This commit adopts `llvm::DenseSet`, `llvm::DenseMap` and
`std::unordered_set` where needed.
It also tweaks some `llvm::SmallVector` sizes.
It brings an additional ~3% improvement during lifting.
This commit disables GVN in order to improve lifting performance.
It also increases `EarlyCSEMssaOptCap` so we keep catching all the jump
target we used to.
Lifting /bin/bash goes from ~29mins to ~5mins
Before this fix, graph preparation sometimes left extra long-edges not
connected to anything in. They were not visible after the SVG
exportation, but they were making edge routing worse (by forcing
horizontal lanes to appear where they shouldn't be - thus spacing nodes
further away from each other than necessary).
Since the `orderEdges` function no longer consumes the graph, we can
finally make the restoration step separate - leading to more explicitly
stated boundaries as to where the graph is a DAG, and where it is not.
Also removed a few entities to avoid duplication in the full names:
- `yield::layout::sugiyama::LayoutOrientation`
-> `yield::layout::sugiyama::Orientation`
- `yield::layout::sugiyama::layout(...)`
-> `yield::layout::sugiyama::compute(...)`
Add support for tracing onto the PipelineC. This is done by:
1. Creating wrapper functions for each PipelineC function with the
script in `scripts/PipelineC_add_tracing.py`. These will call a
special function called `wrap` which will ultimately call a method
with a `_` prepended to the name
2. Conversion of all PipelineC methods in `PipelineC.cpp` to `static`
and their rename with a `_` in front, in order for them to work with
the wrapper function in (1)
3. Generation of 2 additional include files, one for types and one for
functions, to be used by users of tracing files in order to have
introspection.
These steps allow the creation of a trace file with the use of the
`REVNG_C_API_TRACE_PATH` environment variable. The traces can then be
used in conjunction with the `revng trace run` and `revng trace
inspect` commands.
Many Pipeline objects (Globals, Ranks, Kinds, Targets in a TargetList)
were ordered by load order, this makes the playback of traces difficult.
Moreover the order of loading of pipeline files also influences the
ordering of other elements (Step, Analys{is,esLists}). These are also
ordered to prevent incosistent ordering.
Switch from specifying `libraries` and `pipelines` in `rp_initialize`
and `rp_manager_create` to the use of command-line options that are to
be passed via `argc` and `argv` in `rp_initialize`.
This commit drops the `CallToLifted` tag in favor of a
`getCallToIsolatedFunction` function which checks if the call has the
`IsolatedFunction` tag. The function also assumes that indirect calls
are calls to lifted functions.
SimplifyCFG with instruction sinking enabled sometimes can merge two
call sites that are identical at LLVM IR level, but originally come from
two distinct addresses in the original binary.
The nomerge attribute prevents this type of problematic deduplication.
We used to only color instructions in the same BasicBlock as the call to
`newpc`. However, an instruction in the original binary can span
multiple BasicBlocks.
We used to jump to the dispatcher, however we only sink to an
unreachable `anypc`. We were threfore significantly complicating the CFG
in case of indirect function calls.
Alessandro Di Federico