| Section: UVM 1.2 Release Notes | |
| Topic: General Overview | |
| This is release 1.2 of the Universal Verification Methodology (UVM) | |
| reference implementation from Accellera. This distribution includes a | |
| SystemVerilog based class library, examples, UVM Stardard Reference | |
| (PDF and HTML), and a User's Guide. | |
| The class library is a reference implementation of the standard. It, | |
| the examples, and User Guide provide guidance to users of the | |
| standard. For additional information on the Accellera standard, see | |
| <http://www.accellera.org/activities/vip> | |
| A good source for information, answers, and resources is | |
| <http://www.accellera.org/community/uvm>. It contains forums on | |
| methodology, tool specific issues, and commercial announcements. It | |
| also has links to additional resources, Accellera sponsored events, | |
| and other UVM events. | |
| Topic: What's Changed | |
| Release 1.2 contains enhancements and fixes to errata. | |
| UVM 1.2 fixes the following enhancements and issues reported under the | |
| Mantis bug-tracking system. Issues affecting the reference | |
| documentation (API changes or semantic changes) are follwed with a | |
| plus sign (+). Issues affecting backward compatibility are folled with | |
| an asterisk (*) | |
| For detailed information on a specific issue, refer to the Mantis | |
| database at http://www.eda.org/svdb/view.php?id=<id>, where <id> is | |
| the Mantis id. | |
| Topic: 1.1 to 1.2 Migration Script | |
| Many of the backwards compatibility concerns caused when transitioning | |
| from UVM version 1.1 to 1.2 can be addressed with a reasonably simple | |
| search-and-replace in user code. To that end, the ~uvm11-to-uvm12.pl~ | |
| is provided. | |
| The following command will update all SystemVerilog source files | |
| located in the current working directory and any sub-directory: | |
| | % uvm11-to-uvm12.pl --write | |
| See the ~--help~ command-line option for additional details. | |
| The script may not be able to automatically identify all cases of | |
| backwards incompatibility, and the user may need to fix some | |
| manually. Because most of these issues will result in a compile-time | |
| error, they should be easy to identify. | |
| Group: Mantis 3407 (+) | |
| uvm_event should be a parameterized class | |
| The ~uvm_event~ class was split into two classes, an abstract | |
| <uvm_event_base>, and a parameterized <uvm_event#(T)>. | |
| This should have minimal backwards compatibility concerns, as any code | |
| using ~uvm_event~ without explicitly declaring a ~#(T)~ will default | |
| to ~uvm_event#(uvm_object)~. | |
| Group: Mantis 3472 (*) | |
| Deprecate get_config/set_config interface for configuration | |
| The methods ~set_config_int~, ~set_config_string~ and | |
| ~set_config_object~ (and their ~get~ counterparts) are now deprecated | |
| in favor of <uvm_config_int>, <uvm_config_string> and | |
| <uvm_config_object>. | |
| Topic: Backwards Compatibility | |
| Care must be taken when switching from ~set_config_object~ and | |
| ~get_config_object~ to <uvm_config_object>, as the ~clone~ | |
| functionality of the old API no longer exists. | |
| Group: Mantis 3557 | |
| ref guide doesn't show names of UVM runtime phases | |
| The documentation for <UVM Common Phases> and <UVM Run-Time Phases> | |
| has been updated to include a definition of how phases are named. | |
| Group: Mantis 3586 (+) | |
| UVM_DISABLE_AUTO_ITEM_RECORDING should be a run_time option | |
| <uvm_sqr_if_base#(REQ,RSP)::disable_auto_item_recording> and | |
| <uvm_sqr_if_base#(REQ,RSP)::is_auto_item_recording_enabled> have been | |
| added to the <uvm_sqr_if_base#(REQ,RSP)> class, allowing the driver | |
| runtime control over whether or not a sequencer should automatically | |
| record the items which flow through the sequence item port. | |
| Group: Mantis 3693 (+) | |
| Setting enum values by name from command line doesn't work | |
| The <uvm_enum_wrapper#(T)> utility class has been added, which provides the | |
| ability to retrieve an enum value when provided with a string matching | |
| the ~enum.name()~ for that value. | |
| This class is used by the <`uvm_field_enum> macro when performing | |
| auto-configuration, such that values can be set using | |
| ~uvm_config_db#(string)~ (or the <uvm_cmdline_processor::+uvm_set_config_string> plusarg). | |
| Group: Mantis 3741 (+) | |
| Phase's default sequence should also have get_config string support | |
| A new plusarg has been added (<uvm_cmdline_processor::+uvm_set_default_sequence>) which allows | |
| for the ~default_sequence~ a sequencer executes in | |
| <uvm_sequencer_base::start_phase_sequence> to be set directly from the command line. | |
| General support for setting ~default_sequence~ as a string was not | |
| provided, as during runtime the ~default_sequence~ should always be set using | |
| either <uvm_sequence_base> or <uvm_object_wrapper>. | |
