Added HiFi optimized mean and where ops. #6483
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,170 @@ | ||
| /* | ||
| * Copyright (c) Meta Platforms, Inc. and affiliates. | ||
| * All rights reserved. | ||
| * | ||
| * This source code is licensed under the BSD-style license found in the | ||
| * LICENSE file in the root directory of this source tree. | ||
| */ | ||
|
|
||
| #include <executorch/kernels/portable/cpu/util/kernel_ops_util.h> | ||
| #include <executorch/kernels/portable/cpu/util/reduce_util.h> | ||
| #include <executorch/runtime/kernel/kernel_includes.h> | ||
| #include <executorch/runtime/platform/assert.h> | ||
|
|
||
| #include <executorch/backends/cadence/hifi/kernels/kernels.h> | ||
|
|
||
| using exec_aten::ScalarType; | ||
| using exec_aten::Tensor; | ||
| using executorch::aten::RuntimeContext; | ||
| using executorch::runtime::ArrayRef; | ||
| using torch::executor::Error; | ||
| using torch::executor::optional; | ||
|
|
||
| namespace cadence { | ||
| namespace impl { | ||
| namespace HiFi { | ||
| namespace native { | ||
|
|
||
| int prepare_data( | ||
| const Tensor& in, | ||
| Tensor& out, | ||
| optional<ArrayRef<int64_t>> dim_list, | ||
| int* inp_shape, | ||
| int* out_shape, | ||
| int* p_axis, | ||
|
Contributor
There was a problem hiding this comment. Nit: What is |
||
| int num_inp_dims, | ||
| int num_out_dims) { | ||
| for (int i = 0; i < num_inp_dims; i++) { | ||
| inp_shape[i] = in.size(i); | ||
| } | ||
|
|
||
| for (int i = 0; i < num_out_dims; i++) { | ||
| out_shape[i] = out.size(i); | ||
| } | ||
|
Comment on lines
+37
to
+43
Contributor
There was a problem hiding this comment. Nit: can we make this a helper function that accepts a const Tensor& and mutable int* |
||
|
|
||
| int num_axis_dims = 0; | ||
| for (const auto& d : dim_list.value()) { | ||
| if (d < 0) { | ||
| p_axis[num_axis_dims] = num_inp_dims + d; | ||
| num_axis_dims++; | ||
| } else { | ||
| p_axis[num_axis_dims] = d; | ||
| num_axis_dims++; | ||
| } | ||
| } | ||
|
|
||
| return num_axis_dims; | ||
| } | ||
|
|
||
| Tensor& mean_dim_out( | ||
| RuntimeContext& ctx, | ||
| const Tensor& in, | ||
| optional<ArrayRef<int64_t>> dim_list, | ||
| bool keepdim, | ||
| optional<ScalarType> dtype, | ||
| Tensor& out) { | ||
| ET_KERNEL_CHECK( | ||
| ctx, | ||
| torch::executor::check_mean_dim_args(in, dim_list, keepdim, dtype, out), | ||
| InvalidArgument, | ||
| out); | ||
|
|
||
| ET_KERNEL_CHECK( | ||
| ctx, | ||
| torch::executor::resize_reduction_out(in, dim_list, keepdim, out) == | ||
| Error::Ok, | ||
| InvalidArgument, | ||
| out); | ||
|
|
||
| constexpr auto name = "mean.out"; | ||
| constexpr int kNnlibMaxDim = 4; | ||
|
|
||
| bool optimized = 1; | ||
|
Contributor
There was a problem hiding this comment. Nit: Please use |
||
|
|
||
| if (out.scalar_type() != ScalarType::Float) | ||
| optimized = 0; | ||
|
|
||
| if (in.dim() > kNnlibMaxDim) | ||
| optimized = 0; | ||
|
|
||
| if (optimized) { | ||
| float* __restrict__ p_out = out.mutable_data_ptr<float>(); | ||
| const float* __restrict__ p_inp = | ||
| (const float* __restrict__)in.const_data_ptr<float>(); | ||
|
|
||
| int num_elm = in.numel(); | ||
|
|
||
| int num_inp_dims = in.dim(); | ||
| int num_out_dims = out.dim(); | ||
|
|
||
| int inp_shape[kNnlibMaxDim]; | ||
| int out_shape[kNnlibMaxDim]; | ||
| int p_axis[kNnlibMaxDim]; | ||
|
|
||
| for (int i = 0; i < kNnlibMaxDim; i++) { | ||
| out_shape[i] = 1; | ||
| inp_shape[i] = 1; | ||
| p_axis[i] = 1; | ||
| } | ||
|
|
||
| int num_axis_dims = prepare_data( | ||
| in, | ||
| out, | ||
| dim_list, | ||
| inp_shape, | ||
| out_shape, | ||
| p_axis, | ||
| num_inp_dims, | ||
| num_out_dims); | ||
|
|
||
| if (num_axis_dims == num_inp_dims) { | ||
| num_out_dims = 1; | ||
| out_shape[0] = 1; | ||
| } | ||
|
|
||
| int scratch_size = xa_nn_reduce_getsize_nhwc( | ||
| -3, inp_shape, num_inp_dims, p_axis, num_axis_dims, 1); | ||
|
|
||
| void* __restrict__ p_scratch_in = (void* __restrict__)malloc(scratch_size); | ||
|
Contributor
There was a problem hiding this comment. Let's use temporary allocator instead of malloc please. |
||
|
|
||
| xa_nn_reduce_mean_4D_f32_f32( | ||
|
Contributor
There was a problem hiding this comment. Let's add a check for the output of nnlib api. |
||
| p_out, | ||
| out_shape, | ||
| p_inp, | ||
| inp_shape, | ||
| p_axis, | ||
| num_out_dims, | ||
| num_inp_dims, | ||
| num_axis_dims, | ||
| p_scratch_in); | ||
|
|
||
| return out; | ||
| } | ||
|
Comment on lines
+90
to
+142
Contributor
There was a problem hiding this comment. Can we move this to a separate inline function? (like hifi_optimized_mean) |
||
|
|
||
| ET_SWITCH_REALHB_TYPES(in.scalar_type(), ctx, name, CTYPE_IN, [&] { | ||
| ET_SWITCH_FLOATH_TYPES(out.scalar_type(), ctx, name, CTYPE_OUT, [&] { | ||
| CTYPE_OUT* out_data = out.mutable_data_ptr<CTYPE_OUT>(); | ||
| const size_t num = torch::executor::get_reduced_dim_product(in, dim_list); | ||
|
|
||
| for (size_t out_ix = 0; out_ix < out.numel(); ++out_ix) { | ||
| CTYPE_OUT sum = 0; | ||
| if (in.numel() > 0) { | ||
| sum = torch::executor::map_reduce_over_dim_list<CTYPE_IN, CTYPE_OUT>( | ||
| [](CTYPE_IN v) { return static_cast<CTYPE_OUT>(v); }, | ||
| [](CTYPE_OUT outv, CTYPE_OUT acc) { return acc + outv; }, | ||
| in, | ||
| dim_list, | ||
| out_ix); | ||
| } | ||
| out_data[out_ix] = sum / static_cast<float>(num); | ||
| } | ||
| }); | ||
| }); | ||
|
|
||
| return out; | ||
| } | ||
|
|
||
| } // namespace native | ||
| } // namespace HiFi | ||
| } // namespace impl | ||
| } // namespace cadence | ||
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.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
const Tensor& out