[PATCH] D98633: [LV] Move exact FP math check out of Requirements.

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
fhahn added reviewers: Ayal, hsaito, ebrevnov, anemet, lebedev.ri.
Herald added subscribers: hiraditya, nemanjai.
fhahn requested review of this revision.
Herald added a project: LLVM.

We know if the loop contains FP instructions preventing vectorization
after we are done with legality checks. This patch updates the code the
check for un-vectorizable FP operations earlier, to avoid unnecessarily
running the cost model and picking a vectorization factor. It also makes
the code more direct and moves the check to a position where similar
checks are done.

I might be missing something, but I don't see any reason to handle this
check differently to other, similar checks.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D98633

Files:
  llvm/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h
  llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp
  llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
  llvm/test/Transforms/LoopVectorize/PowerPC/reg-usage.ll


Index: llvm/test/Transforms/LoopVectorize/PowerPC/reg-usage.ll
===================================================================
--- llvm/test/Transforms/LoopVectorize/PowerPC/reg-usage.ll
+++ llvm/test/Transforms/LoopVectorize/PowerPC/reg-usage.ll
@@ -234,7 +234,7 @@
   %x.05 = phi ppc_fp128 [ %d, %entry ], [ %sub, %for.body ]
   %arrayidx = getelementptr inbounds ppc_fp128, ppc_fp128* %n, i32 %i.06
   %0 = load ppc_fp128, ppc_fp128* %arrayidx, align 8
-  %sub = fsub ppc_fp128 %x.05, %0
+  %sub = fsub fast ppc_fp128 %x.05, %0
   %inc = add nsw i32 %i.06, 1
   %exitcond = icmp eq i32 %inc, 2048
   br i1 %exitcond, label %for.end, label %for.body
Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -9579,6 +9579,21 @@
     return false;
   }
 
+  if (!Requirements.canVectorizeFPMath(Hints)) {
+    ORE->emit([&]() {
+      auto *ExactFPMathInst = Requirements.getExactFPInst();
+      return OptimizationRemarkAnalysisFPCommute(DEBUG_TYPE, "CantReorderFPOps",
+                                                 ExactFPMathInst->getDebugLoc(),
+                                                 ExactFPMathInst->getParent())
+             << "loop not vectorized: cannot prove it is safe to reorder "
+                "floating-point operations";
+    });
+    LLVM_DEBUG(dbgs() << "LV: loop not vectorized: cannot prove it is safe to "
+                         "reorder floating-point operations\n");
+    Hints.emitRemarkWithHints();
+    return false;
+  }
+
   bool UseInterleaved = TTI->enableInterleavedAccessVectorization();
   InterleavedAccessInfo IAI(PSE, L, DT, LI, LVL.getLAI());
 
Index: llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp
+++ llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp
@@ -250,16 +250,6 @@
     Function *F, Loop *L, const LoopVectorizeHints &Hints) {
   const char *PassName = Hints.vectorizeAnalysisPassName();
   bool Failed = false;
-  if (ExactFPMathInst && !Hints.allowReordering()) {
-    ORE.emit([&]() {
-      return OptimizationRemarkAnalysisFPCommute(
-                 PassName, "CantReorderFPOps", ExactFPMathInst->getDebugLoc(),
-                 ExactFPMathInst->getParent())
-             << "loop not vectorized: cannot prove it is safe to reorder "
-                "floating-point operations";
-    });
-    Failed = true;
-  }
 
   // Test if runtime memcheck thresholds are exceeded.
   bool PragmaThresholdReached =
Index: llvm/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h
===================================================================
--- llvm/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h
+++ llvm/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h
@@ -189,6 +189,11 @@
 
   bool doesNotMeet(Function *F, Loop *L, const LoopVectorizeHints &Hints);
 
+  Instruction *getExactFPInst() { return ExactFPMathInst; }
+  bool canVectorizeFPMath(const LoopVectorizeHints &Hints) const {
+    return !ExactFPMathInst || Hints.allowReordering();
+  }
+
 private:
   unsigned NumRuntimePointerChecks = 0;
   Instruction *ExactFPMathInst = nullptr;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D98633.330677.patch
Type: text/x-patch
Size: 3386 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210315/b07a8b85/attachment.bin>