 |
+
channels,
|
| 1155 |
+
num_levels,
|
| 1156 |
+
num_query,
|
| 1157 |
+
num_point,
|
| 1158 |
+
grad_value,
|
| 1159 |
+
grad_sampling_loc,
|
| 1160 |
+
grad_attn_weight);
|
| 1161 |
+
break;
|
| 1162 |
+
case 64:
|
| 1163 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 64>
|
| 1164 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1165 |
+
0, stream>>>(
|
| 1166 |
+
num_kernels,
|
| 1167 |
+
grad_col,
|
| 1168 |
+
data_value,
|
| 1169 |
+
data_spatial_shapes,
|
| 1170 |
+
data_level_start_index,
|
| 1171 |
+
data_sampling_loc,
|
| 1172 |
+
data_attn_weight,
|
| 1173 |
+
batch_size,
|
| 1174 |
+
spatial_size,
|
| 1175 |
+
num_heads,
|
| 1176 |
+
channels,
|
| 1177 |
+
num_levels,
|
| 1178 |
+
num_query,
|
| 1179 |
+
num_point,
|
| 1180 |
+
grad_value,
|
| 1181 |
+
grad_sampling_loc,
|
| 1182 |
+
grad_attn_weight);
|
| 1183 |
+
break;
|
| 1184 |
+
case 128:
|
| 1185 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 128>
|
| 1186 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1187 |
+
0, stream>>>(
|
| 1188 |
+
num_kernels,
|
| 1189 |
+
grad_col,
|
| 1190 |
+
data_value,
|
| 1191 |
+
data_spatial_shapes,
|
| 1192 |
+
data_level_start_index,
|
| 1193 |
+
data_sampling_loc,
|
| 1194 |
+
data_attn_weight,
|
| 1195 |
+
batch_size,
|
| 1196 |
+
spatial_size,
|
| 1197 |
+
num_heads,
|
| 1198 |
+
channels,
|
| 1199 |
+
num_levels,
|
| 1200 |
+
num_query,
|
| 1201 |
+
num_point,
|
| 1202 |
+
grad_value,
|
| 1203 |
+
grad_sampling_loc,
|
| 1204 |
+
grad_attn_weight);
|
| 1205 |
+
break;
|
| 1206 |
+
case 256:
|
| 1207 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 256>
|
| 1208 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1209 |
+
0, stream>>>(
|
| 1210 |
+
num_kernels,
|
| 1211 |
+
grad_col,
|
| 1212 |
+
data_value,
|
| 1213 |
+
data_spatial_shapes,
|
| 1214 |
+
data_level_start_index,
|
| 1215 |
+
data_sampling_loc,
|
| 1216 |
+
data_attn_weight,
|
| 1217 |
+
batch_size,
|
| 1218 |
+
spatial_size,
|
| 1219 |
+
num_heads,
|
| 1220 |
+
channels,
|
| 1221 |
+
num_levels,
|
| 1222 |
+
num_query,
|
| 1223 |
+
num_point,
|
| 1224 |
+
grad_value,
|
| 1225 |
+
grad_sampling_loc,
|
| 1226 |
+
grad_attn_weight);
|
| 1227 |
+
break;
|
| 1228 |
+
case 512:
|
| 1229 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 512>
|
| 1230 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1231 |
+
0, stream>>>(
|
| 1232 |
+
num_kernels,
|
| 1233 |
+
grad_col,
|
| 1234 |
+
data_value,
|
| 1235 |
+
data_spatial_shapes,
|
| 1236 |
+
data_level_start_index,
|
| 1237 |
+
data_sampling_loc,
|
| 1238 |
+
data_attn_weight,
|
| 1239 |
+
batch_size,
|
| 1240 |
+
spatial_size,
|
| 1241 |
+
num_heads,
|
| 1242 |
+
channels,
|
| 1243 |
+
num_levels,
|
| 1244 |
+
num_query,
|
| 1245 |
+
num_point,
|
| 1246 |
+
grad_value,
|
| 1247 |
+
grad_sampling_loc,