| Group: Mantis 3783 | |
| Remove all $display calls from BCL, except report_server | |
| All messaging is now routed via <uvm_report_server>. | |
| Group: Mantis 3887 (*) | |
| request facility to identify unused factory overrides | |
| The ~uvm_factory~ class was split into two classes, an abstract | |
| <uvm_factory>, a default implementation of <uvm_default_factory>. | |
| This allows for alternate factories (including those with additional | |
| debug functionality) to be created. | |
| Control over the global factory was | |
| migrated to the new <uvm_coreservice_t> class. | |
| Topic: Backwards Compatibility: | |
| The undocumented ~uvm_pkg::factory~ variable has been removed, as it | |
| was unsafe during static initialization, and incompatible with the | |
| <uvm_coreservice_t> class. References to this variable need to be | |
| upgraded to retrieve the factory via <uvm_factory::get>. | |
| Group: Mantis 3893 (+) | |
| Feature request to create a non-hierarchical objection | |
| <uvm_objection::set_propagate_mode> has been added to provide control | |
| over whether or not objections propagate. | |
| Group: Mantis 3901 | |
| Feature request to prevent objection from throwing | |
| decrement-below-zero error if the count is 0 | |
| An error is no longer produced if <uvm_objection::drop_objection> is | |
| called with a count of ~0~, when the current object count is ~0~. | |
| Group: Mantis 4032 (+) | |
| Not possible to undo a factory override. | |
| A factory override can now be "undone" by supplying a type | |
| as an override for itself. | |
| i.e. | |
| | // Perform an override | |
| | set_type_override_by_type(old_type::get_type(), new_type::get_type()); | |
| | // Undo the override | |
| | set_type_override_by_type(old_type::get_type(), old_type::get_type()); | |
| Group: Mantis 4091 | |
| bug in the ~get_full_name~ method of <uvm_reg_file>. | |
| The ~get_full_name~ method of <uvm_reg_file> now includes the top-level block name. | |
| Group: Mantis 4160 | |
| uvm_comparer not using verbosity setting for print_msg() | |
| The <uvm_comparer::print_msg> method now obeys the | |
| <uvm_comparer::verbosity> setting. | |
| Group: Mantis 4181 (+) | |
| transaction order unclear and cant be changed when bus/reg size differ | |
| The ordering of transactions generated when the size of a | |
| register/field is different than the size of the bus can now be | |
| controlled via the new <uvm_reg_transaction_order_policy> class. | |
| Group: Mantis 4253 | |
| Volatile fields should always need updating | |
| Volatile register fields will always cause a write(), even if mirrored | |
| and desired values are the same. | |
| Group: Mantis 4269 (*) | |
| Add UVM_ prefix to enumerated values | |
| Various top-level enumeration values (such as those declared in | |
| <uvm_sequence_state_enum>) did not have the ~UVM_~ prefix. This would | |
| cause compilation errors when the ~uvm_pkg~ was wildcard imported into | |
| a scope which had already declared the same names. These enumerations | |
| have been corrected to include the proper ~UVM_~ prefix. | |
| Topic: Backawards Compatibility | |
| References to the values within <uvm_sequence_state_enum> and | |
| <uvm_sequencer_arb_mode> must now be prefixed with ~UVM_~. | |
| Group: Mantis 4306 | |
| Bad reference to "record_field_int" in reference guide | |
| A bad reference to ~record_field_int~ has been corrected to | |
| ~record_field~ inside of <uvm_recorder>. | |
| Group: Mantis 4307 | |
| Bad reference to "DEC" in reference guide | |
| Bad references to "DEC" have been corrected to <UVM_DEC> within | |
| <uvm_object::do_print> and <uvm_object::do_record>. | |
| Group: Mantis 4366 (+) | |
| uvm_sequence_library documentation missing from html documentation | |
| The <uvm_sequence_library> class has been updated from "beta" to | |
| "official" status. As such, it is now documented in the reference | |
| guide. | |
| Group: Mantis 4378 | |
| Documentation of uvm_component::set_inst_override_by_type() | |
| incorrectly documents uvm_factory::set_inst_override_by_type() | |
| The documentation of <uvm_component::set_inst_override_by_type> as | |
| been updated to reflect the proper ordering of the parameters to | |
| <uvm_factory::set_inst_override_by_type>. | |
| Group: Mantis 4382 (+) | |
| some symbols in uvm_radix_enum undocumented | |
| Documentation was added for <UVM_UNFORMAT2>, <UVM_UNFORMAT4>, | |
| <UVM_REAL>, <UVM_REAL_DEC>, and <UVM_REAL_EXP> within the | |
| <uvm_radix_enum>. | |
| Group: Mantis 4389 (+) | |
| Deprecate hierarchical propagation of objections; add mode bit in | |
| interim | |
| <Mantis 3893 (+)> provides the mode bit for controlling propagation of | |
| objections. Hierarchical propagation was not outright deprecated in | |
| this release. | |
| Group: Mantis 4391 | |
| Redundant code in uvm_port_base | |
| Some redundant code in the <uvm_port_base#(IF)> was removed. | |
