3DNOW! Command Concise Reference
Directive description algorithm
FEMMS quickly switches 3DNOW! / MMX instructions and floating point command status.
PAVGUSB Mmreg1, the Mmreg2 / MEM64 8-bit unsigned compression integer summary. Mmreg1 [63:56] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [55:48] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [47:40] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [39:32] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [31:24] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [23:16] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
Mmreg1 [15:08] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
mmreg1 [07:00] = (mmReg1 [63:56] mmreg2 / mem64 [63:56] 1) / 2
PF2ID mmReg1, MmReg2 / MEM64 converts the compressed floating point to 32-bit compressed integer. IF (mmReg2 / MEM64 [31: 0]> = 231)
MmReg1 [31: 0] = 7FFF_FFFFH
Else
{
IF (mmReg2 / MEM64 [31: 0] <= -231)
Mmreg1 [31: 0] = 8000_0000h
Else
Mmreg1 [31: 0] = INT (mmReg2 / MEM64 [31: 0])
}
IF (mmReg2 / mem64 [63:32]> = 231)
Mmreg1 [63:32] = 7FFF_FFFFH
Else
{
IF (mmReg2 / mem64 [63:32] <= -231)
mmreg1 [63:32] = 8000_0000h
Else
mmreg1 [63:32] = int (mmReg2 / mem64 [63:32])
}
PFACC mmReg1, MmReg2 / MEM64 compressed floating point number is accumulated. Mmreg1 [31:00] = mmreg1 [31:00] mmreg1 [63:32]
Mmreg1 [63:32] = mmreg2 / mem64 [31:00] mmreg2 / mem64 [63:32]
PfAdd MmReg1, MmReg2 / MEM64 compressed floating point number is added. Mmreg1 [31:00] = mmreg1 [31:00] mmreg2 / mem64 [31:00]
Mmreg1 [63:32] = mmreg1 [63:32] mmreg2 / mem64 [63:32]
PFCMPEQ MmReg1, MmReg2 / MEM64 compares whether the compressed floating point is equal. IF (mmReg1 [31:00] == mmreg2 / mem64 [31:00])
mmreg1 [31:00] = fff_ffffh
Else
mmreg1 [31:00] = 0000_0000h
IF (mmReg1 [63:32] == mmreg2 / mem64 [63:32])
mmreg1 [63:32] = fff_ffffh
Else
MmReg1 [63:32] = 0000_0000HPFCMPGE MmReg1, MmReg2 / MEM64 is "greater than or equal" for compressed floating point numbers. IF (mmReg1 [31:00]> = mmreg2 / mem64 [31:00])
mmreg1 [31:00] = fff_ffffh
Else
mmreg1 [31:00] = 0000_0000h
IF (mmReg1 [63:32]> = mmreg2 / mem64 [63:32])
mmreg1 [63:32] = fff_ffffh
Else
mmreg1 [63:32] = 0000_0000h
PFCMPGT MmReg1, MmReg2 / MEM64 is "greater than" on whether the compressed floating point number is "greater than". IF (mmReg1 [31:00]> mmreg2 / mem64 [31:00])
mmreg1 [31:00] = fff_ffffh
Else
mmreg1 [31:00] = 0000_0000h
IF (mmReg1 [63:32]> mmreg2 / mem64 [63:32])
mmreg1 [63:32] = fff_ffffh
Else
mmreg1 [63:32] = 0000_0000h
Pfmax mmReg1, MmReg2 / MEM64 compressed floating point count number of maximum values. IF (mmReg1 [31:00]> mmreg2 / mem64 [31:00])
mmreg1 [31:00] = mmreg1 [31:00]
Else
Mmreg1 [31:00] = mmreg2 / mem64 [31:00]
IF (mmReg1 [63:32]> mmreg2 / mem64 [63:32])
Reg1 [63:32] = mmReg1 [63:32]
Else
Reg1 [63:32] = mmReg2 / MEM64 [63:32]
