Add support for reverse pointer induction variables. These are loops that contain pointers that count backwards.

For example, this is the hot loop in BZIP:

  do {
    m = *--p;
    *p = ( ... );
  } while (--n);



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@173219 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 195 insertions, 0 deletions

diff --git a/test/Transforms/LoopVectorize/bzip_reverse_loops.ll b/test/Transforms/LoopVectorize/bzip_reverse_loops.ll
new file mode 100644
index 0000000..431e422
--- /dev/null
+++ b/test/Transforms/LoopVectorize/bzip_reverse_loops.ll
@@ -0,0 +1,71 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -S -enable-if-conversion | 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"
+target triple = "x86_64-apple-macosx10.8.0"
+
+;CHECK: fc
+;CHECK: load <4 x i16>
+;CHECK-NEXT: shufflevector <4 x i16>
+;CHECK: select <4 x i1>
+;CHECK: store <4 x i16>
+;CHECK: ret
+define void @fc(i16* nocapture %p, i32 %n, i32 %size) nounwind uwtable ssp {
+entry:
+  br label %do.body
+
+do.body:                                          ; preds = %cond.end, %entry
+  %n.addr.0 = phi i32 [ %n, %entry ], [ %dec, %cond.end ]
+  %p.addr.0 = phi i16* [ %p, %entry ], [ %incdec.ptr, %cond.end ]
+  %incdec.ptr = getelementptr inbounds i16* %p.addr.0, i64 -1
+  %0 = load i16* %incdec.ptr, align 2, !tbaa !0
+  %conv = zext i16 %0 to i32
+  %cmp = icmp ult i32 %conv, %size
+  br i1 %cmp, label %cond.end, label %cond.true
+
+cond.true:                                        ; preds = %do.body
+  %sub = sub i32 %conv, %size
+  %phitmp = trunc i32 %sub to i16
+  br label %cond.end
+
+cond.end:                                         ; preds = %do.body, %cond.true
+  %cond = phi i16 [ %phitmp, %cond.true ], [ 0, %do.body ]
+  store i16 %cond, i16* %incdec.ptr, align 2, !tbaa !0
+  %dec = add i32 %n.addr.0, -1
+  %tobool = icmp eq i32 %dec, 0
+  br i1 %tobool, label %do.end, label %do.body
+
+do.end:                                           ; preds = %cond.end
+  ret void
+}
+
+;CHECK: example1
+;CHECK: load <4 x i32>
+;CHECK-NEXT: shufflevector <4 x i32>
+;CHECK: select <4 x i1>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define void @example1(i32* nocapture %a, i32 %n, i32 %wsize) nounwind uwtable ssp {
+entry:
+  br label %do.body
+
+do.body:                                          ; preds = %do.body, %entry
+  %n.addr.0 = phi i32 [ %n, %entry ], [ %dec, %do.body ]
+  %p.0 = phi i32* [ %a, %entry ], [ %incdec.ptr, %do.body ]
+  %incdec.ptr = getelementptr inbounds i32* %p.0, i64 -1
+  %0 = load i32* %incdec.ptr, align 4, !tbaa !3
+  %cmp = icmp slt i32 %0, %wsize
+  %sub = sub nsw i32 %0, %wsize
+  %cond = select i1 %cmp, i32 0, i32 %sub
+  store i32 %cond, i32* %incdec.ptr, align 4, !tbaa !3
+  %dec = add nsw i32 %n.addr.0, -1
+  %tobool = icmp eq i32 %dec, 0
+  br i1 %tobool, label %do.end, label %do.body
+
+do.end:                                           ; preds = %do.body
+  ret void
+}
+
+!0 = metadata !{metadata !"short", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}
+!3 = metadata !{metadata !"int", metadata !1}
diff --git a/test/Transforms/LoopVectorize/ptr_loops.ll b/test/Transforms/LoopVectorize/ptr_loops.ll
new file mode 100644
index 0000000..25599f8
--- /dev/null
+++ b/test/Transforms/LoopVectorize/ptr_loops.ll
@@ -0,0 +1,74 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -S -enable-if-conversion | 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"
+target triple = "x86_64-apple-macosx10.8.0"
+
+@A = global [36 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35], align 16
+@B = global [36 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35], align 16
