TGEMV_MX_ACC

1// PTO-INSTRUCTION: {"assembly":["TGEMV_MX_ACC <bundle operands>"],"block":["BSTART.TGEMVMX.ACC AType","B.DATR BType, RMode, Sat (optional; BType defaults to AType)","B.FPATR PreQuantMode, ReluMode, GroupNCode, RowMaxEn, GroupMaxEn, RowMaxInit, MaxAbsEn, TransA, TransB, CScaleEn (exactly one)","B.DIM LB0 M (optional, default 1; TGEMV permits only M=1)","B.DIM LB1 N (optional, default 1)","B.DIM LB2 K (optional, default 1)","B.IOT ordered Local mathematical sources: C CUBE_M16/M32 accumulator matching A with encoded selector distinct from DstTile, A CUBE_M16/M32 primary, optional A scale, B CUBE_N8 primary, optional B scale","B.IOT D matching A's CUBE_M16/M32 layout with a distinct encoded destination index, optional RowMaxOut, optional GroupMaxOut destinations","B.IOT/B.IOR postprocess operands selected by B.FPATR","BSTOP or the next BSTART completion boundary"],"catalog_indices":[102],"catalog_records":[{"arguments":[{"operand":"destination0"},{"operand":"source0"},{"operand":"source1"},{"operand":"source2"},{"operand":"source3"},{"operand":"source4"}],"command_mnemonic":"BSTART.TGEMVMX.ACC","contract_status":"reviewed-complete","datr_contract":{"allowed_nonzero_fields":["DataType","RMode","Sat"],"pad_union":"must-zero"},"disposition":"accepted-direct-operation","effect_contract":"TGEMV_MX_ACC","family":"CUBE","fault_contract":"ExecuteTileInstruction","function":22,"legality_handler":"TileOperandsLegal_TGEMV_MX_ACC","name":"TGEMV_MX_ACC","operands":[{"field":"destination0","role":"destination"},{"field":"source0","role":"accumulator"},{"field":"source1","role":"left-vector"},{"field":"source2","role":"row-scale"},{"field":"source3","role":"right-matrix"},{"field":"source4","role":"column-scale"}],"restart_contract":"CompleteBundleAtWithAcceptedApplicabilityRules","semantic_handler":"TGEMV_MX_ACC","state_effects":["operand:destination0:destination","operand:source0:accumulator","operand:source1:left-vector","operand:source2:row-scale","operand:source3:right-matrix","operand:source4:column-scale"]}],"classification":["matrix-and-matrix-vector","matrix-vector"],"contract":{"block_composition":["BSTART.TGEMVMX.ACC AType","B.DATR BType, RMode, Sat (optional; BType defaults to AType)","B.FPATR PreQuantMode, ReluMode, GroupNCode, RowMaxEn, GroupMaxEn, RowMaxInit, MaxAbsEn, TransA, TransB, CScaleEn (exactly one)","B.DIM LB0 M (optional, default 1; TGEMV permits only M=1)","B.DIM LB1 N (optional, default 1)","B.DIM LB2 K (optional, default 1)","B.IOT ordered Local mathematical sources: C CUBE_M16/M32 accumulator matching A with encoded selector distinct from DstTile, A CUBE_M16/M32 primary, optional A scale, B CUBE_N8 primary, optional B scale","B.IOT D matching A's CUBE_M16/M32 layout with a distinct encoded destination index, optional RowMaxOut, optional GroupMaxOut destinations","B.IOT/B.IOR postprocess operands selected by B.FPATR","BSTOP or the next BSTART completion boundary"],"canonical_assembly":["TGEMV_MX_ACC <bundle operands>"],"defaults":["Encoded DataType is always AType. Omitted B.DATR preserves AType as BType, selects RNE, and disables saturation.","Omitted LB0, LB1, and LB2 default M, N, and K independently to one; TGEMV fixes M to one.","Exactly one all-zero B.FPATR selects no conversion, activation, or reduction; B.IOR and auxiliary B.IOT operands exist only when a selected postprocess mode requires them.","Local A uses persistent CUBE_M16 or CUBE_M32, Local B uses persistent CUBE_N8, and D is newly allocated in A's M layout; Local C also uses A's M layout. M is fixed to one; N and K are arbitrary positive values independent of