| Group: Mantis 4416 (*) | |
| Remove the confusing uvm_severity type | |
| The confusing ~uvm_severity_type~ was deprecated, and replaced | |
| internally using <uvm_severity>. | |
| Topic: Backwards Compatibility | |
| This change should have minimal effect on user code, as the old type | |
| was simply deprecated, as opposed to removed. The <1.1 to 1.2 Migration Script> | |
| can be used to replace any occurances of the old | |
| type in user code. | |
| Group: Mantis 4418 | |
| set_report_verbosity doc unclear about setting value or threshold | |
| The documentation of | |
| <uvm_report_object::set_report_verbosity_level>, | |
| <uvm_report_object::set_report_id_verbosity>, and | |
| <uvm_report_object::set_report_severity_id_verbosity> were updated to | |
| provide a clearer definition of exactly what the ~verbosity~ values | |
| implied. | |
| Group: Mantis 4425 | |
| BUSY flag not cleared when register model is reset | |
| The register model was updated to clear the BUSY flag when reset. | |
| Group: Mantis 4430 (+) | |
| Refactor start_phase_sequence logic to be able to kill sequence on | |
| phase ending | |
| The method <uvm_sequencer_base::stop_phase_sequence> was added which | |
| acts as the bookend to <uvm_sequencer_base::start_phase_sequence>. It | |
| will call <uvm_sequence_base::kill> on the ~default_sequence~ if the | |
| sequence is still executing when the phase ends. | |
| Topic: Backwards Compatibility | |
| While technically acting different than the previous implementation, | |
| it is unlikely that this will result in any backwards incompatibility | |
| challenges, as the old mechanism simply terminated the process which | |
| launched the ~default_sequence~, resulting in a "zombie" sequence. | |
| Group: Mantis 4431 (*)(+) | |
| Enhance uvm_sequence::starting_phase | |
| The ~uvm_sequence_base::starting_phase~ variable was publicly accessible | |
| and not protected from writes-after-reads. This would result in | |
| strange behavior (including FATALS and HANGS) if the value was changed | |
| between a ~raise~ and ~drop~ of the objection within the phase. | |
| The variable was removed, and is now only accessible via the | |
| <uvm_sequence_base::set_starting_phase> and | |
| <uvm_sequence_base::get_starting_phase> accessors. An error will be | |
| thrown if there are any attempts to 'set' the starting phase after it | |
| has been retrieved via a 'get'. | |
| Topic: Backwards Compatibility | |
| This change could potentially effect users, as references to the | |
| ~uvm_sequence_base::starting_phase~ variable will now miscompile. | |
| The <1.1 to 1.2 Migration Script> can be used to search-and-replace | |
| within the user code, updating it to the new methods. | |
| Group: Mantis 4432 (+) | |
| Provide auto-object-to-phase in uvm_sequence_base | |
| The <uvm_sequence_base::set_automatic_phase_objection> method was | |
| added, simplifying the code required to object to the ~starting_phase~ | |
| inside of a phase. | |
| Group: Mantis 4433 (*) | |
| Clean up the number of objections created by phasing | |
| Objections only make sense in <uvm_task_phase>s which are of the | |
| <UVM_PHASE_IMP> type. For all other phases and types, the objections | |
| have been removed, and any attempt to access the objection will result | |
| in an error. | |
| Topic: Backwards Compatibility | |
| While technically acting different than the previous implementation, | |
| it is unlikely that this will result in any backwards incompatibility | |
| challenges. Any user code which would cause the error to occur would | |
| have been silently ignored by the library in the past, which meant | |
| that whatever the user was ~trying~ to do wouldn't have worked. | |
| Group: Mantis 4434 (+) | |
| Provide get_objection_count in uvm_phase | |
| The method <uvm_phase::get_objection_count> was added, providing the | |
| user with a cleaner mechanism for dropping all raised objections. | |
| | // Old Code: | |
| | uvm_objection phase_done; | |
| | phase_done = phase.get_objection(); | |
| | phase_done.drop_objection(this, | |
| | "dropping all objections", | |
| | phase_done.get_objection_count(this)); | |
| | | |
| | // New Code: | |
| | phase.drop_objection(this, | |
| | "dropping all objections", | |
| | phase.get_objection_count(this)); | |
| | | |
| Group: Mantis 4439 (+) | |
| seq_item_port.put_response is not documented | |
| The previously undocumented <uvm_sqr_if_base#(REQ,RSP)::put_response> | |
| was added to the standard, providing a non-blocking mechanism for | |
| sending response information from a driver. | |
| Group: Mantis 4440 | |
| jump logic too early in phase state pipeline | |
| The previous implementation of the ~jump~ logic <uvm_phase> caused | |
| jumps during the ~phase_ended~ callback to be ignored. This has been | |
| corrected. | |
| Group: Mantis 4469 (+) | |
| Provide schedule introspection via get_predecessor/successor_nodes | |
