Improving build time by removing the gfx11xx and host code from rccl_float8.h (#1789)

* removing extra build time by removing the gfx11xx arch from using hip_fp8

---------

Co-authored-by: Marzieh Berenjkoub <mberenjk@amd.com>
This commit is contained in:
mberenjk
2025-07-09 14:03:47 -05:00
committed by GitHub
vanhempi 9c89573580
commit 697bee4ee8
3 muutettua tiedostoa jossa 6 lisäystä ja 39 poistoa
+2 -2
Näytä tiedosto
@@ -201,10 +201,10 @@ namespace RcclUnitTesting
case ncclUint32: ss << scalarsPerRank.U4[this->globalRank]; break;
case ncclInt64: ss << scalarsPerRank.I8[this->globalRank]; break;
case ncclUint64: ss << scalarsPerRank.U8[this->globalRank]; break;
case ncclFloat8e4m3: ss << scalarsPerRank.F1[this->globalRank]; break;
case ncclFloat8e4m3: ss << (float)scalarsPerRank.F1[this->globalRank]; break;
case ncclFloat32: ss << scalarsPerRank.F4[this->globalRank]; break;
case ncclFloat64: ss << scalarsPerRank.F8[this->globalRank]; break;
case ncclFloat8e5m2: ss << scalarsPerRank.B1[this->globalRank]; break;
case ncclFloat8e5m2: ss << (float)scalarsPerRank.B1[this->globalRank]; break;
case ncclBfloat16: ss << scalarsPerRank.B2[this->globalRank]; break;
default: ss << "(UNKNOWN)";
}