Add a Specific Implementation for RISC-V ISA AI Extensions
Opened this issue · 16 comments
ML2CPP can generate specific C++ code with RISC-V extensions allowing all scikit-learn, pytorch, caret ML models to be deployable natively on this platform.
RISC-V extensions have the advantage to be non-proprietary.
This is a place holder for following the recent efforts on designing RISC-V specific extensions for AI/ML.
No public hardware is available yet.
Sifive Intelligence X280 CPU is an interesting start point. https://www.sifive.com/blog/introduction-to-the-sifive-intelligence-x280
First SBCs with Sifive CPU hit the market (No AI Extensions ;)
SiFive Tech Talk on Accelerating AI: Past, Present, and Future
by Krste Asanovic
https://www.youtube.com/watch?v=8n2HLp2gtYs
“V” Vector Extension Proposal
Krste Asanovic, UC Berkeley & SiFive
RISC-V Workshop Barcelona May 7th 2018
Vector ISA
Presentation by Roger Espasa at Esperanto Technologies on May 7, 2018 at the RISC-V Workshop in Barcelona, hosted by Barcelona Supercomputing Center and Universitat Politecnica de Catalunya in Barcelona, Spain.
Try using c++ Standard Data-parallel vector library when available (Allow all SIMD implementations : MMX, SSE, AVX-512, ARM Neon, IBM, ...)
https://en.cppreference.com/w/cpp/experimental/simd
https://github.com/VcDevel/std-simd
https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=libstdc%2B%2B-v3/include/experimental/simd;hb=HEAD
antoine@z600:~$ vdir /usr/include/c++/11/experimental/bits/
fs_dir.h lfts_config.h simd_builtin.h simd.h
simd_scalar.h fs_fwd.h net.h simd_converter.h
simd_math.h simd_x86_conversions.h fs_ops.h numeric_traits.h
simd_detail.h simd_neon.h simd_x86.h fs_path.h
shared_ptr.h simd_fixed_size.h simd_ppc.h string_view.tcc
SiFive P670 and P470 RISC-V processors feature RISC-V Vector Extensions
RISC-V has standardized a series of standard extensions beyond the integer base instructions which can be implemented or omitted as desired depending on the design goals (e.g. energy/area/performance/storage goals).
https://stackoverflow.com/questions/47405717/dot-product-of-vectors-with-simd
May be ml2cpp already generates code that a smart compiler can vectorize automatically with SIMD instructions. This is also true for all hardware architectures.
ml2cpp only generated source code, is not responsible of the environment this code will be compiled/used in. PERIOD.