feat: refactor asm2asm

[AsterDY]

What this PR for

Refactor asm2asm tool and golang-importing templates. We do this to mainly achieve two goals:

Support traceback when native C functions get panic or profiled

In the past, we generated Plan 9 assembly codes by asm2asm (https://github.com/chenzhuoyu/asm2asm) tools. But since Go ASM can't recognize many 'SP-writing' instructions like addq $xx, %rsp generated from clang, the imported c functions have incorrect pcsp funcdata and will fatally crash once panic or profiling happens inside C codes (even if it is recoverable). Thus, we need to calculate correct pcsp funcdata and import them into go runtime -- which is merely possible under the mechanism of Go ASM.

Better performance

On the other hand, Go ASM doesn't support register-based ABI yet (although it has been supported by pure go functions), which causes obviously performance decline of imported C functions (estimated about 5%~20%). If we can drop the denpency of Go ASM and import C functions in the form of pure machines, dynamically, we can gain a lot of flexibility and improvement on ABI conversion.

How doest it work

In compiling time, we refactor asm2asm to calculate and generate all the function information we need:

• c machine codes, stored in a Go byte slice;
• c function entries (both exported and unexported);
• c function data, at present includes pcsp、max stack size, and text size.

In initial time, we implement ABI convertor (loader/WrapGoC()) to generate conversion codes:

• check if Go stackguard is large enough for specific C function red zone + max stack size -- If it is not, call runtime.morstack();
• spill any pointer-kind arguments onto stack for Go GC tracing;
• exchange Go argument register (rax, rbx ...) with C argument register (rdi, rsi, ...), using stack slots if need (more than 3 arguements)
• load C func entry and call it (movq + call)

and register both C function and Go stub function. At present, we register every C functions as NoPreempt trait and reserve 32B red-zone for them, to avoid unexpected problems caused by different ABI

Benchmark

• Get/Set

name                      old time/op    new time/op    delta
SetOne_Sonic-16             3.35µs ± 7%    3.57µs ±15%     ~     (p=0.278 n=9+10)
GetOne_Sonic-16             1.69µs ±13%    1.35µs ± 2%  -20.26%  (p=0.000 n=9+10)

name                      old speed      new speed      delta
SetOne_Sonic-16           3.90GB/s ± 7%  3.68GB/s ±13%     ~     (p=0.278 n=9+10)
GetOne_Sonic-16           7.59GB/s ±17%  9.67GB/s ± 2%  +27.43%  (p=0.000 n=10+10)

name                      old alloc/op   new alloc/op   delta
GetOne_Sonic-16              24.0B ± 0%     24.0B ± 0%     ~     (all equal)
SetOne_Sonic-16             1.58kB ± 0%    1.58kB ± 0%     ~     (all equal)

name                      old allocs/op  new allocs/op  delta
GetOne_Sonic-16               1.00 ± 0%      1.00 ± 0%     ~     (all equal)
SetOne_Sonic-16               17.0 ± 0%      17.0 ± 0%     ~     (all equal)

• Encode

name                                    old time/op    new time/op    delta
Encoder_Generic_Sonic-16                  40.8µs ±16%    39.5µs ±10%     ~     (p=0.243 n=9+10)
Encoder_Binding_Sonic-16                  6.78µs ± 2%    6.79µs ± 3%     ~     (p=0.842 n=9+10)
Encoder_Binding_Sonic_Fast-16             5.84µs ± 2%    5.80µs ± 3%     ~     (p=0.353 n=10+10)
Encoder_Generic_Sonic_Fast-16             27.5µs ± 4%    25.9µs ± 2%   -5.99%  (p=0.000 n=10+10)

name                                    old speed      new speed      delta
Encoder_Generic_Sonic-16                 321MB/s ±14%   331MB/s ± 9%     ~     (p=0.243 n=9+10)
Encoder_Binding_Sonic-16                1.92GB/s ± 2%  1.92GB/s ± 3%     ~     (p=0.842 n=9+10)
Encoder_Binding_Sonic_Fast-16           2.23GB/s ± 2%  2.25GB/s ± 3%     ~     (p=0.353 n=10+10)
Encoder_Generic_Sonic_Fast-16            474MB/s ± 4%   503MB/s ± 2%   +6.29%  (p=0.000 n=10+10)

