Rewrite SCEV's backedge taken count computation.

Patch by Michele Scandale!

Rewrite of the functions used to compute the backedge taken count of a
loop on LT and GT comparisons.

I decided to split the handling of LT and GT cases becasue the trick
"a > b == -a < -b" in some cases prevents the trip count computation
due to the multiplication by -1 on the two operands of the
comparison. This issue comes from the conservative computation of
value range of SCEVs: taking the negative SCEV of an expression that
have a small positive range (e.g. [0,31]), we would have a SCEV with a
fullset as value range.

Indeed, in the new rewritten function I tried to better handle the
maximum backedge taken count computation when MAX/MIN expression are
used to handle the cases where no entry guard is found.

Some test have been modified in order to check the new value correctly
(I manually check them and reasoning on possible overflow the new
values seem correct).

I finally added a new test case related to the multiplication by -1
issue on GT comparisons.

llvm-svn: 194116

2 files changed, 7 insertions, 7 deletions

diff --git a/llvm/test/Transforms/IndVarSimplify/loop_evaluate_6.ll b/llvm/test/Transforms/IndVarSimplify/loop_evaluate_6.ll
index da38de538f7..af01fe53864 100644
--- a/llvm/test/Transforms/IndVarSimplify/loop_evaluate_6.ll
+++ b/llvm/test/Transforms/IndVarSimplify/loop_evaluate_6.ll
@@ -1,9 +1,4 @@
 ; RUN: opt < %s -indvars -loop-deletion -S | grep phi | count 1
-; XFAIL: *
-
-; Indvars can't evaluate this loop, because ScalarEvolution can't compute
-; an exact trip count, because it doesn't know if dividing by the stride will
-; have a remainder. It could be done with more aggressive VRP though.
 
 define i32 @test(i32 %x_offs) nounwind readnone {
 entry:
diff --git a/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll b/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
index 507f695e67c..057669277cc 100644
--- a/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
+++ b/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
@@ -223,13 +223,18 @@ entry:
   %halfLim = ashr i32 %limit, 2
   br label %loop
 
-; Test cloning an or, which is not an OverflowBinaryOperator.
+; This test originally checked that the OR instruction was cloned. Now the
+; ScalarEvolution is able to understand the loop evolution and that '%iv' at the
+; end of the loop is an even value. Thus '%val' is computed at the end of the
+; loop and the OR instruction is replaced by an ADD keeping the result
+; equivalent.
 ;
 ; CHECK: loop:
 ; CHECK: phi i64
 ; CHECK-NOT: sext
-; CHECK: or i64
+; CHECK: icmp slt i32
 ; CHECK: exit:
+; CHECK: add i64
 loop:
   %iv = phi i32 [ 0, %entry], [ %iv.next, %loop ]
   %t1 = sext i32 %iv to i64