- 1988463 Rename [Cl|Cpu]GemmConvolution to [Cl|Gpu]GemmConv2d by Georgios Pinitas · 3 years ago
- 73df931 Address comments on avoiding releasing weights if used by multiple functions by Giorgio Arena · 3 years ago
- df5bcb6 Enable fast_math on CpuGemmConvolution by Georgios Pinitas · 2 years, 11 months ago
- ebbd529 Update the heuristic to call direct convolution in clConv2d by Gian Marco Iodice · 2 years, 11 months ago
- f5d51f3 Retain weights in ClGemm when reconfiguring the operator with retention by Georgios Pinitas · 2 years, 11 months ago
- 45fbf9f Avoid releasing weights if they are used by multiple functions by Georgios Pinitas · 3 years ago
- d4a5bc5 Ensure correct transformed matrices are used in CpuGemmConvolution by Georgios Pinitas · 3 years ago
- fa1db17 Ensure that correct transformed matrices are used in CpuFullyConnected by Georgios Pinitas · 3 years ago
- d761a3e Fix performance regression due to clFinish() by Gian Marco Iodice · 3 years ago
- 288d3cb Fix compiler error in CLActivationLayer by Pablo Marquez Tello · 3 years ago
- acb4d93 Fix compiler error in GCC 7.4 + Ubuntu 16 by Pablo Marquez Tello · 3 years ago
- 3f74389 Remove 21.08 deprecated code by Freddie Liardet · 3 years ago
- 0f6ca4b Report error for unsupported non-constant weights in CpuFullyConnected by Michele Di Giorgio · 3 years ago
- 6269f87 Fix depthwise convolution assembly kernels by Freddie Liardet · 3 years ago
- 6634194 Avoid over-allocation of temporary buffers within CpuWinogradConv2d by Georgios Pinitas · 3 years ago
- 4164814 Implement Operator API by Georgios Pinitas · 3 years ago
- 6e90c12 Add missing limits include by Freddie Liardet · 3 years ago
- ba2dd82 Benchmark and set default LWS for GEMM, Direct convolution and Winograd by Giorgio Arena · 3 years ago
- 06d1efd Port CLConvolutionLayer by Sheri Zhang · 3 years ago
- 529b5a2 Port ClFullyConnected to new API by Georgios Pinitas · 3 years ago
- a0efe69 Reintroduce implementation of NEConvolutionLayer::get_convolution_method by Michele Di Giorgio · 3 years ago
- 9a1e28b Compilation issue: neon=1 armv8.2 on Android with NDKr18beta1 by Gian Marco Iodice · 3 years ago
- a76e403 Fix A55 performance constant for fp16 hybrid gemm kernel by Georgios Pinitas · 3 years ago
- b55f8e8 Port NEConvolutionLayer by Michalis Spyrou · 3 years ago
- 8e2f64f Create custom flags for enabling fp16 support by Georgios Pinitas · 3 years ago
- a86433a Reduce binary footprint of CpuConvertFullyConnectedWeightsKernel by Michele Di Giorgio · 3 years ago
- 120c87b Fix bare metal build issues by Freddie Liardet · 3 years ago
- bb5d42c Fix cpu GEMM fp16 issue by Freddie Liardet · 3 years ago
- f81f51c Reorganize the kernels into nhwc, nchw and common folders by Adnan AlSinan · 3 years ago
- 97426a7 Remove generated kernels that overlap hand-written ones by Georgios Pinitas · 3 years ago
- 9805583 Fix memory lifetime issue by Georgios Pinitas · 3 years ago
- d87aded Port CLGEMMConvolutionLayer by Manuel Bottini · 3 years ago
- 4718706 Dispatch Conv2d using the Direct method when necessary by Georgios Pinitas · 3 years ago
- 387f80e Update GEMM assembly performance parameters by Georgios Pinitas · 3 years ago
- 5e61cb0 Add missing limits include by Freddie Liardet · 3 years ago
- e92c23e Fix allocation of prepare tensor on ClWinogradConv2d by Georgios Pinitas · 3 years ago
- 7075fe2 Reorganize the kernels into nhwc, nchw and common folders by Adnan AlSinan · 3 years ago
- 22f5ed5 Avoid allocation of auxiliary memory in CpuGemmConvolution by Georgios Pinitas · 3 years ago
- 8bd2556 Fix vector_length identification mechanism for SVE by Georgios Pinitas · 3 years ago
- d9cdf14 Port NEFullyConnectedLayer to memory injecting interface by Michele Di Giorgio · 3 years, 1 month ago
- a8297fb Pass fast math flag for correct GEMM3D validation support by Georgios Pinitas · 3 years ago
- e7012a6 Fix bare metal build error by Freddie Liardet · 3 years ago
- 69a9ac4 Expose fast_math mode for GEMM through BFloat16 by Georgios Pinitas · 3 years ago
- 1fa27ad Inject temporary tensors to pack in they don't exist in CpuSoftmax by Georgios Pinitas · 3 years ago
- f4e84fb Port ClGemmLowp to new API by Georgios Pinitas · 3 years ago
- 767dbf9 Fix oclgrind int overflow warning by Freddie Liardet · 3 years ago
- 4ee8b15 Update GEMM assembly kernels by Georgios Pinitas · 3 years ago
- 59fd7a7 Add layer data to JSON output by Freddie Liardet · 3 years, 1 month ago
- 0f033df Include limits header to fix errors reported by GCC11 by Michele Di Giorgio · 3 years ago
- 2b147ee Avoid multiple Rhs matrix transformation on ClGemm by Georgios Pinitas · 3 years ago
