Visual6502wiki/6502 datapath: Difference between revisions

From NESdev Wiki
Jump to navigationJump to search
(copy wikitext from archive.org)
 
m (fix intrawikis)
 
Line 1: Line 1:
This page discusses the 6502 datapath, using the terminology from [[Hanson's Block Diagram]] and is probably best understood by [http://www.pagetable.com/?p=39 reference to it]
This page discusses the 6502 datapath, using the terminology from [[Visual6502wiki/Hanson's Block Diagram]] and is probably best understood by [http://www.pagetable.com/?p=39 reference to it]


We're interested in which datapath control signals are active in each of the two phases.
We're interested in which datapath control signals are active in each of the two phases.
Line 5: Line 5:
A full cycle consists of phi1 and phi2.  When we say a signal is "effective", we mean it actually does something.
A full cycle consists of phi1 and phi2.  When we say a signal is "effective", we mean it actually does something.


All datapath control signals are latched during phi2; they are set mostly from opcode and timing data, but also some internal state.  We work broadly from left to right. (Which is right to left on [[Balazs' schematic and documents|Balazs' schematic]])
All datapath control signals are latched during phi2; they are set mostly from opcode and timing data, but also some internal state.  We work broadly from left to right. (Which is right to left on [[Visual6502wiki/Balazs' schematic and documents|Balazs' schematic]])


=== External busses and signals ===
=== External busses and signals ===
Line 33: Line 33:
: The S register is actually two latches in series.  This makes it possible to read a value from SB and write a value to ADL at the same time.  On phi2, the value from the "in" latch is forwarded to the "out" latch (and onto the driven bus, if any).
: The S register is actually two latches in series.  This makes it possible to read a value from SB and write a value to ADL at the same time.  On phi2, the value from the "in" latch is forwarded to the "out" latch (and onto the driven bus, if any).


(Note the [[6502 datapath control timing fix|two "tuning fork" structures]], which have contacts
(Note the [[Visual6502wiki/6502 datapath control timing fix|two "tuning fork" structures]], which have contacts
either on the top or bottom, which select whether X, Y, A write SB
either on the top or bottom, which select whether X, Y, A write SB
and DB only during phi1, or slightly longer, during "not phi2". We think this might be a timing fix, or an option left open until after silicon showed which choice worked best)
and DB only during phi1, or slightly longer, during "not phi2". We think this might be a timing fix, or an option left open until after silicon showed which choice worked best)

Latest revision as of 01:29, 11 February 2022

This page discusses the 6502 datapath, using the terminology from Visual6502wiki/Hanson's Block Diagram and is probably best understood by reference to it

We're interested in which datapath control signals are active in each of the two phases.

A full cycle consists of phi1 and phi2. When we say a signal is "effective", we mean it actually does something.

All datapath control signals are latched during phi2; they are set mostly from opcode and timing data, but also some internal state. We work broadly from left to right. (Which is right to left on Balazs' schematic)

External busses and signals

DOR is latched from DB during phi1, and driven onto the data pins in phi2, if a write is done (and, on the 6501, only if the asynchronous DBE signal is on).

DL is latched during phi2, and then put on ADL, ADH, or DB on the next phi1; during phi2, the old value in DL is put on that bus.

ABL and ABH can be loaded from ADL and ADH respectively during phi1; they are put on the address pins in that same phi1, and stay there until changed again.

R/#W is latched during phi2, and then delayed until phi1, where it is output.

Address values

ADL/ABL, ADH/ABH
We already saw these. Effective on the next phi1.
0/ADL0, 0/ADL1, 0/ADL2, 0/ADH0, 0/ADH(1-7)
These set the interrupt vector fetch address, and the zero page and stack high address. Effective on phi2 and the next phi1.


The register file

Y/SB, X/SB, SB/Y, SB/X
Move the X and Y registers from/to the SB. Latched on phi2, just like everything else; effective on the next phi1.
SB/S, S/S, effective on the next phi1.
S/SB, S/ADL, effective on phi2 and the next phi1.
The S register is actually two latches in series. This makes it possible to read a value from SB and write a value to ADL at the same time. On phi2, the value from the "in" latch is forwarded to the "out" latch (and onto the driven bus, if any).

(Note the two "tuning fork" structures, which have contacts either on the top or bottom, which select whether X, Y, A write SB and DB only during phi1, or slightly longer, during "not phi2". We think this might be a timing fix, or an option left open until after silicon showed which choice worked best)

ALU inputs

SB/ADD, 0/ADD, nDB/ADD, DB/ADD, ADL/ADD
Two options for one side, three for the other. Effective on the next phi1.

ALU operation selection

ANDS, EORS, ORS, 1/ADDC, SRS, SUMS, DAA, DSA
Select the ALU operation. Effective on the next phi1 and phi2.

(The overflow and carry out signals AVR and ACR are output from the ALU back to the control logic, latched during phi2, used in phi1. The decimal carries are picked up at phi2 as well).

ALU output register

The ALU output register (ADD) is written during phi2. The value can be used the next cycle:

ADD/SB7, ADD/SB(0-6), ADD/ADL, effective on phi2 and the next phi1.
The ADL output is for address calculations. For output to SB, the top bit is handled separately for rotate right instructions: the ALU always computes a zero there; by not driving it to the bus a one will be read.
SB/AC, effective on the next phi1.
Lines 1-3,5-7 are fed through the decimal adjust first, to finish the proper BCD add/subtract result if necessary, before writing it to the accumulator.
AC/SB, AC/DB, effective on the next phi1.
Write the A reg back to one of the busses.

