forked from kareefardi/caravel_openframe_project
-
Notifications
You must be signed in to change notification settings - Fork 3
/
NOTES
351 lines (301 loc) · 15.1 KB
/
NOTES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
Notes on building the caravel_openframe_project:
=================================================
Tim Edwards
May 1, 2023 to August 29, 2023
This is an example project of a picoRV32 processor and SoC that
roughly matches the management core used on first Efabless SkyWater
OpenMPW. However, this project is intended to synthesize into the
project area of the Caravel Openframe.
The "NOTES" file is a collection of notes I made while trying
to get the openlane build to work cleanly from end to end.
See the README file for the final set of step-by-step instructions
for the build.
=======================================
Build notes:
(1) Building using local (not containerized) OpenROAD and Openlane.
OpenROAD was pulled from git and compiled.
Required libboost to be compiled and installed from source,
because cmake insisted on a boost library newer than that
supported by my OS.
Openlane was pulled from git and built.
(2) cd to openlane/openframe_project_wrapper
(3) Quick experimental build: ~/gits/openlane/flow.tcl
Determined that switch "-ignore_mismatches" is required to
bypass the error and exit caused by my tool versions.
(4) Do: ~/gits/openlane/flow.tcl -ignore_mismatches
Needed to fix one syntax issue; now it complains I need PDK_ROOT set
(5) Do:
setenv PDK_ROOT /usr/share/pdk
setenv PDK sky130A
~/gits/openlane/flow.tcl -ignore_mismatches
(6) Removed CLOCK_TREE_SYNTH from config.json because it is apparently
deprecated.
(7) "click" module needed for python3. But it exists. Why does the
odbpy/lefutil.py script fail, then?
Hacked my way around this by adding
import sys
sys.path.append('/usr/lib/python3.10/site-packages')
import click
to the lefutil.py script. But there must be a proper way to configure
the python path correctly for openroad.
(8) This actually ran. . . verilator (synthesis step 1) found errors, so
need to fix those. Except. . . failed to generate report due to
failure to load module "report", which is another path issue
('/usr/lib64/python3.10/site-packages')
Except this time, it still doesn't work. Can import "report", can
do "from report import report" but cannot do "from report.report import
Report"---no such thing. Need to pip install "reports" ???
(in /home/tim/gits/openlane/scripts/generate_reports.py)
Ahh. . . It is not a standard python package but local to openlane.
Also: The underlying problem is that there is a python package "report"
installed on my system, and this one is being preferred.
Hack solution: "pip uninstall" didn't work, so I just renamed "report"
in the offending library directory so it won't be found.
Back to the original issue: see runs/<run>/logs/synthesis/verilator.log
Problem---No defines.v, no OPENFRAME_IO_PADS defined.
Wrong set of files in config.json. defines.v appears twice and needs
to be removed.
Now verilator complains about all sorts of non-issues. How to disable
verilator? Set RUN_VERILATOR to 0 in the config file.
(9) Synthesis fails on call to standard cell "sky130_fd_sc_hd__einvn_8".
How to solve this? Tried "STD_CELL_LIBRARY_OPT" but that didn't work.
Setting "SYNTH_READ_BLACKBOX_LIB" worked.
(10) Now fails on the SRAM, which is not surprising.
---Added openframe_example.soc_mem.SRAM_0 and
openframe_example.soc_mem.SRAM_1 to macro.cfg but presumably need more.
---Added filenames for the SRAM block to VERILOG_FILES_BLACKBOX, EXTRA_LEFS,
and EXTRA_GDS_FILES (probably need EXTRA_LIBS, too).
(11) "$::env(PDK_ROOT)" doesn't work in the config.json file.
"::env(PDK_ROOT)" doesn't work, either. Punting for now and
inserting full path.
(12) Now it complains there is no port "vssd1" on
sky130_sram_2kbyte_1rw1r_32x512_8 of SRAM_1 in mem_wb.v. What gives?
Testing. . . it will also complain about vccd1 and about SRAM_0, just
generates errors from the bottom of the file up for some reason.
But is USE_POWER_PINS being changed between reading the library and
reading the design? But "SYNTH_USE_PG_PINS_DEFINES" is set to
"USE_POWER_PINS". Try not setting that?
Okay, no good, so try *not* defining it in the netlists file (or
rather, defining it only if SIM is defined). . . and bingo.
(13) "Out of bounds on some signals" error, need to check log file.
Seems not happy about the OR'd bits in the array assignment
assign gpio_all_dat_o[i] = |(gpio_dat_o[i][`OPENFRAME_IO_PADS-1:0]);
??
Disabled the out-of-bound check in openlane tcl_commands/checkers.tcl
Next set of checks appears to be more relevant.
Fixed all of them but there are "multiple driver" errors on tri-state
buffers, so disabling the "check_synth_misc" check, too.
(14) Failed on "proc_rom"---This is a command in yosys/synth.tcl, and it seems
that yosys fails to understand it. Maybe by default yosys is being run
from my original installed version which is a bit out of date?