- d844c08 Port CLIm2ColKernel to ClIm2ColKernel by Manuel Bottini · 3 years ago
- 0b27133 Disabled DirectConv method for NCHW and kernel > 15 by Pablo Marquez Tello · 3 years ago
- 29599d0 Port NEGEMMConvolutionLayer by Manuel Bottini · 3 years ago
- 13ef176 Improve filelist for GPU by Georgios Pinitas · 3 years ago
- 7b23732 Port CLCol2ImKernel to ClCol2ImKernel by Manuel Bottini · 3 years ago
- c38ca38 Fix CL kernel compilation failure by Michalis Spyrou · 3 years ago
- a387e27 Add in-place calculation support for CL elementwise arithmetic kernels by Sheri Zhang · 3 years, 1 month ago
- 96f977e Port NEWinogradConvolutionLayer by Michalis Spyrou · 3 years, 1 month ago
- 04b39e8 Limit the LOOP_UNROLLING on the kernel height by Gian Marco Iodice · 3 years ago
- 24b8920 Port NECol2ImKernel by Manuel Bottini · 3 years, 1 month ago
- da81675 Remove redundant implementations of Add/Sub operators by Georgios Pinitas · 3 years, 1 month ago
- 4411e1f Change CLConvolution selection to run Direct approach on large kernels by Georgios Pinitas · 3 years, 1 month ago
- dd23f2a Add LayerData to all nodes by Freddie Liardet · 3 years, 1 month ago
- cfac51c Port NEGEMMLowp Part 2 by Manuel Bottini · 3 years, 1 month ago
- 06ac6e4 Add basic Operator interface by Georgios Pinitas · 3 years ago
- 4587594 Validate unsupported data types with runtime information by Georgios Pinitas · 3 years ago
- 5e28181 Fix manual LOOP_UNROLLING by Gian Marco Iodice · 3 years ago
- 9002899 Port NEIm2ColKernel by Manuel Bottini · 3 years, 1 month ago
- 5ee0d95 Improve implementation selection speed of CpuElementwiseUnary by Georgios Pinitas · 3 years ago
- 8155c02 Rework OpenCL Depthwise Convolution by Gian Marco Iodice · 3 years, 3 months ago
- 2eb5d16 Align kernel/operator header layout by Georgios Pinitas · 3 years, 1 month ago
- 4dfc553 Port NEGEMM to memory injecting interface (Part 3) by Michele Di Giorgio · 3 years, 1 month ago
- bc4e311 Implement FP GPU depthwise convolution 1x1 kernel for in-place computation by SiCongLi · 3 years, 1 month ago
- 6683165 Add quantization helper functions for OpenCL by Georgios Pinitas · 3 years, 1 month ago
- 2ef59b9 Reduce binary size footprint of CpuGemmInterleave4x4Kernel by Michele Di Giorgio · 3 years, 1 month ago
- f7f6450 Adjust minimum DDK version for manual unroll by Giorgio Arena · 3 years, 1 month ago
- 399f623 Port ClGemmLowpOutputStage operator to new interface by Georgios Pinitas · 3 years, 1 month ago
- c63b722 Revert "Rework OpenCL Depthwise Convolution" by Gian Marco Iodice · 3 years, 1 month ago
- 4a578b9 Port the ClGemmLowp kernels to the new API by Georgios Pinitas · 3 years, 1 month ago
- 53832b2 Port NEGEMM to memory injecting interface (Part 2) by Michele Di Giorgio · 3 years, 1 month ago
- 72b5687 Enable global pooling optimization on OpenCL by Gian Marco Iodice · 3 years, 1 month ago
- 93b75e0 Port NEGEMM to memory injecting interface (Part 1) by Michele Di Giorgio · 3 years, 1 month ago
- 5fdde99 Improve selection speed of CPU implementations by Georgios Pinitas · 3 years, 1 month ago
- 4a95bba Set up the framework to choose the default LWS by Giorgio Arena · 3 years, 1 month ago
- 731fe66 Simplify CpuInfo logic by Georgios Pinitas · 3 years, 1 month ago
- b1496e6 Add code to detect Mali(TM) G31 by Pablo Marquez Tello · 3 years, 1 month ago
- d7316eb Port NEGEMMConv2d to memory injecting interface by Michele Di Giorgio · 3 years, 1 month ago
- cd060c4 Rename pooling implementation folder to pool2d by Georgios Pinitas · 3 years, 1 month ago
- 5f6fdc1 Rework gemmlowp reshaped_only_rhs using the new macros by Giorgio Arena · 3 years, 1 month ago
- 561c176 Rework OpenCL Depthwise Convolution by Gian Marco Iodice · 3 years, 3 months ago
- 20fca52 Create core library using high priority operators by Michalis Spyrou · 3 years, 1 month ago
- 1d35927 Add in-place computation for elementwise operations by Sheri Zhang · 3 years, 1 month ago
- ae58bdf Port NEGEMMLowp Part 1 by Manuel Bottini · 3 years, 1 month ago
- 2db3a99 Fix Winograd heuristic in CLConvolutionLayer by Gian Marco Iodice · 3 years, 1 month ago
- ef5aac6 Add FP16 support to CLRemap by Freddie Liardet · 3 years, 1 month ago
- 8266ae5 Rework the CLConvolutionLayer heuristic by Gian Marco Iodice · 3 years, 1 month ago
- d02d5ed Integrate improved CPU depthwise convolution kernels by Michele Di Giorgio · 3 years, 6 months ago
- a341791 Remove implementation headers from NESoftmaxLayer public header by Georgios Pinitas · 3 years, 1 month ago
- 08302c1 Add CPU discovery capabilities. by Georgios Pinitas · 3 years, 1 month ago
- c6f4ec3 Port CLWinogradConvolutionLayer with ClWinogradConv2d by Manuel Bottini · 3 years, 2 months ago