LoongArch64: add lsx support

[Xiao-Tao]

Binutils(2.41) and GCC(14.1.0) complete LSX and LASX support
More details: https://github.com/loongson/build-tools/wiki
After using LSX, the performance is improved by 5-10%.

Some reference documents for loongarch64 architecture LSX：
LoongArch SX Vector Intrinsics： https://gcc.gnu.org/onlinedocs/gcc/LoongArch-SX-Vector-Intrinsics.html
Unofficial LoongArch Intrinsics Guide： https://jia.je/unofficial-loongarch-intrinsics-guide/

Currently, it has passed all tests in LoongArch architecture 2k2000(only support LSX), 3C5000, 3A6000 environment

LoongArch64 2K2000 machine benchmark test results

main:
Loading /home/zhoumt/ada/build/_deps/url-dataset-src/out.txt
2025-07-10T09:46:19+08:00
Running ./benchmarks/benchdata
Run on (2 X 1400 MHz CPU s)
CPU Caches:
  L1 Instruction 64 KiB (x2)
  L1 Data 64 KiB (x2)
  L2 Unified 2048 KiB (x1)
Load Average: 0.30, 0.20, 0.18
ada spec: Ada follows whatwg/url
bad urls: ---------------------
ada---count of bad URLs       26
servo/url---count of bad URLs 26
whatwg---count of bad URLs    26
-------------------------------

bytes/URL: 86.859205
curl : OMITTED
input bytes: 8688092
number of URLs: 100025
performance counters: Enabled
rust version : 1.82.0
zuri : OMITTED
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href             120769756 ns    120679607 ns            6 GHz=1.3967 cycle/byte=19.54 cycles/url=1.69723k instructions/byte=43.5421 instructions/cycle=2.22835 instructions/ns=3.11235 instructions/url=3.78203k ns/url=1.21517k speed=71.993M/s time/byte=13.8902ns time/url=1.20649us url/s=828.848k/s
BasicBench_AdaURL_aggregator_href   95760157 ns     95627690 ns            7 GHz=1.39657 cycle/byte=15.2875 cycles/url=1.32786k instructions/byte=34.3532 instructions/cycle=2.24714 instructions/ns=3.1383 instructions/url=2.98389k ns/url=950.801 speed=90.8533M/s time/byte=11.0068ns time/url=956.038ns url/s=1.04598M/s
BasicBench_AdaURL_CanParse          63078339 ns     63040431 ns           11 GHz=1.39484 cycle/byte=10.1341 cycles/url=880.236 instructions/byte=23.955 instructions/cycle=2.36382 instructions/ns=3.29715 instructions/url=2.08072k ns/url=631.065 speed=137.818M/s time/byte=7.25596ns time/url=630.247ns url/s=1.58668M/s


LSX:
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href             107859129 ns    107757937 ns            6 GHz=1.39661 cycle/byte=17.4756 cycles/url=1.51792k instructions/byte=38.1925 instructions/cycle=2.18548 instructions/ns=3.05226 instructions/url=3.31737k ns/url=1.08686k speed=80.626M/s time/byte=12.4029ns time/url=1.07731us url/s=928.238k/s
BasicBench_AdaURL_aggregator_href   83822883 ns     83799530 ns            8 GHz=1.39607 cycle/byte=13.4636 cycles/url=1.16944k instructions/byte=29.004 instructions/cycle=2.15425 instructions/ns=3.00749 instructions/url=2.51926k ns/url=837.663 speed=103.677M/s time/byte=9.64533ns time/url=837.786ns url/s=1.19362M/s
BasicBench_AdaURL_CanParse          51950801 ns     51938152 ns           13 GHz=1.39482 cycle/byte=8.34759 cycles/url=725.065 instructions/byte=18.6054 instructions/cycle=2.22883 instructions/ns=3.10883 instructions/url=1.61605k ns/url=519.825 speed=167.278M/s time/byte=5.97808ns time/url=519.252ns url/s=1.92585M/s

LoongArch64 3C5000 machine benchmark test results

main:

Loading /home/zhoumt/ada/build/_deps/url-dataset-src/out.txt
2025-07-09T20:47:58+08:00
Running ./benchmarks/benchdata
Run on (64 X 2200 MHz CPU s)
CPU Caches:
  L1 Instruction 64 KiB (x64)
  L1 Data 64 KiB (x64)
  L2 Unified 256 KiB (x64)
  L3 Unified 16384 KiB (x4)
Load Average: 0.00, 0.00, 0.00
ada spec: Ada follows whatwg/url
bad urls: ---------------------
ada---count of bad URLs       26
servo/url---count of bad URLs 26
whatwg---count of bad URLs    26
-------------------------------

bytes/URL: 86.859205
curl : OMITTED
input bytes: 8688092
number of URLs: 100025
performance counters: Enabled
rust version : 1.82.0
zuri : OMITTED
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href              70237332 ns     70219095 ns           10 GHz=2.19717 cycle/byte=17.9387 cycles/url=1.55814k instructions/byte=43.1562 instructions/cycle=2.40577 instructions/ns=5.28588 instructions/url=3.74852k ns/url=709.157 speed=123.728M/s time/byte=8.08222ns time/url=702.015ns url/s=1.42447M/s
BasicBench_AdaURL_aggregator_href   52536374 ns     52522557 ns           13 GHz=2.19698 cycle/byte=13.2874 cycles/url=1.15414k instructions/byte=34.1956 instructions/cycle=2.57353 instructions/ns=5.65398 instructions/url=2.97021k ns/url=525.33 speed=165.416M/s time/byte=6.04535ns time/url=525.094ns url/s=1.90442M/s
BasicBench_AdaURL_CanParse          33083323 ns     33083240 ns           21 GHz=2.19776 cycle/byte=8.37497 cycles/url=727.443 instructions/byte=23.6957 instructions/cycle=2.82935 instructions/ns=6.21822 instructions/url=2.05819k ns/url=330.993 speed=262.613M/s time/byte=3.80788ns time/url=330.75ns url/s=3.02343M/s


