ibex/vendor/lowrisc_ip/dvsim/SynCfg.py
Rupert Swarbrick f35a407906 Update lowrisc_ip to lowRISC/opentitan@5cae0cf1
Update code from upstream repository
https://github.com/lowRISC/opentitan to revision
5cae0cf1fac783e0d0df8c8597bf65322a696a56

* Allow different assertion "backends" in prim_assert.sv (Rupert
  Swarbrick)
* [prim_prince/doc] Update documentation (Michael Schaffner)
* [prim_prince] Add option to instantiate a registers half-way
  (Michael Schaffner)
* [prim_cipher_pkg] Reuse sbox4_8bit to build wider sbox layers
  (Michael Schaffner)
* [dv/prim] add PRESENT testbench (Udi Jonnalagadda)
* [uvmdvgen] Scoreboard update. (Srikrishna Iyer)
* [flash_ctrl dv] Fix V1 tests (Srikrishna Iyer)
* [prim_cipher_pkg] Replicate common subfunctions for other widths
  (Michael Schaffner)
* [prim/present] fix PRESENT decryption bugs (Udi Jonnalagadda)
* [prim/present] fix some PRESENT encryption bugs (Udi Jonnalagadda)
* [dv] Add get_mem DPI function to Verilator simutil (Stefan
  Wallentowitz)
* [lint/entropy_src] Add the entropy source to the lint regression
  (Michael Schaffner)
* [style-lint] Fix some common style lint warnings (Michael Schaffner)
* first set of security checks added to D2 checklist (Scott Johnson)
* [fpv/tooling] add FPV class extension in dvsim (Cindy Chen)
* [dvsim/lint] Minor fixes for printout issues and result parser
  status (Michael Schaffner)
* [syn] Print detailed messages to .md if publication is disabled
  (Michael Schaffner)
* [prim_util] Do not use $clog2() in Xcelium (Philipp Wagner)
* [prim] Update ResetValue parameter in prim_flop_2sync (Timothy Chen)
* Modified some command-line arguments for DSim (Aimee Sutton)
* [prim_util] Make prim_util a package (Philipp Wagner)
* [dv] Move mem checking to scb (Weicai Yang)
* [lint] Make PINCONNECTEMPTY Verilator waiver common (Philipp Wagner)
* [prim] - Fix generic flash enum reference (Timothy Chen)
* [prim_ram_*adv] Mark cfg port as unused (Philipp Wagner)
* [prim_fifo_sync] Use vbits() for simpler code (Philipp Wagner)
* [prim_flash] Add reset to held_part (Eunchan Kim)
* [lint] Add more lint waivers (Philipp Wagner)
* [dv] Add random backdoor for csr_hw_reset (Weicai Yang)
* [dv] Add set_freq_khz in clk_rst_if (Weicai Yang)
* [prim] Close GAPI file handle in primgen (Philipp Wagner)
* [fpv/prim_packer] fix CI failure due to index out of bound (Cindy
  Chen)
* [prim_arbiter_*] Propagate parameter changes (Michael Schaffner)
* [prim_arbiter_tree] Fix incorrect arbitration behavior (Michael
  Schaffner)
* [prim_arbiter_ppc] Add more FPV fairness checks (Michael Schaffner)
* [prim_ram*] Add an assertion that checks wmask consistency (Michael
  Schaffner)
* [memutil] Increase max memory width to 256bit (Tom Roberts)
* [flash] - Add flash info page support (Timothy Chen)

Signed-off-by: Rupert Swarbrick <rswarbrick@lowrisc.org>
2020-07-01 14:54:13 +01:00

379 lines
14 KiB
Python

# Copyright lowRISC contributors.
# Licensed under the Apache License, Version 2.0, see LICENSE for details.
# SPDX-License-Identifier: Apache-2.0
r"""
Class describing synthesis configuration object
"""
import logging as log
from pathlib import Path
import hjson
from tabulate import tabulate
from OneShotCfg import OneShotCfg
from utils import print_msg_list, subst_wildcards
class SynCfg(OneShotCfg):
"""Derivative class for synthesis purposes.
