diff options
author | Gunes Bayir <gunes.bayir@arm.com> | 2024-07-02 15:45:01 +0100 |
---|---|---|
committer | Gunes Bayir <gunes.bayir@arm.com> | 2024-07-02 16:00:11 +0000 |
commit | a3f238a44d9f306c77be0177f13d22ae3f3bcc57 (patch) | |
tree | 44bf40fb59fb8c4452d65d25e3a967c035bc6863 /src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp | |
parent | f92b0fffa0d32dc08340c1abfa1a7f09c6e53795 (diff) | |
download | ComputeLibrary-a3f238a44d9f306c77be0177f13d22ae3f3bcc57.tar.gz |
Revert "Update CPU kernels and add mixed sign GEMM support"
This reverts commit fc94f4d23abd4bc427b701f54ad85282e9ec7872 and 5d6fff041ade7eb44af0945867212f3979be3d3e (because the latter fixes a build failure caused by the former)
Change-Id: I7d07fea8307e9a7033b30874bbb14ba9202b23d8
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11815
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Adnan AlSinan <adnan.alsinan@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp | 17 |
1 files changed, 6 insertions, 11 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp b/src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp index a553f1be9e..f7e108ac16 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp +++ b/src/core/NEON/kernels/arm_conv/pooling/depthfirst_driver.hpp @@ -64,10 +64,10 @@ class DepthfirstDriver : public PoolingCommon<TInput, TOutput> std::unique_ptr<const IDepthfirstStrategy> m_strat; /* Compute the amount of working space required for a single thread. */ - virtual size_t get_working_size_per_thread(unsigned int) const = 0; + virtual size_t get_working_size_per_thread() const = 0; /* Initialise the working space for a thread. */ - virtual void initialise_working_space(void *, unsigned int) const = 0; + virtual void initialise_working_space(void *) const = 0; /* Compute a portion of the output tensor with padding. */ virtual void compute_tile_padded( @@ -148,8 +148,8 @@ class DepthfirstDriver : public PoolingCommon<TInput, TOutput> { // Get and initialise the working space for this thread. void *thread_working_space = - static_cast<uint8_t *>(working_space) + thread_id * this->get_working_size_per_thread(n_channels); - this->initialise_working_space(thread_working_space, n_channels); + static_cast<uint8_t *>(working_space) + thread_id * this->get_working_size_per_thread(); + this->initialise_working_space(thread_working_space); // Construct convenient representations of the input/output tensors. TensorSpec<const TInput *> input_tensor(reinterpret_cast<const TInput *>(input), ld_input_row, ld_input_col); @@ -289,14 +289,9 @@ class DepthfirstDriver : public PoolingCommon<TInput, TOutput> { } - size_t get_working_size(unsigned int n_threads) const override + size_t get_working_size(unsigned int n_threads) const override final { - return this->get_working_size(n_threads, this->m_args.n_channels); - } - - size_t get_working_size(unsigned int n_threads, unsigned int n_channels) const override final - { - return n_threads * this->get_working_size_per_thread(n_channels); + return n_threads * this->get_working_size_per_thread(); } }; |