|
| 1248 |
+
grad_attn_weight);
|
| 1249 |
+
break;
|
| 1250 |
+
case 1024:
|
| 1251 |
+
ms_deformable_col2im_gpu_kernel_shm_blocksize_aware_reduce_v2<scalar_t, 1024>
|
| 1252 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1253 |
+
0, stream>>>(
|
| 1254 |
+
num_kernels,
|
| 1255 |
+
grad_col,
|
| 1256 |
+
data_value,
|
| 1257 |
+
data_spatial_shapes,
|
| 1258 |
+
data_level_start_index,
|
| 1259 |
+
data_sampling_loc,
|
| 1260 |
+
data_attn_weight,
|
| 1261 |
+
batch_size,
|
| 1262 |
+
spatial_size,
|
| 1263 |
+
num_heads,
|
| 1264 |
+
channels,
|
| 1265 |
+
num_levels,
|
| 1266 |
+
num_query,
|
| 1267 |
+
num_point,
|
| 1268 |
+
grad_value,
|
| 1269 |
+
grad_sampling_loc,
|
| 1270 |
+
grad_attn_weight);
|
| 1271 |
+
break;
|
| 1272 |
+
default:
|
| 1273 |
+
if (channels < 64)
|
| 1274 |
+
{
|
| 1275 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v1<scalar_t>
|
| 1276 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1277 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 1278 |
+
num_kernels,
|
| 1279 |
+
grad_col,
|
| 1280 |
+
data_value,
|
| 1281 |
+
data_spatial_shapes,
|
| 1282 |
+
data_level_start_index,
|
| 1283 |
+
data_sampling_loc,
|
| 1284 |
+
data_attn_weight,
|
| 1285 |
+
batch_size,
|
| 1286 |
+
spatial_size,
|
| 1287 |
+
num_heads,
|
| 1288 |
+
channels,
|
| 1289 |
+
num_levels,
|
| 1290 |
+
num_query,
|
| 1291 |
+
num_point,
|
| 1292 |
+
grad_value,
|
| 1293 |
+
grad_sampling_loc,
|
| 1294 |
+
grad_attn_weight);
|
| 1295 |
+
}
|
| 1296 |
+
else
|
| 1297 |
+
{
|
| 1298 |
+
ms_deformable_col2im_gpu_kernel_shm_reduce_v2<scalar_t>
|
| 1299 |
+
<<<GET_BLOCKS(num_actual_kernels, num_threads), num_threads,
|
| 1300 |
+
num_threads*3*sizeof(scalar_t), stream>>>(
|
| 1301 |
+
num_kernels,
|
| 1302 |
+
grad_col,
|
| 1303 |
+
data_value,
|
| 1304 |
+
data_spatial_shapes,
|
| 1305 |
+
data_level_start_index,
|
| 1306 |
+
data_sampling_loc,
|
| 1307 |
+
data_attn_weight,
|
| 1308 |
+
batch_size,
|
| 1309 |
+
spatial_size,
|
| 1310 |
+
num_heads,
|
| 1311 |
+
channels,
|
| 1312 |
+
num_levels,
|
| 1313 |
+
num_query,
|
| 1314 |
+
num_point,
|
| 1315 |
+
grad_value,
|
| 1316 |
+
grad_sampling_loc,
|
| 1317 |
+
grad_attn_weight);
|
| 1318 |
+
}
|
| 1319 |
+
}
|
| 1320 |
+
}
|
| 1321 |
+
cudaError_t err = cudaGetLastError();
|
| 1322 |
+
if (err != cudaSuccess)
|
| 1323 |
+
{
|
| 1324 |
+
printf("error in ms_deformable_col2im_cuda: %s\n", cudaGetErrorString(err));
|
| 1325 |
+
}
|
| 1326 |
+
|
| 1327 |
+
}
|
flake.nix
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
description = "Flake for deformable_detr kernels";
|
| 3 |
+
|
| 4 |
+
inputs = {
|
| 5 |
+
kernel-builder.url = "git+ssh://[email protected]/huggingface/kernel-builder";
|
| 6 |
+
};
|
| 7 |
+
|
| 8 |
+
outputs =
|
| 9 |
+
{
|
| 10 |
+
self,
|
| 11 |
+