name                                    old alloc/op   new alloc/op   delta
Encoder_Binding_Sonic-16                  14.1kB ± 0%    14.1kB ± 0%   +0.23%  (p=0.000 n=10+10)
Encoder_Generic_Sonic-16                  13.8kB ± 0%    13.8kB ± 0%     ~     (p=0.154 n=9+9)
Encoder_Generic_Sonic_Fast-16             9.71kB ± 0%    9.69kB ± 0%     ~     (p=0.225 n=10+10)
Encoder_Binding_Sonic_Fast-16             9.83kB ± 0%    9.84kB ± 0%     ~     (p=0.239 n=10+10)

name                                    old allocs/op  new allocs/op  delta
Encoder_Generic_Sonic-16                    4.00 ± 0%      4.00 ± 0%     ~     (all equal)
Encoder_Generic_Sonic_Fast-16               4.00 ± 0%      4.00 ± 0%     ~     (all equal)
Encoder_Binding_Sonic-16                    4.00 ± 0%      4.00 ± 0%     ~     (all equal)
Encoder_Binding_Sonic_Fast-16               4.00 ± 0%      4.00 ± 0%     ~     (all equal)

• Decode

name                                    old time/op    new time/op    delta
Decoder_Binding_Sonic_Fast-16             35.0µs ± 4%    39.2µs ± 6%  +11.80%  (p=0.000 n=10+10)
Decoder_Binding_Sonic-16                  36.3µs ± 3%    39.2µs ± 4%   +8.03%  (p=0.000 n=10+10)
Decoder_Generic_Sonic-16                  91.6µs ± 6%    80.4µs ± 6%  -12.18%  (p=0.000 n=9+10)
Decoder_Generic_Sonic_Fast-16             69.5µs ±10%    59.2µs ± 4%  -14.91%  (p=0.000 n=10+10)

name                                    old speed      new speed      delta
Decoder_Binding_Sonic_Fast-16            372MB/s ± 4%   333MB/s ± 6%  -10.52%  (p=0.000 n=10+10)
Decoder_Binding_Sonic-16                 359MB/s ± 3%   332MB/s ± 4%   -7.43%  (p=0.000 n=10+10)
Decoder_Generic_Sonic-16                 143MB/s ± 7%   162MB/s ± 7%  +13.86%  (p=0.000 n=9+10)
Decoder_Generic_Sonic_Fast-16            189MB/s ±10%   220MB/s ± 4%  +16.85%  (p=0.000 n=10+10)

name                                    old alloc/op   new alloc/op   delta
Decoder_Generic_Sonic_Fast-16             49.0kB ± 0%    49.1kB ± 0%   +0.05%  (p=0.004 n=10+10)
Decoder_Generic_Sonic-16                  56.8kB ± 0%    56.8kB ± 0%     ~     (p=0.984 n=10+9)
Decoder_Binding_Sonic_Fast-16             24.3kB ± 0%    24.3kB ± 0%     ~     (p=0.149 n=10+9)
Decoder_Binding_Sonic-16                  27.3kB ± 0%    27.3kB ± 0%   -0.01%  (p=0.044 n=9+8)

name                                    old allocs/op  new allocs/op  delta
Decoder_Generic_Sonic-16                     723 ± 0%       723 ± 0%     ~     (all equal)
Decoder_Generic_Sonic_Fast-16                313 ± 0%       313 ± 0%     ~     (all equal)
Decoder_Binding_Sonic-16                     137 ± 0%       137 ± 0%     ~     (all equal)
Decoder_Binding_Sonic_Fast-16               34.0 ± 0%      34.0 ± 0%     ~     (all equal)

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (main@b40bbd5). Click here to learn what that means.
The diff coverage is n/a.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

@@           Coverage Diff           @@
##             main     #393   +/-   ##
=======================================
  Coverage        ?   78.01%           
=======================================
  Files           ?       62           
  Lines           ?    10419           
  Branches        ?        0           
=======================================
  Hits            ?     8128           
  Misses          ?     1943           
  Partials        ?      348           

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more