From bdd85395164cfb5e67fb68be05e427079cdb6220 Mon Sep 17 00:00:00 2001 From: momo609 <963372609@qq.com> Date: Tue, 23 May 2023 20:47:50 +0800 Subject: [PATCH 1/4] add npu build.md --- docs/zh_cn/get_started/build.md | 48 +++++++++++++++++++++++++++++++++ 1 file changed, 48 insertions(+) diff --git a/docs/zh_cn/get_started/build.md b/docs/zh_cn/get_started/build.md index 99d2214e90..368119ab04 100644 --- a/docs/zh_cn/get_started/build.md +++ b/docs/zh_cn/get_started/build.md @@ -354,3 +354,51 @@ y = torch.tensor([1, 5, 3]).mlu() w = torch.ones(10).float().mlu() output = sigmoid_focal_loss(x, y, 2.0, 0.25, w, 'none') ``` + +### 在昇腾 NPU 机器编译 mmcv-full + +在编译 mmcv-full 前，需要安装 torch_npu，完整安装教程详见 [PyTorch 安装指南](https://gitee.com/ascend/pytorch/blob/master/docs/zh/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97.md#pytorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97) + +#### 选项 1: 使用 pip 安装 Ascend 编译版本的 mmcv-full + +Ascend 编译版本的 mmcv-full 在 mmcv >= 1.7.0 时已经支持直接 pip 安装 + +```bash +pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/ascend/torch1.8.0/index.html +``` + +#### 选项 2: 使用 NPU 设备源码编译安装 mmcv-full + +- 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv/tree/master) + +```bash +git pull https://github.com/open-mmlab/mmcv/tree/master +``` + +- 编译 + +```bash +MMCV_WITH_OPS=1 MAX_JOBS=8 FORCE_NPU=1 python setup.py build_ext +``` + +- 安装 + +```bash +MMCV_WITH_OPS=1 FORCE_NPU=1 python setup.py develop +``` + +#### 验证 + +```python +import torch +import torch_npu +from mmcv.ops import softmax_focal_loss + +# Init tensor to the NPU +x = torch.randn(3, 10).npu() +y = torch.tensor([1, 5, 3]).npu() +w = torch.ones(10).float().npu() + +output = softmax_focal_loss(x, y, 2.0, 0.25, w, 'none') +print(output) +``` From e19bbe8045e875a280757c8c3541ceb6cbfd3f91 Mon Sep 17 00:00:00 2001 From: momo609 <963372609@qq.com> Date: Tue, 6 Jun 2023 09:31:11 +0800 Subject: [PATCH 2/4] update roi_pool_npu --- docs/zh_cn/get_started/build.md | 16 ++++---- mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp | 45 +++++++++++++++------- 2 files changed, 40 insertions(+), 21 deletions(-) diff --git a/docs/zh_cn/get_started/build.md b/docs/zh_cn/get_started/build.md index 368119ab04..9d0ba3ab78 100644 --- a/docs/zh_cn/get_started/build.md +++ b/docs/zh_cn/get_started/build.md @@ -355,24 +355,24 @@ w = torch.ones(10).float().mlu() output = sigmoid_focal_loss(x, y, 2.0, 0.25, w, 'none') ``` -### 在昇腾 NPU 机器编译 mmcv-full +### 在昇腾 NPU 机器编译 mmcv -在编译 mmcv-full 前，需要安装 torch_npu，完整安装教程详见 [PyTorch 安装指南](https://gitee.com/ascend/pytorch/blob/master/docs/zh/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97.md#pytorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97) +在编译 mmcv 前，需要安装 torch_npu，完整安装教程详见 [PyTorch 安装指南](https://gitee.com/ascend/pytorch/blob/master/docs/zh/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97.md#pytorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97) -#### 选项 1: 使用 pip 安装 Ascend 编译版本的 mmcv-full +#### 选项 1: 使用 NPU 设备源码编译安装 mmcv (推荐方式) -Ascend 编译版本的 mmcv-full 在 mmcv >= 1.7.0 时已经支持直接 pip 安装 +- 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv/tree/master) ```bash -pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/ascend/torch1.8.0/index.html +git pull https://github.com/open-mmlab/mmcv/tree/master ``` -#### 选项 2: 使用 NPU 设备源码编译安装 mmcv-full +#### 选项 2: 使用 pip 安装 Ascend 编译版本的 mmcv -- 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv/tree/master) +Ascend 编译版本的 mmcv-lite 在 mmcv >= 1.7.0 时已经支持直接 pip 安装 ```bash -git pull https://github.com/open-mmlab/mmcv/tree/master +pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/ascend/torch1.8.0/index.html ``` - 编译 diff --git a/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp b/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp index f428311fee..a9191daccd 100644 --- a/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp +++ b/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp @@ -11,19 +11,38 @@ void roi_pool_forward_npu(Tensor input, Tensor rois, Tensor output, int64_t pooled_channel = 1; at::Tensor roi_actual_num = at_npu::native::OpPreparation::ApplyTensor( {}, rois.options().dtype(at::kInt), rois); - OpCommand