"""
def __init__(self, flow_cfg_file, proj_root, args):
super().__init__(flow_cfg_file, proj_root, args)
def __post_init__(self):
super().__post_init__()
# Set the title for synthesis results.
self.results_title = self.name.upper() + " Synthesis Results"
def gen_results_summary(self):
'''
Gathers the aggregated results from all sub configs
'''
# Generate results table for runs.
log.info("Create summary of synthesis results")
results_str = "## " + self.results_title + " (Summary)\n\n"
results_str += "### " + self.timestamp_long + "\n\n"
self.results_summary_md = results_str + "\nNot supported yet.\n"
print(self.results_summary_md)
# Return only the tables
return self.results_summary_md
def _gen_results(self):
# '''
# The function is called after the regression has completed. It looks
# for a regr_results.hjson file with aggregated results from the
# synthesis run. The hjson needs to have the following (potentially
# empty) fields
#
# results = {
# "tool": "dc",
# "top" : <name of toplevel>,
#
# "messages": {
# "flow_errors" : [],
# "flow_warnings" : [],
# "analyze_errors" : [],
# "analyze_warnings" : [],
# "elab_errors" : [],
# "elab_warnings" : [],
# "compile_errors" : [],
# "compile_warnings" : [],
# },
#
# "timing": {
# # per timing group (ususally a clock domain)
# # in nano seconds
# <group> : {
# "tns" : <value>,
# "wns" : <value>,
# "period" : <value>,
# ...
# }
# },
#
# "area": {
# # gate equivalent of a NAND2 gate
# "ge" : <value>,
#
# # summary report, in GE
# "comb" : <value>,
# "buf" : <value>,
# "reg" : <value>,
# "macro" : <value>,
# "total" : <value>,
#
# # hierchical report of first submodule level
# "instances" : {
# <name> : {
# "comb" : <value>,
# "buf" : <value>,
# "reg" : <value>,
# "macro" : <value>,
# "total" : <value>,
# },
# ...
# },
# },
#
# "power": {
# "net" : <value>,
# "int" : <value>,
# "leak" : <value>,
# },
#
# "units": {
# "voltage" : <value>,
# "capacitance" : <value>,
# "time" : <value>,
# "dynamic" : <value>,
# "static" : <value>,
# }
# }
#
# note that if this is a master config, the results will
# be generated using the _gen_results_summary function
# '''
def _create_entry(val, norm=1.0, total=None, perctag="%"):
"""
Create normalized entry with an optional
percentage appended in brackets.
"""
if val is not None and norm is not None:
if total is not None:
perc = float(val) / float(total) * 100.0
entry = "%2.1f %s" % (perc, perctag)
else:
value = float(val) / norm
entry = "%2.1f" % (value)
else:
entry = "--"
return entry
self.result = {}
# Generate results table for runs.
results_str = "## " + self.results_title + "\n\n"
results_str += "### " + self.timestamp_long + "\n"
results_str += "### Synthesis Tool: " + self.tool.upper() + "\n\n"
# TODO: extend this to support multiple build modes
for mode in self.build_modes:
# results_str += "## Build Mode: " + mode.name + "\n\n"
result_data = Path(
subst_wildcards(self.build_dir, {"build_mode": mode.name}) +
'/results.hjson')
log.info("looking for result data file at %s", result_data)
try:
with open(result_data, "r") as results_file:
self.result = hjson.load(results_file, use_decimal=True)
except IOError as err:
log.warning("%s", err)
self.result = {
"messages": {
"flow_errors": ["IOError: %s" % err],
"flow_warnings": [],
"analyze_errors": [],
"analyze_warnings": [],
"elab_errors": [],
"elab_warnings": [],
"compile_errors": [],
"compile_warnings": [],
},
}
# Message summary
# results_str += "### Tool Message Summary\n\n"
if "messages" in self.result:
header = [
"Build Mode", "Flow Warnings", "Flow Errors",
"Analyze Warnings", "Analyze Errors", "Elab Warnings",
"Elab Errors", "Compile Warnings", "Compile Errors"
]
colalign = ("left", ) + ("center", ) * (len(header) - 1)
table = [header]
messages = self.result["messages"]
table.append([
mode.name,
str(len(messages["flow_warnings"])) + " W ",
str(len(messages["flow_errors"])) + " E ",
str(len(messages["analyze_warnings"])) + " W ",
str(len(messages["analyze_errors"])) + " E ",
str(len(messages["elab_warnings"])) + " W ",
str(len(messages["elab_errors"])) + " E ",
str(len(messages["compile_warnings"])) + " W ",
str(len(messages["compile_errors"])) + " E ",
])
if len(table) > 1:
results_str += tabulate(table,
headers="firstrow",
tablefmt="pipe",
colalign=colalign) + "\n\n"
else:
results_str += "No messages found\n\n"
else:
results_str += "No messages found\n\n"
