From mboxrd@z Thu Jan 1 00:00:00 1970 From: erik quanstrom Date: Wed, 5 Feb 2014 07:46:51 -0500 To: edgecomberts@gmail.com, 9fans@9fans.net Message-ID: <3b72cf2290d28c849e1f7a5ffde134a6@mikro.quanstro.net> In-Reply-To: References: <5778ffc079ca4ad7227b37b1158b29f1@proxima.alt.za> MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Subject: Re: [9fans] Inferno and the Parallella Topicbox-Message-UUID: b6c488e8-ead8-11e9-9d60-3106f5b1d025 > Oh, its ok. I like the GSoC idea. I just don't think I'm GSoC material, I'm > hardware type, even if I will be a uni student this year going forward - > "If it draws blood, its hardware" as the old maxim goes. it's great to hear the enthusiam, but sadly, it seems over ambitious. to work with this heterogeneous co-processor with the usual tools, and be any more interesting than a standard arm, i think at least the following needs to be done 1. bootstrap the arm processor get plan 9 running. 1a. program the fpga with adapteva's binary blob. 1b. drivers for a minmal set of devices. 2. write a compiler/assembler/linker for the epiphany multicore; populate /epi/include. a emulator may need to be written. 3. write the libmach hooks for the same 4. write the asm for /sys/src/lib*/epi (or at least libc) 5. decide what kind of operating framework the epi should have, and write the appropriate glue. it's not clear to me that a standard kernel could work at all. (what kind of coherence model is there?) this can't be done by one gsoc student in a summer. and there's the open ended question of how to use the epi coprocessor. a very bright, gifted, experienced, stubborn, and diligent student might have some hope of accomplishing 1/1a or a significant part of 2. but that's a stretch. 3, 4 seem to be properly sized for one student gsoc. 5 is unknown. so, in order to have something usable at the end, one would need 5 students, 5 mentors, someone to do 1b, and sort of a scrum master to help coordinate. i see several serious risks to this idea. a. what if we get less than 5 students, or mentors, or slots? b. sadly, not all students complete the summer. how do we recover if even one person drops out? c. do we have someone qualified to be scrum master for 10 people (5 students and 5 mentors)? with enough time? d. 5 is open ended. this seems too big a leap, given the student success rate is not yet 100%. so if you're a student still excited about this project, reframing the problem so that it stands alone (even if it's just bootstrapping the arm chip) seems like the best option to me. now i could be wrong or overly pessamistic, so i'd love to hear other opinions. - erik