| The <uvm_phase::get_adjacent_predecessor_nodes> and | |
| <uvm_phase::get_adjacent_successor_nodes> methods were adding the | |
| <uvm_phase> api. This provides the user with the ability to | |
| programmatically traverse the entire phase graph. | |
| Group: Mantis 4476 (+) | |
| uvm_sequence fields req/rsp are not documented | |
| The previously undocumented <uvm_sequence#(REQ,RSP)::req> and | |
| <uvm_sequence#(REQ,RSP)::rsp> fields were added to the standard. | |
| Group: Mantis 4488 (+) | |
| sequencer_base should throw error if user allows wait_for_relevant to | |
| return while is_relevant=0 | |
| An error will now be thrown if the sequencer exceeds the allowable | |
| number of returns from <uvm_sequence_base::wait_for_relevant> without | |
| <uvm_sequence_base::is_relevant> returning true. The allowable | |
| maximum number can be controlled via the | |
| <uvm_sequencer_base::set_max_zero_time_wait_relevant_count> method. | |
| Group: Mantis 4513 (+) | |
| Add uvm_mem_mam_policy allocator to uvm_vreg::allocate() | |
| The <uvm_vreg::allocate> API was expanded to include an optional third | |
| ~alloc~ argument. | |
| Group: Mantis 4516 (*) | |
| Remove the unnecessary over-prioritization of uvm_sequence_base in | |
| default_sequence | |
| The previous implementation of <uvm_sequencer_base> over-prioritized | |
| ~default_sequence~ configurations which were of type | |
| <uvm_sequence_base>, such that they would override configurations of | |
| type <uvm_object_wrapper>, even if the wrapper configurations were | |
| done with higher precedence. | |
| Topic: Backwards Compatibility | |
| While technically acting different than the previous implementation, | |
| it is unlikely that this will result in any backwards incompatibility | |
| challenges. The previous implementation would have resulted in | |
| unexpected behavior at best. | |
| Group: Mantis 4517 (+) | |
| Add phase-transition callbacks | |
| The <uvm_phase_cb> and <uvm_phase_cb_pool> were added to provide | |
| additional hooks into the phasing API. | |
| Group: Mantis 4518 (*)(+) | |
| Make UVM_OBJECT_MUST_HAVE_CONSTRUCTOR the default behavior | |
| ~Mantis 3770~ provided the ability to optionally enforce an explicit | |
| constructor with a string-type ~name~ argument in the UVM 1.1a, via | |
| the ~`UVM_OBJECT_MUST_HAVE_CONSTRUCTOR~ symbol. This functionality is | |
| now the ~*default*~ behavior for the UVM. | |
| Topic: Why uvm_object constructors are now mandatory | |
| The UVM recommends that the following constructor be specified for any class | |
| extended from <uvm_object>: | |
| | class my_obj extends uvm_object; | |
| | ... | |
| | `uvm_object_utils(my_obj) | |
| | | |
| | function new(string name = "my_obj") | |
| | super.new(name); | |
| | ... | |
| | endfunction | |
| | ... | |
| | endclass | |
| However, in UVM 1.0 and UVM 1.1, the presence of such a constructor were not enforced | |
| by the library and they were technically optional. | |
| If no constructor was specified, a default constructor was provided by SystemVerilog: | |
| | class my_obj extends uvm_object; | |
| | ... | |
| | `uvm_object_utils(my_obj) | |
| | | |
| | function new(); | |
| | endfunction | |
| | ... | |
| | endclass | |
| Note the important difference: in the default constructor, there is no ~name~ argument. | |
| Because the factory could not rely on the presence of that argument, | |
| <uvm_object> instances were created by calling ~new()~ without any arguments | |
| and subsequently setting the object instance name using <uvm_object::set_name>. | |
| | obj = new(); | |
| | if (name != "") obj.set_name(name); | |
| In most cases, this difference was not significant. | |
| However, for any class that makes use of the value of the ~name~ argument | |
| in the constructor, | |
| | class my_obj extends uvm_object; | |
| | your_obj sub; | |
| | | |
| | `uvm_object_utils(my_obj) | |
| | | |
| | function new(string name = "my_obj") | |
| | super.new(name); | |
| | sub = your_obj::type_id::create({name,".sub"}); | |
| | endfunction | |
| | ... | |
| | endclass | |
| a difference in behavior was observed between | |
| instantiating an object directly: | |
| | my_obj o = new("o"); | |
| and instantiating an object via the | |
| factory: | |
| | my_obj o = my_obj::type_id::create("o"); | |
| The difference could have been worked around by | |
| overriding the <uvm_object::set_name> method to | |
| percolate the effect of the name change. | |
| However, that work-around was often impossible or too late. | |
| In the examples above, the former creates an instance of ~your_obj~ named "o.sub" | |
| whereas the latter creates an instance named "my_obj.sub". | |
| It made it impossible to control the sub-object factory based on the | |
| parent object name. | |
| Instantiating coverage group is another example where the work-around did not work. | |
| Coverage groups cannot be renamed and | |
