[和| Alibaba AL.Labs 阿里巴巴人工智能实验室 AiILabs & TVM PART 1 : ARM32 CPU CONTENT PART 2 : HIFI4 DSP PART 3 : _ PowervVR GPU [和| Alibaba AL.Labs 阿里巴巴人工智能实验室 ARM 32 CPU Resolution Quantization Orize Resolution 1. GEMM Tensorize (10x speed up) 2. HIFI4 Program (don't need dlopen) Serial Communication HIFI4 DSP HIFI4 DSP HIFI4 DSP [和| Alibaba AL.Labs 阿里巴巴人工智能实验室 PowerVR

PRESENTATION AGENDA 人 人 e 人 e@ TVM Q@ AliOs Overview TVM @ AliOs ARM CPU TVM @ AliOos Hexagon DSP TVM @ Alios Intel GPU Misc /NiiOS ! 驱动万物智能 PART ONE TVM Q@ AliOs Overview AiOS 1驱动万物智能 AliOs overview TFLite Quantized Support 1.61X MobilenetVl TFlite TV Hexagon DSP | 1.27X MobilenetV1 upport 1.12X Mobilenet V1 Multimodal Interection CPU (ARM、Intel) 1驱动万物智能 Accelerated Op Library / Others Inference Engine DSP (Qualcomm) PART TWO Alios TVM @ ARM CPU AiOS 1驱动万物智能 Alios TVMQOARM CPU 。 Support TFLite ( Open

TSANWrite tests!Moving up the testing pyramid Unit Tests DSP Regression VST3 Unit Tests Integration E2EMoving up the testing pyramid Unit Tests DSP Regression VST3Changing legacy APIs should you?Breaking Legacy CodeThe desire to standardizeWe have built the same things… ● Lots of similar and unique DSP algorithms ● 7 UI technologies ● 4 installer technologies ● etc….47 xkcd.com/2030/Discover your hidden

Commons Attribution 4.0 International License SYCL Single Source C++ Parallel Programming GPU FPGA DSP Custom Hardware GPU CPU CPU CPU Standard C++ Application Code C++ Libraries ML Frameworks can give better performance on complex apps and libs than hand-coding AI/Tensor HW GPU FPGA DSP Custom Hardware GPU CPU CPU CPU AI/Tensor HW Other BackendsSYCL 2020 is here! Open Standard C++ Application Code SYCL in Embedded Systems, Automotive, and AI Compilation Ingestion FPGA DSP Dedicated Hardware GPU Vision / Inferencing Engine Compiled Code Hardware Acceleration APIs

hikey960 (mali g71) N/A FPGA vta pynq, ultra96 N/A sdaccel Out-of-tree support or WIP: Hexagon DSP (via llvm), Ascend NPU, and more Green: Linaro 96BoardsLinaro for TVM ● Linaro AI/ML group can be organized way ○ Arm - Cortex-A/Cortex-M/Neoverse CPU, Mali GPU, Ethos NPU ○ Qualcomm - Hexagon DSP, Adreno GPU ○ Hisilicon, Xilinx, NXP, TI, ST, Fujitsu, Riken, and etc ● Collaborations between Arm

Circuit Simulation with Wave Digital Filters in C++ Author: Jatin Chowdhury Affiliation: Chowdhury DSP E-Mail: jatin@chowdsp.com Website: chowdsp.com CppCon 2024, Aurora, CO, USA, September 15-20, 2024 Topologies and Multiple Nonlinearities", M. Rest, et. al, DAFx-19, 2016 https://github.com/Chowdhury-DSP/chowdsp_wdf https://arxiv.org/pdf/2210.12554 Performance Comparison: How much time (seconds) is required

looks in the environment variable AUDIODEV for a device to use. If not found, it falls back to /dev/dsp. mode is one of 'r' for read-only (record) access, 'w' for write-only (playback) access and 'rw' for The correspondence is obvious: for example, setfmt() corresponds to the SNDCTL_DSP_SETFMT ioctl, and sync() to SNDCTL_DSP_SYNC (this can be useful when consulting the OSS documentation). If the underlying For example, (fmt, channels, rate) = dsp.setparameters(fmt, channels, rate) is equivalent to fmt = dsp.setfmt(fmt) channels = dsp.channels(channels) rate = dsp.rate(rate) oss_audio_device.bufsize()

looks in the environment variable AUDIODEV for a device to use. If not found, it falls back to /dev/dsp. mode is one of 'r' for read-only (record) access, 'w' for write-only (playback) access and 'rw' for The correspondence is obvious: for example, setfmt() corresponds to the SNDCTL_DSP_SETFMT ioctl, and sync() to SNDCTL_DSP_SYNC (this can be useful when consulting the OSS documentation). If the underlying For example, (fmt, channels, rate) = dsp.setparameters(fmt, channels, rate) is equivalent to fmt = dsp.setfmt(fmt) channels = dsp.channels(channels) rate = dsp.rate(rate) oss_audio_device.bufsize()

looks in the environment variable AUDIODEV for a device to use. If not found, it falls back to /dev/dsp. mode is one of 'r' for read-only (record) access, 'w' for write-only (playback) access and 'rw' for The correspondence is obvious: for example, setfmt() corresponds to the SNDCTL_DSP_SETFMT ioctl, and sync() to SNDCTL_DSP_SYNC (this can be useful when consulting the OSS documentation). If the underlying For example, (fmt, channels, rate) = dsp.setparameters(fmt, channels, rate) is equivalent to fmt = dsp.setfmt(fmt) channels = dsp.channels(channels) rate = dsp.rate(rate) oss_audio_device.bufsize()

looks in the environment variable AUDIODEV for a device to use. If not found, it falls back to /dev/dsp. mode is one of 'r' for read-only (record) access, 'w' for write-only (playback) access and 'rw' for The correspondence is obvious: for example, setfmt() corresponds to the SNDCTL_DSP_SETFMT ioctl, and sync() to SNDCTL_DSP_SYNC (this can be useful when consulting the OSS documentation). If the underlying For example, (fmt, channels, rate) = dsp.setparameters(fmt, channels, rate) is equivalent to fmt = dsp.setfmt(fmt) channels = dsp.channels(channels) rate = dsp.rate(rate) oss_audio_device.bufsize()