# Hierarchical Area report
results_str += "### Circuit Complexity in [kGE]\n\n"
if "area" in self.result:
header = [
"Instance", "Comb ", "Buf/Inv", "Regs", "Macros", "Total",
"Total [%]"
]
colalign = ("left", ) + ("center", ) * (len(header) - 1)
table = [header]
# print top-level summary first
row = ["**" + self.result["top"] + "**"]
try:
kge = float(self.result["area"]["ge"]) * 1000.0
for field in ["comb", "buf", "reg", "macro", "total"]:
row += [
"**" +
_create_entry(self.result["area"][field], kge) +
"**"
]
row += ["**--**"]
table.append(row)
# go through submodules
for name in self.result["area"]["instances"].keys():
if name == self.result["top"]:
continue
row = [name]
for field in ["comb", "buf", "reg", "macro", "total"]:
row += [
_create_entry(
self.result["area"]["instances"][name]
[field], kge)
]
# add percentage of total
row += [
_create_entry(
self.result["area"]["instances"][name][field],
kge, self.result["area"]["total"], "%u")
]
table.append(row)
except TypeError:
results_str += "Gate equivalent is not properly defined\n\n"
if len(table) > 1:
results_str += tabulate(table,
headers="firstrow",
tablefmt="pipe",
colalign=colalign) + "\n\n"
else:
results_str += "No area report found\n\n"
else:
results_str += "No area report found\n\n"
# Timing report
results_str += "### Timing in [ns]\n\n"
if "timing" in self.result and "units" in self.result:
header = ["Clock", "Period", "WNS", "TNS"]
colalign = ("left", ) + ("center", ) * (len(header) - 1)
table = [header]
for clock in self.result["timing"].keys():
row = [clock]
row += [
_create_entry(
self.result["timing"][clock]["period"],
1.0E-09 / float(self.result["units"]["time"])),
_create_entry(
self.result["timing"][clock]["wns"], 1.0E-09 /
float(self.result["units"]["time"])) + " EN",
_create_entry(
self.result["timing"][clock]["tns"], 1.0E-09 /
float(self.result["units"]["time"])) + " EN"
]
table.append(row)
if len(table) > 1:
results_str += tabulate(table,
headers="firstrow",
tablefmt="pipe",
colalign=colalign) + "\n\n"
else:
results_str += "No timing report found\n\n"
else:
results_str += "No timing report found\n\n"
# Power report
results_str += "### Power Estimates in [mW]\n\n"
if "power" in self.result and "units" in self.result:
header = ["Network", "Internal", "Leakage", "Total"]
colalign = ("center", ) * len(header)
table = [header]
try:
self.result["power"]["net"]
power = [
float(self.result["power"]["net"]) *
float(self.result["units"]["dynamic"]),
float(self.result["power"]["int"]) *
float(self.result["units"]["dynamic"]),
float(self.result["power"]["leak"]) *
float(self.result["units"]["static"])
]
total_power = sum(power)
row = [_create_entry(power[0], 1.0E-3) + " / " +
_create_entry(power[0], 1.0E-3, total_power),
_create_entry(power[1], 1.0E-3) + " / " +
_create_entry(power[1], 1.0E-3, total_power),
_create_entry(power[2], 1.0E-3) + " / " +
_create_entry(power[2], 1.0E-3, total_power),
_create_entry(total_power, 1.0E-3)]
table.append(row)
# in case fp values are NoneType
except TypeError:
results_str += "No power report found\n\n"
if len(table) > 1:
results_str += tabulate(table,
headers="firstrow",
tablefmt="pipe",
colalign=colalign) + "\n\n"
else:
results_str += "No power report found\n\n"
# Append detailed messages if they exist
# Note that these messages are omitted in publication mode
hdr_key_pairs = [("Flow Warnings", "flow_warnings"),
("Flow Errors", "flow_errors"),
("Analyze Warnings", "analyze_warnings"),
("Analyze Errors", "analyze_errors"),
("Elab Warnings", "elab_warnings"),
("Elab Errors", "elab_errors"),
("Compile Warnings", "compile_warnings"),
("Compile Errors", "compile_errors")]
has_msg = False
for _, key in hdr_key_pairs:
if key in self.result['messages']:
has_msg = True
break
if has_msg and not self.args.publish:
results_str += "\n### Errors and Warnings for Build Mode `'" + mode.name + "'`\n"
for hdr, key in hdr_key_pairs:
msgs = self.result['messages'].get(key)
results_str += print_msg_list("#### " + hdr, msgs, self.max_msg_count)
# TODO: add support for pie / bar charts for area splits and
# QoR history
self.results_md = results_str
# Write results to the scratch area
self.results_file = self.scratch_path + "/results_" + self.timestamp + ".md"
log.info("Detailed results are available at %s", self.results_file)
with open(self.results_file, 'w') as f:
f.write(self.results_md)
return self.results_md