[LLVMbugs] [Bug 845] NEW: Eliminate cyclic dependencies between libraries

bugzilla-daemon at cs.uiuc.edu bugzilla-daemon at cs.uiuc.edu
Wed Jul 26 11:32:08 PDT 2006


http://llvm.org/bugs/show_bug.cgi?id=845

           Summary: Eliminate cyclic dependencies between libraries
           Product: libraries
           Version: 1.0
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Core LLVM classes
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sabre at nondot.org


We want to eliminate cycles in the dependency graph of the LLVM libraries.  llvm-config reports this, 
and Reid has compiled this information.  To get this info, use:

list of which libs/objs depend on which other ones:
GenLibDeps.pl -flat /path/to/Debug/lib > LibDeps.flat

full dependency listing including symbols:
GenLibDeps.pl -why /path/to/Debug/lib | c++filt > LibDeps.whyå

Here are some simple dependencies that can be snipped, this may be out-of-date:

The main cycle is this:
  libLLVMAnalysis.a -> libLLVMCodeGen.a -> libLLVMScalarOpts.a ->
  libLLVMSelectionDAG.a -> libLLVMTarget.a -> libLLVMTransformUtils.a ->
  libLLVMipa.a -> libLLVMAnalysis.a

libLLVMAnalysis.a shouldn't depend on libLLVMCodeGen.a
libLLVMSelectionDAG.a shouldn't depend on libLLVMTransformUtils.a (it uses it for critical edge 
breaking) 
libLLVMTransformUtils.a shouldn't use LLVMipa (it uses/updates call-graph stuff)
libLLVMTarget.a uses libLLVMCodeGen.a for MVT::getVectorType  (codegen should use target, not the 
other way around)

LLVMIA64 uses intrinsic lowering?  It shouldn't.

Maybe snippable:
libLLVMCodeGen uses Analysis only for loopinfo?

TransformUtils shouldn't depend on ScalarOpts.  It currently does to get LowerSelectID, LowerSwitchID, 
PromoteMemoryToRegisterID.  We need to make a pass be able to "update" another pass without having 
a strong dependency on that pass existing.

Thanks to Reid for collecting much of this data.

-Chris



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.



More information about the llvm-bugs mailing list