[llvm] 5b250a2 - [Analysis][LoopVectorize] do not form reductions of pointers

Sanjay Patel via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 19 11:03:50 PST 2021


Author: Sanjay Patel
Date: 2021-02-19T14:01:57-05:00
New Revision: 5b250a27ec7822aa0a32abb696cb16c2cc60149c

URL: https://github.com/llvm/llvm-project/commit/5b250a27ec7822aa0a32abb696cb16c2cc60149c
DIFF: https://github.com/llvm/llvm-project/commit/5b250a27ec7822aa0a32abb696cb16c2cc60149c.diff

LOG: [Analysis][LoopVectorize] do not form reductions of pointers

This is a fix for https://llvm.org/PR49215 either before/after
we make a verifier enhancement for vector reductions with D96904.

I'm not sure what the current thinking is for pointer math/logic
in IR. We allow icmp on pointer values. Therefore, we match min/max
patterns, so without this patch, the vectorizer could form a vector
reduction from that sequence.

But the LangRef definitions for min/max and vector reduction
intrinsics do not allow pointer types:
https://llvm.org/docs/LangRef.html#llvm-smax-intrinsic
https://llvm.org/docs/LangRef.html#llvm-vector-reduce-umax-intrinsic

So we would crash/assert at some point - either in IR verification,
in the cost model, or in codegen. If we do want to allow this kind
of transform, we will need to update the LangRef and all of those
parts of the compiler.

Differential Revision: https://reviews.llvm.org/D97047

Added: 
    llvm/test/Transforms/LoopVectorize/reduction-ptr.ll

Modified: 
    llvm/lib/Analysis/IVDescriptors.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/IVDescriptors.cpp b/llvm/lib/Analysis/IVDescriptors.cpp
index 6204331a2e9b..601dc8dcdf10 100644
--- a/llvm/lib/Analysis/IVDescriptors.cpp
+++ b/llvm/lib/Analysis/IVDescriptors.cpp
@@ -243,11 +243,14 @@ bool RecurrenceDescriptor::AddReductionVar(PHINode *Phi, RecurKind Kind,
   if (RecurrenceType->isFloatingPointTy()) {
     if (!isFloatingPointRecurrenceKind(Kind))
       return false;
-  } else {
+  } else if (RecurrenceType->isIntegerTy()) {
     if (!isIntegerRecurrenceKind(Kind))
       return false;
     if (isArithmeticRecurrenceKind(Kind))
       Start = lookThroughAnd(Phi, RecurrenceType, VisitedInsts, CastInsts);
+  } else {
+    // Pointer min/max may exist, but it is not supported as a reduction op.
+    return false;
   }
 
   Worklist.push_back(Start);

diff  --git a/llvm/test/Transforms/LoopVectorize/reduction-ptr.ll b/llvm/test/Transforms/LoopVectorize/reduction-ptr.ll
new file mode 100644
index 000000000000..5cae61638f31
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/reduction-ptr.ll
@@ -0,0 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s  -loop-vectorize -force-vector-width=4 -S | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+
+; Reductions of pointer types are not supported.
+
+define void @PR49215(i32* %p, i32* %q) {
+; CHECK-LABEL: @PR49215(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[G:%.*]] = phi i32* [ [[P:%.*]], [[ENTRY]] ], [ [[UMIN:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[CMP2:%.*]] = icmp ult i32* [[Q:%.*]], [[G]]
+; CHECK-NEXT:    [[UMIN]] = select i1 [[CMP2]], i32* [[Q]], i32* [[G]]
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], undef
+; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT:%.*]], label [[FOR_BODY]]
+; CHECK:       loopexit:
+; CHECK-NEXT:    [[UMIN_LCSSA:%.*]] = phi i32* [ [[UMIN]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[PHI_CAST:%.*]] = ptrtoint i32* [[UMIN_LCSSA]] to i64
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %g = phi i32* [ %p, %entry ], [ %umin, %for.body ]
+  %cmp2 = icmp ult i32* %q, %g
+  %umin = select i1 %cmp2, i32* %q, i32* %g
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, undef
+  br i1 %exitcond, label %loopexit, label %for.body
+
+loopexit:
+  %phi.cast = ptrtoint i32* %umin to i64
+  ret void
+}


        


More information about the llvm-commits mailing list