cmd; - cmd.Name("RoiPoolingWithArgMax") - .Input(input) - .Input(rois) - .Input(roi_actual_num) - .Output(output) - .Output(argmax) - .Attr("pooled_h", pooled_height_64) - .Attr("pooled_w", pooled_width_64) - .Attr("spatial_scale_h", spatial_scale) - .Attr("spatial_scale_w", spatial_scale) - .Attr("pool_channel", pooled_channel) - .Run(); + if (input.sizes()[1] % 16 == 0) { + OpCommand cmd; + cmd.Name("RoiPoolingWithArgMax") + .Input(input) + .Input(rois) + .Input(roi_actual_num) + .Output(output) + .Output(argmax) + .Attr("pooled_h", pooled_height_64) + .Attr("pooled_w", pooled_width_64) + .Attr("spatial_scale_h", spatial_scale) + .Attr("spatial_scale_w", spatial_scale) + .Attr("pool_channel", pooled_channel) + .Run(); + + } else { + OpCommand cmd; + cmd.Name("RoiPoolingWithArgMax") + .Input(input) + .Input(rois) + .Input(roi_actual_num) + .Output(output) + .Output(argmax) + .Attr("pooled_h", pooled_height_64) + .Attr("pooled_w", pooled_width_64) + .Attr("spatial_scale_h", spatial_scale) + .Attr("spatial_scale_w", spatial_scale) + .Attr("pool_channel", pooled_channel) + .Attr("_exclude_engines", (string) "AiCore") + .Run(); + } + } void roi_pool_backward_npu(Tensor grad_output, Tensor rois, Tensor argmax, From aa5a76ee03dd3a1026fd3fe4390be45d2bfb8d6f Mon Sep 17 00:00:00 2001 From: momo609 <963372609@qq.com> Date: Thu, 8 Jun 2023 13:05:09 +0800 Subject: [PATCH 3/4] update build.md --- docs/zh_cn/get_started/build.md | 14 +++----------- 1 file changed, 3 insertions(+), 11 deletions(-) diff --git a/docs/zh_cn/get_started/build.md b/docs/zh_cn/get_started/build.md index 9d0ba3ab78..a2e6e7302b 100644 --- a/docs/zh_cn/get_started/build.md +++ b/docs/zh_cn/get_started/build.md @@ -359,20 +359,12 @@ output = sigmoid_focal_loss(x, y, 2.0, 0.25, w, 'none') 在编译 mmcv 前，需要安装 torch_npu，完整安装教程详见 [PyTorch 安装指南](https://gitee.com/ascend/pytorch/blob/master/docs/zh/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97.md#pytorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97) -#### 选项 1: 使用 NPU 设备源码编译安装 mmcv (推荐方式) +#### 使用 NPU 设备源码编译安装 mmcv -- 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv/tree/master) +- 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv.git) ```bash -git pull https://github.com/open-mmlab/mmcv/tree/master -``` - -#### 选项 2: 使用 pip 安装 Ascend 编译版本的 mmcv - -Ascend 编译版本的 mmcv-lite 在 mmcv >= 1.7.0 时已经支持直接 pip 安装 - -```bash -pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/ascend/torch1.8.0/index.html +git pull https://github.com/open-mmlab/mmcv.git ``` - 编译 From 433b4c80cc08de0151568ae21aaabe88ad1d0f5e Mon Sep 17 00:00:00 2001 From: momo609 <963372609@qq.com> Date: Thu, 8 Jun 2023 19:44:15 +0800 Subject: [PATCH 4/4] update build.md --- docs/zh_cn/get_started/build.md | 10 +++++++++- mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp | 1 - 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/docs/zh_cn/get_started/build.md b/docs/zh_cn/get_started/build.md index a2e6e7302b..8c2ad15d64 100644 --- a/docs/zh_cn/get_started/build.md +++ b/docs/zh_cn/get_started/build.md @@ -359,7 +359,7 @@ output = sigmoid_focal_loss(x, y, 2.0, 0.25, w, 'none') 在编译 mmcv 前，需要安装 torch_npu，完整安装教程详见 [PyTorch 安装指南](https://gitee.com/ascend/pytorch/blob/master/docs/zh/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97/PyTorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97.md#pytorch%E5%AE%89%E8%A3%85%E6%8C%87%E5%8D%97) -#### 使用 NPU 设备源码编译安装 mmcv +#### 选项 1: 使用 NPU 设备源码编译安装 mmcv (推荐方式) - 拉取 [MMCV 源码](https://github.com/open-mmlab/mmcv.git) @@ -379,6 +379,14 @@ MMCV_WITH_OPS=1 MAX_JOBS=8 FORCE_NPU=1 python setup.py build_ext MMCV_WITH_OPS=1 FORCE_NPU=1 python setup.py develop ``` +#### 选项 2: 使用 pip 安装 Ascend 编译版本的 mmcv + +Ascend 编译版本的 mmcv 在 mmcv >= 1.7.0 时已经支持直接 pip 安装 + +```bash +pip install mmcv -f https://download.openmmlab.com/mmcv/dist/ascend/torch1.8.0/index.html +``` + #### 验证 ```python diff --git a/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp b/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp index a9191daccd..75f39caaa8 100644 --- a/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp +++ b/mmcv/ops/csrc/pytorch/npu/roi_pool_npu.cpp @@ -42,7 +42,6 @@ void roi_pool_forward_npu(Tensor input, Tensor rois, Tensor output, .Attr("_exclude_engines", (string) "AiCore") .Run(); } - } void roi_pool_backward_npu(Tensor grad_output, Tensor rois, Tensor argmax,