| must be instantiated in the encapsulating object's constructor. | |
| The following example would create coverage groups that are always named "my_obj" | |
| when using the object factory. | |
| | class my_obj extends uvm_object; | |
| | covergroup my_cg(string name); | |
| | option.name = name; | |
| | ... | |
| | endgroup | |
| | | |
| | `uvm_object_utils(my_obj) | |
| | | |
| | function new(string name = "my_obj") | |
| | super.new(name); | |
| | my_cg = new(name); | |
| | endfunction | |
| | ... | |
| | endclass | |
| Because the object factory is such a fundamental element of UVM, | |
| it is important that it be able to rely on the presence of a ~name~ argument | |
| in object constructors. | |
| Topic: Backwards Compatibility | |
| In UVM 1.2, the UVM object factory now requires that uvm_object have a | |
| constructor with a name argument. Should this constructor be missing, | |
| a syntax error will be issued. | |
| However, because much of the factory code is hidden in the utility macros, | |
| the syntax error may not point to the ultimate cause of the problem | |
| and may be confusing the diagnose. | |
| For example, this simple file | |
| | `include "uvm_macros.svh" | |
| | program test; | |
| | | |
| | import uvm_pkg::*; | |
| | | |
| | class my_obj extends uvm_object; | |
| | `uvm_object_utils(my_obj) | |
| | endclass | |
| | | |
| | endprogram | |
| will produce the following error | |
| messages: | |
| Questa: | |
| | TBD | |
| IUS: | |
| |file: test.sv | |
| |`uvm_object_utils(my_obj) | |
| | | | |
| |ncvlog: *E,TOOMAC (test.sv,8|24): too many actual arguments [10.2.2][10.3(IEEE)]. | |
| VCS: | |
| | Error-[TMAFTC] Too many arguments to function/task call | |
| | $UVM_HOME/src/base/uvm_registry.svh, 197 | |
| | "my_obj::new(name)" | |
| | The above function/task call is done with more arguments than needed. | |
| | | |
| | | |
| | Error-[TMAFTC] Too many arguments to function/task call | |
| | test.sv, 6 | |
| | "my_obj::new(name)" | |
| | The above function/task call is done with more arguments than needed. | |
| When using VCS, ignore the first message reported in uvm_registry.svh | |
| and focus on the subsequent one identifying the invalid constructor being called. | |
| The reported line number will be on the ~endclass~ token of the class | |
| that is missing the required constructor. | |
| Topic: Code migration script | |
| Suitable constructors may be automatically added to classes that are missing them | |
| by using the ~add_uvm_object_new.pl~ script. | |
| The following command will update all SystemVerilog source files | |
| located in the current working directory and any sub-directory: | |
| | % add_uvm_object_new.pl --write | |
| See the ~--help~ command-line option for more details. | |
| The script may not be able to automatically identify all classes | |
| with missing constructors, | |
| or users may prefer to manually modify their source files. | |
| The script is also unable to fix unsuitable constructors; | |
| those will need to be fixed manually. | |
| Because the error is a compile-time error, it will be easy to identify | |
| any remaining changes that are required. | |
| Topic: Migrating code using UVM 1.2 | |
| To ease the process of migrating code to UVM 1.2, | |
| the old (incorrect) behavior remain available. | |
| To restore the old behavior, it is necessary to define | |
| the `UVM_OBJECT_DO_NOT_NEED_CONSTRUCTOR symbol when compiling the UVM library: | |
| | % ... +define+UVM_OBJECT_DO_NOT_NEED_CONSTRUCTOR uvm_pkg.sv | |
| Users must be aware that this may result in difference in behavior in code that | |
| now relies on the correct implementation of the UVM object factory. | |
| Topic: What will happen in UVM 1.3? | |
| In UVM 1.3, the UVM object factory will have the corrected behavior *ONLY*. | |
| The old (incorrect) behavior will no longer remain available. | |
| This is to encourage users to migrate their code forward | |
| and enable VIP providers to eventually rely on the correct | |
| behavior of the UVM object factory. | |
| Group: Mantis 4519 (+) | |
| Enable all flavors of phase jumping | |
| The <uvm_phase>'s ~jump~ API was extended to include | |
| <uvm_phase::set_jump_phase> and <uvm_phase::end_prematurely>. These | |
| methods allow for a "softer" ~jump~ wherein the phase's execution is | |
| not immediately terminated. | |
| Group: Mantis 4532 | |
| uvm_reg_access_seq ignoring regs with one RO field | |
| The <uvm_reg_access_seq> implementation was incorrectly skipping | |
| registers which contained ~any~ 'RO' fields, as opposed to registers | |
| which had ~only~ 'RO' fields. This behavior has been corrected. | |
| Group: Mantis 4533 | |
| initialized memory on DPI side may survive reset and will cause bad | |
| behaviour | |
| The C-side of the UVM's internal DPI calls was not safe during a | |
| simulation reset, and could potentially cause memory leaks and bad | |
| behavior. This has been corrected. | |
| Group: Mantis 4542 | |
