Thread by @TubeTimeUS

so an early 6502 microprocessor has been imaged! the first versions from 1976 lacked a ROR (rotate right) instruction, so let's dig in and see what changed. 🧵

https://twitter.com/Siliconinsid/status/1433515926901727236

they look very similar at first glance. on the left is the 1976 revision A, and on the right is a later revision D (mfg by Rockwell, in this case).

here they are, imported into GIMP, and roughly overlaid. there are a few pads that have been moved around (mostly on the left) and the pad structures have changed.

the most obvious place to start looking for functional differences is the instruction decode ROM array.

unfortunately the rev A image has a bunch of gnarly stitching artifacts. this makes analysis a lot harder.

after careful study, i could only find four bits that were added, highlighted in red in the image below.

checking the Visual 6502, this corresponds with the op-ror net!

in the rev A 6502, it was just never decoded. this whole column never goes active. but it appears that it goes somewhere...

so it goes...nowhere? it's hard to trace because there are tons of stitching artifacts. the first little vertical branch is just polysilicon that ends and isn't used. the longer part disappears into some logic.

zooming in, take a look at the trace (rev A on the left, rev D on the right). they've repurposed it in rev D but they left the contact via behind (center).

in rev D, that disconnected contact is called node 182.

in the schematic that net has something to do with branches. i don't know why that would have been brought out of the instruction ROM in the original--it was clearly not decoded anyway.

hard to tell but that old trace also connected to (or seems to have been involved with) op-T3-branch. so it was some unused branch logic.

what's clear from this brief investigation is...

THIS IS NOT A BUG

the original, first revision 6502 did not have a "bugged" ROR instruction. it simply had no ROR instruction at all! the logic to implement it did not exist. the corresponding ROM decode line was dead logic for some unused branch instruction variant.

all of this lines up with my various conversations with Bill Mensch. apparently Chuck Peddle thought a ROR wasn't necessary and they didn't include it in the original design. it wasn't until later, when customers demanded it, that they added the instruction.

so if you tried to run an ROR instruction on a rev A 6502, what you got was...a partially decoded instruction, just like the other weird ones that typically didn't do anything, halted the CPU, or had marginally useful results.

later on, for the CMOS 65C02, Bill decoded those "extra instructions" to NOPs. chip designers sure don't like undefined behavior.