Limiting gep merging to fix the performance problem described in

https://llvm.org/bugs/show_bug.cgi?id=23163.

Gep merging sometimes behaves like a reverse CSE/LICM optimization,
which has negative impact on performance. In this patch we restrict
gep merging to happen only when the indexes to be merged are both consts,
which ensures such merge is always beneficial.

The patch makes gep merging only happen in very restrictive cases.
It is possible that some analysis/optimization passes rely on the merged
geps to get better result, and we havn't notice them yet. We will be ready
to further improve it once we see the cases.

Differential Revision: http://reviews.llvm.org/D8911

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@235455 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
index 3b46156..f62941f 100644 (file)
--- a/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1467,6 +1467,11 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
         // normalized.
         if (SO1->getType() != GO1->getType())
           return nullptr;
+        // Only do the combine when GO1 and SO1 are both constants. Only in
+        // this case, we are sure the cost after the merge is never more than
+        // that before the merge.
+        if (!isa<Constant>(GO1) || !isa<Constant>(SO1))
+          return nullptr;
         Sum = Builder->CreateAdd(SO1, GO1, PtrOp->getName()+".sum");
       }
 

diff --git a/test/Transforms/InstCombine/descale-zero.ll b/test/Transforms/InstCombine/descale-zero.ll
index 4347be4..e69de29 100644 (file)
--- a/test/Transforms/InstCombine/descale-zero.ll
+++ b/test/Transforms/InstCombine/descale-zero.ll
@@ -1,20 +0,0 @@
-; RUN: opt < %s -instcombine -S | FileCheck %s
-
-target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64-apple-macosx10.10.0"
-
-define internal i8* @descale_zero() {
-entry:
-; CHECK: load i8*, i8** inttoptr (i64 48 to i8**), align 16
-; CHECK-NEXT: ret i8*
-  %i16_ptr = load i16*, i16** inttoptr (i64 48 to i16**), align 16
-  %num = load i64, i64* inttoptr (i64 64 to i64*), align 64
-  %num_times_2 = shl i64 %num, 1
-  %num_times_2_plus_4 = add i64 %num_times_2, 4
-  %i8_ptr = bitcast i16* %i16_ptr to i8*
-  %i8_ptr_num_times_2_plus_4 = getelementptr i8, i8* %i8_ptr, i64 %num_times_2_plus_4
-  %num_times_neg2 = mul i64 %num, -2
-  %num_times_neg2_minus_4 = add i64 %num_times_neg2, -4
-  %addr = getelementptr i8, i8* %i8_ptr_num_times_2_plus_4, i64 %num_times_neg2_minus_4
-  ret i8* %addr
-}

diff --git a/test/Transforms/InstCombine/getelementptr.ll b/test/Transforms/InstCombine/getelementptr.ll
index 9bb1b12..276ada9 100644 (file)
--- a/test/Transforms/InstCombine/getelementptr.ll
+++ b/test/Transforms/InstCombine/getelementptr.ll
@@ -104,8 +104,8 @@ define i32* @test7(i32* %I, i64 %C, i64 %D) {
         %B = getelementptr i32, i32* %A, i64 %D
         ret i32* %B
 ; CHECK-LABEL: @test7(
-; CHECK: %A.sum = add i64 %C, %D
-; CHECK: getelementptr i32, i32* %I, i64 %A.sum
+; CHECK: %A = getelementptr i32, i32* %I, i64 %C
+; CHECK: %B = getelementptr i32, i32* %A, i64 %D
 }
 
 define i8* @test8([10 x i32]* %X) {