per-PE TSize. Required E8M0 scales remain ordinary row-major Tiles.","TransA=0 and TransB=0 select no logical transpose. TGEMV requires both controls to remain zero.","C and D are both mandatory. In decoded blocks C's six-bit relative selector must differ from zero-extended DstTile before rename; direct Tile calls require destination TileIndex to differ from accumulator TileIndex."],"encoding_class":"selector-encoded-block-operation","examples":["BSTART.TGEMVMX.ACC AType; B.DATR BType, RMode, Sat (optional; BType defaults to AType); B.FPATR PreQuantMode, ReluMode, GroupNCode, RowMaxEn, GroupMaxEn, RowMaxInit, MaxAbsEn, TransA, TransB, CScaleEn (exactly one); B.DIM LB0 M (optional, default 1; TGEMV permits only M=1); B.DIM LB1 N (optional, default 1); B.DIM LB2 K (optional, default 1); B.IOT ordered Local mathematical sources: C CUBE_M16/M32 accumulator matching A with encoded selector distinct from DstTile, A CUBE_M16/M32 primary, optional A scale, B CUBE_N8 primary, optional B scale; B.IOT D matching A's CUBE_M16/M32 layout with a distinct encoded destination index, optional RowMaxOut, optional GroupMaxOut destinations; B.IOT/B.IOR postprocess operands selected by B.FPATR; BSTOP or the next BSTART completion boundary"],"exceptions":["A reserved DataType or fixed-bit mismatch raises Fault_IllegalInstruction before block state changes.","Missing, duplicate, or non-Matrix B.FPATR use raises Fault_BundleControl before allocation or payload effects.","Illegal types, dimensions, masks, binding streams, descriptors, shapes, capacities, aliases, readiness, or postprocess values raise Fault_TileLegality before source snapshots and effects."],"field_contracts":{},"field_zero_meanings":{},"legality":["The carrier selects exactly CUBE Function 22 and TileOperation_TGEMV_MX_ACC.","Local A uses persistent CUBE_M16 or CUBE_M32, Local B uses persistent CUBE_N8, and D is newly allocated in A's M layout; Local C also uses A's M layout. M is fixed to one; N and K are arbitrary positive values independent of per-PE TSize. Required E8M0 scales remain ordinary row-major Tiles.","C and D are both mandatory. In decoded blocks C's six-bit relative selector must differ from zero-extended DstTile before rename; direct Tile calls require destination TileIndex to differ from accumulator TileIndex.","TGEMV is Local-only: TransA and TransB are zero and every effective Shared binding rejects before effects.","Each matrix side independently requires an E8M0 scale exactly when its MX input type is not FP16 or BF16. C is one explicit Local MxN accumulator source and D is a distinct newly published destination; C's encoded relative selector and D's zero-extended DstTile hand must differ before rename. M is fixed to one and every Shared binding is illegal.","Every common nonzero four-bit PE_MASK is legal; all four PEs complete cooperative Shared readiness while only selected PEs allocate and publish. Mask zero is a strict no-op before descriptor reads, faults, allocation, readiness checks, or lifetime effects.","B.DATR permits only BType, RMode, and Sat. Exactly one B.FPATR is mandatory and closes the conditional postprocess schema."],"memory_effects":["none"],"operands":[{"field":"destination0","role":"destination"},{"field":"source0","role":"accumulator"},{"field":"source1","role":"left-vector"},{"field":"source2","role":"row-scale"},{"field":"source3","role":"right-matrix"},{"field":"source4","role":"column-scale"}],"ordering":["Complete schema, field, type, dimension, descriptor, shape, capacity, readiness, alias, and allocation preflight precedes every source snapshot and destination