Mentions légales du service

Skip to content
Snippets Groups Projects

Repository graph

You can move around the graph by using the arrow keys.
Select Git revision
  • master default protected
1 result
Created with Raphaël 2.2.024Aug23222113Jul94314Jun24May2318Apr1310adding microsofa examplesmastermasteradding dumb examplecleaningcleaningbug fix in ApolloStatementMetadataAdderStoreversion of FixParallelLoopFunction that works only in -00...cleaning runtimecleaningcleaningcleaningcleaningcleaningchange name of ApolloMemAccess_test to ApolloInstrumentMemAccessbug fix: when generating the ir-llvm in O3 directly, these passes were crashing -> correctedremoving printsoptimizing the building and coloring of the graph - we do not use CSR format for the adjacency matrix anymore - these optim accelerates greatly the building of the graphexample usage of my passesbug fix with the metadata islocal and renaming of variablesusing new metadata apollomem.pragma.local / shared / private in ApolloLoopExtractStrictly to rename variables accordinglyusing mem2reg properly in pass InlineCallsApolloStatementstrying to apply mem2reg + inlinefunction on the functions called from the apollo loops -> I have trouble using mem2reg --- files forgotten in previous committrying to apply mem2reg + inlinefunction on the functions called from the apollo loops -> I have trouble using mem2regusing #pragma apollo dcop private <var_name> to mark a variable as private to the parallel loop, so as to create a local copy inside the loop - + cleanup + it works in parallel on my example toto_timestep + WARNING: there a a lot of unused passes I should still remove from the projectcleanup + bug fixes : now it is OK with sequential execution calling inner loop function several times on different slices of the iteration space, but result is false in // as some private variables of the inner loop (namely %j, %nj) are passed by reference when we would like them to be passed by valueadding a runtime library to run parallel version using openMP : for the moment the result is correct only using 1 thread...bug fix : call to apollo_create_and_color_dependency_graph was not inserted at the right place in instrumented versionusing std::set<uintptr_t> (ordered set) to store addresses accessed during instrumentation phaseApolloLoopParallelVersion OK + adding a call to apollo_create_and_color_dependency_graph at the end of instrumented version in pass ApolloMemAccess_testcreation of the data structures to represent a colored graph + continuing ApolloLoopParallelVersionadding color loop in parallel version of the apollo loops - WARNING: not finished, outer (color) loop is OK, but inner loop still needs to be modifiedlib for runtime instrumentation and memoisation of loop with constant memory access pattern over time steps