Introduce -dot-cfg-mssa option which creates dot-cfg style file with mssa comments included in source

[jamieschmeiser]

Summary

Expand the print-memoryssa and print<memoryssa> passes with a new hidden
option -cfg-dot-mssa that names a file. When set, a dot-cfg style file
will be generated into the named file with the memoryssa comments retained
and those blocks containing them shown in light pink. The option does
nothing in isolation.

[dblaikie]

Needs some test coverage?

llvm/include/llvm/Analysis/CFGPrinter.h

152–156

These could/should be llvm::function_ref, since they're not being captured/outliving the function call.

Could also skip the -> void since it's implicit (& the one statement lambda body seems clear enough that it's not returning anything, and the functor type written a line above has the void return type spelled out), if you like.

[jamieschmeiser]

Respond to review comments: use llvm::function_ref, add unit test.
Fix option name and MemoryUse search string.
Regarding "-> void", I prefer to have the return type specified.

[dblaikie]

Seems pretty reasonable to me - @asbirlea could you check the test case to see the expected output, etc, makes sense? MSSA itself is not so much my wheelhouse so I'm not rightly sure myself.

llvm/include/llvm/Analysis/CFGPrinter.h

31

This can now be removed/replaced with STLExtras.h

[asbirlea]

Resulting dot file for the test looks good.

[dblaikie]

Sounds good to me then!

[jamieschmeiser]

Respond to review comments: replace header include.

[anhtuyen]

Reopen due to a failed build on builder polly-x86_64-linux
Full details are available at:

https://urldefense.proofpoint.com/v2/url?u=http-3A__lab.llvm.org-3A8011_-23builders_10_builds_615&d=DwICaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=Nr2SnYQn80wzIpml4veK2C-U4ilVZzi0kWYCc2WSLoA&m=CQgGRYR_3xNbiZpMevJ_fOqMziVnPdj-Qa8sJXA1NHQ&s=2jJHm84n7dRHKQQ9ax_D-LQiQ6lf_9ypNbnuph-cCAs&e=

Buildbot URL: https://urldefense.proofpoint.com/v2/url?u=http-3A__lab.llvm.org-3A8011_&d=DwICaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=Nr2SnYQn80wzIpml4veK2C-U4ilVZzi0kWYCc2WSLoA&m=CQgGRYR_3xNbiZpMevJ_fOqMziVnPdj-Qa8sJXA1NHQ&s=JFUUxRP6j5j0BPDw9yCc6lMViVmJFWrmcTQN13xbtEE&e=
Worker for this Build: polly-x86_64-gce1

[jamieschmeiser]

Fix build problems revealed when delivering changes.

llvm/include/llvm/Analysis/CFGPrinter.h

 #include "llvm/Analysis/HeatUtils.h"
 #include "llvm/IR/CFG.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/Support/FormatVariadic.h"
 #include "llvm/Support/GraphWriter.h"
+#include <functional>
 
 namespace llvm {
 class CFGViewerPass : public PassInfoMixin<CFGViewerPass> {
 public:
   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
 };
 
 class CFGOnlyViewerPass : public PassInfoMixin<CFGOnlyViewerPass> {
static std::string getSimpleNodeLabel(const BasicBlock *Node, DOTFuncInfo *) {
 
     std::string Str;
     raw_string_ostream OS(Str);
 
     Node->printAsOperand(OS, false);
     return OS.str();
   }
 
-  static std::string getCompleteNodeLabel(const BasicBlock *Node,
-                                          DOTFuncInfo *) {
+  static void eraseComment(std::string &OutStr, unsigned &I, unsigned Idx) {
+    OutStr.erase(OutStr.begin() + I, OutStr.begin() + Idx);
+    --I;
+  }
+
+  static std::string getCompleteNodeLabel(
+      const BasicBlock *Node, DOTFuncInfo *,
+      llvm::function_ref<void(raw_string_ostream &, const BasicBlock &)>
+          HandleBasicBlock = [](raw_string_ostream &OS,
+                                const BasicBlock &Node) -> void { OS << Node; },
+      llvm::function_ref<void(std::string &, unsigned &, unsigned)>
+          HandleComment = eraseComment) {
     enum { MaxColumns = 80 };
     std::string Str;
     raw_string_ostream OS(Str);
 
     if (Node->getName().empty()) {
       Node->printAsOperand(OS, false);
       OS << ":";
     }
 
