An Indigo2 Compiles for an Hour — IRIX 6.2, MIPSpro, SpeedShop and the n32 ABI · 1080p Ambient

An xwsh window on an Indigo2 IMPACT R10000, September 1996, light text on SGI slate. Nothing is written here — this is the toolchain working. A 3-D finite-difference wave model has to be rebuilt: seven C files through MIPSpro 7.1, and then a link that refuses to happen. The refusal is the most IRIX thing there is. Everything compiles clean as n32 mips4, and ld says FATAL 12, expecting n32 objects. The culprit is not an object at all — it is a small in-house archive somebody built in January for the old R4400 box, with no ABI flag in its Makefile at all, so it took the o32 default. elfdump reads the flags word out of both and the whole thing is settled in one line. Then the interesting half: it runs, SpeedShop says 88 percent of the time is in one function, and the rest of the hour is spent taking that seriously. -O3, then -OPT:roundoff=2, then -IPA, with the disassembler open each time — MIPS IV madd.s appears, the loads get hoisted ahead of their uses, and fd_swap vanishes into its caller entirely. 34.4 seconds down to 18.6, and not one line of source changed. Checkpoint, restart, tar to DAT, log out. An hour of sparse, patient instrumental — music only, no typing sounds. 1 hour, 1920×1080, with sound. Music from #Uppbeat (free for Creators!): https://uppbeat.io/music/tracks/simon... License code: M8QSDAWQYWLDAMYB #sgi #irix #indigo2 #mips #r10000 #mipspro #speedshop #elf #compiling #optimization #1990s #retrocomputing #workstation #busyworklabs #musiconly #focus #deepwork #programming