Pfmin mmReg1, MmReg2 / MEM64 compressed floating point count number minimum. IF (mmReg1 [31:00] mmreg1 [31:00] = mmreg1 [31:00] Else Mmreg1 [31:00] = mmreg2 / mem64 [31:00] IF (mmReg1 [63:32] mmreg1 [63:32] = mmreg1 [63:32] Else Mmreg1 [63:32] = mmreg2 / mem64 [63:32] Pfmul mmreg1, MmReg2 / MEM64 compressed floating point number multiplication operation. mmreg1 [31:00] = mmreg1 [31:00] * mmreg2 / mem64 [31:00] Mmreg1 [63:32] = mmreg1 [63:32] * mmreg2 / mem64 [63:32] The approximate value of the PFRCP mmreg1 and MmReg2 / MEM64 compressed floating point count. Mmreg1 [31:00] = 1.0F / mmreg2 / mem64 [31:00] mmreg1 [63:32] = 1.0F / mmreg2 / mem64 [31:00] PFRCPIT1 MmReg1, MmReg2 / MEM64 The first step of compressed floating point numbers to the fewer. The second step of iteration of the reciprocal (or the countdown of floating point square root) of PFRCPIT2 mmReg1, MmReg2 / MEM64 compressed floating point number. Pfrsqit1 mmReg1, MmReg2 / MEM64 The first step of the reciprocal iteration of the square root of the floating point number. Pfrsqrt mmReg1, MmReg2 / MEM64 finds an approximate value of the reciprocal of floating point square roots. mmreg1 [31:00] = 1.0F / SQRT (mmReg2 / MEM64 [31:00]) Mmreg1 [63:32] = 1.0F / SQRT (mmReg2 / MEM64 [31:00]) PFSUB mmReg1, MmReg2 / MEM64 compressed floating point number subtraction operation. MmReg1 [31:00] = mmreg1 [31: 0] - mmreg2 / mem64 [31:00] Mmreg1 [63:32] = mmreg1 [63:32] - mmreg2 / mem64 [63:32] Pfsubr mmreg1, MmReg2 / MEM64 compressed floating point number reverse subtraction operation. Mmreg1 [31:00] = mmreg2 / mem64 [31:00] - mmreg1 [31:00] Mmreg1 [63:32] = mmreg2 / mem64 [63:32] - mmreg1 [63:32] PI2FD MmReg1, MmReg2 / MEM64 converts compressed 32-bit integers into floating point numbers. mmreg1 [31:00] = float (mmreg2 / mem64 [31:00]) mmreg1 [63:32] = float (mmreg2 / mem64 [63:32]) Pmulhrw mmreg1, MREG2 / MEM64 multiplied 16-bit symbolic compression integers and looped. Save 16-bit high. Mmreg1 [63:48] = (mmReg1 [63:48] * mmreg2 / mem64 [63:48] 8000H) >> 16 Mmreg1 [47:32] = (mmreg1 [47:32] * mmreg2 / mem64 [47:32] 8000H) >> 16 Mmreg1 [31:16] = (mmReg1 [31:16] * mmreg2 / mem64 [31:16] 8000H) >> 16 MmReg1 [15:00] = (mmreg1 [15:00] * mmreg2 / mem64 [15:00] 8000H) >> 16 Prefetch (W) MEM8 data prefetch instructions put the data to be used in the L1 data cache. 3DNOW! Expanded instruction concise reference Directive description algorithm PF2IW MEMREG1, MEMREG2 / MEM64 converts the compressed floating point number to 16-bit integers. IF (mmReg2 / MEM64 [31:00]> = 215) mmreg1 [31:00] = 0000_7FFFH Else { IF (mmReg2 / MEM64 [31:00] <= -215) mmreg1 [31:00] = fff_8000h Else Mmreg1 [31:00] = INT (mmreg2 / mem64 [31:00]) } IF (mmReg2 / mem64 [63:32]> = 215) Mmreg1 [63:32] = 0000_7fffh Else { IF (mmReg2 / MEM64 [63:32] <= -215) mmreg1 [63:32] = fff_8000h Else mmreg1 [63:32] = int (mmReg2 / mem64 [63:32]) } PFNAcC mmreg1, mmreg2 / MEM64 compressed floating point number is reduced by mmreg1 [31:00] = mmreg1 [31:00] - mmreg1 [63:32] Mmreg1 [63:32] = MemReg2 / MEM64 [31: 0] - Memreg2 / MEM64 [63:32] Pfpnacc mmReg1, MREG2 / MEM64 compressed floating point number mixed or decrease mmreg1 [31:00] = mmReg1 [31:00] - mmreg1 [63:32] Mmreg1 [63:32] = MemReg2 / MEM64 [31:00] Memreg2 / MEM64 [63:32] PI2FW mmreg1, mmReg2 / MEM64 compressed 16-bit integer transformation into floating point number mmreg1 [31:00] = float (mmReg2 / MEM64 [15:00]) mmreg1 [63:32] = float (mmreg2 / mem64 [47:32]) Pswapd mmreg1, mmreg2 / MEM64 compression exchange double word mmreg1 [63:32] = mmReg2 / mem64 [31:00] mmreg1 [31:00] = mmreg2 / mem64 [63:32]
