triton.experimental.gluon.language.amd.cdna4.scaled_upcast
- triton.experimental.gluon.language.amd.cdna4.scaled_upcast(src, scale, elem_type, axis=None, _semantic=None)
Upcast an fp4 or fp8 tensor and fold raw E8M0 scale payload into the CDNA4 scaled-upcast op.
The
scaletensor must use raw E8M0 payload inint8oruint8, and must already have the expanded output shape and scaled-upcast result layout. For fp4 inputs, that is the canonical unpacked layout implied bysrcandaxis.elem_typemust befp16orbf16.