Merge pull request #37 from edwin7026/coverage_calc_time

Decrease coverage calculation time

CHANGELOG.md

@@ -2,13 +2,19 @@
 
 This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.12.0] - 2022-04-15
+- Parallelized coverage computation.
+- Added feature to remove coverpoints when hit.
+- Added CLI option to specify number of processes to be spawned.
+- Added CLI option to turn on/off feature to remove hit coverpoints.
+
 ## [0.11.0] - 2022-04-03
 - Added plugins to use new rvopcode format
 - Added CLI option to setup rvopcode plugin
 
 ## [0.10.2] - 2022-03-15
 - Added method to generate data patterns for bitmanip instructions
--
+
 ## [0.10.1] - 2022-02-10
 - Added vxsat to supported csr_regs
 - Added comments to coverpoint functions for P-ext

docs/source/isac_cov_calc.rst

@@ -0,0 +1,48 @@
+********************************
+Performance improvements to ISAC
+********************************
+
+Coverage computation in ISAC is an iterative process where all coverpoints
+are evaluated for every instruction. This causes an exponential increase in coverage
+computation time as the number of coverpoints increases. Following are methods leveraged to produce
+performance improvements to ISAC.
+
+Parallelized coverage calculation
+#################################
+
+Parallelization is one way to reduce the time taken to compute the coverage.
+Parallelization of coverage computation is achieved by the parallelization of ``compute_per_line()``
+method in ``coverage.py`` file. The implementation resorts to the use of queues to relay statistics of hit
+coverpoint back to the main process.
+
+Usage
+*****
+The number of processes to be spawned for coverage computation can be provided using,
+``--procs`` or ``-p`` option while running ISAC. As an example, ::
+    
+    riscv_isac --verbose info coverage -d -t add-01.log --parser-name c_sail --decoder-name internaldecoder -o coverage.rpt --sig-label begin_signature end_signature --test-label rvtest_code_begin rvtest_code_end -e add-01.elf -c dataset.cgf -c rv32i.cgf -x 32 -l add --procs 3
+
+The above command partitions the supplied CGF file into three different sets based on the coverlabels. Three processes would be
+spawned and are supplied with a part of the partitioned CGF file dictionary. Each of the processes are asynchronously supplied with ``instructionObject`` 
+objects via independent queues. The processes compute coverage independently against the decoded ``instructionObject`` with its own set of coverlabels 
+and its respective coverpoints. The updated cgf dictionary and coverpoint hit statistics are relayed back to the main process through a queue.
+The cgf dictionary and the statistics are merged before continuing further.
+
+The default number of processes spawned is 1
+
+Removal of Hit Coverpoints from CGF Dictionary
+##############################################
+
+For architectural testing, it is enough if a coverpoint is hit just once and has impact on the signature. When this feature 
+is enabled, coverpoints would be deleted from the working CGF dictionary once hit to avoid evaluating it for all subsequent
+instructions. 
+
+Usage
+*****
+This feature can be enabled using the ``--no-count`` option while running ISAC. As an example, ::
+
+    riscv_isac --verbose info coverage -d -t add-01.log --parser-name c_sail --decoder-name internaldecoder -o coverage.rpt --sig-label begin_signature end_signature --test-label rvtest_code_begin rvtest_code_end -e add-01.elf -c dataset.cgf -c rv32i.cgf -x 32 -l add --no-count
+
+When ``--no-count`` option supplied, the hit coverpoints are stored and are then deleted from the CGF dictionary at the end of every iteration
+
+This feature is off by default.

riscv_isac/__init__.py

@@ -4,4 +4,4 @@
 
 __author__ = """InCore Semiconductors Pvt Ltd"""
 __email__ = '[email protected]'
-__version__ = '0.11.0'
+__version__ = '0.12.0'