-    OS << *Node;
+    HandleBasicBlock(OS, *Node);
     std::string OutStr = OS.str();
     if (OutStr[0] == '\n')
       OutStr.erase(OutStr.begin());
 
     // Process string output to make it nicer...
     unsigned ColNum = 0;
     unsigned LastSpace = 0;
     for (unsigned i = 0; i != OutStr.length(); ++i) {
       if (OutStr[i] == '\n') { // Left justify
         OutStr[i] = '\\';
         OutStr.insert(OutStr.begin() + i + 1, 'l');
         ColNum = 0;
         LastSpace = 0;
       } else if (OutStr[i] == ';') {             // Delete comments!
         unsigned Idx = OutStr.find('\n', i + 1); // Find end of line
-        OutStr.erase(OutStr.begin() + i, OutStr.begin() + Idx);
-        --i;
+        HandleComment(OutStr, i, Idx);
       } else if (ColNum == MaxColumns) { // Wrap lines.
         // Wrap very long names even though we can't find a space.
         if (!LastSpace)
           LastSpace = i;
         OutStr.insert(LastSpace, "\\l...");
         ColNum = i - LastSpace;
         LastSpace = 0;
         i += 3; // The loop will advance 'i' again.

llvm/lib/Analysis/MemorySSA.cpp

 #include "llvm/ADT/None.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/iterator.h"
 #include "llvm/ADT/iterator_range.h"
 #include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/CFGPrinter.h"
 #include "llvm/Analysis/IteratedDominanceFrontier.h"
 #include "llvm/Analysis/MemoryLocation.h"
 #include "llvm/Config/llvm-config.h"
 #include "llvm/IR/AssemblyAnnotationWriter.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instruction.h"

 #include <iterator>
 #include <memory>
 #include <utility>
 
 using namespace llvm;
 
 #define DEBUG_TYPE "memoryssa"
 
+static cl::opt<std::string>
+    DotCFGMSSA("dot-cfg-mssa",
+               cl::value_desc("file name for generated dot file"),
+               cl::desc("file name for generated dot file"), cl::init(""));
+
 INITIALIZE_PASS_BEGIN(MemorySSAWrapperPass, "memoryssa", "Memory SSA", false,
                       true)
 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
 INITIALIZE_PASS_END(MemorySSAWrapperPass, "memoryssa", "Memory SSA", false,
                     true)
 
 INITIALIZE_PASS_BEGIN(MemorySSAPrinterLegacyPass, "print-memoryssa",
MemorySSAPrinterLegacyPass::MemorySSAPrinterLegacyPass() : FunctionPass(ID) {
   initializeMemorySSAPrinterLegacyPassPass(*PassRegistry::getPassRegistry());
 }
 
 void MemorySSAPrinterLegacyPass::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.setPreservesAll();
   AU.addRequired<MemorySSAWrapperPass>();
 }
 