| uvm_test_done should be made "const" | |
| The undocumented and deprecated ~uvm_pkg::uvm_test_done~ reference was | |
| made ~const~ to prevent users from accidently changing it. | |
| Topic: Backwards Compatibility | |
| The ~uvm_test_done~ variable is not part of the UVM | |
| standard, but may have been used by users coming from the OVM. As | |
| such, this change does not have any backward compatibility concerns | |
| for the UVM standard. | |
| Group: Mantis 4550 | |
| are uvm tlm connections with non-uvm_object a legal/supported use | |
| model? | |
| The ~TLM~ documentation was expanded to explicitly allow | |
| non-<uvm_object>-based types within their parameterization. | |
| Group: Mantis 4553 (+) | |
| Infinite recursion in uvm_reg_field::get_access | |
| An error message within the <uvm_reg_field::get_access> method was | |
| recursively calling <uvm_reg_field::get_access>, resulting in an | |
| infinite recursion. This behavior was corrected. | |
| Additionally, while fixing this mantis is was determined that the map | |
| of field accesses defined in <uvm_reg_field::set_access> was lacking a | |
| name for the scenario caused when a 'RO' field was added to a 'WO' | |
| map. The "NOACCESS" policy was added to handle this scenario. | |
| Group: Mantis 4548 | |
| cleanup static dpi issues | |
| This mantis wrapped various fixes within the C-side of DPI used within | |
| the UVM. | |
| Group: Mantis 4566 | |
| comparer.show_max can cause check to pass when length mismatches | |
| A bug in the <uvm_comparer> class caused queues of different sizes to | |
| false-pass when the <uvm_comparer::show_max> value was not equal to | |
| '1'. This behavior has been corrected. | |
| Group: Mantis 4567 | |
| uvm_reg::do_predict ignoring UVM_NOT_OK | |
| The ~uvm_reg::do_predict~ method was overwriting <UVM_NOT_OK> status | |
| under certain circumstances, and replacing it with <UVM_IS_OK>. This | |
| behavior has been corrected. | |
| Group: Mantis 4570 (*)(+) | |
| Messaging System Revamp | |
| The reporting system was overhauled to use an object-based approach to | |
| passing information (as opposed to passing many fields multiple | |
| times). This simultaneously expanded the functionality of the | |
| messaging API while providing an easier path for adding functionality | |
| to this section of the standard in the future. | |
| Topic: Backwards Compatibility | |
| The ~uvm_report_server::process_report~ and | |
| ~uvm_report_server::compose_message~ methods have been replaced by the | |
| object-based <uvm_report_server::execute_report_message> and | |
| <uvm_report_server::compose_report_message> methods. User code which | |
| used the old methods will not behave as expected with the new | |
| implementation. | |
| Topic: Migrating code using UVM 1.2 | |
| To ease the process of migrating code to UVM 1.2, the old behavior | |
| remains available to the user. To restore the old behavior, it is | |
| necessary to define the `UVM_DEPRECATED_REPORTING symbol when | |
| compiling the UVM library: | |
| | % ... +define+UVM_DEPRECATED_REPORTING uvm_pkg.sv | |
| Users must be aware that this will result in different behavior in | |
| code that now relies on the new implementation of <uvm_report_server>. | |
| Group: Mantis 4572 | |
| packing/unpacking issues | |
| The ~uvm_packer::pack_int~ method (which has been renamed to | |
| <uvm_packer::pack_field>) was incorrectly implemented, causing big | |
| endian packs of 64+ bits to be preocessed incorrectly. Additionally, | |
| the <`uvm_unpack_array> implementation did not work properly when | |
| <uvm_packer::use_metadata> was set to '0'. These behaviors have been | |
| corrected. | |
| Group: Mantis 4587 | |
| Bugs in UVM 1.1-c: uvm_reg_tlm_adapter::reg2bus() | |
| Various fixes we provided for <uvm_reg_tlm_adapter::reg2bus> such that | |
| iw would properly adhere to <uvm_tlm_gp> rules for ~length~, | |
| ~byte_enable~, and ~byte_enable_length~. | |
| Group: Mantis 4588 | |
| in-built sequence is not equipped to exclude a block from the test | |
| sequence | |
| The <uvm_reg_hw_reset_seq> was updated to ensure sub-blocks were | |
| properly excluded when relevant ~NO_REG_TEST~ attributes were set. | |
| Group: Mantis 4590 | |
| update uvm_hdl.c to allow VCS/IUS to perform backdoor access to VHDL | |
| The capability to allow VHDL UVM_REG backdoor access using VCS and IUS | |
| were added. | |
| Group: Mantis 4606 | |
| uvm_report_server is not really extendable/chainable | |
| In a fix similar to <Mantis 3887 (*)>, ~uvm_report_server~ | |
| was also split into two classes, an abstract <uvm_report_server> and a | |
| default implementation of <uvm_default_report_server>. This allows | |
| <uvm_report_server> to be fully extended, and multiple report servers | |
| can be chained together using a "delegate" pattern. | |
| Control over the global report server was | |
| migrated to the new <uvm_coreservice_t> class. | |