Commented "proc_rom" out of the synth script and continued.
(15) Failed because config file was set with SYNTH_ELABORATE_ONLY. Fixed.
(16) Tried to read verilog of the SRAM; failed to black-box it. Maybe put back
the original brackets in the blackbox definition in config.json?
(Did not make a difference)
Following suggestion and adding "/// sta-blackbox" to the verilog file,
but that's in the PDK, so not a preferred solution.
(Might also be because the "extra liberty files" setting was not in
the config file?)
(17) Next error: invalid command name "report_parasitic_annoation" in
multi_corner.tcl. Commenting the line
(openlane/scripts/openroad/sta/multi_corner.tcl)
Now is running STA, maybe successfully. . .
Seems to have hung up on something. Produced an output almost
immediately, after which it is just grinding away.
Ran 10 hours and is still going in the morning, so something is
wrong.
NOTE: This happened on "report_clock_skew", and tracking that
down to openroad/sta/multi_corner.tcl, there's a comment that
says OpenROAD hangs if the "report_clock_skew" command is run
on a clockless design. Presumably there's something similar
going on, like the clock isn't specified correctly. Disabling
the report for now.
(18) Floorplan---same issue with python paths for importing "click" as
above (see (7)). Needed in defutil.py, apply_def_template.py,
manual_macro_place.py
(19) Floorplan---complains that some pins are not on the manufacturing
grid, specifically por_l/porb_h/porb_l. This comes from the
DEF file at fixed_dont_change/openframe_project_wrapper.def.
Magic shouldn't be putting things at off-grid positions!
In fact, it shouldn't even be possible. . . Except that
distance units have been set at 1000, not 200. But magic did,
and generated non-integer entries in the DEF file to boot, so
this will need to be investigated. Actually, there were rather
a lot of such entries in the file at 1/2 grid points. I think
that this is due to centerlines, and the DEF writer is not
paying attention to the grid limit settings in "cifoutput"
although it is assumed to be producing equivalent output.
(20) Macro placement: Macros not found. Used names
"openframe_example.soc_mem.SRAM_0" (and "1") in macro.cfg.
What should this be instead?
Missing a hiearchy layer; should be
"openframe_example.soc_mem.mem.SRAM_0"
After that. . . got all the way to step 7!
(21) Global placement: "click" again. . .
(22) Detail placement failed during/after clock tree synthesis.
Chances are good that this is due to the incorrect floorplan
DEF (tracks and sites do not cover the entire die area as
they should)? Need to look at output.
Result---The placement actually covers the area in spite of
the incorrect list of tracks/sites, so that doesn't seem to
be the underlying problem.
(23) Continuing 5/9/2023 after break due to travel. Committed all work
so here's a recap:
Do:
setenv PDK_ROOT /usr/share/pdk
setenv PDK sky130A
cd openlane/openframe_project_wrapper
~/gits/openlane/flow.tcl -ignore_mismatches
(had updated sky130A PDK and needed to redo the "/// sta-blackbox"
comment line in sky130_sram_2kbyte_1rw1r_32x512_8.v)
(24) Continuing 8/11/2023 after the openlane team worked over this.
Copied over all openlane configuration files. The openlane
flow has now been divided into multiple blocks (picoRV32 core,
clock routing, DLL, and then the top level wrapper) which are
hardened hierarchically. Macros were added which connect to the
VCCD1/VSSD1 busses (although preferably there should be additional
connections to VCCD/VSSD and VCCD2/VSSD2 for robustness).
Note that I am attempting to keep the *unbuilt* sources in the
original "main" repository branch, while working on a build in
a "build" branch. It should be possible to keep a minimum set of
files in the (committed) upstream repository while being able to
generate the final layout through a series of known repeatable
steps.
(25) NOTE: The issue with having to rewrite the openlane scripts for
the "click" package is probably best resolved by running
in a python virtual environment, even though I consider that
solution akin to docker and dislike it immensely.
(26) Continuing 8/28/2023 after another hiatus.
Testing with the following steps:
(1) Setup
setenv PDK_ROOT /usr/share/pdk
setenv PDK sky130A
(2) Synthesize the digital_locked_loop macro.
cd openlane/digital_locked_loop
~/gits/openlane/flow.tcl -ignore_mismatches
--> There is an immediate issue that verilator pulls all .v
files from the PDK, and is catching the newly-added
black-box files. Temporarily renaming them to avoid
a conflict.
--> Another issue is that fakediode_2 is calling a base
cell sky130_fd_sc_hd__diode that does not exist.
Changing name to from diode to diode_2 (need to fix in
the PDK install).
--> Same thing with fill_12 calling base cell fill that
does not exist; this one just needs to be deleted.
--> And verilator just seems to be overly eager to find
errors where they don't exist. Any way to turn it
off??? (Yes---set env variable "RUN_VERILATOR" to 0)
--> Failed due to no SYNTH_CAP_LOAD defined?
Defined it and set to 52 based on an entry in caravel.
