Commit 2ef80ce8 authored by Hans Wennborg's avatar Hans Wennborg
Browse files

Merging r308847: 

------------------------------------------------------------------------
r308847 | mkazantsev | 2017-07-23 08:40:19 -0700 (Sun, 23 Jul 2017) | 24 lines

[SCEV] Limit max size of AddRecExpr during evolving

When SCEV calculates product of two SCEVAddRecs from the same loop, it
tries to combine them into one big AddRecExpr. If the sizes of the initial
SCEVs were `S1` and `S2`, the size of their product is `S1 + S2 - 1`, and every
operand of the resulting SCEV is combined from operands of initial SCEV and
has much higher complexity than they have.

As result, if we try to calculate something like:
  %x1 = {a,+,b}
  %x2 = mul i32 %x1, %x1
  %x3 = mul i32 %x2, %x1
  %x4 = mul i32 %x3, %x2
  ...
The size of such SCEVs grows as `2^N`, and the arguments
become more and more complex as we go forth. This leads
to long compilation and huge memory consumption.

This patch sets a limit after which we don't try to combine two
`SCEVAddRecExpr`s into one. By default, max allowed size of the
resulting AddRecExpr is set to 16.

Differential Revision: https://reviews.llvm.org/D35664

------------------------------------------------------------------------

llvm-svn: 310629
parent 375efaaa

llvm/lib/Analysis/ScalarEvolution.cpp

+11 −0
Original line number Diff line number Diff line
@@ -162,6 +162,11 @@ static cl::opt<unsigned> 
                cl::desc("Maximum depth of recursive SExt/ZExt"),

                cl::init(8));



static cl::opt<unsigned>

    MaxAddRecSize("scalar-evolution-max-add-rec-size", cl::Hidden,

                  cl::desc("Max coefficients in AddRec during evolving"),

                  cl::init(16));



//===----------------------------------------------------------------------===//

//                           SCEV class definitions

//===----------------------------------------------------------------------===//
@@ -2878,6 +2883,12 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops, 
      if (!OtherAddRec || OtherAddRec->getLoop() != AddRecLoop)

        continue;



      // Limit max number of arguments to avoid creation of unreasonably big

      // SCEVAddRecs with very complex operands.

      if (AddRec->getNumOperands() + OtherAddRec->getNumOperands() - 1 >

          MaxAddRecSize)

        continue;



      bool Overflow = false;

      Type *Ty = AddRec->getType();

      bool LargerThan64Bits = getTypeSizeInBits(Ty) > 64;

llvm/test/Analysis/ScalarEvolution/max-addrec-size.ll

0 → 100644
+33 −0
Original line number Diff line number Diff line 
; RUN: opt -analyze -scalar-evolution -scalar-evolution-max-add-rec-size=3 < %s | FileCheck %s



; Show that we are able to avoid creation of huge SCEVs by capping the max

; AddRec size.

define i32 @test_01(i32 %a, i32 %b) {



; CHECK-LABEL: Classifying expressions for: @test_01

; CHECK-NEXT:    %iv = phi i32 [ %a, %entry ], [ %iv.next, %loop ]

; CHECK-NEXT:    -->  {%a,+,%b}<%loop> U: full-set S: full-set

; CHECK-NEXT:    %iv.next = add i32 %iv, %b

; CHECK-NEXT:    -->  {(%a + %b),+,%b}<%loop> U: full-set S: full-set

; CHECK-NEXT:    %x1 = mul i32 %iv, %iv.next

; CHECK-NEXT:    -->  {((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop> U: full-set S: full-set

; CHECK-NEXT:    %x2 = mul i32 %x1, %x1

; CHECK-NEXT:    -->  ({((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop> * {((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop>) U: full-set S: full-set

; CHECK-NEXT:    %x3 = mul i32 %x2, %x1

; CHECK-NEXT:    -->  ({((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop> * {((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop> * {((%a + %b) * %a),+,(((2 * %a) + (2 * %b)) * %b),+,(2 * %b * %b)}<%loop>) U: full-set S: full-set



entry:

  br label %loop



loop:

  %iv = phi i32 [ %a, %entry ], [ %iv.next, %loop ]

  %iv.next = add i32 %iv, %b

  %cond = icmp slt i32 %iv.next, 1000

  br i1 %cond, label %loop, label %exit



exit:

  %x1 = mul i32 %iv, %iv.next

  %x2 = mul i32 %x1, %x1

  %x3 = mul i32 %x2, %x1

  ret i32 %x3

}