| Topic: Backwards Compatibility | |
| If users had previously extended <uvm_report_server>, then their code | |
| will likely no longer compile because they are not fully implementing | |
| the ~pure virtual~ interface which the report server now provides. | |
| The fix to reinstate the previous functionality is to extend the | |
| <uvm_default_report_server> instead of the <uvm_report_server>. | |
| The <1.1 to 1.2 Migration Script> can be used to replace any | |
| occurances of the old type in user code. | |
| Group: Mantis 4617 | |
| uvm_report_cb::add_by_name example needs review | |
| A bad example in the documentation of ~uvm_report_cb::add_by_name~ has | |
| been fixed. | |
| Group: Mantis 4618 (*) | |
| resource lookup broken/undefined when fieldname has wildcards | |
| Meta characters (globs or regular expressions) within the ~fieldname~ | |
| of a resource are no longer interpretted by the resource pool. | |
| Topic: Backwards Compatibility | |
| The previous behavior for meta characters within the field name was | |
| not fully defined. | |
| Group: Mantis 4621 | |
| printing of int fields will mask Z values with X | |
| The <uvm_printer::print_field> and <uvm_printer::print_field_int> | |
| methods were incorrectly implemented, causing 'Z' values to appear as | |
| 'X' when printed. This behavior has been corrected. | |
| Group: Mantis 4622 | |
| uvm_phase::unsync() does not work | |
| The implementation of <uvm_phase::unsync> was updated to properly remove | |
| the sync relationship. | |
| Group: Mantis 4643 | |
| uvm_dpi_get_next_arg_c doesnt fully adhere to -1800-lrm | |
| The internal method ~uvm_dpi_get_next_arg_c~ was updated to completely | |
| adhere to the 1800 lrm. | |
| Group: Mantis 4649 (+)(*) | |
| Improve memory performance of 'bitstream' interfaces for | |
| report/record/etc. | |
| The new <uvm_integral_t> was standardized, which is a less | |
| memory-expensive packed vector than the <uvm_bitstream_t>. Support | |
| for this new type was added to the <uvm_printer>, <uvm_packer>, | |
| <uvm_recorder> and <uvm_comparer>, as well as the auto-configuration | |
| provided by the <`uvm_field_*macros>. | |
| Topic: Backwards Compatibility | |
| In order to provide a consistent API, the ~uvm_printer::print_int~ | |
| method was deprecated and replaced by <uvm_printer::print_field_int>. | |
| Group: Mantis 4666 | |
| uvm_config_db::set expects to be invoked in process context | |
| The implementation of <uvm_config_db::set> was improperly assuming | |
| that the method would be invoked within a process context | |
| (i.e. process.self() != null). This behavior has been corrected. | |
| Group: Mantis 4677 | |
| remove uvm_global_copy_map | |
| The internal and undocumented ~uvm_global_copy_map~ was removed and | |
| replaced by a simple associative array. | |
| Group: Mantis 4686 (+) | |
| predict updated not just the mirrored field | |
| The documentation of the <uvm_reg::predict> and <uvm_reg_field::predict> | |
| methods was clarified to indicate that the methods update ~both~ the | |
| mirrored and desired values. | |
| Group: Mantis 4690 | |
| Typo in uvm_sequence_base.vh (mehod should be method) | |
| A typo in the documentation of <uvm_sequence_base> was corrected. | |
| Group: Mantis 4711 | |
| uvm macros violate 1800lrm iterator requirements | |
| The <`uvm_field_*macros> were updated to obey 1800 LRM requirements | |
| for the iterators of arrays. | |
| Group: Mantis 4731 | |
| Static initialization races | |
| The fix to <Mantis 4606> accidently included a static race condition | |
| which was fixed. | |
| Group: Mantis 4732 (+) | |
| uvm_phase_state initializes to incorrect value | |
| The new phase state <UVM_PHASE_UNINITIALIZED> was added to the | |
| <uvm_phase_state>. This new state is the default state for all | |
| phases after construction. | |
| Group: Mantis 4734 | |
| Bug in uvm_reg_map::get_size(); | |
| The implementation of ~uvm_reg_map::get_size~ had a typo which | |
| resulted in the value of the last offset added being used instead of | |
| the value of the largest offset. This behavior has been corrected. | |
| Group: Mantis 4738 | |
| DPI Code does not use UVM Messaging | |
| Similar to <Mantis 3783>, the messages generated from the C-side | |
| implementation of the UVM dpi code have been refactored to route | |
| through standard UVM messaging. | |
| Group: Mantis 4743 (+) | |
| Promote get_next_item, try_next_item, item_done, get, peek, and put to | |
| be a part of the standard | |
| The documentation of the <uvm_sequencer#(REQ,RSP)::Sequencer Interface> | |
| was expanded to include the implementations of the various methods. | |
| Group: Mantis 4744 (*)(+) | |
| Recording System Revamp | |
| Similar to <Mantis 4570 (*)(+)>, the recording system was overhauled | |
| to use an object-based approach to connecting to transaction | |
| databases. This resulted in two new classes <uvm_tr_stream> and | |
