whitequark changed the topic of #glasgow to: glasgow debug tool · code https://github.com/whitequark/Glasgow · logs https://freenode.irclog.whitequark.org/glasgow
<_whitenotifier-9> [whitequark/Glasgow] whitequark pushed 2 commits to master [+0/-0/±2] https://git.io/fhd3A
<_whitenotifier-9> [whitequark/Glasgow] whitequark e809080 - access.direct.arguments: support defaults for width ranges.
<_whitenotifier-9> [whitequark/Glasgow] whitequark def4973 - applet.audio_output: support up to 16 channels.
<_whitenotifier-9> [Glasgow] Success. The Travis CI build passed - https://travis-ci.org/whitequark/Glasgow/builds/495207036?utm_source=github_status&utm_medium=notification
<_whitenotifier-9> [whitequark/Glasgow] whitequark pushed 3 commits to master [+0/-0/±5] https://git.io/fhdsf
<_whitenotifier-9> [whitequark/Glasgow] whitequark b1b8af0 - access.direct.arguments: support default width.
<_whitenotifier-9> [whitequark/Glasgow] whitequark 756808a - cli: leave formatting of sample CLI invocations alone.
<_whitenotifier-9> [whitequark/Glasgow] whitequark f70d556 - applet.audio_output: fix frequency offset on multichannel output.
<_whitenotifier-9> [Glasgow] Success. The Travis CI build passed - https://travis-ci.org/whitequark/Glasgow/builds/495208479?utm_source=github_status&utm_medium=notification
<Ultrasauce> time for an opus decoder?
jevinskie has joined #glasgow
jevinskie has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<whitequark> lol no
jevinskie has joined #glasgow
ylamarre has joined #glasgow
jevinskie has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<TD-Linux> the way to do an opus decoder would be a picorv32 core running libopus
<sorear> can libopus decode in real time on a 10-20 MIPS core with 16 kB RAM
<TD-Linux> it uses ~35MHz on an A53 so no
<sorear> we can probably get some core running that's less terribly inefficient than picorv32 but the memory limit is hard
<TD-Linux> yeah I know it needs less than 256kB but I don't remember exactly how much
<TD-Linux> also I guess that RAM is also code space, so no go :(
<sorear> might work on revB which has the up5k (128kB of SPRAM + 16 kB DPRAM)
<sorear> the up5k has much slower fabric though
<sorear> wonderful pranks: file formats described by RFCs, where the only explicitly normative section is _checks notes_ 157 printed pages of uuencoded gzip data
<whitequark> what
<sorear> oh sorry base64 not uuencode
<sorear> libopus is … a lot of code in a lot of files
<sorear> has anyone written a nice single-file decoder
<TD-Linux> yeah we were responsible
<TD-Linux> there is no stbopus if that's what you want
<TD-Linux> there are alternate implementations though
<sorear> stb?
<whitequark> google stb_image
<sorear> thanks
<sorear> there *is* a stb_vorbis
<sorear> i found a random TI appnote where their DSP runs a SWB vorbis decoder at 10MHz; so the fabric speed is unlikely to be a major problem if you can add the needed dsp bits
<whitequark> hum
* sorear suspects that a pipelined, folded-datapath rv32i (e.g. 16-bit ALU, all instructions take 2 cycles) could be Pareto-better than picorv32; note that folding means you have half as many instructions in a given length pipeline -> much less hazard logic