AddedInstrns sets for the first machine instruction. It is hard
to ensure that the right order is preserved, and sure enough, the
order was broken. Instead, use a separate set for the function entry.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2318 91177308-0d34-0410-b5e6-96231b3b80d8
to copy the FP arg. to an integer. Necessary so that the
register allocator has two different live ranges for the FP value
and the int. argument.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2314 91177308-0d34-0410-b5e6-96231b3b80d8
AddedInstrns sets for the first machine instruction. It is hard
to ensure that the right order is preserved, and sure enough, the
order was broken. Instead, use a separate set for the function entry.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2312 91177308-0d34-0410-b5e6-96231b3b80d8
(This can be improved to avoid making the initial pass over the method.)
Also, ensure automatic vars and reg. spills areas are not extended
if their sizes are used for computing some other offset.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2310 91177308-0d34-0410-b5e6-96231b3b80d8
* Verify that load, store, and GEP instructions indices are correct, because
they _continually_ bite me on this pool allocations stuff
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2309 91177308-0d34-0410-b5e6-96231b3b80d8
* Implement checking that a binary operator's two sides are the same type
* Actually check that an instruction does not have a name if it has a void
return type.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2305 91177308-0d34-0410-b5e6-96231b3b80d8
put it into the AsmWriter. This code is kinda gross and could probably be
cleaned up, but not now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2299 91177308-0d34-0410-b5e6-96231b3b80d8
We now use an InstVisitor to delegate to different cases that we are
interested in handling. We also fix the FIXME's by adding users to the
worklist when appropriate.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2292 91177308-0d34-0410-b5e6-96231b3b80d8
* Allow structs with negative offsets. This enables the em3d benchmark to
be made typesafe. In this case, the struct had an array as the first
element, so a negative index was ok (the expr was -8 + 8x)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2271 91177308-0d34-0410-b5e6-96231b3b80d8
* Fix bug printing method types (oops when converting previous code)
* Move some of the code from Constant::getStrValue implementations here.
This allows us to do a MUCH MUCH better job printing out complex constant
values (think an array of structure pointers), because we print symbolic
type names instead of structural. The assistance when debugging is immense
Eventually, Constant::getStrValue should be removed (it existed before because
the asmwriter file was not in VMCore), and everyone should go through the
asmwriter that want's equivalent functionality.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2268 91177308-0d34-0410-b5e6-96231b3b80d8
* Make the function pointer argument explicit for a call nodes
* Eliminate unreachable global values
* Merge call nodes that are identical
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2266 91177308-0d34-0410-b5e6-96231b3b80d8
* Fold call nodes that are indistinguishable for each other. This is a big
win for external functions like sqrt, which would multiply dramatically
before.
* Global nodes with no edges to or from them are now eliminated from the graph.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2257 91177308-0d34-0410-b5e6-96231b3b80d8
