-
Notifications
You must be signed in to change notification settings - Fork 5.8k
[XPU] update xhpc to support VL model pretraining and inference #75870
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
Merged
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit
Hold shift + click to select a range
2514d73
update XPU ReLU grad callers to the new API signature (#75321)
bigwhite37 5f8487f
[XPU] Update XHPC to 20250828 (#74978)
xinruiM 50a6707
[XPU] support index_elementwise_get kernel (#75486)
cqulilujia acefdcb
[XPU] Update XHPC to 20251014 and add some dim check in FlashAttnKern...
ZibinGuo File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
[XPU] support index_elementwise_get kernel (#75486)
- Loading branch information
commit 50a67078e97a88e222767d96241e171eb2bc8efa
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
171 changes: 171 additions & 0 deletions
paddle/phi/kernels/xpu/index_elementwise_get_kernel.cc
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,171 @@ | ||
| // Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved. | ||
| // | ||
| // Licensed under the Apache License, Version 2.0 (the "License"); | ||
| // you may not use this file except in compliance with the License. | ||
| // You may obtain a copy of the License at | ||
| // | ||
| // http://www.apache.org/licenses/LICENSE-2.0 | ||
| // | ||
| // Unless required by applicable law or agreed to in writing, software | ||
| // distributed under the License is distributed on an "AS IS" BASIS, | ||
| // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| // See the License for the specific language governing permissions and | ||
| // limitations under the License. | ||
|
|
||
| #include "paddle/phi/kernels/index_elementwise_get_kernel.h" | ||
|
|
||
| #include "paddle/phi/backends/xpu/xpu_context.h" | ||
| #include "paddle/phi/core/kernel_registry.h" | ||
| #include "paddle/phi/kernels/funcs/index_elementwise.h" | ||
| #include "paddle/phi/kernels/funcs/stride_utils.h" | ||
|
|
||
| namespace phi { | ||
| template <typename T, typename Context, typename IndexT = int> | ||
| void XPUIndexElementwiseGetKernel(const Context& dev_ctx, | ||
| const DenseTensor& input, | ||
| const std::vector<const DenseTensor*>& index, | ||
| const std::vector<int64_t>& input_dims, | ||
| const std::vector<int64_t>& input_strides, | ||
| const std::vector<int64_t>& index_dims, | ||
| const std::vector<int64_t>& index_strides, | ||
| const int64_t slice_offset, | ||
| DenseTensor* output) { | ||
| int64_t numel = 0; | ||
| int64_t num_indices = 0; | ||
| std::vector<int64_t> shape_tmp; | ||
| std::vector<int64_t> stride_tmp; | ||
| funcs::cal_shape_stride(index_dims, &num_indices, &shape_tmp, &stride_tmp); | ||
|
|
||
| auto sizes = std::array<int64_t, DDim::kMaxRank>{}; | ||
| auto strides = std::array<int64_t, DDim::kMaxRank>{}; | ||
| for (int64_t i = 0; i < num_indices; i++) { | ||
| sizes[i] = index_dims[i]; | ||
| strides[i] = index_strides[i]; | ||
| } | ||
| std::array<int64_t*, 3> strides_array; | ||
| std::vector<int64_t> desired_shape; | ||
| std::array<std::vector<int64_t>, 3> strides_vec; | ||
| funcs::IndexGetStride<3>(input_dims, | ||
| input_strides, | ||
| phi::SizeOf(input.dtype()), | ||
| std::vector<int64_t>(), | ||
| std::vector<int64_t>(), | ||
| phi::SizeOf(input.dtype()), | ||
| shape_tmp, | ||
| stride_tmp, | ||
| phi::SizeOf(index[0]->dtype()), | ||
| &desired_shape, | ||
| &strides_array, | ||
| &numel, | ||
| strides_vec); | ||
| const int64_t N = output->numel(); | ||
| PADDLE_ENFORCE_GE( | ||
| N, 0, common::errors::InvalidArgument("Output numel must >= 0")); | ||
| PADDLE_ENFORCE_LE( | ||
| N, | ||
| std::numeric_limits<int32_t>::max(), | ||
| common::errors::InvalidArgument("Output numel must <= INT32_MAX")); | ||
|
|
||
| dev_ctx.template Alloc<T>(output); | ||
