[InstCombine] add restrict reassoc for the powi(X,Y) / X

[vfdff]

add restrict reassoc for the powi(X,Y) / X according the discuss on PR69998

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Allen (vfdff)

Changes

add restrict reassoc for the powi(X,Y) / X according the discuss on PR69998

---

Full diff: https://github.com/llvm/llvm-project/pull/86428.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp (+14-14)
• (modified) llvm/test/Transforms/InstCombine/powi.ll (+14-3)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index 7be7923175e497..3e49395ad3703d 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -604,6 +604,18 @@ Instruction *InstCombinerImpl::foldPowiReassoc(BinaryOperator &I) {
       Y->getType() == Z->getType())
     return createPowiExpr(I, *this, X, Y, Z);
 
+  // powi(X, Y) / X --> powi(X, Y-1)
+  // This is legal when (Y - 1) can't wraparound, in which case reassoc and nnan
+  // are required.
+  // TODO: Multi-use may be also better off creating Powi(x,y-1)
+  if (I.hasAllowReassoc() && I.hasNoNaNs() &&
+      match(Op0, m_OneUse(m_AllowReassoc(m_Intrinsic<Intrinsic::powi>(
+                     m_Specific(Op1), m_Value(Y))))) &&
+      willNotOverflowSignedSub(Y, ConstantInt::get(Y->getType(), 1), I)) {
+    Constant *NegOne = ConstantInt::getAllOnesValue(Y->getType());
+    return createPowiExpr(I, *this, Op1, Y, NegOne);
+  }
+
   return nullptr;
 }
 
@@ -1886,20 +1898,8 @@ Instruction *InstCombinerImpl::visitFDiv(BinaryOperator &I) {
     return replaceInstUsesWith(I, Pow);
   }
 
-  // powi(X, Y) / X --> powi(X, Y-1)
-  // This is legal when (Y - 1) can't wraparound, in which case reassoc and nnan
-  // are required.
-  // TODO: Multi-use may be also better off creating Powi(x,y-1)
-  if (I.hasAllowReassoc() && I.hasNoNaNs() &&
-      match(Op0, m_OneUse(m_Intrinsic<Intrinsic::powi>(m_Specific(Op1),
-                                                       m_Value(Y)))) &&
-      willNotOverflowSignedSub(Y, ConstantInt::get(Y->getType(), 1), I)) {
-    Constant *NegOne = ConstantInt::getAllOnesValue(Y->getType());
-    Value *Y1 = Builder.CreateAdd(Y, NegOne);
-    Type *Types[] = {Op1->getType(), Y1->getType()};
-    Value *Pow = Builder.CreateIntrinsic(Intrinsic::powi, Types, {Op1, Y1}, &I);
-    return replaceInstUsesWith(I, Pow);
-  }
+  if (Instruction *FoldedPowi = foldPowiReassoc(I))
+    return FoldedPowi;
 
   return nullptr;
 }
diff --git a/llvm/test/Transforms/InstCombine/powi.ll b/llvm/test/Transforms/InstCombine/powi.ll
index 43e34c889106e1..6c0575e8b71971 100644
--- a/llvm/test/Transforms/InstCombine/powi.ll
+++ b/llvm/test/Transforms/InstCombine/powi.ll
@@ -313,7 +313,7 @@ define double @fdiv_pow_powi(double %x) {
 ; CHECK-NEXT:    [[DIV:%.*]] = fmul reassoc nnan double [[X:%.*]], [[X]]
 ; CHECK-NEXT:    ret double [[DIV]]
 ;
-  %p1 = call double @llvm.powi.f64.i32(double %x, i32 3)
+  %p1 = call reassoc double @llvm.powi.f64.i32(double %x, i32 3)
   %div = fdiv reassoc nnan double %p1, %x
   ret double %div
 }
@@ -323,7 +323,7 @@ define float @fdiv_powf_powi(float %x) {
 ; CHECK-NEXT:    [[DIV:%.*]] = call reassoc nnan float @llvm.powi.f32.i32(float [[X:%.*]], i32 99)
 ; CHECK-NEXT:    ret float [[DIV]]
 ;
-  %p1 = call float @llvm.powi.f32.i32(float %x, i32 100)
+  %p1 = call reassoc float @llvm.powi.f32.i32(float %x, i32 100)
   %div = fdiv reassoc nnan float %p1, %x
   ret float %div
 }
@@ -347,10 +347,21 @@ define double @fdiv_pow_powi_multi_use(double %x) {
 define float @fdiv_powf_powi_missing_reassoc(float %x) {
 ; CHECK-LABEL: @fdiv_powf_powi_missing_reassoc(
 ; CHECK-NEXT:    [[P1:%.*]] = call float @llvm.powi.f32.i32(float [[X:%.*]], i32 100)
-; CHECK-NEXT:    [[DIV:%.*]] = fdiv nnan float [[P1]], [[X]]
+; CHECK-NEXT:    [[DIV:%.*]] = fdiv reassoc nnan float [[P1]], [[X]]
 ; CHECK-NEXT:    ret float [[DIV]]
 ;
   %p1 = call float @llvm.powi.f32.i32(float %x, i32 100)
+  %div = fdiv reassoc nnan float %p1, %x
+  ret float %div
+}
+
+define float @fdiv_powf_powi_missing_reassoc1(float %x) {
+; CHECK-LABEL: @fdiv_powf_powi_missing_reassoc1(
+; CHECK-NEXT:    [[P1:%.*]] = call reassoc float @llvm.powi.f32.i32(float [[X:%.*]], i32 100)
+; CHECK-NEXT:    [[DIV:%.*]] = fdiv nnan float [[P1]], [[X]]
+; CHECK-NEXT:    ret float [[DIV]]
+;
+  %p1 = call reassoc float @llvm.powi.f32.i32(float %x, i32 100)
   %div = fdiv nnan float %p1, %x
   ret float %div
 }

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[dtcxzyw]

LGTM. Thanks!