[SCEV] Limit AddRec "simplifications" to avoid combinatorial explosions

[mkazantsev]

Summary

SCEV's transform that turns {A1,+,A2,+,...,+,An}<L> * {B1,+,B2,+,...,+,Bn}<L> into
a single AddRec of size 2n+1 with complex combinatorial coefficients can easily
trigger exponential growth of the SCEV (in case if nothing gets folded and simplified).
We tried to restrain this transform using the option scalar-evolution-max-add-rec-size,
but its default value seems to be insufficiently small: the test attached to this patch
with default value of this option 16 has a SCEV of >3M symbols (when printed out).

This patch reduces the simplification limit. It is not a cure to combinatorial
explosions, but at least it reduces this corner case to something more or less
reasonable.

[sanjoy]

lgtm

test/Analysis/ScalarEvolution/binomial-explision.ll

19

Let's remove these undefs from the test case (to make it 100% obvious that the test has nothing to do with optimizing undefs).

lib/Analysis/ScalarEvolution.cpp

 static cl::opt<unsigned>
     MaxExtDepth("scalar-evolution-max-ext-depth", cl::Hidden,
                 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));
+                  cl::init(8));
 
 //===----------------------------------------------------------------------===//
 //                           SCEV class definitions
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 // Implementation of the SCEV class.
 //

test/Analysis/ScalarEvolution/binomial-explision.ll

+; RUN: opt -analyze -scalar-evolution < %s | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1"
+
+; Check that we don't have unreasonably huge SCEVs and in particular only a
+; reasonable amount of AddRecs in the notation of %tmp19. If we "simplify" SCEVs
+; too aggressively, we may end up with huge nested expressions.
+define void @test() {
+
+; CHECK: %tmp19 = mul i32 %tmp17, %tmp18
+; CHECK: ((((
+; CHECK-NOT: (((((
+; CHECK: %tmp20 = add i32 %tmp19, undef
+
+bb:
+  br label %bb1
+
+bb1:                                              ; preds = %bb3, %bb
+  %tmp = phi i64 [ undef, %bb ], [ %tmp22, %bb3 ]
+  %tmp2 = phi i32 [ undef, %bb ], [ %tmp4, %bb3 ]
+  br label %bb5
+
+bb3:                                              ; preds = %bb5
+  %tmp4 = add i32 %tmp2, undef
+  br label %bb1
+
+bb5:                                              ; preds = %bb5, %bb1
+  %tmp6 = phi i32 [ %tmp23, %bb5 ], [ %tmp2, %bb1 ]
+  %tmp7 = sub i32 -119, %tmp6
+  %tmp8 = mul i32 %tmp7, undef
+  %tmp9 = sub i32 -120, %tmp6
+  %tmp10 = mul i32 %tmp8, %tmp9
+  %tmp11 = mul i32 undef, %tmp10
+  %tmp12 = sub i32 -121, %tmp6
+  %tmp13 = mul i32 %tmp10, %tmp12
+  %tmp14 = mul i32 %tmp11, %tmp13
+  %tmp15 = sub i32 -122, %tmp6
+  %tmp16 = mul i32 %tmp13, %tmp15
+  %tmp17 = mul i32 %tmp14, %tmp16
+  %tmp18 = mul i32 %tmp16, undef
+  %tmp19 = mul i32 %tmp17, %tmp18
+  %tmp20 = add i32 %tmp19, undef
+  %tmp21 = sext i32 %tmp20 to i64
+  %tmp22 = add i64 undef, %tmp21
+  %tmp23 = add i32 %tmp6, 7
+  br i1 undef, label %bb5, label %bb3
+}