kernel-builder,
|
| 12 |
+
}:
|
| 13 |
+
kernel-builder.lib.genFlakeOutputs ./.;
|
| 14 |
+
}
|
torch-ext/deformable_detr/__init__.py
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import List
|
| 2 |
+
import torch
|
| 3 |
+
|
| 4 |
+
from ._ops import ops
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
def ms_deform_attn_backward(
|
| 8 |
+
value: torch.Tensor,
|
| 9 |
+
spatial_shapes: torch.Tensor,
|
| 10 |
+
level_start_index: torch.Tensor,
|
| 11 |
+
sampling_loc: torch.Tensor,
|
| 12 |
+
attn_weight: torch.Tensor,
|
| 13 |
+
grad_output: torch.Tensor,
|
| 14 |
+
im2col_step: int,
|
| 15 |
+
) -> List[torch.Tensor]:
|
| 16 |
+
return ops.ms_deform_attn_backward(
|
| 17 |
+
value,
|
| 18 |
+
spatial_shapes,
|
| 19 |
+
level_start_index,
|
| 20 |
+
sampling_loc,
|
| 21 |
+
attn_weight,
|
| 22 |
+
grad_output,
|
| 23 |
+
im2col_step,
|
| 24 |
+
)
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def ms_deform_attn_forward(
|
| 28 |
+
value: torch.Tensor,
|
| 29 |
+
spatial_shapes: torch.Tensor,
|
| 30 |
+
level_start_index: torch.Tensor,
|
| 31 |
+
sampling_loc: torch.Tensor,
|
| 32 |
+
attn_weight: torch.Tensor,
|
| 33 |
+
im2col_step: int,
|
| 34 |
+
) -> torch.Tensor:
|
| 35 |
+
return ops.ms_deform_attn_forward(
|
| 36 |
+
value,
|
| 37 |
+
spatial_shapes,
|
| 38 |
+
level_start_index,
|
| 39 |
+
sampling_loc,
|
| 40 |
+
attn_weight,
|
| 41 |
+
im2col_step,
|
| 42 |
+
)
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
__all__ = ["ms_deform_attn_forward", "ms_deform_attn_backward"]
|
torch-ext/deformable_detr/layers.py
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import List, Union, Tuple
|
| 2 |
+
|
| 3 |
+
from torch import Tensor
|
| 4 |
+
from torch.autograd import Function
|
| 5 |
+
from torch.autograd.function import once_differentiable
|
| 6 |
+
import torch.nn as nn
|
| 7 |
+
|
| 8 |
+
from ._ops import ops
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
class MultiScaleDeformableAttentionFunction(Function):
|
| 12 |
+
@staticmethod
|
| 13 |
+
def forward(
|
| 14 |
+
context,
|
| 15 |
+
value: Tensor,
|
| 16 |
+
value_spatial_shapes: Tensor,
|
| 17 |
+
value_level_start_index: Tensor,
|
| 18 |
+
sampling_locations: Tensor,
|
| 19 |
+
attention_weights: Tensor,
|
| 20 |
+
im2col_step: int,
|
| 21 |
+
):
|
| 22 |
+
context.im2col_step = im2col_step
|
| 23 |
+
output = ops.ms_deform_attn_forward(
|
| 24 |
+
value,
|
| 25 |
+
value_spatial_shapes,
|
| 26 |
+
value_level_start_index,
|
| 27 |
+
sampling_locations,
|
| 28 |
+
attention_weights,
|
| 29 |
+
context.im2col_step,
|
| 30 |
+
)
|
| 31 |
+
context.save_for_backward(
|
| 32 |
+
value,
|
| 33 |
+
value_spatial_shapes,
|
| 34 |
+
value_level_start_index,
|
| 35 |
+
sampling_locations,
|
| 36 |
+
attention_weights,
|
| 37 |
+
)
|
| 38 |
+
return output
|