+class DOTFuncMSSAInfo {
+private:
+  const Function &F;
+  MemorySSAAnnotatedWriter MSSAWriter;
+
+public:
+  DOTFuncMSSAInfo(const Function &F, MemorySSA &MSSA)
+      : F(F), MSSAWriter(&MSSA) {}
+
+  const Function *getFunction() { return &F; }
+  MemorySSAAnnotatedWriter &getWriter() { return MSSAWriter; }
+};
+
+template <>
+struct GraphTraits<DOTFuncMSSAInfo *> : public GraphTraits<const BasicBlock *> {
+  static NodeRef getEntryNode(DOTFuncMSSAInfo *CFGInfo) {
+    return &(CFGInfo->getFunction()->getEntryBlock());
+  }
+
+  // nodes_iterator/begin/end - Allow iteration over all nodes in the graph
+  using nodes_iterator = pointer_iterator<Function::const_iterator>;
+
+  static nodes_iterator nodes_begin(DOTFuncMSSAInfo *CFGInfo) {
+    return nodes_iterator(CFGInfo->getFunction()->begin());
+  }
+
+  static nodes_iterator nodes_end(DOTFuncMSSAInfo *CFGInfo) {
+    return nodes_iterator(CFGInfo->getFunction()->end());
+  }
+
+  static size_t size(DOTFuncMSSAInfo *CFGInfo) {
+    return CFGInfo->getFunction()->size();
+  }
+};
+
+template <>
+struct DOTGraphTraits<DOTFuncMSSAInfo *> : public DefaultDOTGraphTraits {
+
+  DOTGraphTraits(bool IsSimple = false) : DefaultDOTGraphTraits(IsSimple) {}
+
+  static std::string getGraphName(DOTFuncMSSAInfo *CFGInfo) {
+    return "MSSA CFG for '" + CFGInfo->getFunction()->getName().str() +
+           "' function";
+  }
+
+  std::string getNodeLabel(const BasicBlock *Node, DOTFuncMSSAInfo *CFGInfo) {
+    return DOTGraphTraits<DOTFuncInfo *>::getCompleteNodeLabel(
+        Node, nullptr,
+        [CFGInfo](raw_string_ostream &OS, const BasicBlock &BB) -> void {
+          BB.print(OS, &CFGInfo->getWriter(), true, true);
+        },
+        [](std::string &S, unsigned &I, unsigned Idx) -> void {
+          std::string Str = S.substr(I, Idx - I);
+          StringRef SR = Str;
+          if (SR.count(" = MemoryDef(") || SR.count(" = MemoryPhi(") ||
+              SR.count("MemoryUse("))
+            return;
+          DOTGraphTraits<DOTFuncInfo *>::eraseComment(S, I, Idx);
+        });
+  }
+
+  static std::string getEdgeSourceLabel(const BasicBlock *Node,
+                                        const_succ_iterator I) {
+    return DOTGraphTraits<DOTFuncInfo *>::getEdgeSourceLabel(Node, I);
+  }
+
+  /// Display the raw branch weights from PGO.
+  std::string getEdgeAttributes(const BasicBlock *Node, const_succ_iterator I,
+                                DOTFuncMSSAInfo *CFGInfo) {
+    return "";
+  }
+
+  std::string getNodeAttributes(const BasicBlock *Node,
+                                DOTFuncMSSAInfo *CFGInfo) {
+    return getNodeLabel(Node, CFGInfo).find(';') != std::string::npos
+               ? "style=filled, fillcolor=lightpink"
+               : "";
+  }
+};
+
 bool MemorySSAPrinterLegacyPass::runOnFunction(Function &F) {
   auto &MSSA = getAnalysis<MemorySSAWrapperPass>().getMSSA();
+  if (DotCFGMSSA != "") {
+    DOTFuncMSSAInfo CFGInfo(F, MSSA);
+    WriteGraph(&CFGInfo, "", false, "MSSA", DotCFGMSSA);
+  } else
     MSSA.print(dbgs());
+
   if (VerifyMemorySSA)
     MSSA.verifyMemorySSA();
   return false;
 }
 
 AnalysisKey MemorySSAAnalysis::Key;
 
 MemorySSAAnalysis::Result MemorySSAAnalysis::run(Function &F,
bool MemorySSAAnalysis::Result::invalidate(
   auto PAC = PA.getChecker<MemorySSAAnalysis>();
   return !(PAC.preserved() || PAC.preservedSet<AllAnalysesOn<Function>>()) ||
          Inv.invalidate<AAManager>(F, PA) ||
          Inv.invalidate<DominatorTreeAnalysis>(F, PA);
 }
 
 PreservedAnalyses MemorySSAPrinterPass::run(Function &F,
                                             FunctionAnalysisManager &AM) {
+  auto &MSSA = AM.getResult<MemorySSAAnalysis>(F).getMSSA();
+  if (DotCFGMSSA != "") {
+    DOTFuncMSSAInfo CFGInfo(F, MSSA);
+    WriteGraph(&CFGInfo, "", false, "MSSA", DotCFGMSSA);
+  } else {
     OS << "MemorySSA for function: " << F.getName() << "\n";
-  AM.getResult<MemorySSAAnalysis>(F).getMSSA().print(OS);
+    MSSA.print(OS);
+  }
 
   return PreservedAnalyses::all();
 }
 
 PreservedAnalyses MemorySSAVerifierPass::run(Function &F,
                                              FunctionAnalysisManager &AM) {
   AM.getResult<MemorySSAAnalysis>(F).getMSSA().verifyMemorySSA();
 

