[LoopDataPrefetch] Add missed LoopSimplify dependence for prefetch pass

SCEVExpander::expandCodeFor may expand add recurrences for loop with a preheader,
so we should make LoopDataPrefetch dependent on LoopSimplify.
This patch will try to fix : https://bugs.llvm.org/show_bug.cgi?id=43784

Reviewed By: Meinersbur

Differential Revision: https://reviews.llvm.org/D108448

diff --git a/llvm/lib/Transforms/Scalar/LoopDataPrefetch.cpp b/llvm/lib/Transforms/Scalar/LoopDataPrefetch.cpp
index a5d7835..77d7660 100644 (file)
--- a/llvm/lib/Transforms/Scalar/LoopDataPrefetch.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopDataPrefetch.cpp
@@ -29,6 +29,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/ScalarEvolutionExpander.h"
 #include "llvm/Transforms/Utils/ValueMapper.h"
@@ -127,6 +128,8 @@ public:
     AU.addPreserved<DominatorTreeWrapperPass>();
     AU.addRequired<LoopInfoWrapperPass>();
     AU.addPreserved<LoopInfoWrapperPass>();
+    AU.addRequiredID(LoopSimplifyID);
+    AU.addPreservedID(LoopSimplifyID);
     AU.addRequired<OptimizationRemarkEmitterWrapperPass>();
     AU.addRequired<ScalarEvolutionWrapperPass>();
     AU.addPreserved<ScalarEvolutionWrapperPass>();
@@ -143,6 +146,7 @@ INITIALIZE_PASS_BEGIN(LoopDataPrefetchLegacyPass, "loop-data-prefetch",
 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
 INITIALIZE_PASS_DEPENDENCY(OptimizationRemarkEmitterWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass)
 INITIALIZE_PASS_END(LoopDataPrefetchLegacyPass, "loop-data-prefetch",

diff --git a/llvm/test/CodeGen/AArch64/O3-pipeline.ll b/llvm/test/CodeGen/AArch64/O3-pipeline.ll
index 8d03f7d..1d6eca6 100644 (file)
--- a/llvm/test/CodeGen/AArch64/O3-pipeline.ll
+++ b/llvm/test/CodeGen/AArch64/O3-pipeline.ll
@@ -25,6 +25,7 @@
 ; CHECK-NEXT:       Simplify the CFG
 ; CHECK-NEXT:       Dominator Tree Construction
 ; CHECK-NEXT:       Natural Loop Information
+; CHECK-NEXT:       Canonicalize natural loops
 ; CHECK-NEXT:       Lazy Branch Probability Analysis
 ; CHECK-NEXT:       Lazy Block Frequency Analysis
 ; CHECK-NEXT:       Optimization Remark Emitter

