Add TargetMachine::addEarlyAsPossiblePasses.

[jlebar]

Summary

This is a hook to allow TargetMachine to install passes at the
EP_EarlyAsPossible PassManagerBuilder extension point.

[chandlerc]

I feel like it would be useful to implement this at least for one target (I'm assuming NVPTX) to show the use case before landing it...

[jlebar]

In D18614#394117, @chandlerc wrote:

I feel like it would be useful to implement this at least for one target (I'm assuming NVPTX) to show the use case before landing it...

Yes, see D18616.

(The problem, it seems to me, is that we ask people to aggressively split up patches but then don't have a good way to give reviewers the full context. I can manually go in to Phabricator and set the depends-on metadata, but when I'm posting a dozen small patches, the relative time cost of doing that is huge. It needs to Happen Automatically.)

[jlebar]

Chandler, are you happy with this? It seemed like the main problem was that I hadn't linked to D18616 originally.

[jlebar]

Chandler, friendly ping.

[jlebar]

Friendly ping.

[• koriakin]

include/llvm/Target/TargetMachine.h

223

Where do you call this function?

[jlebar]

include/llvm/Target/TargetMachine.h

223

It's called from clang, or whatever is setting up the pass manager. D18617.

[rnk]

Shouldn't this patch also modify the pass pipeline to call this hook at some point?

include/llvm/Target/TargetMachine.h

220–221

Maybe this should say "before all mid-level optimization passes". I initially assumed you wanted this to run before IR-level codegen passes (CodeGenPrepare, LSR, etc), but that's what we have today.

[jlebar]

In D18614#414054, @rnk wrote:

Shouldn't this patch also modify the pass pipeline to call this hook at some point?

AFAICT, it needs to be done outside of llvm proper, see D18617.

include/llvm/Target/TargetMachine.h

220–221

I actually do want to run before IR-level codegen passes, see D18617. I agree that the comment isn't clear, I will change it.

[jlebar]

Clarify comment.

[rnk]

lgtm

include/llvm/Target/TargetMachine.h

223

As discussed, we should call it from opt as well, then. opt should have a TargetMachine somewhere.

[jlebar]

Call addEarlyAsPossiblePasses from opt.

include/llvm/Target/TargetMachine.h

public:
 
   /// \brief Get a \c TargetIRAnalysis appropriate for the target.
   ///
   /// This is used to construct the new pass manager's target IR analysis pass,
   /// set up appropriately for this target machine. Even the old pass manager
   /// uses this to answer queries about the IR.
   virtual TargetIRAnalysis getTargetIRAnalysis();
 
+  /// Add target-specific function passes that should be run as early as
+  /// possible in the optimization pipeline.  Most TargetMachines have no such
+  /// passes.
+  virtual void addEarlyAsPossiblePasses(PassManagerBase &) {}
+
   /// These enums are meant to be passed into addPassesToEmitFile to indicate
   /// what type of file to emit, and returned by it to indicate what type of
   /// file could actually be made.
   enum CodeGenFileType {
     CGFT_AssemblyFile,
     CGFT_ObjectFile,
     CGFT_Null         // Do not emit any output.
   };

tools/opt/opt.cpp

 }
 
 /// This routine adds optimization passes based on selected optimization level,
 /// OptLevel.
 ///
 /// OptLevel - Optimization Level
 static void AddOptimizationPasses(legacy::PassManagerBase &MPM,
                                   legacy::FunctionPassManager &FPM,
-                                  unsigned OptLevel, unsigned SizeLevel) {
+                                  TargetMachine *TM, unsigned OptLevel,
+                                  unsigned SizeLevel) {
   if (!NoVerify || VerifyEach)
     FPM.add(createVerifierPass()); // Verify that input is correct
 
   PassManagerBuilder Builder;
   Builder.OptLevel = OptLevel;
   Builder.SizeLevel = SizeLevel;
 
   if (DisableInline) {
static void AddOptimizationPasses(legacy::PassManagerBase &MPM,
   // If option wasn't forced via cmd line (-vectorize-loops, -loop-vectorize)
   else if (!Builder.LoopVectorize)
     Builder.LoopVectorize = OptLevel > 1 && SizeLevel < 2;
 
