[PATCH] D107621: [FuncSpec] Move invariant computation for spec cost out of loop (NFC-ish)

[Chuanqi Xu via Phabricator via llvm-commits]

ChuanqiXu created this revision.
ChuanqiXu added a reviewer: SjoerdMeijer.
Herald added subscribers: snehasish, ormris, hiraditya.
ChuanqiXu requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Noticed that the computation for function specialization cost of a function wouldn't change during the traversal of the arguments for the function. We could hoist the computation out of the traversal.  I observed about ~1% improvement on compile time for spec2017. But I guess it may not be precise. This should be NFC and fine.

Test Plan: check-all, spec2017 int rate


https://reviews.llvm.org/D107621

Files:
  llvm/lib/Transforms/IPO/FunctionSpecialization.cpp


Index: llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
===================================================================
--- llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
+++ llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
@@ -334,6 +334,19 @@
 
     LLVM_DEBUG(dbgs() << "FnSpecialization: Try function: " << F->getName()
                       << "\n");
+
+    // Determine if it would be profitable to create a specialization of the
+    // function where the argument takes on the given constant value. If so,
+    // add the constant to Constants.
+    auto FnSpecCost = getSpecializationCost(F);
+    if (!FnSpecCost.isValid()) {
+      LLVM_DEBUG(dbgs() << "FnSpecialization: Invalid specialisation cost.\n");
+      return false;
+    }
+
+    LLVM_DEBUG(dbgs() << "FnSpecialization: func specialisation cost: ";
+               FnSpecCost.print(dbgs()); dbgs() << "\n");
+
     // Determine if we should specialize the function based on the values the
     // argument can take on. If specialization is not profitable, we continue
     // on to the next argument.
@@ -351,7 +364,7 @@
       // be set to false by isArgumentInteresting (that function only adds
       // values to the Constants list that are deemed profitable).
       SmallVector<Constant *, 4> Constants;
-      if (!isArgumentInteresting(&A, Constants, IsPartial)) {
+      if (!isArgumentInteresting(&A, Constants, FnSpecCost, IsPartial)) {
         LLVM_DEBUG(dbgs() << "FnSpecialization: Argument is not interesting\n");
         continue;
       }
@@ -535,6 +548,7 @@
   /// argument.
   bool isArgumentInteresting(Argument *A,
                              SmallVectorImpl<Constant *> &Constants,
+                             const InstructionCost &FnSpecCost,
                              bool &IsPartial) {
     Function *F = A->getParent();
 
@@ -576,18 +590,6 @@
       return false;
     }
 
-    // Determine if it would be profitable to create a specialization of the
-    // function where the argument takes on the given constant value. If so,
-    // add the constant to Constants.
-    auto FnSpecCost = getSpecializationCost(F);
-    if (!FnSpecCost.isValid()) {
-      LLVM_DEBUG(dbgs() << "FnSpecialization: Invalid specialisation cost.\n");
-      return false;
-    }
-
-    LLVM_DEBUG(dbgs() << "FnSpecialization: func specialisation cost: ";
-               FnSpecCost.print(dbgs()); dbgs() << "\n");
-
     for (auto *C : PossibleConstants) {
       LLVM_DEBUG(dbgs() << "FnSpecialization: Constant: " << *C << "\n");
       if (ForceFunctionSpecialization) {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D107621.364702.patch
Type: text/x-patch
Size: 2577 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210806/ef33f3c4/attachment.bin>