effect.","D and every enabled reduction output publish as one atomic group; rejection publishes none and successful sources persist."],"standalone_opcode":false,"state_effects":["Multiply the scaled matrix and vector and accumulate into the supplied Tile.","After complete preflight, execute TGEMV_MX_ACC with the operand bindings listed above; destination definedness changes only as specified by that handler.","For Local execution, publish D with A's CUBE_M16 or CUBE_M32 layout and final output dtype; ordinary Bias, MX scales, and enabled reduction auxiliaries keep their operation-owned layouts.","C is snapshotted before multiplication and remains descriptor-and-payload unchanged after success or rejection."]},"depends_on":["PTO-BLOCK-MODEL-SCHEMA-PROFILE-ENCODING"],"engine":"CUBE","id":"PTO-TILE-TGEMV-MX-ACC","mnemonic":"TGEMV_MX_ACC","summary":"Multiply the scaled matrix and vector and accumulate into the supplied Tile.","surface":"tile"}2// PTO-REVIEW: {"review_method":"formal-definition-read","outcome":"FORMAL-COMPLETE","reviewed_fields":["assembly","encoding","defaults","operation","state","memory","ordering","faults","reserved"]}3// NDF-BEGIN: PTO-TGEMV-MX-ACC-CONTRACT-0014// ndf: kind=contract level=L1 layer=tile status=accepted5// TGEMV_MX_ACC MUST select CUBE Function 22 and TileOperation_TGEMV_MX_ACC.6// Its encoded DataType is AType; omitted B.DATR MUST preserve AType as7// BType, omitted LB0/LB1/LB2 MUST default M/N/K independently to one,8// Local N and K MUST be positive and independent of per-PE TSize; M9// MUST equal one.10// Each matrix side independently requires an E8M0 scale exactly when its MX11// input type is not FP16 or BF16. C is one explicit Local MxN accumulator12// source and D is a distinct newly published destination. C's encoded relative13// selector MUST differ from zero-extended DstTile before physical rename. M is fixed to one and every Shared binding is14// illegal.15// TransA and TransB MUST remain zero and every Shared binding MUST reject.16// Every common nonzero Local PE mask is legal; mask zero is a strict no-op.17// C and D MUST both be explicit; direct Tile calls MUST also use different18// destination and accumulator TileIndex values. C MUST persist unchanged.19// Exactly one B.FPATR MUST close postprocess defaults and operands.20// Complete schema, type, shape, capacity, definedness, readiness, alias,21// and allocation preflight MUST precede source snapshots and effects;22// successful outputs MUST publish atomically and sources MUST persist.23// NDF-END: PTO-TGEMV-MX-ACC-CONTRACT-00124// DOC-BEGIN: decode25readonly func InstructionContractOperation_TGEMV_MX_ACC()26    => TileOperation27begin28    return TileOperation_TGEMV_MX_ACC;29end;30// DOC-END: decode31// DOC-BEGIN: operation32readonly func InstructionContractCubeFunction_TGEMV_MX_ACC()33    => integer {0..31}34begin35    return 22;36end;37
38readonly func InstructionContractSharedOperandsAllowed_TGEMV_MX_ACC()39    => boolean40begin41    return FALSE;42end;43
44readonly func InstructionContractOperandsLegal_TGEMV_MX_ACC(45    destination: TileIndex,46    accumulator: TileIndex,47    left_vector: TileIndex,48    row_scale: TileIndex,49    right_matrix: TileIndex,50    column_scale: TileIndex) => boolean51begin52    return TileOperandsLegal_TGEMV_MX_ACC(53        destination,54        accumulator,55        left_vector,56        row_scale,57        right_matrix,58        column_scale);59end;60
61readonly func InstructionContractHandler_TGEMV_MX_ACC()62    => TileSemanticHandler63begin64    return TileHandler_TGEMV_MX_ACC;65end;66// DOC-END: operation67