| 39 |
+
|
| 40 |
+
@staticmethod
|
| 41 |
+
@once_differentiable
|
| 42 |
+
def backward(context, grad_output):
|
| 43 |
+
(
|
| 44 |
+
value,
|
| 45 |
+
value_spatial_shapes,
|
| 46 |
+
value_level_start_index,
|
| 47 |
+
sampling_locations,
|
| 48 |
+
attention_weights,
|
| 49 |
+
) = context.saved_tensors
|
| 50 |
+
grad_value, grad_sampling_loc, grad_attn_weight = ops.ms_deform_attn_backward(
|
| 51 |
+
value,
|
| 52 |
+
value_spatial_shapes,
|
| 53 |
+
value_level_start_index,
|
| 54 |
+
sampling_locations,
|
| 55 |
+
attention_weights,
|
| 56 |
+
grad_output,
|
| 57 |
+
context.im2col_step,
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
return grad_value, None, None, grad_sampling_loc, grad_attn_weight, None
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
class MultiScaleDeformableAttention(nn.Module):
|
| 64 |
+
def forward(
|
| 65 |
+
self,
|
| 66 |
+
value: Tensor,
|
| 67 |
+
value_spatial_shapes: Tensor,
|
| 68 |
+
value_spatial_shapes_list: List[Tuple],
|
| 69 |
+
level_start_index: Tensor,
|
| 70 |
+
sampling_locations: Tensor,
|
| 71 |
+
attention_weights: Tensor,
|
| 72 |
+
im2col_step: int,
|
| 73 |
+
):
|
| 74 |
+
return MultiScaleDeformableAttentionFunction.apply(
|
| 75 |
+
value,
|
| 76 |
+
value_spatial_shapes,
|
| 77 |
+
level_start_index,
|
| 78 |
+
sampling_locations,
|
| 79 |
+
attention_weights,
|
| 80 |
+
im2col_step,
|
| 81 |
+
)
|
torch-ext/ms_deform_attn_cpu.cpp
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************************************
|
| 7 |
+
* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
|
| 8 |
+
**************************************************************************************************
|
| 9 |
+
*/
|
| 10 |
+
|
| 11 |
+
#include <vector>
|
| 12 |
+
|
| 13 |
+
#include <ATen/ATen.h>
|
| 14 |
+
#include <ATen/cuda/CUDAContext.h>
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
at::Tensor
|
| 18 |
+
ms_deform_attn_cpu_forward(
|
| 19 |
+
const at::Tensor &value,
|
| 20 |
+
const at::Tensor &spatial_shapes,
|
| 21 |
+
const at::Tensor &level_start_index,
|
| 22 |
+
const at::Tensor &sampling_loc,
|
| 23 |
+
const at::Tensor &attn_weight,
|
| 24 |
+
const int im2col_step)
|
| 25 |
+
{
|
| 26 |
+
AT_ERROR("Not implement on cpu");
|
| 27 |
+
}
|
| 28 |
+
|
| 29 |
+
std::vector<at::Tensor>
|
| 30 |
+
ms_deform_attn_cpu_backward(
|
| 31 |
+
const at::Tensor &value,
|
| 32 |
+
const at::Tensor &spatial_shapes,
|
| 33 |
+
const at::Tensor &level_start_index,
|
| 34 |
+
const at::Tensor &sampling_loc,
|
| 35 |
+
const at::Tensor &attn_weight,
|
| 36 |
+
const at::Tensor &grad_output,
|
| 37 |
+
const int im2col_step)
|
| 38 |
+
{
|
| 39 |
+
AT_ERROR("Not implement on cpu");
|
| 40 |
+
}
|
torch-ext/ms_deform_attn_cpu.h
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
/*!