+
+;CHECK:_Z5test1v
+;CHECK: load <4 x i32>
+;CHECK: shufflevector <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @_Z5test1v() nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %0, %1
+  %p.02 = phi i32* [ getelementptr inbounds ([36 x i32]* @A, i64 0, i64 18), %0 ], [ %4, %1 ]
+  %b.01 = phi i32* [ getelementptr inbounds ([36 x i32]* @B, i64 0, i64 0), %0 ], [ %5, %1 ]
+  %2 = load i32* %b.01, align 4
+  %3 = shl nsw i32 %2, 1
+  store i32 %3, i32* %p.02, align 4
+  %4 = getelementptr inbounds i32* %p.02, i64 -1
+  %5 = getelementptr inbounds i32* %b.01, i64 1
+  %6 = icmp eq i32* %4, getelementptr ([36 x i32]* @A, i64 128102389400760775, i64 3)
+  br i1 %6, label %7, label %1
+
+; <label>:7                                       ; preds = %1
+  ret i32 0
+}
+
+;CHECK:_Z5test2v
+;CHECK: load <4 x i32>
+;CHECK: shufflevector <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @_Z5test2v() nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %0, %1
+  %p.02 = phi i32* [ getelementptr inbounds ([36 x i32]* @A, i64 0, i64 25), %0 ], [ %3, %1 ]
+  %b.01 = phi i32* [ getelementptr inbounds ([36 x i32]* @B, i64 0, i64 2), %0 ], [ %4, %1 ]
+  %2 = load i32* %b.01, align 4
+  store i32 %2, i32* %p.02, align 4
+  %3 = getelementptr inbounds i32* %p.02, i64 -1
+  %4 = getelementptr inbounds i32* %b.01, i64 1
+  %5 = icmp eq i32* %4, getelementptr inbounds ([36 x i32]* @A, i64 0, i64 18)
+  br i1 %5, label %6, label %1
+
+; <label>:6                                       ; preds = %1
+  ret i32 0
+}
+
+;CHECK:_Z5test3v
+;CHECK: load <4 x i32>
+;CHECK: shufflevector <4 x i32>
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @_Z5test3v() nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %0, %1
+  %p.02 = phi i32* [ getelementptr inbounds ([36 x i32]* @A, i64 0, i64 29), %0 ], [ %3, %1 ]
+  %b.01 = phi i32* [ getelementptr inbounds ([36 x i32]* @B, i64 0, i64 5), %0 ], [ %4, %1 ]
+  %2 = load i32* %b.01, align 4
+  store i32 %2, i32* %p.02, align 4
+  %3 = getelementptr inbounds i32* %p.02, i64 -1
+  %4 = getelementptr inbounds i32* %b.01, i64 1
+  %5 = icmp eq i32* %3, getelementptr ([36 x i32]* @A, i64 128102389400760775, i64 3)
+  br i1 %5, label %6, label %1
+
+; <label>:6                                       ; preds = %1
+  ret i32 0
+}
diff --git a/test/Transforms/LoopVectorize/struct_access.ll b/test/Transforms/LoopVectorize/struct_access.ll
new file mode 100644
index 0000000..de65d0d
--- /dev/null
+++ b/test/Transforms/LoopVectorize/struct_access.ll
@@ -0,0 +1,50 @@
+; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -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"
+target triple = "x86_64-apple-macosx10.9.0"
+
+%struct.coordinate = type { i32, i32 }
+
+; Make sure that we don't generate a wide load when accessing the struct.
+; struct coordinate {
+;  int x;
+;  int y;
+; };
+;
+;
+; int foo(struct coordinate *A, int n) {
+;
+;   int sum = 0;
+;   for (int i = 0; i < n; ++i)
+;     sum += A[i].x;
+;
+;   return sum;
+; }
+
+;CHECK: @foo
+;CHECK-NOT: load <4 x i32>
+;CHECK: ret
+define i32 @foo(%struct.coordinate* nocapture %A, i32 %n) nounwind uwtable readonly ssp {
+entry:
+  %cmp4 = icmp sgt i32 %n, 0
+  br i1 %cmp4, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %sum.05 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %x = getelementptr inbounds %struct.coordinate* %A, i64 %indvars.iv, i32 0
+  %0 = load i32* %x, align 4, !tbaa !0
+  %add = add nsw i32 %0, %sum.05
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
+  ret i32 %sum.0.lcssa
+}
+
+!0 = metadata !{metadata !"int", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}