[llvm] 7e4cf58 - [InstCombine] Add general constant support to eq/ne icmp(add(X,C1),add(Y,C2)) -> icmp(add(X,C1-C2),Y) fold

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-03-15T14:17:38Z
New Revision: 7e4cf582cffdc287a0d6ee354e90a4fbdc5a696d

URL: https://github.com/llvm/llvm-project/commit/7e4cf582cffdc287a0d6ee354e90a4fbdc5a696d
DIFF: https://github.com/llvm/llvm-project/commit/7e4cf582cffdc287a0d6ee354e90a4fbdc5a696d.diff

LOG: [InstCombine] Add general constant support to eq/ne icmp(add(X,C1),add(Y,C2)) -> icmp(add(X,C1-C2),Y) fold

A further extension for Issue #32161

For eq/ne comparisons - the sign mismatch and bounds constraints are redundant, so if the that fold fails, fallback and just fold the constants directly.

https://alive2.llvm.org/ce/z/cdodNQ

The loop rotation test change looks mostly benign - the backend doesn't seem to suffer? https://gcc.godbolt.org/z/dErMY78To

Differential Revision: https://reviews.llvm.org/D121551

Added: 
    

Modified: 
    llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
    llvm/test/Transforms/InstCombine/icmp.ll
    llvm/test/Transforms/PhaseOrdering/loop-rotation-vs-common-code-hoisting.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 651d564763074..2d2a998b750fe 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4134,6 +4134,13 @@ Instruction *InstCombinerImpl::foldICmpBinOp(ICmpInst &I,
         return new ICmpInst(Pred, A, NewAdd);
       }
     }
+    Constant *Cst1, *Cst2;
+    if (match(B, m_ImmConstant(Cst1)) && match(D, m_ImmConstant(Cst2)) &&
+        ICmpInst::isEquality(Pred)) {
+      Constant *Diff = ConstantExpr::getSub(Cst2, Cst1);
+      Value *NewAdd = Builder.CreateAdd(C, Diff);
+      return new ICmpInst(Pred, A, NewAdd);
+    }
   }
 
   // Analyze the case when either Op0 or Op1 is a sub instruction.

diff  --git a/llvm/test/Transforms/InstCombine/icmp.ll b/llvm/test/Transforms/InstCombine/icmp.ll
index fe47d040b76d2..30da87fbd4f08 100644
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -1565,9 +1565,8 @@ define <2 x i1> @icmp_add20_eq_add57_undef(<2 x i32> %x, <2 x i32> %y) {
 
 define <2 x i1> @icmp_add20_eq_add57_vec_nonsplat(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_add20_eq_add57_vec_nonsplat(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 20, i32 19>
-; CHECK-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[Y:%.*]], <i32 57, i32 58>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[Y:%.*]], <i32 37, i32 39>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <2 x i32> [[TMP1]], [[X:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add <2 x i32> %x, <i32 20, i32 19>
@@ -1614,9 +1613,8 @@ define <2 x i1> @icmp_sub57_ne_sub20_vec_undef(<2 x i32> %x, <2 x i32> %y) {
 
 define <2 x i1> @icmp_sub57_ne_sub20_vec_nonsplat(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_sub57_ne_sub20_vec_nonsplat(
-; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -57, i32 -58>
-; CHECK-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[Y:%.*]], <i32 -20, i32 -21>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ne <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[Y:%.*]], <i32 37, i32 37>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne <2 x i32> [[TMP1]], [[X:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add <2 x i32> %x, <i32 -57, i32 -58>

diff  --git a/llvm/test/Transforms/PhaseOrdering/loop-rotation-vs-common-code-hoisting.ll b/llvm/test/Transforms/PhaseOrdering/loop-rotation-vs-common-code-hoisting.ll
index 561c2fc399037..97d11283ed46b 100644
--- a/llvm/test/Transforms/PhaseOrdering/loop-rotation-vs-common-code-hoisting.ll
+++ b/llvm/test/Transforms/PhaseOrdering/loop-rotation-vs-common-code-hoisting.ll
@@ -76,19 +76,21 @@ define void @_Z4loopi(i32 %width) {
 ; ROTATE-NEXT:    [[CMP:%.*]] = icmp slt i32 [[WIDTH:%.*]], 1
 ; ROTATE-NEXT:    br i1 [[CMP]], label [[RETURN:%.*]], label [[FOR_COND_PREHEADER:%.*]]
 ; ROTATE:       for.cond.preheader:
-; ROTATE-NEXT:    [[SUB:%.*]] = add nsw i32 [[WIDTH]], -1
 ; ROTATE-NEXT:    [[CMP13_NOT:%.*]] = icmp eq i32 [[WIDTH]], 1
-; ROTATE-NEXT:    br i1 [[CMP13_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY:%.*]]
+; ROTATE-NEXT:    br i1 [[CMP13_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
+; ROTATE:       for.body.preheader:
+; ROTATE-NEXT:    [[TMP0:%.*]] = add i32 [[WIDTH]], -2
+; ROTATE-NEXT:    br label [[FOR_BODY:%.*]]
 ; ROTATE:       for.cond.cleanup:
 ; ROTATE-NEXT:    tail call void @f0()
 ; ROTATE-NEXT:    tail call void @f2()
 ; ROTATE-NEXT:    br label [[RETURN]]
 ; ROTATE:       for.body:
-; ROTATE-NEXT:    [[I_04:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_COND_PREHEADER]] ]
+; ROTATE-NEXT:    [[I_04:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
 ; ROTATE-NEXT:    tail call void @f0()
 ; ROTATE-NEXT:    tail call void @f1()
 ; ROTATE-NEXT:    [[INC]] = add nuw nsw i32 [[I_04]], 1
-; ROTATE-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[SUB]]
+; ROTATE-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[I_04]], [[TMP0]]
 ; ROTATE-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]]
 ; ROTATE:       return:
 ; ROTATE-NEXT:    ret void