From 26369ba67b697621e80a15df28badf90c1cc8f0f Mon Sep 17 00:00:00 2001
From: Dorit Nuzman <dorit.nuzman@intel.com>
Date: Wed, 14 Aug 2019 05:22:20 +0000
Subject: [PATCH] [LV] Fold-tail flag

This is the compiler-flag equivalent of the Predicate pragma
(https://reviews.llvm.org/D65197), to direct the vectorizer to fold the
remainder-loop into the main-loop using predication.

Differential Revision: https://reviews.llvm.org/D66108

Reviewers: Ayal, hsaito, fhahn, SjoerdMeije
llvm-svn: 368801
---
 lib/Transforms/Vectorize/LoopVectorize.cpp    | 18 ++++++++++++-----
 .../LoopVectorize/X86/tail_loop_folding.ll    | 20 ++++++++++++++++++-
 2 files changed, 32 insertions(+), 6 deletions(-)

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 1f8cbf7f340..20305416079 100644
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -177,6 +177,14 @@ static cl::opt<unsigned> TinyTripCountVectorThreshold(
              "value are vectorized only if no scalar iteration overheads "
              "are incurred."));
 
+// Indicates that an epilogue is undesired, predication is preferred.
+// This means that the vectorizer will try to fold the loop-tail (epilogue)
+// into the loop and predicate the loop body accordingly.
+static cl::opt<bool> PreferPredicateOverEpilog(
+    "prefer-predicate-over-epilog", cl::init(false), cl::Hidden,
+    cl::desc("Indicate that an epilogue is undesired, predication should be "
+             "used instead."));
+
 static cl::opt<bool> MaximizeBandwidth(
     "vectorizer-maximize-bandwidth", cl::init(false), cl::Hidden,
     cl::desc("Maximize bandwidth when selecting vectorization factor which "
@@ -906,7 +914,7 @@ enum ScalarEpilogueLowering {
   CM_ScalarEpilogueNotAllowedLowTripLoop,
 
   // Loop hint predicate indicating an epilogue is undesired.
-  CM_ScalarEpilogueNotNeededPredicatePragma
+  CM_ScalarEpilogueNotNeededUsePredicate
 };
 
 /// LoopVectorizationCostModel - estimates the expected speedups due to
@@ -4804,9 +4812,9 @@ Optional<unsigned> LoopVectorizationCostModel::computeMaxVF() {
   switch (ScalarEpilogueStatus) {
   case CM_ScalarEpilogueAllowed:
     return computeFeasibleMaxVF(TC);
-  case CM_ScalarEpilogueNotNeededPredicatePragma:
+  case CM_ScalarEpilogueNotNeededUsePredicate:
     LLVM_DEBUG(
-        dbgs() << "LV: vector predicate hint found.\n"
+        dbgs() << "LV: vector predicate hint/switch found.\n"
                << "LV: Not allowing scalar epilogue, creating predicated "
                << "vector loop.\n");
     break;
@@ -7298,8 +7306,8 @@ getScalarEpilogueLowering(Function *F, Loop *L, LoopVectorizeHints &Hints,
       (F->hasOptSize() ||
        llvm::shouldOptimizeForSize(L->getHeader(), PSI, BFI)))
     SEL = CM_ScalarEpilogueNotAllowedOptSize;
-  else if (Hints.getPredicate())
-    SEL = CM_ScalarEpilogueNotNeededPredicatePragma;
+  else if (PreferPredicateOverEpilog || Hints.getPredicate()) 
+    SEL = CM_ScalarEpilogueNotNeededUsePredicate;
 
   return SEL;
 }
diff --git a/test/Transforms/LoopVectorize/X86/tail_loop_folding.ll b/test/Transforms/LoopVectorize/X86/tail_loop_folding.ll
index d7767385c78..eb0b499f512 100644
--- a/test/Transforms/LoopVectorize/X86/tail_loop_folding.ll
+++ b/test/Transforms/LoopVectorize/X86/tail_loop_folding.ll
@@ -1,4 +1,5 @@
 ; RUN: opt < %s -loop-vectorize -S | FileCheck %s
+; RUN: opt < %s -loop-vectorize -prefer-predicate-over-epilog -S | FileCheck -check-prefix=PREDFLAG %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
@@ -13,7 +14,15 @@ define dso_local void @tail_folding_enabled(i32* noalias nocapture %A, i32* noal
 ; CHECK:  %index.next = add i64 %index, 8
 ; CHECK:  %12 = icmp eq i64 %index.next, 432
 ; CHECK:  br i1 %12, label %middle.block, label %vector.body, !llvm.loop !0
-
+; PREDFLAG-LABEL: tail_folding_enabled(
+; PREDFLAG:  vector.body:
+; PREDFLAG:  %wide.masked.load = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(
+; PREDFLAG:  %wide.masked.load1 = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(
+; PREDFLAG:  %8 = add nsw <8 x i32> %wide.masked.load1, %wide.masked.load
+; PREDFLAG:  call void @llvm.masked.store.v8i32.p0v8i32(
+; PREDFLAG:  %index.next = add i64 %index, 8
+; PREDFLAG:  %12 = icmp eq i64 %index.next, 432
+; PREDFLAG:  br i1 %12, label %middle.block, label %vector.body, !llvm.loop !0
 entry:
   br label %for.body
 
@@ -40,6 +49,15 @@ define dso_local void @tail_folding_disabled(i32* noalias nocapture %A, i32* noa
 ; CHECK-NOT:  @llvm.masked.load.v8i32.p0v8i32(
 ; CHECK-NOT:  @llvm.masked.store.v8i32.p0v8i32(
 ; CHECK:      br i1 %44, label {{.*}}, label %vector.body
+; PREDFLAG-LABEL: tail_folding_disabled(
+; PREDFLAG:  vector.body:
+; PREDFLAG:  %wide.masked.load = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(
+; PREDFLAG:  %wide.masked.load1 = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(
+; PREDFLAG:  %8 = add nsw <8 x i32> %wide.masked.load1, %wide.masked.load
+; PREDFLAG:  call void @llvm.masked.store.v8i32.p0v8i32(
+; PREDFLAG:  %index.next = add i64 %index, 8
+; PREDFLAG:  %12 = icmp eq i64 %index.next, 432
+; PREDFLAG:  br i1 %12, label %middle.block, label %vector.body, !llvm.loop !4
 entry:
   br label %for.body