diff --git a/llvm/test/Transforms/LoopDataPrefetch/AArch64/pr43784.ll b/llvm/test/Transforms/LoopDataPrefetch/AArch64/pr43784.ll
new file mode 100644 (file)
index 0000000..c33e31a
--- /dev/null
+++ b/llvm/test/Transforms/LoopDataPrefetch/AArch64/pr43784.ll
@@ -0,0 +1,118 @@
+; RUN: opt -mcpu=kryo -loop-simplify -loop-data-prefetch -max-prefetch-iters-ahead=1000 -min-prefetch-stride=16 -S < %s | FileCheck %s
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+%struct._Chv = type { i32, i32, i32, i32, i32, i32, i32*, i32*, double*, %struct._DV, %struct._Chv* }
+%struct._DV = type { i32, i32, i32, double* }
+
+declare double* @f_entries() local_unnamed_addr
+
+define i32 @f(%struct._Chv* %chv, i32 %npivot, i32* %pivotsizes, i32* %sizes) local_unnamed_addr {
+if.end:
+  switch i32 undef, label %sw.default [
+    i32 1, label %sw.epilog
+    i32 3, label %sw.epilog
+    i32 4, label %sw.epilog
+    i32 5, label %sw.epilog
+    i32 6, label %sw.epilog
+    i32 7, label %sw.epilog
+  ]
+
+sw.default:                                       ; preds = %if.end
+  unreachable
+
+sw.epilog:                                        ; preds = %if.end, %if.end, %if.end, %if.end, %if.end, %if.end
+  br label %if.end12
+
+if.end12:                                         ; preds = %sw.epilog
+  %nD13 = getelementptr inbounds %struct._Chv, %struct._Chv* %chv, i64 0, i32 1
+  %0 = load i32, i32* %nD13, align 4
+  %nU15 = getelementptr inbounds %struct._Chv, %struct._Chv* %chv, i64 0, i32 3
+  %1 = load i32, i32* %nU15, align 4
+  %add17 = add i32 %1, %0
+  %call18 = call double* @f_entries()
+  switch i32 undef, label %sw.epilog2454 [
+    i32 3, label %sw.bb213
+  ]
+
+sw.bb213:                                         ; preds = %if.end12
+  switch i32 undef, label %sw.epilog2454 [
+    i32 0, label %sw.bb214
+  ]
+
+sw.bb214:                                         ; preds = %sw.bb213
+  br label %if.then220
+
+if.then220:                                       ; preds = %sw.bb214
+  %type230 = getelementptr inbounds %struct._Chv, %struct._Chv* %chv, i64 0, i32 4
+  %2 = load i32, i32* %type230, align 8
+  br label %if.else319
+
+if.else319:                                       ; preds = %if.then220
+  switch i32 %2, label %sw.epilog2454 [
+    i32 2, label %for.cond372.preheader
+  ]
+
+for.cond372.preheader:                            ; preds = %if.else319
+  %cmp3734342 = icmp sgt i32 %npivot, 0
+  br i1 %cmp3734342, label %for.body374.lr.ph, label %sw.epilog2454
+
+for.body374.lr.ph:                                ; preds = %for.cond372.preheader
+  br label %for.body374
+
+for.body374:                                      ; preds = %for.body374.lr.ph
+  %arrayidx376 = getelementptr inbounds i32, i32* %pivotsizes, i64 0
+  %3 = load i32, i32* %arrayidx376, align 4
+  %add377 = add i32 %3, 0
+  br label %for.body381.lr.ph
+
+for.body381.lr.ph:                                ; preds = %for.body374
+  %cmp3874327 = icmp slt i32 %add377, %add17
+  br i1 %cmp3874327, label %for.body381.us, label %for.body381.preheader
+
+for.body381.preheader:                            ; preds = %for.body381.lr.ph
+  ret i32 undef
+
+for.body381.us:                                   ; preds = %for.cond386.for.end421_crit_edge.us, %for.body381.lr.ph
+  %indvars.iv4991.us = phi i64 [ %indvars.iv.next4992.us, %for.cond386.for.end421_crit_edge.us ], [ undef, %for.body381.lr.ph ]
+  %stride226.54337.us = phi i32 [ %dec423.us, %for.cond386.for.end421_crit_edge.us ], [ %add17, %for.body381.lr.ph ]
+  %kstart225.54336.us = phi i32 [ %add422.us, %for.cond386.for.end421_crit_edge.us ], [ 0, %for.body381.lr.ph ]
+  %4 = trunc i64 %indvars.iv4991.us to i32
+  %5 = xor i32 %4, -1
+  %add382.us = add i32 %add377, %5
+  %sub383.us = add i32 %add382.us, %kstart225.54336.us
+  br label %for.body388.us
+
+; CHECK: for.body388.us:
+; CHECK: call void @llvm.prefetch
+for.body388.us:                                   ; preds = %for.inc418.us, %for.body381.us
+  %kk224.3.in4330.us = phi i32 [ %sub383.us, %for.body381.us ], [ %kk224.3.us, %for.inc418.us ]
+  %jj223.34329.us = phi i32 [ %add377, %for.body381.us ], [ %inc419.us, %for.inc418.us ]
+  %kk224.3.us = add nsw i32 %kk224.3.in4330.us, 1
+  %mul389.us = shl nsw i32 %kk224.3.us, 1
+  %idxprom390.us = sext i32 %mul389.us to i64
+  %arrayidx391.us = getelementptr inbounds double, double* %call18, i64 %idxprom390.us
+  %6 = load double, double* %arrayidx391.us, align 8
+  %call396.us = call double @Zabs(double %6)
+  br label %for.inc418.us
+
+for.inc418.us:                                    ; preds = %for.body388.us
+  %inc419.us = add nsw i32 %jj223.34329.us, 1
+  %cmp387.us = icmp slt i32 %inc419.us, %add17
+  br i1 %cmp387.us, label %for.body388.us, label %for.cond386.for.end421_crit_edge.us
+
+for.cond386.for.end421_crit_edge.us:              ; preds = %for.inc418.us
+  %add422.us = add nsw i32 %kstart225.54336.us, %stride226.54337.us
+  %dec423.us = add nsw i32 %stride226.54337.us, -1
+  %indvars.iv.next4992.us = add nsw i64 %indvars.iv4991.us, 1
+  br label %for.body381.us
+
+sw.epilog2454:                                    ; preds = %for.cond372.preheader, %if.else319, %sw.bb213, %if.end12
+  ret i32 0
+}
+
+declare double @llvm.fabs.f64(double)
+declare dso_local double @Zabs(double) local_unnamed_addr
+declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg)
+