SLPVectorizer: Change the order in which new instructions are added to the function.

We are not working on a DAG and I ran into a number of problems when I enabled the vectorizations of 'diamond-trees' (trees that share leafs).
* Imroved the numbering API.
* Changed the placement of new instructions to the last root.
* Fixed a bug with external tree users with non-zero lane.
* Fixed a bug in the placement of in-tree users.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@182508 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 104 insertions, 1 deletions

diff --git a/test/Transforms/SLPVectorizer/X86/crash_povray.ll b/test/Transforms/SLPVectorizer/X86/crash_povray.ll
new file mode 100644
index 0000000000..7ef8df49f0
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/crash_povray.ll
@@ -0,0 +1,34 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7
+
+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"
+
+%struct.hoge = type { double, double, double}
+
+define void @zot(%struct.hoge* %arg) {
+bb:
+  %tmp = load double* undef, align 8
+  %tmp1 = fsub double %tmp, undef
+  %tmp2 = load double* undef, align 8
+  %tmp3 = fsub double %tmp2, undef
+  %tmp4 = fmul double %tmp3, undef
+  %tmp5 = fmul double %tmp3, undef
+  %tmp6 = fsub double %tmp5, undef
+  %tmp7 = getelementptr inbounds %struct.hoge* %arg, i64 0, i32 1
+  store double %tmp6, double* %tmp7, align 8
+  %tmp8 = fmul double %tmp1, undef
+  %tmp9 = fsub double %tmp8, undef
+  %tmp10 = getelementptr inbounds %struct.hoge* %arg, i64 0, i32 2
+  store double %tmp9, double* %tmp10, align 8
+  br i1 undef, label %bb11, label %bb12
+
+bb11:                                             ; preds = %bb
+  br label %bb14
+
+bb12:                                             ; preds = %bb
+  %tmp13 = fmul double undef, %tmp2
+  br label %bb14
+
+bb14:                                             ; preds = %bb12, %bb11
+  ret void
+}
diff --git a/test/Transforms/SLPVectorizer/X86/in-tree-user.ll b/test/Transforms/SLPVectorizer/X86/in-tree-user.ll
new file mode 100644
index 0000000000..69dc8897d6
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/in-tree-user.ll
@@ -0,0 +1,50 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-macosx10.7.0"
+
+@.str = private unnamed_addr constant [6 x i8] c"bingo\00", align 1
+
+; We can't vectorize when the roots are used inside the tree.
+;CHECK: @in_tree_user
+;CHECK-NOT: load <2 x double>
+;CHECK: ret
+define void @in_tree_user(double* nocapture %A, i32 %n) {
+entry:
+  %conv = sitofp i32 %n to double
+  br label %for.body
+
+for.body:                                         ; preds = %for.inc, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
+  %0 = shl nsw i64 %indvars.iv, 1
+  %arrayidx = getelementptr inbounds double* %A, i64 %0
+  %1 = load double* %arrayidx, align 8
+  %mul1 = fmul double %conv, %1
+  %mul2 = fmul double %mul1, 7.000000e+00
+  %add = fadd double %mul2, 5.000000e+00
+  %BadValue = fadd double %add, %add    ; <------------------ In tree user.
+  %2 = or i64 %0, 1
+  %arrayidx6 = getelementptr inbounds double* %A, i64 %2
+  %3 = load double* %arrayidx6, align 8
+  %mul8 = fmul double %conv, %3
+  %mul9 = fmul double %mul8, 4.000000e+00
+  %add10 = fadd double %mul9, 9.000000e+00
+  %cmp11 = fcmp ogt double %add, %add10
+  br i1 %cmp11, label %if.then, label %for.inc
+
+if.then:                                          ; preds = %for.body
+  %call = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([6 x i8]* @.str, i64 0, i64 0))
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body, %if.then
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 100
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.inc
+  ret void
+}
+
+declare i32 @printf(i8* nocapture, ...)
+
diff --git a/test/Transforms/SLPVectorizer/X86/multi_user.ll b/test/Transforms/SLPVectorizer/X86/multi_user.ll
index aaa6063fde..d4d4d28950 100644
--- a/test/Transforms/SLPVectorizer/X86/multi_user.ll
+++ b/test/Transforms/SLPVectorizer/X86/multi_user.ll
@@ -12,8 +12,8 @@ target triple = "x86_64-apple-macosx10.7.0"
 ;}
 
 ;CHECK: @foo
-;CHECK: insertelement <4 x i32>
 ;CHECK: load <4 x i32>
+;CHECK: insertelement <4 x i32>
 ;CHECK: add <4 x i32>
 ;CHECK: store <4 x i32>
 ;CHECK: ret
diff --git a/test/Transforms/SLPVectorizer/X86/ordering.ll b/test/Transforms/SLPVectorizer/X86/ordering.ll
new file mode 100644
index 0000000000..588e115f5f
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/ordering.ll
@@ -0,0 +1,19 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7
+
+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"
+
+define void @updateModelQPFrame(i32 %m_Bits) {
+entry:
+  %0 = load double* undef, align 8
+  %mul = fmul double undef, %0
+  %mul2 = fmul double undef, %mul
+  %mul4 = fmul double %0, %mul2
+  %mul5 = fmul double undef, 4.000000e+00
+  %mul7 = fmul double undef, %mul5
+  %conv = sitofp i32 %m_Bits to double
+  %mul8 = fmul double %conv, %mul7
+  %add = fadd double %mul4, %mul8
+  %cmp11 = fcmp olt double %add, 0.000000e+00
+  ret void
+}