Teach SCEVExpander to expand arithmetic involving pointers into GEP

instructions. It attempts to create high-level multi-operand GEPs,
though in cases where this isn't possible it falls back to casting
the pointer to i8* and emitting a GEP with that. Using GEP instructions
instead of ptrtoint+arithmetic+inttoptr helps pointer analyses that
don't use ScalarEvolution, such as BasicAliasAnalysis.

Also, make the AddrModeMatcher more aggressive in handling GEPs.
Previously it assumed that operand 0 of a GEP would require a register
in almost all cases. It now does extra checking and can do more
matching if operand 0 of the GEP is foldable. This fixes a problem
that was exposed by SCEVExpander using GEPs.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72093 91177308-0d34-0410-b5e6-96231b3b80d8

test/Transforms/LoopStrengthReduce/2009-04-28-no-reduce-mul.ll

@@ -1,4 +1,5 @@
-; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis | grep {mul.*%lsr.iv} | count 2
+; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis \
+; RUN:   | grep {getelementptr.*%lsr.iv.*%lsr.iv.*<i32\\*>}
 ; The multiply in bb2 must not be reduced to an add, as the sext causes the
 ; %1 argument to become negative after a while.
 ; ModuleID = '<stdin>'