LSX:
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href              61362122 ns     61356556 ns           11 GHz=2.19716 cycle/byte=15.6198 cycles/url=1.35673k instructions/byte=37.825 instructions/cycle=2.4216 instructions/ns=5.32066 instructions/url=3.28545k ns/url=617.49 speed=141.6M/s time/byte=7.06214ns time/url=613.412ns url/s=1.63023M/s
BasicBench_AdaURL_aggregator_href   45419574 ns     45418890 ns           15 GHz=2.1973 cycle/byte=11.4936 cycles/url=998.327 instructions/byte=28.8651 instructions/cycle=2.51141 instructions/ns=5.51831 instructions/url=2.5072k ns/url=454.342 speed=191.288M/s time/byte=5.22772ns time/url=454.075ns url/s=2.20228M/s
BasicBench_AdaURL_CanParse          26432103 ns     26432046 ns           26 GHz=2.19764 cycle/byte=6.69288 cycles/url=581.338 instructions/byte=18.3647 instructions/cycle=2.74392 instructions/ns=6.03016 instructions/url=1.59515k ns/url=264.528 speed=328.695M/s time/byte=3.04233ns time/url=264.254ns url/s=3.78423M/s

LoongArch64 3A6000 machine benchmark test results

main:
Loading /home/zhoumt/ada/build/_deps/url-dataset-src/out.txt
2025-07-09T20:28:22+08:00
Running ./benchmarks/benchdata
Run on (8 X 2500 MHz CPU s)
CPU Caches:
  L1 Instruction 64 KiB (x8)
  L1 Data 64 KiB (x8)
  L2 Unified 256 KiB (x8)
  L3 Unified 16384 KiB (x1)
Load Average: 0.25, 0.06, 0.02
ada spec: Ada follows whatwg/url
bad urls: ---------------------
ada---count of bad URLs       26
servo/url---count of bad URLs 26
whatwg---count of bad URLs    26
-------------------------------

bytes/URL: 86.859205
curl : OMITTED
input bytes: 8688092
number of URLs: 100025
performance counters: Enabled
rust version : 1.85.0
zuri : OMITTED
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href              47189615 ns     47152658 ns           15 GHz=2.49793 cycle/byte=13.7227 cycles/url=1.19194k instructions/byte=38.2253 instructions/cycle=2.78555 instructions/ns=6.95809 instructions/url=3.32022k ns/url=477.174 speed=184.255M/s time/byte=5.42727ns time/url=471.409ns url/s=2.1213M/s
BasicBench_AdaURL_aggregator_href   30084600 ns     30060977 ns           23 GHz=2.49791 cycle/byte=8.65737 cycles/url=751.972 instructions/byte=27.1609 instructions/cycle=3.13732 instructions/ns=7.83674 instructions/url=2.35918k ns/url=301.041 speed=289.016M/s time/byte=3.46002ns time/url=300.535ns url/s=3.3274M/s
BasicBench_AdaURL_CanParse          21901334 ns     21884383 ns           32 GHz=2.498 cycle/byte=6.29039 cycles/url=546.378 instructions/byte=20.2599 instructions/cycle=3.22077 instructions/ns=8.04546 instructions/url=1.75976k ns/url=218.727 speed=397M/s time/byte=2.51889ns time/url=218.789ns url/s=4.57061M/s


LSX:
--------------------------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------
BasicBench_AdaURL_href              44818268 ns     44785516 ns           16 GHz=2.49801 cycle/byte=13.0466 cycles/url=1.13322k instructions/byte=35.7832 instructions/cycle=2.74272 instructions/ns=6.85134 instructions/url=3.1081k ns/url=453.648 speed=193.993M/s time/byte=5.15482ns time/url=447.743ns url/s=2.23342M/s
BasicBench_AdaURL_aggregator_href   29356161 ns     29334797 ns           24 GHz=2.49795 cycle/byte=8.44307 cycles/url=733.358 instructions/byte=25.3689 instructions/cycle=3.0047 instructions/ns=7.50558 instructions/url=2.20352k ns/url=293.584 speed=296.17M/s time/byte=3.37644ns time/url=293.275ns url/s=3.40977M/s
BasicBench_AdaURL_CanParse          18897944 ns     18883385 ns           37 GHz=2.49848 cycle/byte=5.43229 cycles/url=471.845 instructions/byte=18.1011 instructions/cycle=3.33212 instructions/ns=8.32526 instructions/url=1.57224k ns/url=188.852 speed=460.092M/s time/byte=2.17348ns time/url=188.787ns url/s=5.29698M/s

[lemire]

@Xiao-Tao

Could we get CI tests to work?
https://github.com/simdutf/simdutf/blob/master/.github/workflows/loongarch64-gcc-14.2.yml

[lemire]

@Xiao-Tao I see that CI was merged, from a separate PR.

Could you sync with our main branch?

[Xiao-Tao]

Thank you, CI has been re-triggered and all tests passed. @lemire @anonrig

[anonrig]

Thanks for the contribution @Xiao-Tao