Switch to latest toolchain

[mactcp]

I've switched to the latest toolchain. It works great, the read speed was unchanged. The newer compiler is producing slightly better code for writing (readDataLoop function), and with a slight adjustment to the timing, that is benchmarking a fraction faster.

I got these benchmark results:
LC III+: Read 1234 KB/sec, write 861 KB/sec (was 848 KB/sec with old toolchain)
Quadra 650: Read 1143 KB/sec, write 1162 KB/sec (was 950 KB/sec)
PowerMac 7100/80: Read 1266 KB/sec, write 1273 KB/sec (was 1253 KB/sec)
PowerMac G3/266: Read 1273 KB/sec, write 1307 KB/sec (was 1286 KB/sec)

[mactcp]

I noticed a regression in write speed when a change moved the readDataLoop function from 16 byte aligned to 8 byte aligned, so I've forced it to be 16 byte aligned.

I'll check and make sure the writeDataLoop isn't also sensitive to 16 byte alignment.

[mactcp]

writeDataLoop isn't sensitive to 16 byte alignment, so this is good to go.

[erichelgeson]

Thanks!

[erichelgeson]

It looks like taking the pin off still defaults to toolchain-gccarmnoneeabi 1.60301.0 (6.3.1) https://github.com/erichelgeson/BlueSCSI/runs/5613341596?check_suite_focus=true#step:8:27

Locally for me since I had 1.90301.200702 installed it chose that - which was slower compared to 1.60301.0.

@mactcp which toolchain did yours run locally? We should pin it to that explicitly.

Versions: https://registry.platformio.org/tools/platformio/toolchain-gccarmnoneeabi/versions

[mactcp]

@erichelgeson I'm using 1.90301.200702

[erichelgeson]

Got it - was a build issue on my side with something cached - a full clean and rebuild I see the same results as you now. I'm going to push up an explicit pin to avoid any build oddities.