Merge branch 'release_33' of git://github.com/llvm-mirror/llvm into embtk-support-release-3.3

5 files changed, 164 insertions, 22 deletions

diff --git a/test/Transforms/LoopUnroll/scevunroll.ll b/test/Transforms/LoopUnroll/scevunroll.ll
index 99b3a7d861..308a036316 100644
--- a/test/Transforms/LoopUnroll/scevunroll.ll
+++ b/test/Transforms/LoopUnroll/scevunroll.ll
@@ -66,13 +66,16 @@ exit2:
 
 ; SCEV properly unrolls multi-exit loops.
 ;
+; SCEV cannot currently unroll this loop.
+; It should ideally detect a trip count of 5.
+; rdar:14038809 [SCEV]: Optimize trip count computation for multi-exit loops.
 ; CHECK: @multiExit
-; CHECK: getelementptr i32* %base, i32 10
-; CHECK-NEXT: load i32*
-; CHECK: br i1 false, label %l2.10, label %exit1
-; CHECK: l2.10:
-; CHECK-NOT: br
-; CHECK: ret i32
+; CHECKFIXME: getelementptr i32* %base, i32 10
+; CHECKFIXME-NEXT: load i32*
+; CHECKFIXME: br i1 false, label %l2.10, label %exit1
+; CHECKFIXME: l2.10:
+; CHECKFIXME-NOT: br
+; CHECKFIXME: ret i32
 define i32 @multiExit(i32* %base) nounwind {
 entry:
   br label %l1
@@ -170,3 +173,38 @@ for.body87:
   br label %for.body87
 }
 
+; PR16130: clang produces incorrect code with loop/expression at -O2
+; rdar:14036816 loop-unroll makes assumptions about undefined behavior
+;
+; The loop latch is assumed to exit after the first iteration because
+; of the induction variable's NSW flag. However, the loop latch's
+; equality test is skipped and the loop exits after the second
+; iteration via the early exit. So loop unrolling cannot assume that
+; the loop latch's exit count of zero is an upper bound on the number
+; of iterations.
+;
+; CHECK: @nsw_latch
+; CHECK: for.body:
+; CHECK: %b.03 = phi i32 [ 0, %entry ], [ %add, %for.cond ]
+; CHECK: return:
+; CHECK: %b.03.lcssa = phi i32 [ %b.03, %for.body ], [ %b.03, %for.cond ]
+define void @nsw_latch(i32* %a) nounwind {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.cond, %entry
+  %b.03 = phi i32 [ 0, %entry ], [ %add, %for.cond ]
+  %tobool = icmp eq i32 %b.03, 0
+  %add = add nsw i32 %b.03, 8
+  br i1 %tobool, label %for.cond, label %return
+
+for.cond:                                         ; preds = %for.body
+  %cmp = icmp eq i32 %add, 13
+  br i1 %cmp, label %return, label %for.body
+
+return:                                           ; preds = %for.body, %for.cond
+  %b.03.lcssa = phi i32 [ %b.03, %for.body ], [ %b.03, %for.cond ]
+  %retval.0 = phi i32 [ 1, %for.body ], [ 0, %for.cond ]
+  store i32 %b.03.lcssa, i32* %a, align 4
+  ret void
+}
diff --git a/test/Transforms/LoopUnroll/unloop.ll b/test/Transforms/LoopUnroll/unloop.ll
index 5a9cacda44..9a938cc287 100644
--- a/test/Transforms/LoopUnroll/unloop.ll
+++ b/test/Transforms/LoopUnroll/unloop.ll
@@ -21,8 +21,8 @@ outer:
 inner:
   %iv = phi i32 [ 0, %outer ], [ %inc, %tail ]
   %inc = add i32 %iv, 1
-  %wbucond = call zeroext i1 @check()
-  br i1 %wbucond, label %outer.backedge, label %tail
+  call zeroext i1 @check()
+  br i1 true, label %outer.backedge, label %tail
 
 tail:
   br i1 false, label %inner, label %exit
@@ -126,25 +126,27 @@ return:
 ; Ensure that only the middle loop is removed and rely on verify-loopinfo to
 ; check soundness.
 ;
-; CHECK: @unloopDeepNested
+; This test must be disabled until trip count computation can be optimized...
+; rdar:14038809 [SCEV]: Optimize trip count computation for multi-exit loops.
+; CHECKFIXME: @unloopDeepNested
 ; Inner-inner loop control.
-; CHECK: while.cond.us.i:
-; CHECK: br i1 %cmp.us.i, label %next_data.exit, label %while.body.us.i
-; CHECK: if.then.us.i:
-; CHECK: br label %while.cond.us.i
+; CHECKFIXME: while.cond.us.i:
+; CHECKFIXME: br i1 %cmp.us.i, label %next_data.exit, label %while.body.us.i
+; CHECKFIXME: if.then.us.i:
+; CHECKFIXME: br label %while.cond.us.i
 ; Inner loop tail.
-; CHECK: if.else.i:
-; CHECK: br label %while.cond.outer.i
+; CHECKFIXME: if.else.i:
+; CHECKFIXME: br label %while.cond.outer.i
 ; Middle loop control (removed).
-; CHECK: valid_data.exit:
-; CHECK-NOT: br
-; CHECK: %cmp = call zeroext i1 @check()
+; CHECKFIXME: valid_data.exit:
+; CHECKFIXME-NOT: br
+; CHECKFIXME: %cmp = call zeroext i1 @check()
 ; Outer loop control.
-; CHECK: copy_data.exit:
-; CHECK: br i1 %cmp38, label %if.then39, label %while.cond.outer
+; CHECKFIXME: copy_data.exit:
+; CHECKFIXME: br i1 %cmp38, label %if.then39, label %while.cond.outer
 ; Outer-outer loop tail.
-; CHECK: while.cond.outer.outer.backedge:
-; CHECK: br label %while.cond.outer.outer
+; CHECKFIXME: while.cond.outer.outer.backedge:
+; CHECKFIXME: br label %while.cond.outer.outer
 define void @unloopDeepNested() nounwind {
 for.cond8.preheader.i:
   %cmp113.i = call zeroext i1 @check()
diff --git a/test/Transforms/LoopVectorize/lcssa-crash.ll b/test/Transforms/LoopVectorize/lcssa-crash.ll
index 06b3b08aa0..de6be54849 100644
--- a/test/Transforms/LoopVectorize/lcssa-crash.ll
+++ b/test/Transforms/LoopVectorize/lcssa-crash.ll
@@ -27,3 +27,14 @@ for.end.i.i.i:
   unreachable
 }
 
