NVIDIA SASS Instructions' history

sass_instructions.md

SASS Instructions' history, extracted from CUDA Binary Utilities - Instruction Set Reference and with links to a reverse-engineered ISA when available:

• Maxwell (Compute Capability 5.x, ~2014) and Pascal (CC 6.x, ~2016)
  • Instructions (108): ATOM ATOMS B2R BAR BFE BFI BPT BRA BRK BRX CAL CCTL CCTLL CCTLT CONT CS2R CSET CSETP DADD DFMA DMNMX DMUL DSET DSETP EXIT F2F F2I FADD FCHK FCMP FFMA FLO FMNMX FMUL FSET FSETP FSWZADD HADD2 HFMA2 HMUL2 HSET2 HSETP2 I2F I2I IADD IADD3 ICMP IMAD IMADSP IMNMX IMUL ISCADD ISET ISETP JCAL JMP JMX LD LDC LDG LDL LDS LEA LOP LOP3 MEMBAR MOV MUFU NOP P2R PBK PCNT PEXIT POPC PRET PRMT PSET PSETP R2B R2P RED RET RRO S2R SEL SHF SHFL SHL SHR SSY ST STG STL STS SUATOM SULD SURED SUST SYNC TEX TEXS TLD TLD4 TLD4S TLDS TXQ VOTE XMAD
• Volta (CC 7.0 and 7.2, ~2017)
  • Added (48): ATOMG BMOV BMSK BREAK BREV BSSY BSYNC CALL DEPBAR ERRBAR FADD32I FFMA32I FMUL32I FRND FSEL GETLMEMBASE HADD2_32I HFMA2_32I HMMA HMUL2_32I I2IP IABS IADD32I IDP IDP4A IMMA IMUL32I ISCADD32I KILL LEPC LOP32I MATCH MOV32I NANOSLEEP PLOP3 PMTRIG QSPC RPCMOV RTT SETCTAID SETLMEMBASE SGXT TMML TXD VABSDIFF VABSDIFF4 WARPSYNC YIELD
  • Removed (26): BFE BFI BRK CAL CONT CSET CSETP DMNMX DSET FCMP ICMP IMADSP ISET JCAL PBK PCNT PEXIT PRET PSET RRO SSY SYNC TEXS TLD4S TLDS XMAD
• Turing (CC 7.5, ~2018)
  • Added (33): BMMA BRXU JMXU LDSM MOVM R2UR S2UR UBMSK UBREV UCLEA UFLO UIADD3 UIADD3.64 UIMAD UISETP ULDC ULEA ULOP ULOP3 ULOP32I UMOV UP2UR UPLOP3 UPOPC UPRMT UPSETP UR2UP USEL USGXT USHF USHL USHR VOTEU
• Ampere (CC 8.0, ~2020) and Ada (CC 8.6, ~2022)
  • Added (8): DMMA F2IP HMNMX2 I2FP LDGDEPBAR LDGSTS REDUX UF2FP
  • Removed (2): R2B RTT
• Hopper (CC 9.0, ~2022)
  • Added (36): ACQBULK BGMMA CGAERRBAR ELECT ENDCOLLECTIVE FENCE HGMMA IGMMA LDGMC PREEXIT QGMMA REDAS REDG STAS STSM SYNCS UBLKCP UBLKPF UBLKRED UCGABAR_ARV UCGABAR_WAIT ULEPC USETMAXREG UTMACCTL UTMACMDFLUSH UTMALDG UTMAPF UTMAREDG UTMASTG VHMNMX VIADD VIADDMNMX VIMNMX VIMNMX3 WARPGROUP WARPGROUPSET
  • Removed (1): RED
• Blackwell (CC 10.0 and 12.0, ~2024)
  • Added (48): ACQSHMINIT CREDUX CS2UR FADD2 FFMA2 FHADD FHFMA FMNMX3 FMUL2 LDCU LDT LDTM OMMA QMMA STT STTM UF2F UF2I UF2IP UFADD UFFMA UFMNMX UFMUL UFRND UFSEL UFSET UFSETP UGETNEXTWORKID UI2F UI2FP UI2I UI2IP UIABS UIMNMX UMEMSETS UREDGR USTGR UTCATOMSWS UTCBAR UTCCP UTCHMMA UTCIMMA UTCOMMA UTCQMMA UTCSHIFT UVIADD UVIMNMX UVIRTCOUNT
  • Removed (8): BGMMA BMMA HGMMA IGMMA QGMMA ULDC WARPGROUP WARPGROUPSET

The reverse-engineered ISA further mentions these instructions not listed above: AL2P ALD ARRIVES CSMTEST F2FP FOOTPRINT IPA ISBERD LDTRAM OUT PIXLD SUQUERY USETSHMSZ