No idea what the units are. . . Or why it needs to
be defined (might need an openlane update).
--> Now stuck again with the "click" module issue, in
io_place.py (fixed) and diodes.py (fixed) and
wire_lengths.py (fixed) and power_utils.py (fixed)
--> Started step 30 (stream-out with klayout) but klayout
isn't running in batch. Why would you stream out with
both klayout and magic? Just exiting from klayout
manually (but how to fix. . . and is this step even
needed?).
Back-annotated note: (1) No, the klayout step is not
needed and exiting klayout continues without issue.
(2) Failure to run klayout in batch mode is a self-
inflicted problem caused by using a script to launch
klayout but failing to handle arguments. Reworked by
adding the directory for klayout's shared object
libraries to ld.so.conf.d and running ldconfig, and
replacing the script with a symbolic link to the
executable.
--> Step 37, cvc_rv does not exist. I have the source, but
it is not installed---and I cannot install it due to
inscrutable errors. (set env variable RUN_CVC = 0
to disable)
--> Finally, the flow completed (although is it correct?)
(there are max fanout violations reported.)
(3) Synthesize the picosoc macro.
cd openlane/picosoc
~/gits/openlane/flow.tcl -ignore_mismatches
--> Cannot find LEF of digital_locked_loop.
Copied the LEF file manually (but why isn't this done
automatically?)
--> Cannot find GDS of digital_locked_loop.
Copied the GDS file manually (but why isn't this done
automatically?)
--> Need to disable verilator
--> Cannot find .lib of digital_locked_loop.
None was created, so had to copy from the working files
Back-annotated comment: This can be corrected simply
by changing STA_WRITE_LIB to 1.
--> Cannot find gl/.v file of digital_locked_loop
Copied the .v file manually (but why isn't this done
automatically?)
--> Needs SYNTH_CAP_LOAD defined (see above)
--> Yosys error: picorv32 redefined ??
Added "ifndef PnR"..."endif" around file includes
--> Needs digital_locked_loop.min.spef in
signoff/digital_locked_loop/openlane-signoff/spef/
(added nom and max for good measure)
--> Now running and spending extensive time in step 16
(global routing resizer design optimizations)
Letting it run. (Started 5:18pm)
--> And then it died on a totally obscure syntax error of
a line in picosoc.nl.v "wire [38:0] \% Xz ;". Seems
to have been derived from the pnl.v line which has
"wire [38:0] \J\xfd6\x88 ;". Not clear where this
comes from but tool updates (OpenROAD in particular)
are probably in order (wanders off for several hours
on a quest to rebuild OpenROAD. . .)
--> (Returns after finally coercing OpenROAD to compile
again) Now I get a different obscure error earlier
in the sequence:
File "/home/tim/gits/openlane/scripts/odbpy/reader.py",
line 37, in __init__
odb.read_def(self.db, def_in)
File "odb_py.py", line 9553, in read_def
TypeError: in method 'read_def', argument 1 of type 'odb::dbTech *'
Something has gone very wrong here and it is not clear
to me what, other than that it was caused by recompiling
OpenROAD.
(27) Continuing 8/29/2023 after updating OpenROAD *and* openlane.
(1) Create python virtual environment this time to avoid issue
with the "click" package.
python3 -m venv ./venv
./venv/bin/python3 -m pip install --upgrade --no-cache-dir -r \
/home/tim/gits/openlane/requrements.txt
source ./venv/bin/activate.csh
--> This works!
(2) Run ~/gits/openlane/flow.tcl -ignore_mismatches
--> RUN_VERILATOR superceded by RUN_LINTER (fixed)
--> yosys fails on command "proc_rom". . .
Need yosys update now. . . (Goes off and rebuilds yosys)
(3) (Hey, I'm back) Trying again with yosys updated.
Run ~/gits/openlane/flow.tcl -ignore_mismatches
--> Noted that SYNTH_CAP_LOAD is deprecated. Not sure
why that didn't show up before. . . ? Changed to
OUTPUT_CAP_LOAD as requested, for subsequent runs.
--> Now running and at step 15 and going well so far.
--> Now running and at step 26 and going well so far.
--> Now running and at step 35 and going well so far.
--> Success!
(28) Now run the project wrapper:
(Will presumably need to copy files from the picosoc synthesis
to the top level before this will work.)
Copy picosoc LEF, GDS, .lib, and GL verilog;
cd openlane/openframe_project_wrapper
~/gits/openlane/flow.tcl -ignore_mismatches
--> Need to also copy the GDS of vccd1_connection
and vssd1_connection from the "_work" branch.
--> Apparently these files can't be compressed,
either.
--> Also need gate-level .v files for vccd1 and
vssd1 connection.
--> Forgot the picosoc SPEF files. (min, max, nom)
--> Now running without problems.
--> Observation: "Generating lef with Magic" seems
to be taking a rather long time. (But not
ridiculously long, so probably okay.)
Final solution generated but there are hold violations at the
typical corner. (Single violation clk on gpio_in[38] to
openframe_example/_34243_)