The Program Counter

ADH/PCH, PCH/PCH, PCL/PCL, ADL/PCL
select whether to use the current PC, or take a new value from the internal address busses. Effective on the next phi1.
PCH/DB, PCL/DB, PCH/ADH, PCL/ADL
write the PC to one of the busses. Effective on phi2 and the next phi1.
I/PC, effective during the next phi1 and phi2.
Increment the PC, or not. When incrementing, the new value is put on ADL,ADH because there are no internal latches in the PC incrementer. For every instruction, the first two bytes are fetched (during execution of the previous instruction); I/PC peeks ahead (or back, if you want to look at it that way) to the next instruction that is predecoded, so it can skip incrementing PC if that is a one-byte instruction.
P/DB
Write the flag values to the DB; effective on phi2 and the next phi1. The DB can be read to set the flag values as well; it is read during phi2, and then latched in the flag register on the next phi1.
SB/DB, SB/ADH
Connect two busses together. Effective on phi2 and the next phi1.

Precharge

All internal busses (SB, DB, ADL, ADH) are driven high during phi2, as a sort of precharge. In fact commonly they are also driven by data signals during phi2, causing an intermediate voltage to appear on the bus.

A note on signal naming

In our Javascript simulation the datapath control signals are tabulated according to Hanson's names, but in the layout they are named with a prefix according to their position across the chip. So

  • SSB, SADL, SBS, SS

will be found as

  • dpc4_SSB,dpc5_SADL,dpc6_SBS,dpc7_SS

in the source. See also the table below.

As Balazs used another naming scheme in his very useful but incomplete schematic, we should also cross-reference his names:

Balazs Hanson JSSim note
R1x7 Y/SB dpc0_YSB drive sb from y
R1x6 SB/Y dpc1_SBY load y from sb
R1x5 X/SB dpc2_XSB drive sb from x
R1x4 SB/X dpc3_SBX load x from sb
R1x2 S/SB dpc4_SSB drive sb from stack pointer
R1x1 S/ADL dpc5_SADL drive adl from stack pointer
R1x3 SB/S dpc6_SBS load stack pointer from sb
? S/S dpc7_SS recirculate stack pointer
R2x1 notDB/ADD dpc8_nDBADD alu b side: select not-idb input
R2x2 DB/ADD dpc9_DBADD alu b side: select idb input
R2x3 ADL/ADD dpc10_ADLADD alu b side: select adl input
R2x4 (??) SB/ADD dpc11_SBADD alu a side: select sb
R2x5 0/ADD dpc12_0ADD alu a side: select zero
R2x6 ORS dpc13_ORS alu op: a or b
R2x7 SRS dpc14_SRS alu op: logical right shift
R2x8 ANDS dpc15_ANDS alu op: a and b
R2x9 EORS dpc16_EORS alu op: a xor b (?)
R2x12 SUMS dpc17_SUMS alu op: a plus b (?)
? DAA dpc18_#DAA decimal related (inverted)
R2x14,7 ADD/SB(7) dpc19_ADDSB7 alu to sb bit 7 only
R2x14 ADD/SB(0-6) dpc20_ADDSB06 alu to sb bits 6-0 only
R2x15 ADD/ADL dpc21_ADDADL alu to adl
R2x20,6 DSA dpc22_#DSA decimal related/SBC only (inverted)
R3x4 SB/AC dpc23_SBAC (optionally decimal-adjusted) sb to acc
R3x1 AC/SB dpc24_ACSB acc to sb
R3x3 SB/DB dpc25_SBDB sb pass-connects to idb (bi-directionally)
R3x2 AC/DB dpc26_ACDB acc to idb
R3x0 SB/ADH dpc27_SBADH sb pass-connects to adh (bi-directionally)
R3x5,0 0/ADH0 dpc28_0ADH0 zero to adh0 bit0 only
R3x5 0/ADH(1-7) dpc29_0ADH17 zero to adh bits 7-1 only
R4x2 ADH/PCH dpc30_ADHPCH load pch from adh
R4x3 PCH/PCH dpc31_PCHPCH load pch from pch incremented
R4x4 PCH/ADH dpc32_PCHADH drive adh from pch incremented
R4x1 PCH/DB dpc33_PCHDB drive idb from pch incremented
!! PCLC dpc34_PCLC pch carry in and pcl FF detect?
Carry PCHC dpc35_PCHC pcl 0x?F detect - half-carry
notCarry I/PC dpc36_#IPC pcl carry in (inverted)
R5x1 PCL/DB dpc37_PCLDB drive idb from pcl incremented
R5x4 PCL/ADL dpc38_PCLADL drive adl from pcl incremented
R5x3 PCL/PCL dpc39_PCLPCL load pcl from pcl incremented
R5x2 ADL/PCL dpc40_ADLPCL load pcl from adl
Dkx2 DL/ADL dpc41_DL/ADL pass-connect adl to mux node driven by idl
Dkx3 DL/ADH dpc42_DL/ADH pass-connect adh to mux node driven by idl
Dkx1 DL/DB dpc43_DL/DB pass-connect idb to mux node driven by idl