| age |
author |
description |
| 3 weeks ago |
utke on macbook |
refactor loop clases, make the init/condition/update optional to reflect better the XAIF schemadefault tip |
| 3 weeks ago |
utke on macbook |
a new helper |
| 7 weeks ago |
hlcm |
check hasElement before calling get for theEVp2LCGVpMap |
| 7 weeks ago |
hlcm |
add debug printing for Keys/Elements in GuardedMap |
| 7 weeks ago |
hlcm |
Replace uses of map with uses of GuardedMap class |
| 7 weeks ago |
hlcm |
Add a GuardedMap template for use instead of regular map; With map, elements are inserted during index operations. Now, elements must be gotten and added explicitly, and an error will be raised if the element is not there |
| 8 weeks ago |
utke on macbook |
typo |
| 8 weeks ago |
utke on macbook |
adapt message texts and manage lists |
| 8 weeks ago |
utke on macbook |
move handleExplicitUnresolved to the new 6th sweep because the CallGraphVertex determines what is missing in the 5th but only later we descend in the 5th sweep down to the BasicBlocks to create the adjoint counterparts so we can't yet do the handleExplicitUnresolved in the 5th sweep and have to postpone to the 6th |
| 8 weeks ago |
utke on macbook |
debug message |
| 8 weeks ago |
utke on macbook |
rename for now because this is not finished and we need the 6th sweep for the unresolved entries found in simple loops |
| 8 weeks ago |
utke on macbook |
mutual pointers between adjoint and original SubroutineCallAlg instances; rename methods; pass through reversal type; make handleArrayAccessIndices public |
| 8 weeks ago |
utke on macbook |
refactor containers for push related constructs into a PushContainer class; instances of that class are put nto a map keyed of by the type of reversal we are doing; also refactor more elements into the BasicBlockElementAlg; renamed appropriately and change access to reflect the current functionality. |
| 8 weeks ago |
utke on macbook |
add heuristic for heuristic Push/Pop |
| 8 weeks ago |
utke on macbook |
refactor to base class + rename + reorder apparently wrong print XML logic |
| 8 weeks ago |
utke on macbook |
refactor to base class |
| 8 weeks ago |
utke on macbook |
rename |
| 8 weeks ago |
utke on macbook |
rename |
| 8 weeks ago |
utke on macbook |
fill in expanded info |
| 8 weeks ago |
utke on macbook |
expanded information |
| 8 weeks ago |
utke on macbook |
typo |
| 3 months ago |
utke on macbook |
wrong name |
| 4 months ago |
utke on macbook |
eclipse |
| 4 months ago |
utke on macbook |
both adjoint variants |
| 4 months ago |
utke on macbook |
separate the allocation calls between preaccumulation and propagation |
| 7 months ago |
utke on macbook |
branch merge |
| 7 months ago |
utke on macbook |
fix up build problem when a .cpp source file is removed but the incremental build picks up the orhpaned .o file that should be removed. So derive the .o names directly from the list of the .cpp files in the src subdirs. |
| 7 months ago |
lyonsam |
branch merge |
| 7 months ago |
lyonsam |
minor comment/typo/debugging stuff |
| 7 months ago |
lyonsam |
relatively large change separating logic for flattening from LCG construction; |
| 7 months ago |
lyonsam |
moved some flags and runElimination from BasicBlockAlg to Sequence |
| 8 months ago |
lyonsam |
moved propagation logic in BBP from BasicBlockAlg to Sequence; also some cleanup |
| 8 months ago |
lyonsam |
Assignments created on the fly for SSA are no longer created with new |
| 8 months ago |
lyonsam |
moved logic for flattening sequential inlinable assignments into sequences (from AssignmentAlg::algorithm_action_2 to BasicBlockAlg::algorithm_action_2) and simplified |
| 8 months ago |
lyonsam |
debugging, typedefs, cleanup |
| 8 months ago |
lyonsam |
debugging improvements |
| 8 months ago |
lyonsam |
moved independent/dependent vertex list population from BasicBlockAlg to Sequence |
| 7 months ago |
utke on macbook |
factor out logic for source/target type propagators, add logic to not mangle the name when the arguments represent array slices (e.g. a(:,3) as opposed to just a or a(1,1) which is a scalar) because syntactic inlining for non-scalar arguments means appending a loopindex to the inlinable's argument so it would produce syntactically incorrect a(:,3)(loopIndex). |
| 7 months ago |
utke on macbook |
add argument shape information as a suffix to the name (needed for vector mode) |
| 7 months ago |
utke on macbook |
allow for appending a suffix (name mangeling after we inspected all the arguments) |
| 7 months ago |
utke on macbook |
let DerivativePropagatorEntry instance represent themselves as InlinableSubroutineCall instances, add a -H switch to the AlgConfig |
| 7 months ago |
utke on macbook |
let DerivativePropagatorEntry instance represent themselves as InlinableSubroutineCall instances, behavior switched by ourHidDPsAsICs |
| 7 months ago |
utke on macbook |
let DerivativePropagatorEntry instance represent themselves as InlinableSubroutineCall instances |
| 7 months ago |
utke on macbook |
for convenience check and turn expressions into single variable/constants |
| 8 months ago |
lyonsam |
refactoring of auxiliary assignments for partial evaluation: |
| 8 months ago |
lyonsam |
new NameCreator for auxiliary variables |
| 8 months ago |
lyonsam |
spelling fix: auxilliary->auxiliary |
| 8 months ago |
lyonsam |
assignments added to front or back of a sequence as part of transformation (such as accumulation assignments) now created without algorithm object by default |
| 8 months ago |
lyonsam |
minor debugging improvements |
| 8 months ago |
lyonsam |
getMaxVertex() now checks for uniqueness |
| 8 months ago |
lyonsam |
PushPop: CombinedGraph refinement |
| 8 months ago |
lyonsam |
typo fixes |
| 8 months ago |
lyonsam |
PrivateLinearizedComputationalGraph, RemainderGraph, and AccumulationGraph: creation, debugging, and GraphViz display improvements |
| 9 months ago |
lyonsam |
branch merge |
| 9 months ago |
lyonsam |
PushPop: building the CombinedGraph |
| 9 months ago |
lyonsam |
association of aux vars with PLCG vertices, association of maximal expression vertices with PLCG edges for direct copy edges; Graphviz label tweak for PLCG vertices |
| 9 months ago |
lyonsam |
improved debugging and GraphViz drawings for PLCGs and RemainderGraphs |
| 9 months ago |
lyonsam |
PLCG vertices now maintain a set of associated expression vertices |
| 9 months ago |
lyonsam |
added -B option for specifying SVG viewer (defaults to firefox) |
| 9 months ago |
lyonsam |
useful set typedefs |