llvm/test/Analysis/MemorySSA/print-dot-cfg-mssa.ll

+; RUN: opt -basic-aa -print-memoryssa -dot-cfg-mssa=out.dot -enable-new-pm=0 -analyze < %s 2>&1 > /dev/null
+;RUN: FileCheck %s -input-file=out.dot
+; RUN: opt -aa-pipeline=basic-aa -passes='print<memoryssa>' -dot-cfg-mssa=out.dot < %s 2>&1 > /dev/null
+;RUN: FileCheck %s -input-file=out.dot
+
+; Test -dot-cfg-mssa option for -print-memoryssa.
+; Test is based on following C code with some forwarding basic blocks
+; added to show that only those blocks with memory ssa comments
+; are colourized.
+
+;void g();
+
+;int f(int *p, int *q, int *r) {
+;  int i = 0;
+;  if (*r)
+;    i = 1;
+;  else
+;    g();
+;  *p = *q + 1;
+;  if (i)
+;    ++i;
+;  return *q;
+;}
+
+define signext i32 @f(i32* %p, i32* %q, i32* %r) {
+entry:
+  br label %bb1
+
+bb1:
+  %p.addr = alloca i32*, align 8
+  %q.addr = alloca i32*, align 8
+  %r.addr = alloca i32*, align 8
+  %i = alloca i32, align 4
+  store i32* %p, i32** %p.addr, align 8
+  store i32* %q, i32** %q.addr, align 8
+  store i32* %r, i32** %r.addr, align 8
+  %0 = bitcast i32* %i to i8*
+  store i32 0, i32* %i, align 4
+  %1 = load i32*, i32** %r.addr, align 8
+  %2 = load i32, i32* %1, align 4
+  %tobool = icmp ne i32 %2, 0
+  br i1 %tobool, label %if.then, label %if.else
+
+if.then:
+  store i32 1, i32* %i, align 4
+  br label %bb2
+
+bb2:
+  br label %if.end
+
+if.else:
+  call void bitcast (void (...)* @g to void ()*)()
+  br label %if.end
+
+if.end:
+  %3 = load i32*, i32** %q.addr, align 8
+  %4 = load i32, i32* %3, align 4
+  %add = add nsw i32 %4, 1
+  %5 = load i32*, i32** %p.addr, align 8
+  store i32 %add, i32* %5, align 4
+  %6 = load i32, i32* %i, align 4
+  %tobool1 = icmp ne i32 %6, 0
+  br i1 %tobool1, label %if.then2, label %if.end3
+
+if.then2:
+  %7 = load i32, i32* %i, align 4
+  %inc = add nsw i32 %7, 1
+  br label %bb3
+
+bb3:
+  store i32 %inc, i32* %i, align 4
+  br label %if.end3
+
+if.end3:
+  br label %bb4
+
+bb4:
+  %8 = load i32*, i32** %q.addr, align 8
+  %9 = load i32, i32* %8, align 4
+  %10 = bitcast i32* %i to i8*
+  ret i32 %9
+}
+
+declare void @g(...)
+
+; CHECK: digraph "MSSA"
+; CHECK-NEXT: label="MSSA";
+; CHECK: {{Node0x.* [shape=record,label="{entry:.*}"]}}
+; CHECK: {{[shape=record,style=filled, fillcolor=lightpink,label="{bb1:.*1 = MemoryDef(liveOnEntry).*2 = MemoryDef(1).*3 = MemoryDef(2).*4 = MemoryDef(3).*MemoryUse(3).*MemoryUse(liveOnEntry).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{if.then:.*5 = MemoryDef(4).*}"]}}
+; CHECK {{[shape=record,label="{bb2:.*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{if.else:.*6 = MemoryDef(4).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{if.end:.*10 = MemoryPhi({bb2,5},{if.else,6})/*MemoryUse(2).*MemoryUse(10).*MemoryUse(1).*7 = MemoryDef(10).*MemoryUse(10).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{if.then2:.*MemoryUse(10).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{bb3:.*8 = MemoryDef(7).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{if.end3:.*9 = MemoryPhi({if.end,7},{bb3,8}).*}"]}}
+; CHECK {{[shape=record,style=filled, fillcolor=lightpink,label="{bb4:.*MemoryUse(2).*MemoryUse(7).*}"]}}
+