| using XPUType = typename XPUTypeTrait<T>::Type; | ||
| using XPUTypeIndexT = typename XPUTypeTrait<IndexT>::Type; | ||
|
|
||
| // passed vector params for XPU | ||
| std::vector<const XPUTypeIndexT*> index_ptrs_vec; | ||
| std::vector<int64_t> index_numel_vec; | ||
| for (int i = 0; i < num_indices; i++) { | ||
| // since XPU WRAPPER_CHECK_PTR only supports original GM ptrs, so we pass | ||
| // the IndexT* type ptrs, which is different from the CPU/GPU's char* ptr. | ||
| index_ptrs_vec.push_back( | ||
| reinterpret_cast<const XPUTypeIndexT*>(index[i]->data<IndexT>())); | ||
| // index_numel_vec is for the length of WRAPPER_CHECK_PTR | ||
| index_numel_vec.push_back(index[i]->numel()); | ||
| } | ||
| std::vector<int64_t> sizes_vec = | ||
| std::vector<int64_t>(sizes.begin(), sizes.begin() + num_indices); | ||
| std::vector<int64_t> orig_strides_vec = | ||
| std::vector<int64_t>(strides.begin(), strides.begin() + num_indices); | ||
| std::vector<std::vector<int64_t>> strides_vec_vec = | ||
| std::vector<std::vector<int64_t>>(strides_vec.begin(), strides_vec.end()); | ||
|
|
||
| const char* in_ptr = | ||
| reinterpret_cast<const char*>(input.data<T>()) + slice_offset; | ||
| char* out_ptr = reinterpret_cast<char*>(output->data<T>()); | ||
|
|
||
| // for checkptr and checksum in XPU | ||
| int64_t data_size_in = input.Holder()->size() - input.meta().offset; | ||
| int64_t data_size_out = output->Holder()->size() - output->meta().offset; | ||
|
|
||
| bool is_get = true; | ||
| int r = xpu::index_elementwise_tensor<XPUType, XPUTypeIndexT>( | ||
| dev_ctx.x_context(), | ||
| reinterpret_cast<const XPUType*>(in_ptr), // XPU ptr | ||
| reinterpret_cast<XPUType*>(out_ptr), // XPU ptr | ||
| index_ptrs_vec, // vec of XPU ptrs | ||
| index_numel_vec, // CPU vec | ||
| desired_shape, // CPU vec | ||
| sizes_vec, // CPU vec | ||
| orig_strides_vec, // CPU vec | ||
| strides_vec_vec, // CPU vec | ||
| N, // int64_t | ||
| data_size_in, // int64_t | ||
| data_size_out, // int64_t | ||
| is_get); // true for get, false for put | ||
| PADDLE_ENFORCE_XDNN_SUCCESS(r, "index_elementwise_tensor_get"); | ||
| } | ||
|
|
||
| template <typename T, typename Context> | ||
| void IndexElementwiseGetKernel(const Context& dev_ctx, | ||
| const DenseTensor& x, | ||
| const std::vector<const DenseTensor*>& index, | ||
| const std::vector<int64_t>& input_dims, | ||
| const std::vector<int64_t>& input_strides, | ||
| const std::vector<int64_t>& index_dims, | ||
| const std::vector<int64_t>& index_strides, | ||
| const int64_t slice_offset, | ||
| const bool accumulate, | ||
| const bool is_combined, | ||
| DenseTensor* out) { | ||
| const auto& index_type = index[0]->dtype(); | ||
| PADDLE_ENFORCE_EQ(index_type == phi::DataType::INT64, | ||
| true, | ||
| common::errors::InvalidArgument( | ||
| "Index holds the wrong type, it holds [%s], but " | ||
| "desires to be [%s].", | ||
| index_type, | ||
| phi::DataType::INT64)); | ||
|
|
||
| auto out_dims = out->dims(); | ||
| if (out_dims.size() > 0) { | ||
| std::vector<int64_t> output_dims(input_dims); | ||
| out->Resize(phi::make_ddim(output_dims)); | ||
| } | ||
| dev_ctx.template Alloc<T>(out); | ||
| if (out->numel() == 0) return; | ||
| XPUIndexElementwiseGetKernel<T, Context, int64_t>(dev_ctx, | ||
| x, | ||
| index, | ||
| input_dims, | ||
| input_strides, | ||
| index_dims, | ||
| index_strides, | ||
| slice_offset, | ||
| out); | ||
| } | ||
|
|
||
| } // namespace phi | ||
|
|
||
| PD_REGISTER_KERNEL(index_elementwise_get, | ||
| XPU, | ||
| ALL_LAYOUT, | ||
| phi::IndexElementwiseGetKernel, | ||
| bool, | ||
| float, | ||
| double, | ||
| int, | ||
| int8_t, | ||
| int64_t, | ||
| int16_t, | ||
| uint8_t, | ||
| phi::float16, | ||
| phi::bfloat16) {} |
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.