7,647 karma · joined November 25, 2013
I was capturing QuickDraw library calls - the low level graphics primitives, to figure out where the graphics time in apps was going and found out sometimes excel did it 9 times
Of course users didn't see it more than once, but our hardware made all that wasted time run faster
I did the architectural design for the SuperMac cards. I figured out what needed to be accelerated, dropping code into people's machines to see where the cycles were going. Others did the physical design for the first 2 cards, I did the design of the chip in the Thunder and later cards (designed the data paths and state machines and a full simulation, someone else actually laid the gates)
If your card has a SQD01 on it it's my work. It peaks at 1.5Gb/s on solid fills
One of the other bugs (the Quark/ATM one) was also because of the programmers were worried about writing over stuff that hadn't been completely erased, the Quark guys wrote a string with 2 spaces at the end through a box that masked the end of the string, the ATM font renderer saw it couldn't fit the text so it split it in half and tried again so it drew N/2 N/4 N/8 ... strings. It spent all it's time in the 68k's multiply instructions figuring out how wide the strings (and substrings) were, our fancy 24-bit character rendering hardware was an afterthought
To be fair excel would erase places white that it wanted to write up to 9 times before it drew any black pixels, we made that very fast! we didn't tell them :-)
At the time 24-bit framebuffers were so slow that before we built graphics acceleration hardware people would switch back to 8-bit to get stuff done, making 24-bit/true colour your daily driver was a big step forward.
This happens on every memory access, so the thing you want to avoid is ping ponging writeable cache lines between CPUs (especially before you have a chance to actually write it) - LL/SC instructions sit on top of these protocols and allow instructions to tell is a cache line had been "stolen" before you have a chance to write it
Even wore our portion was originally in Berkeley, the border was moved 100 years or so ago so that someone could open the closest bar to the campus when Berkeley had more stringent liquor licensing
Used to be in the early days that some people depended on how the original verilog interpreter ordered events, it was a silly thing (models would only run on one simulator, cause of lots of angst).
'<=' assignment fixed a lot of these problems, using it correctly means that you can model synchronous logic without caring about event ordering (at the cost of an extra copy and an extra event which can be mostly optimised away by a compiler).
In combination 'always @(*)' and '=', and assign give you reliable combinatorial logic.
In real world logic a lot of event ordering is non deterministic - one signal can appear before/after another depending on temperature all in all it's best not to design depending it if you possibly can, do it right and you don't care about event ordering, let your combinatorial circuits waggle around as their inputs change and catch the result in flops synchronously.
IMHO Verilog's main problems are that it: a) mixes flops and wires in a confusing way, and b) if you stay away from the synthesisable subset lets you do things that do depend on event ordering that can get you into trouble (but you need that sometimes to build test benches)
68000s however had a problem: they couldn't return correctly from a page (MMU) fault (68010s fixed that) for a pre-VM (pre BSD or SVR2) UNIX world - however you could get around this with a few smarts