+; PR16139
+define void @test2(i8* %x) {
+entry:
+  indirectbr i8* %x, [ label %L0, label %L1 ]
+
+L0:
+  br label %L0
+
+L1:
+  ret void
+}
diff --git a/test/Transforms/LoopVectorize/no_outside_user.ll b/test/Transforms/LoopVectorize/no_outside_user.ll
new file mode 100644
index 0000000000..6f0357c5e5
--- /dev/null
+++ b/test/Transforms/LoopVectorize/no_outside_user.ll
@@ -0,0 +1,41 @@
+; RUN: opt -S -loop-vectorize -force-vector-unroll=1 -force-vector-width=2 < %s | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
+
+@f = common global i32 0, align 4
+@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
+@c = common global i32 0, align 4
+@a = common global i32 0, align 4
+@b = common global i32 0, align 4
+@e = common global i32 0, align 4
+
+; We used to vectorize this loop. But it has a value that is used outside of the
+; and is not a recognized reduction variable "tmp17".
+
+; CHECK-NOT: <2 x i32>
+
+define i32 @main()  {
+bb:
+  %b.promoted = load i32* @b, align 4
+  br label %.lr.ph.i
+
+.lr.ph.i:
+  %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
+  %tmp2 = icmp sgt i32 %tmp8, 10
+  br i1 %tmp2, label %bb16, label %bb10
+
+bb10:
+  br label %bb16
+
+bb16:
+  %tmp17 = phi i32 [ 0, %bb10 ], [ 1, %.lr.ph.i ]
+  %tmp18 = add nsw i32 %tmp8, 1
+  %tmp19 = icmp slt i32 %tmp18, 4
+  br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
+
+f1.exit.loopexit:
+  %.lcssa = phi i32 [ %tmp17, %bb16 ]
+  ret i32 %.lcssa
+}
+
+
diff --git a/test/Transforms/LoopVectorize/value-ptr-bug.ll b/test/Transforms/LoopVectorize/value-ptr-bug.ll
new file mode 100644
index 0000000000..f376656f07
--- /dev/null
+++ b/test/Transforms/LoopVectorize/value-ptr-bug.ll
@@ -0,0 +1,50 @@
+; RUN: opt -S -loop-vectorize -force-vector-width=4 -force-vector-unroll=1 -dce -instcombine < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+
+; PR16073
+
+; Because we were caching value pointers accross a function call that could RAUW
+; we would generate an undefined value store below:
+; SCEVExpander::expandCodeFor would change a value (the start value of an
+; induction) that we cached in the induction variable list.
+
+; CHECK: test_vh
+; CHECK-NOT: store <4 x i8> undef
+
+define void @test_vh(i32* %ptr265, i32* %ptr266, i32 %sub267) {
+entry:
+  br label %loop
+
+loop:
+  %inc = phi i32 [ %sub267, %entry ], [ %add, %loop]
+  %ext.inc = sext i32 %inc to i64
+  %add.ptr265 = getelementptr inbounds i32* %ptr265, i64 %ext.inc
+  %add.ptr266 = getelementptr inbounds i32* %ptr266, i64 %ext.inc
+  %add = add i32 %inc, 9
+  %cmp = icmp slt i32 %add, 140
+  br i1 %cmp, label %block1, label %loop
+
+block1:
+  %sub267.lcssa = phi i32 [ %add, %loop ]
+  %add.ptr266.lcssa = phi i32* [ %add.ptr266, %loop ]
+  %add.ptr265.lcssa = phi i32* [ %add.ptr265, %loop ]
+  %tmp29 = bitcast i32* %add.ptr265.lcssa to i8*
+  %tmp30 = bitcast i32* %add.ptr266.lcssa to i8*
+  br label %do.body272
+
+do.body272:
+  %row_width.5 = phi i32 [ %sub267.lcssa, %block1 ], [ %dec, %do.body272 ]
+  %sp.4 = phi i8* [ %tmp30, %block1 ], [ %incdec.ptr273, %do.body272 ]
+  %dp.addr.4 = phi i8* [ %tmp29, %block1 ], [ %incdec.ptr274, %do.body272 ]
+  %incdec.ptr273 = getelementptr inbounds i8* %sp.4, i64 1
+  %tmp31 = load i8* %sp.4, align 1
+  %incdec.ptr274 = getelementptr inbounds i8* %dp.addr.4, i64 1
+  store i8 %tmp31, i8* %dp.addr.4, align 1
+  %dec = add i32 %row_width.5, -1
+  %cmp276 = icmp eq i32 %dec, 0
+  br i1 %cmp276, label %loop.exit, label %do.body272
+
+loop.exit:
+  ret void
+}