mscclpp

mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-12 17:26:04 +00:00

Files

Binyang Li e412804eab Improve signal/wait performance and fix barrier issue (#499 )

Remove __assert_fail for release build. This will reduce the number of
PTX instructions inside the loop. Also Trying to resolve this issue
reported in #497. Reduce the number of PTX instructions from 8 to 6.
8 ranks signal/wait will reduce from 3.2us->2.8us on NDv5
Also NDEBUG flag is confused here, sometime it will not be set. Use
customized flag for debug build.

Here is current PTX:
```
      ld.u64  %rd12, [%rd2+-24];
      mov.u64         %rd13, %rd12;
      mov.u64         %rd11, %rd13;
      ld.acquire.sys.b64 %rd10,[%rd11];
      setp.lt.u64     %p1, %rd10, %rd3;
      @%p1 bra        $L__BB2_1;
```

If we change to `asm volatile("ld.global.acquire.sys.b64 %0, [%1];" :
"=l"(flag) : "l"(flag_addr));` will reduce to 4 instructions. We can get
2.1 us for 8 ranks signal/wait
```
        ld.u64  %rd9, [%rd1+-24];
        ld.global.acquire.sys.b64 %rd8, [%rd9];
        setp.lt.u64     %p1, %rd8, %rd2;
        @%p1 bra        $L__BB2_1;
```

2025-04-16 14:22:10 -07:00

CMakeLists.txt

Fix assert declaration & add a compile test (#303 )

2024-05-20 02:39:30 +00:00

compile_tests.cu

Improve signal/wait performance and fix barrier issue (#499 )

2025-04-16 14:22:10 -07:00

core_tests.cc

Add Context and Endpoint classes to enable non-Communicator use-cases (#166 )

2023-09-06 13:10:04 +08:00

cuda_utils_tests.cc

Add GpuBuffer class (#423 )