[PATCH] D67764: [LV] Runtime checks with OptForSize

[Sjoerd Meijer via Phabricator via llvm-commits]

SjoerdMeijer updated this revision to Diff 221156.
SjoerdMeijer added a comment.

Thanks. The runtime check is now done earlier.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D67764/new/

https://reviews.llvm.org/D67764

Files:
  llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
  llvm/test/Transforms/LoopVectorize/runtime-check-optsize.ll


Index: llvm/test/Transforms/LoopVectorize/runtime-check-optsize.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/LoopVectorize/runtime-check-optsize.ll
@@ -0,0 +1,67 @@
+; RUN: opt < %s -O1 -disable-basicaa -S -o - | FileCheck %s
+
+; This function has the OptSize attribute, and requires runtime memory checks.
+; Check that we nicely bail, i.e. don't vectorise this loop (and
+; don't run in an assert) because we don't want to grow code-size.
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define dso_local void @f_w2(i64* noalias %x_p, i64* noalias %y_p, i64* noalias %z_p) #0 {
+; CHECK-LABEL: @f_w2(
+; CHECK-NOT:   vector.body:
+entry:
+  %x_p.addr = alloca i64*, align 8
+  %y_p.addr = alloca i64*, align 8
+  %z_p.addr = alloca i64*, align 8
+  %i = alloca i32, align 4
+  store i64* %x_p, i64** %x_p.addr, align 8
+  store i64* %y_p, i64** %y_p.addr, align 8
+  store i64* %z_p, i64** %z_p.addr, align 8
+  %0 = bitcast i32* %i to i8*
+  store i32 0, i32* %i, align 4
+  br label %for.cond
+
+for.cond:
+  %1 = load i32, i32* %i, align 4
+  %cmp = icmp slt i32 %1, 128
+  br i1 %cmp, label %for.body, label %for.cond.cleanup
+
+for.cond.cleanup:
+  %2 = bitcast i32* %i to i8*
+  br label %for.end
+
+for.body:
+  %3 = load i64*, i64** %x_p.addr, align 8
+  %4 = load i32, i32* %i, align 4
+  %idxprom = sext i32 %4 to i64
+  %arrayidx = getelementptr inbounds i64, i64* %3, i64 %idxprom
+  %5 = load i64, i64* %arrayidx, align 8
+  %6 = load i64*, i64** %y_p.addr, align 8
+  %7 = load i32, i32* %i, align 4
+  %idxprom1 = sext i32 %7 to i64
+  %arrayidx2 = getelementptr inbounds i64, i64* %6, i64 %idxprom1
+  %8 = load i64, i64* %arrayidx2, align 8
+  %add = add nsw i64 %5, %8
+  %9 = load i64*, i64** %z_p.addr, align 8
+  %10 = load i32, i32* %i, align 4
+  %idxprom3 = sext i32 %10 to i64
+  %arrayidx4 = getelementptr inbounds i64, i64* %9, i64 %idxprom3
+  store i64 %add, i64* %arrayidx4, align 8
+  br label %for.inc
+
+for.inc:
+  %11 = load i32, i32* %i, align 4
+  %inc = add nsw i32 %11, 1
+  store i32 %inc, i32* %i, align 4
+  br label %for.cond, !llvm.loop !10
+
+for.end:
+  ret void
+}
+
+attributes #0 = { nounwind optsize uwtable "target-cpu"="core-avx2" "target-features"="+avx,+avx2" }
+
+!10 = distinct !{!10, !11, !12}
+!11 = !{!"llvm.loop.vectorize.width", i32 2}
+!12 = !{!"llvm.loop.vectorize.enable", i1 true}
Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7555,6 +7555,11 @@
   // Use the cost model.
   LoopVectorizationCostModel CM(SEL, L, PSE, LI, &LVL, *TTI, TLI, DB, AC, ORE,
                                 F, &Hints, IAI);
+
+  // Don't vectorize if we expect code growth and compile for min size.
+  if (F->hasOptSize() && CM.runtimeChecksRequired())
+    return false;
+
   CM.collectValuesToIgnore();
 
   // Use the planner for vectorization.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D67764.221156.patch
Type: text/x-patch
Size: 3138 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190921/1a4b5ab3/attachment.bin>