Refactor drmemtrace scheduler to specialize code by mapping option #6831

derekbruening · 2024-05-28T21:20:39Z

Many drmemtrace scheduler routines have big conditionals dispatching on mapping options. We may be able to simplify the code if we refactor to use subclasses or templates to specialize scheduler code based on mapping options.

This first step toward separating key scheduler code into separate subclasses by mode of operation adds the pImpl idiom to the scheduler: the scheduler_t implementation is moved to a private class scheduler_impl_t, to let us (in a future step) create the appropriate subclass without changing callers who construct a scheduler_t. This also helps to split up and modularize the enormous scheduler class. A future step will move the scheduler_impl_t code into a new file scheduler_impl.cpp. That was separated out to make the diff easier to read here. The bulk of the changes involve: + The name change scheduler_tmpl_t to scheduler_impl_tmpl_t. + Having to qualify enum types as we do not have C++20's using enum. + Having to refer to types in another class now that the impl class is separate from the outer class. + Moving inlined routines out of scheduler.h now that the implementation is private and scheduler_impl_tmpl_t is an incomplete type in scheduler.h. + Some sched_type_t qualifications are removed due to new (private) using statements to make the code more readable. Issue: #6831

Moves the scheduler_impl_tmpl_t implementation out of scheduler.cpp and into a new file scheduler_impl.cpp. Issue: #6831

This first step toward separating key scheduler code into separate subclasses by mode of operation adds the pImpl idiom to the scheduler: the scheduler_t implementation is moved to a private class scheduler_impl_t, to let us (in a future step) create the appropriate subclass without changing callers who construct a scheduler_t. This also helps to split up and modularize the enormous scheduler class. A future step will move the scheduler_impl_t code into a new file scheduler_impl.cpp. That was separated out to make the diff easier to read here. The bulk of the changes involve: + The name change scheduler_tmpl_t to scheduler_impl_tmpl_t. + Having to qualify enum types as we do not have C++20's using enum. + Having to refer to types in another class now that the impl class is separate from the outer class. + Moving inlined routines out of scheduler.h now that the implementation is private and scheduler_impl_tmpl_t is an incomplete type in scheduler.h. + Some sched_type_t qualifications are removed due to new (private) using statements to make the code more readable. Issue: #6831

Moves the scheduler_impl_tmpl_t implementation out of scheduler.cpp and into a new file scheduler_impl.cpp. Issue: #6831

Splits the bulk of pick_next_input() into a new virtual method pick_next_input_try() which is implemented by new subclasses scheduler_{dynamic,fixed,replay}.cpp. Moves pick_next_input_as_previously() into the _replay version. Issue: #6831

Splits the bulk of pick_next_input() into a new virtual method pick_next_input_for_mode() which is implemented by new subclasses scheduler_{dynamic,fixed,replay}.cpp. Moves pick_next_input_as_previously() into the _replay version. Issue: #6831

Splits the mode-specific middle of next_record() into a new virtual method process_next_record_candidate() which is implemented in the new subclasses scheduler_{dynamic,fixed,replay}.cpp. Issue: #6831

Adds used include files to the 3 scheduler subclasses that were omitted when new overrides were added in step 3. Issue: #6831

…ts (#7079) Splits the mode-specific middle of next_record() into a new virtual method check_for_input_switch() for identifying whether to trigger a switch and call pick_next_input_for_mode(). The new method is implemented in the new subclasses scheduler_{dynamic,fixed,replay}.cpp. Issue: #6831

Splits set_initial_schedule() into separate overrides in the 3 scheduler subclasses, sharing the initial content reading in init() prior to calling set_initial_schedule(). Splits set_output_active(): it is only valid in dynamic. Moves rebalance_queues() and two rebalance vars to dynamic. Moves read_recorded_schedule() and read_and_instantiate_traced_schedule() to replay. Adds more using statements to remove the need for "this->" prefixes on frequent base class member field references. Issue: #6831

derekbruening · 2024-11-14T03:42:11Z

Though file size is not enough by itself, it does give a general, rough indication of modularity.
Before, scheduler.cpp was enormous:

$ wc ../src/clients/drcachesim/scheduler/*
   253    874   7507 ../src/clients/drcachesim/scheduler/flexible_queue.h
  4658  19106 222947 ../src/clients/drcachesim/scheduler/scheduler.cpp
  2178  11132  97930 ../src/clients/drcachesim/scheduler/scheduler.h
   133    656   4943 ../src/clients/drcachesim/scheduler/speculator.cpp
   109    507   3915 ../src/clients/drcachesim/scheduler/speculator.h

After 7 steps (2 not yet merged) it is still large but not nearly as bad:

$ wc ../src/clients/drcachesim/scheduler/*
     253     874    7507 ../src/clients/drcachesim/scheduler/flexible_queue.h
     352    1035   13057 ../src/clients/drcachesim/scheduler/scheduler.cpp
    1284    7144   59422 ../src/clients/drcachesim/scheduler/scheduler.h
    1100    4737   54957 ../src/clients/drcachesim/scheduler/scheduler_dynamic.cpp
     171     767    7798 ../src/clients/drcachesim/scheduler/scheduler_fixed.cpp
    2981   11603  133463 ../src/clients/drcachesim/scheduler/scheduler_impl.cpp
    1150    5089   50449 ../src/clients/drcachesim/scheduler/scheduler_impl.h
     521    2665   28390 ../src/clients/drcachesim/scheduler/scheduler_replay.cpp
     133     656    4943 ../src/clients/drcachesim/scheduler/speculator.cpp
     109     507    3915 ../src/clients/drcachesim/scheduler/speculator.h