   // When #pragma vectorize is on for SLP, do the same as above
   Builder.SLPVectorize =
       DisableSLPVectorization ? false : OptLevel > 1 && SizeLevel < 2;
 
+  // Add target-specific passes that need to run as early as possible.
+  if (TM)
+    Builder.addExtension(
+        PassManagerBuilder::EP_EarlyAsPossible,
+        [&](const PassManagerBuilder &, legacy::PassManagerBase &PM) {
+          TM->addEarlyAsPossiblePasses(PM);
+        });
+
   Builder.populateFunctionPassManager(FPM);
   Builder.populateModulePassManager(MPM);
 }
 
 static void AddStandardLinkPasses(legacy::PassManagerBase &PM) {
   PassManagerBuilder Builder;
   Builder.VerifyInput = true;
   if (DisableOptimizations)
#endif
   for (unsigned i = 0; i < PassList.size(); ++i) {
     if (StandardLinkOpts &&
         StandardLinkOpts.getPosition() < PassList.getPosition(i)) {
       AddStandardLinkPasses(Passes);
       StandardLinkOpts = false;
     }
 
     if (OptLevelO1 && OptLevelO1.getPosition() < PassList.getPosition(i)) {
-      AddOptimizationPasses(Passes, *FPasses, 1, 0);
+      AddOptimizationPasses(Passes, *FPasses, TM.get(), 1, 0);
       OptLevelO1 = false;
     }
 
     if (OptLevelO2 && OptLevelO2.getPosition() < PassList.getPosition(i)) {
-      AddOptimizationPasses(Passes, *FPasses, 2, 0);
+      AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 0);
       OptLevelO2 = false;
     }
 
     if (OptLevelOs && OptLevelOs.getPosition() < PassList.getPosition(i)) {
-      AddOptimizationPasses(Passes, *FPasses, 2, 1);
+      AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 1);
       OptLevelOs = false;
     }
 
     if (OptLevelOz && OptLevelOz.getPosition() < PassList.getPosition(i)) {
-      AddOptimizationPasses(Passes, *FPasses, 2, 2);
+      AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 2);
       OptLevelOz = false;
     }
 
     if (OptLevelO3 && OptLevelO3.getPosition() < PassList.getPosition(i)) {
-      AddOptimizationPasses(Passes, *FPasses, 3, 0);
+      AddOptimizationPasses(Passes, *FPasses, TM.get(), 3, 0);
       OptLevelO3 = false;
     }
 
     const PassInfo *PassInf = PassList[i];
     Pass *P = nullptr;
     if (PassInf->getTargetMachineCtor())
       P = PassInf->getTargetMachineCtor()(TM.get());
     else if (PassInf->getNormalCtor())
#endif
   }
 
   if (StandardLinkOpts) {
     AddStandardLinkPasses(Passes);
     StandardLinkOpts = false;
   }
 
   if (OptLevelO1)
-    AddOptimizationPasses(Passes, *FPasses, 1, 0);
+    AddOptimizationPasses(Passes, *FPasses, TM.get(), 1, 0);
 
   if (OptLevelO2)
-    AddOptimizationPasses(Passes, *FPasses, 2, 0);
+    AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 0);
 
   if (OptLevelOs)
-    AddOptimizationPasses(Passes, *FPasses, 2, 1);
+    AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 1);
 
   if (OptLevelOz)
-    AddOptimizationPasses(Passes, *FPasses, 2, 2);
+    AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 2);
 
   if (OptLevelO3)
-    AddOptimizationPasses(Passes, *FPasses, 3, 0);
+    AddOptimizationPasses(Passes, *FPasses, TM.get(), 3, 0);
 
   if (OptLevelO1 || OptLevelO2 || OptLevelOs || OptLevelOz || OptLevelO3) {
     FPasses->doInitialization();
     for (Function &F : *M)
       FPasses->run(F);
     FPasses->doFinalization();
   }