(no commit message)
[libreriscv.git] / nlnet_2022_opf_isa_wg / discussion.mdwn
1 # Questions
2
3 you applied to the 2022-08 open call from NLnet. We have some questions regarding your project proposal Libre-SOC OpenPOWER ISA WG, but obviously we are incurring some delays due to the deluge of payment requests ;)
4
5 **
6 You requested a neat round sum of 100000 euro.
7 Can you provide some more detail on how you arrived at this amount?
8 Could you provide a breakdown of the main tasks, and the
9 associated effort? What rates did you use?
10 **
11
12 last question first: we've learned (painfully, by losing opportunities
13 and team members) that the prior rates which were around EUR 1500 per
14 person are inadequate to attract the quality we need, and had to double
15 it. I (personally) used to be ok when working out of Taiwan for 3 years
16 on EUR 1200-1500, and Jacob was in student-subsidised accommodation.
17
18 bottom line: 3 people, at EUR 3,000, is actually only 11 months duration.
19 if we include binutils part-time as 0.25 people it's only 10 months.
20
21 these are the tasks:
22
23 * ongoing communication with OPF ISA WG: over 12-18 months,
24 it takes appx 1-2 hours per day of communication time to
25 prepare and answer questions.
26 * preparation (and revision) of RFCs: although they are templatable
27 and partial cut/paste from the wiki the revisions are not, needing
28 ongoing feedback. plus, we will need approximately 20-25 RFCs.
29 * Compliance Test Suites: there are already thousands of unit tests,
30 these need to be expanded for the 8/16/32-bit work (thousands, in
31 each bit-width). Again: several months of work
32 * Example algorithms: strncpy, quicksort, insertionsort, UTF-8
33 validation and conversion to UTF-16; more Audio/Video examples - all of
34 these are critically important showcases. Each example chosen
35 can average around 1-2 weeks work.
36 * Developing and improving the Simulator itself, to confirm correct
37 functionality: again, several months (this is always ongoing)
38 * The Test API: this will be a simpler self-contained task to make
39 it auto-generate Makefiles (and cover other systems), and
40 also by this time we will have cavatools in the mix: approx 8 weeks
41 * binutils needs ongoing updates, an estimated budget covering
42 10-14 weeks would be good.
43
44 **
45 Is there meanwhile news on the requirements of IBM and the ISA WG?
46 **
47
48 somewhat. the page is now open - https://openpower.foundation/isarfc/ -
49 and they have prepared a process/procedure document (legally required
50 to be followed, under the OPF's ByLaws), which is adapting as we're
51 literally the first people to use it.
52
53 **
54 A request for 100k is very large, and the timelines are
55 pretty long too.
56 **
57
58 yes and no. if we assume 3 people (one junior editor, two and a half
59 programmers: simulator, unit tests, binutils) it actually doesn't go far.
60
61 **
62 Can we not take it step by step?
63 **
64
65 EUR 50,000 assuming 3.5 people at EUR 3,000 is actually only 5 months.
66 realistically that would mean we would actually need to begin the
67 submission process on the very next cycle! (2022-10E - 2022-12E would
68 be more likely but cutting it fine)
69
70 **
71 It would be better for us to achieve this incrementally, as in:
72 start with a smaller amount for meeting submission criteria for
73 the block of instructions, deliver initial code, tests,
74 documentation - and when more budget is needed, a new chunk is added.
75 **
76
77 I don't have a problem with that, if you are fine with the extra admin
78 work :) i appreciate it does mean not needing a special EU Auditor,
79 and also an opportunity to review, half-way through.
80
81 What would work on the legal compliance for the development look
82 like? Who would be doing that?
83
84 IBM - or more to the point the OPF ISA WG - requesting that all
85 contributors sign an "Inbound Patent License Agreement". in our
86 case there *aren't* any patents, but we still have to sign an
87 agreement that there aren't any, and, also, that if we *do* create
88 any patents that those will be assigned to the OPF immediately.
89 Perhaps a budget for some legal assistance in reviewing that agreement
90 might be a good idea? NLnet has funded this work under its
91 "Works for the Public Good" mandate: we don't want to be caught out
92 here.
93
94 **
95 How would you manage such a large amount of RFCs, which must
96 be perceived as a denial of service at the WG?
97 **
98
99 carefully! we have been warning them consistently and persistently
100 for 24 months. each RFC when it gets to the "Presentation as
101 actual Changes" stage, will be passed through to compiler and
102 hardware experts for their consideration. IBM has had many many
103 RFCs in-house over the years: this isn't something that's new to
104 them.
105
106 ultimately, though, by comparison with RISC-V having *seventy*
107 unique Technical Working Groups, realistically it is simply
108 the Power ISA that has a lot of catching up to do. we will be
109 the catalyst that drives that... carefully :)
110
111 **
112 Is there infrastructure in place to manage the lifecycle of each RFC?
113 **
114
115 yes. the bugtracker, wiki, and mailing list, and the RFCs themselves
116 are in the git repository that's behind the wiki. full cross-referencing
117 in each has been found over a 4 year period of managing this project.
118
119 then there is also the "main" page tracking *all* RFCs (which will
120 get its own bugreport at some point)
121
122 * https://libre-soc.org/openpower/sv/rfc/
123
124 Example of the cross-referencing so far:
125
126 * https://libre-soc.org/openpower/sv/rfc/ls001/
127 * https://git.libre-soc.org/?p=libreriscv.git;a=history;f=openpower/sv/rfc/ls001.mdwn
128 * https://lists.libre-soc.org/pipermail/libre-soc-dev/2022-October/005344.html
129 * https://bugs.libre-soc.org/show_bug.cgi?id=924, note the discussion
130 * https://libre-soc.org/openpower/sv/rfc/ls001/discussion/
131
132 **
133 How are discussions going to be linked to each RFC?
134 **
135
136 As above example: by a cross-referenced URL in each one, and the
137 standard practice of adding a "discussion" page in the wiki if
138 necessary, although this is often subsumed by the bugtracker.
139
140 **
141 What are the timelines?
142 **
143
144 based on 3.5 people? realistically only around 10 months.
145 (EUR 50,000 cuts that by half).
146
147