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>
[ROCm/rccl commit: 697bee4ee8]
This commit is contained in:
1 szülő
b1470b4e50
commit
1623fcc7a1
3 files changed
+6
-39
No files matched your search
@@ -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)";
|
||||
}
|
||||
|
||||
Reference in new issue
Block a user