LoopIdiom: Replace custom dependence analysis with LoopDependenceAnalysis.

Requires a lot less code and complexity on loop-idiom's side and the more precise analysis can catch more cases, like the one I included as a test case. This also fixes the edge-case miscompilation from PR9481. I'm not entirely sure that all cases are handled that the old checks handled but LDA will certainly become smarter in the future. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@166390 91177308-0d34-0410-b5e6-96231b3b80d8
2025-08-05 13:26:55 +00:00 · 2012-10-21 15:03:07 +00:00
parent bb950854ac
commit 5c6e9ae14e
3 changed files with 128 additions and 74 deletions
--- a/test/Transforms/LoopIdiom/multi-dimensional.ll
+++ b/test/Transforms/LoopIdiom/multi-dimensional.ll
@@ -0,0 +1,49 @@
+; RUN: opt -basicaa -loop-idiom < %s -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"
+target triple = "x86_64-apple-darwin10.0.0"
+
+%struct.ham = type { [2 x [2 x [2 x [16 x [8 x i32]]]]], i32, %struct.zot }
+%struct.zot = type { i32, i16, i16, [2 x [1152 x i32]] }
+
+define void @test1(%struct.ham* nocapture %arg) nounwind {
+bb:
+  br label %bb1
+
+bb1:                                              ; preds = %bb11, %bb
+  %tmp = phi i64 [ 0, %bb ], [ %tmp12, %bb11 ]
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %bb1
+  %tmp3 = phi i64 [ 0, %bb1 ], [ %tmp8, %bb2 ]
+  %tmp4 = getelementptr inbounds %struct.ham* %arg, i64 0, i32 0, i64 0, i64 1, i64 1, i64 %tmp, i64 %tmp3
+  store i32 0, i32* %tmp4, align 4
+  %tmp5 = getelementptr inbounds %struct.ham* %arg, i64 0, i32 0, i64 0, i64 1, i64 0, i64 %tmp, i64 %tmp3
+  store i32 0, i32* %tmp5, align 4
+  %tmp6 = getelementptr inbounds %struct.ham* %arg, i64 0, i32 0, i64 0, i64 0, i64 1, i64 %tmp, i64 %tmp3
+  store i32 0, i32* %tmp6, align 4
+  %tmp7 = getelementptr inbounds %struct.ham* %arg, i64 0, i32 0, i64 0, i64 0, i64 0, i64 %tmp, i64 %tmp3
+  store i32 0, i32* %tmp7, align 4
+  %tmp8 = add i64 %tmp3, 1
+  %tmp9 = trunc i64 %tmp8 to i32
+  %tmp10 = icmp eq i32 %tmp9, 8
+  br i1 %tmp10, label %bb11, label %bb2
+
+bb11:                                             ; preds = %bb2
+  %tmp12 = add i64 %tmp, 1
+  %tmp13 = trunc i64 %tmp12 to i32
+  %tmp14 = icmp eq i32 %tmp13, 16
+  br i1 %tmp14, label %bb15, label %bb1
+
+bb15:                                             ; preds = %bb11
+  ret void
+
+; CHECK: @test1
+; CHECK: bb1:
+; CHECK-NOT: store
+; CHECK: call void @llvm.memset.p0i8.i64
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64
+; CHECK-NOT: store
+; CHECK: br
+}
--- a/test/Transforms/LoopIdiom/sideeffect.ll
+++ b/test/Transforms/LoopIdiom/sideeffect.ll
@@ -0,0 +1,53 @@
+; RUN: opt -basicaa -loop-idiom < %s -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"
+target triple = "x86_64-apple-darwin10.0.0"
+
+; PR9481
+define i32 @test1() nounwind uwtable ssp {
+entry:
+  %a = alloca [10 x i8], align 1
+  br label %for.body
+
+for.cond1.preheader:                              ; preds = %for.body
+  %arrayidx5.phi.trans.insert = getelementptr inbounds [10 x i8]* %a, i64 0, i64 0
+  %.pre = load i8* %arrayidx5.phi.trans.insert, align 1
+  br label %for.body3
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv29 = phi i64 [ 0, %entry ], [ %indvars.iv.next30, %for.body ]
+  call void (...)* @bar() nounwind
+  %arrayidx = getelementptr inbounds [10 x i8]* %a, i64 0, i64 %indvars.iv29
+  store i8 23, i8* %arrayidx, align 1
+  %indvars.iv.next30 = add i64 %indvars.iv29, 1
+  %lftr.wideiv31 = trunc i64 %indvars.iv.next30 to i32
+  %exitcond32 = icmp eq i32 %lftr.wideiv31, 1000000
+  br i1 %exitcond32, label %for.cond1.preheader, label %for.body
+
+for.body3:                                        ; preds = %for.body3, %for.cond1.preheader
+  %0 = phi i8 [ %.pre, %for.cond1.preheader ], [ %add, %for.body3 ]
+  %indvars.iv = phi i64 [ 1, %for.cond1.preheader ], [ %indvars.iv.next, %for.body3 ]
+  call void (...)* @bar() nounwind
+  %arrayidx7 = getelementptr inbounds [10 x i8]* %a, i64 0, i64 %indvars.iv
+  %1 = load i8* %arrayidx7, align 1
+  %add = add i8 %1, %0
+  store i8 %add, i8* %arrayidx7, align 1
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 1000000
+  br i1 %exitcond, label %for.end12, label %for.body3
+
+for.end12:                                        ; preds = %for.body3
+  %arrayidx13 = getelementptr inbounds [10 x i8]* %a, i64 0, i64 2
+  %2 = load i8* %arrayidx13, align 1
+  %conv14 = sext i8 %2 to i32
+  %arrayidx15 = getelementptr inbounds [10 x i8]* %a, i64 0, i64 6
+  %3 = load i8* %arrayidx15, align 1
+  %conv16 = sext i8 %3 to i32
+  %add17 = add nsw i32 %conv16, %conv14
+  ret i32 %add17
+
+; CHECK: @test1
+; CHECK-NOT: @llvm.memset
+}
+
+declare void @bar(...)