Splits eof_or_idle() into separate overrides in the 3 scheduler subclasses. This then allows moving pop_from_ready_queue*() into the dynamic subclass. Moves process_markers(), ready_queue_empty(), and syscall_incurs_switch() into the dynamic subclass. Issue: #6831

derekbruening · 2024-11-14T19:07:42Z

The main goals of this initial round of refactoring are A) simplify the huge methods with if.else splits based on mode; B) split the code up better as scheduler.cpp was very very large. The interfaces between the main class the subclasses could be more clearly separated and crystalized; that will likely be left for a future pass.

Splits set_initial_schedule() into separate overrides in the 3 scheduler subclasses, sharing the initial content reading in init() prior to calling set_initial_schedule(). Makes set_output_active() virtual: it is only valid in dynamic. Moves rebalance_queues() and two rebalance vars to dynamic. Moves read_recorded_schedule() and read_and_instantiate_traced_schedule() to replay. Adds more using statements to remove the need for "this->" prefixes on frequent base class member field references. Issue: #6831

Splits eof_or_idle() into separate overrides in the 3 scheduler subclasses. This then allows moving pop_from_ready_queue*() into the dynamic subclass. Moves process_markers(), ready_queue_empty(), and syscall_incurs_switch() into the dynamic subclass. Issue: #6831

Adds swap_in_input() and swap_out_input() subclass hooks inside set_cur_input(). This allows moving ready queue functions into the dynamic subclass, and will be used there for future live workload tracking for #7067. Moves add_to_ready_queue(), add_to_ready_queue_hold_locks(), add_to_unscheduled_queue(), and print_queue_stats() into the dynamic subclass. Issue: #7067, #6831

derekbruening added Maintainability Component-DrMemtrace labels May 28, 2024

derekbruening mentioned this issue May 28, 2024

i#6822 unscheduled: Add unscheduled-input drmemtrace support #6826

Merged

derekbruening mentioned this issue Nov 11, 2024

i#6831 sched refactor, step 1: Add pImpl idiom #7074

Merged

derekbruening added a commit that referenced this issue Nov 12, 2024

i#6831 sched refactor, step 2: Split out scheduler_impl.cpp

dae1fb8

Moves the scheduler_impl_tmpl_t implementation out of scheduler.cpp and into a new file scheduler_impl.cpp. Issue: #6831

derekbruening mentioned this issue Nov 12, 2024

i#6831 sched refactor, step 2: Split out scheduler_impl.cpp #7077

Merged

derekbruening added a commit that referenced this issue Nov 12, 2024

i#6831 sched refactor, step 2: Split out scheduler_impl.cpp (#7077)

52ba775

Moves the scheduler_impl_tmpl_t implementation out of scheduler.cpp and into a new file scheduler_impl.cpp. Issue: #6831

derekbruening mentioned this issue Nov 12, 2024

i#6831 sched refactor, step 3: Split pick_next_input() #7078

Merged

derekbruening mentioned this issue Nov 13, 2024

i#6831 sched refactor, step 5: Split check for whether to switch inputs #7079

Merged

derekbruening added a commit that referenced this issue Nov 13, 2024

i#6831 sched refactor, step 5: Add missing includes

59d14f1

Adds used include files to the 3 scheduler subclasses that were omitted when new overrides were added in step 3. Issue: #6831

derekbruening mentioned this issue Nov 13, 2024

i#6831 sched refactor, step 4: Add missing includes #7080

Merged

derekbruening added a commit that referenced this issue Nov 13, 2024

i#6831 sched refactor, step 4: Add missing includes (#7080)

88abaaa

Adds used include files to the 3 scheduler subclasses that were omitted when new overrides were added in step 3. Issue: #6831

derekbruening mentioned this issue Nov 14, 2024

i#6831 sched refactor, step 6: Split set_initial_schedule() #7082

Merged

derekbruening mentioned this issue Nov 14, 2024

i#6831 sched refactor, step 7: Split eof_or_idle #7083

Merged

derekbruening mentioned this issue Nov 19, 2024

i#6831 sched refactor: Add set_cur_input() hooks #7089

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refactor drmemtrace scheduler to specialize code by mapping option #6831

Refactor drmemtrace scheduler to specialize code by mapping option #6831

derekbruening commented May 28, 2024

derekbruening commented Nov 14, 2024

derekbruening commented Nov 14, 2024 •

edited

Loading

Refactor drmemtrace scheduler to specialize code by mapping option #6831

Refactor drmemtrace scheduler to specialize code by mapping option #6831

Comments

derekbruening commented May 28, 2024

derekbruening commented Nov 14, 2024

derekbruening commented Nov 14, 2024 • edited Loading

derekbruening commented Nov 14, 2024 •

edited

Loading