[RFC] Improve quantized convolution performance for armv8 architectures (#5754)
* Improve quantized conv2d performance for armv8
Signed-off-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Change-Id: I3a3d29f5332dd9b3354e8e0dfb24677a521f9c8f
* Add ASF header to conv2d_gemm.py
Change-Id: I33853279e39c849ae1b555a9c91d7557985a0a35
* Run clang-format-10 on c++ files
Change-Id: Ieee22f032e595dabfc1616ab33466fcbf8d94365
* Fix pylint errors/warnings
Change-Id: I435d4d7bca7500db99547f4401fdc0d0995a1ff4
* Fix pylint errors/warnings in topi
Change-Id: I2fc1ad8453e9020072ab967c849df5390c2967b5
* Fix legalizations tests for aarch64
Change-Id: I0a67a49a7849f52ef7d57b9292ce9125bbb7cb2c
* Reintroduce conv2d_nhwc_spatial_pack.arm_cpu and int16 cast
Change-Id: I91b67fabd475e90a9b75f2dd5ecfee851265e0bb
* Switch type of legalization depending on the strategy used
Change-Id: I9a03040a8c40a6cd2658ed14c3751e05a8e19f2b
* Revert last commit
Change-Id: Ice34101e358e3ce8ebfb12c58f73e910ba5de8e8
* Fix the auto-tuner by registering the correct schedules
Change-Id: Id9273688b2620e1ea849ab01b4c46af8fbf37fd0
* Address review comments
Change-Id: Ia1755a0af7b6d159072d9f0c93c932c481101e48
* Improve usability and readability of conv2d_gemm_weight_transform
Change-Id: I3333186bbc2fe4054b58ce15d910e3be7b315482
* Change variable name to weight in Conv2DGemmWeightTransformRel
Change-Id: Ifb5f1f33af7512fe67c6b049b20a42a0bb2d26c9
* Fix clang-10 linting errors
Change-Id: I25ccc844d9cee23766096e1daddb6180abc413a6
* Trigger tests
Change-Id: Id37706fb7cf77a87a3cc817ecf8046297d9ca95a
14 files changed: