BSTART.TMATMULMX

1// PTO-INSTRUCTION: {"assembly":["BSTART.TMATMULMX DataType"],"block":[],"catalog_indices":[44],"catalog_records":[{"asm":"BSTART.TMATMULMX DataType","constraints":[{"field":"DataType","operator":"one-of","values":[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,16,17,18,19,20,24,25,26,27,28]}],"encoding":[{"index":0,"mask":"0x07ffffff","match":"0x00431181","width_bits":32}],"encoding_kind":"L32","fields":[{"name":"DataType","pieces":[{"instruction_lsb":27,"value_lsb":0,"width":5}],"signedness":"encoding-defined","width":5}],"form_id":"bstart_tmatmulmx_32_e26cb3874648","length_bits":32,"mnemonic":"BSTART.TMATMULMX","semantic_family":"CMD","semantic_group":"Bundle Split","semantic_handler":"ExecuteBundleStart","semantic_summary":"Starts CUBE Function 4 for the TMATMUL_MX Matrix-matrix complete-bundle operation.","status":"accepted"}],"classification":["execution"],"contract":{"block_composition":["BSTART.TMATMULMX 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 or cooperative group_M (optional, default 1)","B.DIM LB1 N (optional, default 1)","B.DIM LB2 K (optional, default 1)","B.IOS complete right or both matrix operand groups (optional; cooperative mask 1111)","B.IOT ordered Local mathematical sources: 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, optional RowMaxOut, optional GroupMaxOut destinations","B.IOT/B.IOR postprocess operands selected by B.FPATR","BSTOP or the next BSTART completion boundary"],"canonical_assembly":["BSTART.TMATMULMX DataType"],"defaults":["Encoded DataType is always AType. Omitted B.DATR preserves AType as BType, selects RNE, and disables saturation.","Omitted LB0 defaults Local M or cooperative group_M to one; omitted LB1 and LB2 default N and K 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. M, N, and K are arbitrary positive values independent of per-PE TSize. Each side uses group-32 E8M0 or HiF4X2 group-64 U32 scale; Local scales use CUBE_M32 and Shared scales remain ordinary Tiles.","TransA=0 and TransB=0 select no logical transpose. Each nonzero control is legal only when the corresponding primary is Shared."],"encoding_class":"standalone-encoded","examples":["BSTART.TMATMULMX 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 or cooperative group_M (optional, default 1); B.DIM LB1 N (optional, default 1); B.DIM LB2 K (optional, default 1); B.IOS complete right or both matrix operand groups (optional; cooperative mask 1111); B.IOT ordered Local mathematical sources: 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, 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":{"DataType":{"ref":"PTO-FIELD-BLOCK-DATATYPE"}},"field_zero_meanings":{"DataType":"Encoded zero selects FP64."},"legality":["The carrier selects exactly CUBE Function 4 and TileOperation_TMATMUL_MX.","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. M, N, and K are arbitrary positive values independent of per-PE TSize. Each side uses group-32 E8M0 or HiF4X2 group-64 U32 scale; Local scales use CUBE_M32 and Shared scales remain ordinary Tiles.","A Shared primary must satisfy hardware-maintained whole-parent readiness and publication before payload access; fixed-quarter allocation or initialization masks are not a prerequisite. Any cooperative Local-A/Shared-B or Shared-A/Shared-B TMATMUL interprets LB0 as Core-total group_M in 1..128; Shared A has shape group_MxK, Shared B has shape KxN, and PE i uses valid_M=clamp(group_M-i*M_per_PE,0,M_per_PE) with M_per_PE 16 for group_M<=64 and 32 for group_M>=65. TransA and TransB apply only to their corresponding Shared primary. Right-only Shared inherits Local A layout; all-Shared ACC inherits C layout; all-Shared non-ACC selects M16 through M=16 and M32 through M=32.","Each non-FP16/BF16 side requires its assigned scale: group-32 E8M0 for MX FP8/FP4 or group-64 U32 for HiF4X2. Published Shared operands may replace the right group or both matrix groups; supplementary operands and destinations remain Local.","Every cooperative nonzero PE_MASK must be 1111; all four PEs complete Shared readiness, while zero-row PEs suppress every compute-only Local resolution and effect. 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":"DataType","role":"tile element data type selector"}],"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":true,"state_effects":["Start a CUBE Function 4 descriptor with encoded DataType preserved as AType.","At block completion execute TileOperation_TMATMUL_MX using the resolved M, N, K, input types, mathematical operands, and B.FPATR postprocess schema.","Publish the complete output group atomically after successful preflight and computation; do not consume mathematical or postprocess sources.","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.","Successful Shared primary reads leave every Shared descriptor, mask, publication state, payload, and lifetime unchanged."]},"depends_on":["PTO-BLOCK-MODEL-SCHEMA-PROFILE-ENCODING"],"id":"PTO-BLOCK-BSTART-TMATMULMX","mnemonic":"BSTART.TMATMULMX","summary":"Starts CUBE Function 4 for the TMATMUL_MX Matrix-matrix complete-bundle operation.","surface":"block"}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-BSTART-TMATMULMX-CONTRACT-0014// ndf: kind=contract level=L1 layer=block status=accepted5// BSTART.TMATMULMX MUST select CUBE Function 4 and TileOperation_TMATMUL_MX.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 M, N, and K MUST be positive per-PE dimensions independent of TSize;9// cooperative LB0 MUST be Core-total group_M.10// For every cooperative Shared-input TMATMUL, LB0 MUST be Core-total group_M;11// Shared A MUST have shape group_MxK and Shared B MUST have shape KxN.12// Each matrix side independently requires an E8M0 scale exactly when its MX13// input type is not FP16 or BF16. Published Shared operands may replace the14// right group or both matrix groups; supplementary operands and destinations15// remain Local.16// Shared primaries MUST be fully published and ready in all four fixed quarters.17// TransA/TransB MUST apply only to the corresponding Shared primary. Every18// cooperative nonzero PE mask MUST be 1111; zero-row PEs MUST have no Local effect.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-BSTART-TMATMULMX-CONTRACT-00124
25
26// DOC-BEGIN: decode27readonly func InstructionContractMatches_BSTART_TMATMULMX(28    operation: CommandOperation) => boolean29begin30    return operation ==31        CommandOperation_bstart_tmatmulmx_32_e26cb3874648;32end;33// DOC-END: decode34// DOC-BEGIN: operation35readonly func InstructionContractTileOperation_BSTART_TMATMULMX()36    => TileOperation37begin38    return TileOperation_TMATMUL_MX;39end;40
41readonly func InstructionContractCubeFunction_BSTART_TMATMULMX()42    => integer {0..31}43begin44    return 4;45end;46
47readonly func InstructionContractSharedOperandsAllowed_BSTART_TMATMULMX()48    => boolean49begin50    return TRUE;51end;52
53readonly func InstructionContractHandler_BSTART_TMATMULMX()54    => CommandSemanticHandler55begin56    return CommandHandler_ExecuteBundleStart;57end;58// DOC-END: operation59