| <uvm_tr_database>, which provided an object-based implementation of | |
| the undocumented implementation API for the ~uvm_recorder~. | |
| Additionally, a new data policy <uvm_link_base> was added for | |
| expressing arbitrary relationships in a compile-safe manner. | |
| Topic: Backwards Compatibility | |
| The changes made to the <uvm_recorder> API were primarily isolated to | |
| undocumented and non-standardized methods within <uvm_recorder> | |
| class. New examples are provided via <uvm_text_recorder>, | |
| <uvm_text_tr_stream> and <uvm_text_tr_database> to illustrate to | |
| developers how the new system works. | |
| Group: Mantis 4766 (*) | |
| Fix broken `uvm_record_field + `uvm_record_attribute interaction | |
| The fix for this mantis involved creating a new method | |
| <uvm_recorder::use_record_attribute>. The new method allows for a | |
| conditional usage of the <`uvm_record_attribute> macro, allowing users | |
| to create their own recorders without interfering with | |
| vendor-specific recorders | |
| Topic: Backwards Compatibility | |
| Vendors which had provided an implementation of the | |
| <`uvm_record_attribute> macro need to add an implementation of | |
| <uvm_recorder::use_record_attribute> in their recorders to allow | |
| compatibility with the new macro implementation. | |
| Group: Mantis 4767 (+) | |
| `uvm_pack_int and `uvm_unpack_int bypass the uvm_packer policy methods | |
| The previous implementation of the ~`uvm_pack_*~ macros bypassed the | |
| <uvm_packer> policy methods, and directly accessed an undocumented | |
| member variable of the <uvm_packer> class. This implementation was | |
| refactored to use the policy methods without accessing any | |
| undocumented functionality. | |
| Additionally, a new API was added to the <uvm_packer> to provide the | |
| ability to pack/unpack an arbrary number of bits from an array of | |
| bits, bytes or ints. This new API provides easier compatibility with | |
| the SystemVerilog stream operator. | |
| Group: Mantis 4772 | |
| separate uvm_hdl* source into files-by-vendor | |
| Related to <Mantis 4588>, the fix for this mantis separated the hdl | |
| vendor dpi implementations into vendor-specific files. | |
| Group: Mantis 4783 (*) | |
| `uvm_record_* macros bypass uvm_recorder policy class | |
| The previously undocumented ~`uvm_record_*~ macros bypassed the | |
| <uvm_recorder> policy methods, and called undocumented and | |
| non-standard methods. Their implementation has been refactored to use the | |
| <uvm_recorder> policy methods by default. | |
| Topic: Backwards Compatibility | |
| The changes to the macros should be transparent to most end users, | |
| however developers which were supporting the old undocumented macros | |
| may need to update their implementations. | |
| Group: Mantis 4786 (+) | |
| print_real is undocumented | |
| A typo in the documentation of <uvm_printer::print_real> resulted in | |
| the method being absent from the standard. This has been corrected. | |
| Group: Mantis 4784 (+) | |
| `uvm_record_int/string/time/real are undocumented | |
| After fixing <Mantis 4783 (*)>, the previously undocumented | |
| <`uvm_record_int>, <`uvm_record_field>, <`uvm_record_real>, | |
| <`uvm_record_string> and <`uvm_record_time> have been officially added | |
| to the standard. | |
| Group: Mantis 4861 | |
| Adding a uvm_reg_map to an existing regmap instance required both maps | |
| to have the same parent. The behavior was fixed and the bad error | |
| message corrected. | |
| Group: Mantis 4871 | |
| Fixed a problem wherein multiple ~+uvm_set_*~ commands were ignored | |
| after a format error. | |
| Group: Mantis 4882 | |
| The instance name of the phase objection was changed to reflect the | |
| name of the phase. | |
| Group: Mantis 4884 | |
| A performance fix was implemented which avoided the copy of large QDA | |
| within the spell checker. | |
| Group: Mantis 4893 | |
| Fixed a bug within uvm_leaf_scope which could lead to stall of | |
| simulation. | |
| Group: Mantis 4920 | |
| Fixed random stability bug in ~lookup_name~ which caused instability | |
| when the config database was queried. | |
| Group: Mantis 4923 | |
| A check was added for an out-of-bound queue access in uvm_misc.svh. | |
| This check caused no functional change, but prevents certain tools | |
| from throwing a warning when processing the code. | |
| Group: Mantis 4925 | |
| The macros <`UVM_MAX_STREAMBITS>, <`UVM_PACKER_MAX_BYTES> and | |
| <`UVM_DEFAULT_TIMEOUT> were relocated to | |
| src/macros/uvm_global_defines.svh, which is included via | |
| uvm_macros.svh. These changes will make the UVM 1.2 Reference HTML | |
| out of sync with the UVM 1.2 Reference PDF, however this change is | |
| strictly contained to the formatting/location of documentation, not a | |
| functional difference. | |
| Group: Mantis 4931 | |
| Various typos and bad links within the reference guide were corrected. |