|
| 2 |
+
**************************************************************************************************
|
| 3 |
+
* Deformable DETR
|
| 4 |
+
* Copyright (c) 2020 SenseTime. All Rights Reserved.
|
| 5 |
+
* Licensed under the Apache License, Version 2.0 [see LICENSE for details]
|
| 6 |
+
**************************************************************************************************
|
| 7 |
+
* Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
|
| 8 |
+
**************************************************************************************************
|
| 9 |
+
*/
|
| 10 |
+
|
| 11 |
+
#pragma once
|
| 12 |
+
#include <torch/extension.h>
|
| 13 |
+
|
| 14 |
+
at::Tensor
|
| 15 |
+
ms_deform_attn_cpu_forward(
|
| 16 |
+
const at::Tensor &value,
|
| 17 |
+
const at::Tensor &spatial_shapes,
|
| 18 |
+
const at::Tensor &level_start_index,
|
| 19 |
+
const at::Tensor &sampling_loc,
|
| 20 |
+
const at::Tensor &attn_weight,
|
| 21 |
+
const int im2col_step);
|
| 22 |
+
|
| 23 |
+
std::vector<at::Tensor>
|
| 24 |
+
ms_deform_attn_cpu_backward(
|
| 25 |
+
const at::Tensor &value,
|
| 26 |
+
const at::Tensor &spatial_shapes,
|
| 27 |
+
const at::Tensor &level_start_index,
|
| 28 |
+
const at::Tensor &sampling_loc,
|
| 29 |
+
const at::Tensor &attn_weight,
|
| 30 |
+
const at::Tensor &grad_output,
|
| 31 |
+
const int im2col_step);
|
| 32 |
+
|
torch-ext/torch_binding.cpp
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#include <torch/library.h>
|
| 2 |
+
|
| 3 |
+
#include "registration.h"
|
| 4 |
+
#include "torch_binding.h"
|
| 5 |
+
|
| 6 |
+
TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
|
| 7 |
+
ops.def("ms_deform_attn_forward(Tensor value, Tensor spatial_shapes,"
|
| 8 |
+
" Tensor level_start_index, Tensor sampling_loc,"
|
| 9 |
+
" Tensor attn_weight, int im2col_step) -> Tensor");
|
| 10 |
+
ops.impl("ms_deform_attn_forward", torch::kCUDA, &ms_deform_attn_cuda_forward);
|
| 11 |
+
|
| 12 |
+
ops.def("ms_deform_attn_backward(Tensor value, Tensor spatial_shapes,"
|
| 13 |
+
" Tensor level_start_index, Tensor sampling_loc,"
|
| 14 |
+
" Tensor attn_weight, Tensor grad_output,"
|
| 15 |
+
" int im2col_step) -> Tensor[]");
|
| 16 |
+
ops.impl("ms_deform_attn_backward", torch::kCUDA, &ms_deform_attn_cuda_backward);
|
| 17 |
+
}
|
| 18 |
+
|
| 19 |
+
REGISTER_EXTENSION(TORCH_EXTENSION_NAME)
|
torch-ext/torch_binding.h
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
#include <torch/torch.h>
|
| 4 |
+
|
| 5 |
+
at::Tensor ms_deform_attn_cuda_forward(const at::Tensor &value,
|
| 6 |
+
const at::Tensor &spatial_shapes,
|
| 7 |
+
const at::Tensor &level_start_index,
|
| 8 |
+
const at::Tensor &sampling_loc,
|
| 9 |
+
const at::Tensor &attn_weight,
|
| 10 |
+
const int64_t im2col_step);
|
| 11 |
+
|
| 12 |
+
std::vector<at::Tensor> ms_deform_attn_cuda_backward(
|
| 13 |
+
const at::Tensor &value, const at::Tensor &spatial_shapes,
|
| 14 |
+
const at::Tensor &level_start_index, const at::Tensor &sampling_loc,
|
| 15 |
+
const at::Tensor &attn_weight, const at::Tensor &grad_output,
|
| 16 |
+
const int64_t im2col_step);
|