From d4f643a5df29ee85345508e4220fc21e30ab2581 Mon Sep 17 00:00:00 2001
From: Hal Finkel <hfinkel@anl.gov>
Date: Fri, 10 Apr 2015 15:05:02 +0000
Subject: [PATCH] [PowerPC] Prefetching should also consider depth > 1 loops

Iterating over loops from the LoopInfo instance only provides top-level loops.
We need to search the whole tree of loops to find the inner ones.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@234603 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/PowerPC/PPCLoopDataPrefetch.cpp    |  7 +-
 .../PowerPC/loop-data-prefetch-inner.ll       | 66 +++++++++++++++++++
 2 files changed, 71 insertions(+), 2 deletions(-)
 create mode 100644 test/CodeGen/PowerPC/loop-data-prefetch-inner.ll

diff --git a/lib/Target/PowerPC/PPCLoopDataPrefetch.cpp b/lib/Target/PowerPC/PPCLoopDataPrefetch.cpp
index 005bcafe0cf..77d3727c303 100644
--- a/lib/Target/PowerPC/PPCLoopDataPrefetch.cpp
+++ b/lib/Target/PowerPC/PPCLoopDataPrefetch.cpp
@@ -14,6 +14,7 @@
 #define DEBUG_TYPE "ppc-loop-data-prefetch"
 #include "PPC.h"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/ADT/DepthFirstIterator.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/CodeMetrics.h"
@@ -110,8 +111,10 @@ bool PPCLoopDataPrefetch::runOnFunction(Function &F) {
 
   bool MadeChange = false;
 
-  for (LoopInfo::iterator I = LI->begin(), E = LI->end();
-       I != E; ++I) {
+  if (LI->empty())
+    return MadeChange;
+
+  for (auto I = df_begin(*LI->begin()), E = df_end(*LI->begin()); I != E; ++I) {
     Loop *L = *I;
     MadeChange |= runOnLoop(L);
   }
diff --git a/test/CodeGen/PowerPC/loop-data-prefetch-inner.ll b/test/CodeGen/PowerPC/loop-data-prefetch-inner.ll
new file mode 100644
index 00000000000..adcc7b90bc4
--- /dev/null
+++ b/test/CodeGen/PowerPC/loop-data-prefetch-inner.ll
@@ -0,0 +1,66 @@
+; RUN: llc < %s | FileCheck %s
+target datalayout = "E-m:e-i64:64-n32:64"
+target triple = "powerpc64-bgq-linux"
+
+; Function Attrs: nounwind
+define void @foo(double* %x, double* nocapture readonly %y) #0 {
+entry:
+  %scevgep = getelementptr double, double* %x, i64 1599
+  %scevgep20 = getelementptr double, double* %y, i64 1599
+  br label %vector.memcheck
+
+vector.memcheck:                                  ; preds = %for.end, %entry
+  %j.015 = phi i32 [ 0, %entry ], [ %inc7, %for.end ]
+  %bound0 = icmp uge double* %scevgep20, %x
+  %bound1 = icmp uge double* %scevgep, %y
+  %memcheck.conflict = and i1 %bound0, %bound1
+  br i1 %memcheck.conflict, label %middle.block, label %vector.body
+
+vector.body:                                      ; preds = %vector.memcheck, %vector.body
+  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.memcheck ]
+  %0 = getelementptr inbounds double, double* %y, i64 %index
+  %1 = bitcast double* %0 to <4 x double>*
+  %wide.load = load <4 x double>, <4 x double>* %1, align 8
+  %2 = fadd <4 x double> %wide.load, <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
+  %3 = getelementptr inbounds double, double* %x, i64 %index
+  %4 = bitcast double* %3 to <4 x double>*
+  store <4 x double> %2, <4 x double>* %4, align 8
+  %index.next = add i64 %index, 4
+  %5 = icmp eq i64 %index.next, 1600
+  br i1 %5, label %middle.block, label %vector.body
+
+middle.block:                                     ; preds = %vector.body, %vector.memcheck
+  %resume.val = phi i1 [ false, %vector.memcheck ], [ true, %vector.body ]
+  %trunc.resume.val = phi i64 [ 0, %vector.memcheck ], [ 1600, %vector.body ]
+  br i1 %resume.val, label %for.end, label %for.body3
+
+for.body3:                                        ; preds = %middle.block, %for.body3
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body3 ], [ %trunc.resume.val, %middle.block ]
+  %arrayidx = getelementptr inbounds double, double* %y, i64 %indvars.iv
+  %6 = load double, double* %arrayidx, align 8
+  %add = fadd double %6, 1.000000e+00
+  %arrayidx5 = getelementptr inbounds double, double* %x, i64 %indvars.iv
+  store double %add, double* %arrayidx5, align 8
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1600
+  br i1 %exitcond, label %for.end, label %for.body3
+
+for.end:                                          ; preds = %middle.block, %for.body3
+  tail call void @bar(double* %x) #2
+  %inc7 = add nuw nsw i32 %j.015, 1
+  %exitcond16 = icmp eq i32 %inc7, 100
+  br i1 %exitcond16, label %for.end8, label %vector.memcheck
+
+for.end8:                                         ; preds = %for.end
+  ret void
+
+; CHECK-LABEL: @foo
+; CHECK: dcbt
+}
+
+declare void @bar(double*) #1
+
+attributes #0 = { nounwind "target-cpu"="a2q" }
+attributes #1 = { "target-cpu"="a2q" }
+attributes #2 = { nounwind }
+
-- 
2.34.1