sandsifter VS riscv-boom

This has some background.

sandsifter

This is a really clever technique! I was impressed by sandsifter[1] when it originally came out, and this seems an awful lot faster and less prone to false negatives (since it's purely speculative and doesn't require sandsifter's `#PF` hack).

At the risk of unwarranted self-promotion: the other side of this equation is fidelity in software instruction set decoders. x86's massive size and layers of historical complexity make it among the most difficult instruction formats to accurately decode; I've spent a good part of the last two years working on a fuzzer that's discovered thousands of bugs in various popular x86 decoders[2][3].

[1]: https://github.com/xoreaxeaxeax/sandsifter

[2]: https://github.com/trailofbits/mishegos

[3]: https://ww.easychair.org/publications/preprint_download/1LHr

Idea:

If any assembler/disassembler author/team out there wants to produce an assembler/disassembler which is authoritative (difficult to do on x86, because there are so many different possible combinations of instruction encoding, https://github.com/xoreaxeaxeax/sandsifter : "Typically, several million undocumented instructions on your processor will be found, but these generally fall into a small number of different groups.") -- then what they'd do is to create a third program -- which "pits" the output of Assembler A vs. Assembler B, Disassembler A vs. Disassembler B...

That is, between any two assemblers (for the same CPU architecture/instruction set), or any two disassemblers, where are the anomalies?

If we think about an assembler as a simple function, y=f(x), that is, I give it a string of ascii bytes as input (x), and I get a string (1..n) binary bytes as output (y),

Looks like from Appendix D that only 2 bugs were found in BOOM:

> 1. Inaccurate instruction count when minstret is written by software

I don't know what that means, but having minstret written by software was definitely not something I ever tested. In general, perf counters are likely to be undertested.

> 2. Static rounding is ignored for fdiv.s and fsqrt.s

A mistake was made in only listening to the dynamic rounding mode for the fdiv/sqrt unit. This is one of those bugs that is trivially found if you test for it, but it turns out that no benchmarking ever cared about this and from all of the fuzzers I used when I worked on BOOM, NONE of them hit it (including commercial ones...). Ooops.

Fixed here: https://github.com/riscv-boom/riscv-boom/pull/629/files

The two most micro architecturally advanced cores that I know of are BOOM, an out of order RV64GC core with all the features you expect plus sort of weird fancy things like short forward branch predication, and VROOM, another out of order RV64GC core with things like uop fusion and a trace cache.

it is used in the Berkley Out-of-Order RISC-V processor: https://github.com/riscv-boom/riscv-boom

SonicBOOM: https://github.com/riscv-boom/riscv-boom

Some cores are open source and you can see for yourself.

Rename logic from BOOM, a RISC-V core written in a DSL embedded in Scala:

https://github.com/riscv-boom/riscv-boom/blob/1ef2bc6f6c98e5...

From RSD, a core designed for FPGAs written in SystemVerilog:

https://github.com/rsd-devel/rsd/blob/master/Processor/Src/R...

And then there's the recently open-sourced XuanTie C910, which contains this Verilog… which is completely unreadable. Seems like it was produced by some kind of code generator that they didn't open-source?

https://github.com/T-head-Semi/openc910/blob/d4a3b947ec9bb8f...

We don't have Sifive's specifically but we do have the open source cores they've historically used to design their cores: https://github.com/riscv-boom/riscv-boom https://github.com/chipsalliance/rocket-chip

If you look at the decoder (https://github.com/riscv-boom/riscv-boom/blob/master/src/main/scala/exu/decode.scala), you can see that the fence instructions are also marked as "unique" instructions. Only one "unique" instruction is allowed in the pipeline at a time.