[LV] Use vector phis for some secondary induction variables

Previously, we materialized secondary vector IVs from the primary scalar IV,
by offseting the primary to match the correct start value, and then broadcasting
it - inside the loop body. Instead, we can use a real vector IV, like we do for
the primary.

This enables using vector IVs for secondary integer IVs whose type matches the
type of the primary.

Differential Revision: http://reviews.llvm.org/D20932

llvm-svn: 272283

1 files changed, 49 insertions, 2 deletions

diff --git a/llvm/test/Transforms/LoopVectorize/induction.ll b/llvm/test/Transforms/LoopVectorize/induction.ll
index c2d4d96153a..b193a5b4a85 100644
--- a/llvm/test/Transforms/LoopVectorize/induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/induction.ll
@@ -174,8 +174,11 @@ loopexit:
 
 ; CHECK-LABEL: wrappingindvars1
 ; CHECK-LABEL: vector.scevcheck
+; CHECK-LABEL: vector.ph
+; CHECK: %[[START:.*]] = add <2 x i32> %{{.*}}, <i32 0, i32 1>
 ; CHECK-LABEL: vector.body
-; CHECK: add <2 x i32> {{%[^ ]*}}, <i32 0, i32 1>
+; CHECK: %[[PHI:.*]] = phi <2 x i32> [ %[[START]], %vector.ph ], [ %[[STEP:.*]], %vector.body ]
+; CHECK: %[[STEP]] = add <2 x i32> %[[PHI]], <i32 2, i32 2>
 define void @wrappingindvars1(i8 %t, i32 %len, i32 *%A) {
  entry:
   %st = zext i8 %t to i16
@@ -209,8 +212,11 @@ define void @wrappingindvars1(i8 %t, i32 %len, i32 *%A) {
 ; The expression gets converted to ({4 * (zext %t to i32),+,4}).
 ; CHECK-LABEL: wrappingindvars2
 ; CHECK-LABEL: vector.scevcheck
+; CHECK-LABEL: vector.ph
+; CHECK: %[[START:.*]] = add <2 x i32> %{{.*}}, <i32 0, i32 4>
 ; CHECK-LABEL: vector.body
-; CHECK: add <2 x i32> {{%[^ ]*}}, <i32 0, i32 4>
+; CHECK: %[[PHI:.*]] = phi <2 x i32> [ %[[START]], %vector.ph ], [ %[[STEP:.*]], %vector.body ]
+; CHECK: %[[STEP]] = add <2 x i32> %[[PHI]], <i32 8, i32 8>
 define void @wrappingindvars2(i8 %t, i32 %len, i32 *%A) {
 
 entry:
@@ -302,3 +308,44 @@ for.body:
 exit:
   ret void
 }
+
+; IND-LABEL: nonprimary
+; IND-LABEL: vector.ph
+; IND: %[[INSERT:.*]] = insertelement <2 x i32> undef, i32 %i, i32 0
+; IND: %[[SPLAT:.*]] = shufflevector <2 x i32> %[[INSERT]], <2 x i32> undef, <2 x i32> zeroinitializer
+; IND: %[[START:.*]] = add <2 x i32> %[[SPLAT]], <i32 0, i32 42>
+; IND-LABEL: vector.body:
+; IND: %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+; IND: %vec.ind = phi <2 x i32> [ %[[START]], %vector.ph ], [ %step.add, %vector.body ]
+; IND: %step.add = add <2 x i32> %vec.ind, <i32 84, i32 84>
+; IND: %index.next = add i32 %index, 2
+; IND: %[[CMP:.*]] = icmp eq i32 %index.next
+; IND: br i1 %[[CMP]]
+; UNROLL-LABEL: nonprimary
+; UNROLL-LABEL: vector.ph
+; UNROLL: %[[INSERT:.*]] = insertelement <2 x i32> undef, i32 %i, i32 0
+; UNROLL: %[[SPLAT:.*]] = shufflevector <2 x i32> %[[INSERT]], <2 x i32> undef, <2 x i32> zeroinitializer
+; UNROLL: %[[START:.*]] = add <2 x i32> %[[SPLAT]], <i32 0, i32 42>
+; UNROLL-LABEL: vector.body:
+; UNROLL: %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+; UNROLL: %vec.ind = phi <2 x i32> [ %[[START]], %vector.ph ], [ %step.add1, %vector.body ]
+; UNROLL: %step.add = add <2 x i32> %vec.ind, <i32 84, i32 84>
+; UNROLL: %step.add1 = add <2 x i32> %vec.ind, <i32 168, i32 168>
+; UNROLL: %index.next = add i32 %index, 4
+; UNROLL: %[[CMP:.*]] = icmp eq i32 %index.next
+; UNROLL: br i1 %[[CMP]]
+define void @nonprimary(i32* nocapture %a, i32 %start, i32 %i, i32 %k) {
+for.body.preheader:
+  br label %for.body
+
+for.body:
+  %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ %i, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i32, i32* %a, i32 %indvars.iv
+  store i32 %indvars.iv, i32* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 42
+  %exitcond = icmp eq i32 %indvars.iv.next, %k
+  br i1 %exitcond, label %exit, label %for.body
+
+exit:
+  ret void
+}