[PATCH] D132996: [InstCombine] complete fold extractvalue (any_mul_with_overflow X, -1)

[Chenbing.Zheng via Phabricator via llvm-commits]

Chenbing.Zheng updated this revision to Diff 457162.
Chenbing.Zheng edited the summary of this revision.
Chenbing.Zheng added a comment.

add alive2 in summary


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D132996/new/

https://reviews.llvm.org/D132996

Files:
  llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
  llvm/test/Transforms/InstCombine/with_overflow.ll


Index: llvm/test/Transforms/InstCombine/with_overflow.ll
===================================================================
--- llvm/test/Transforms/InstCombine/with_overflow.ll
+++ llvm/test/Transforms/InstCombine/with_overflow.ll
@@ -793,13 +793,10 @@
   ret <4 x i8> %r
 }
 
-; TODO: partly failed to match vector constant with poison element
-
 define <4 x i8> @smul_neg1_vec_poison(<4 x i8> %x, <4 x i1>* %p) {
 ; CHECK-LABEL: @smul_neg1_vec_poison(
-; CHECK-NEXT:    [[M:%.*]] = tail call { <4 x i8>, <4 x i1> } @llvm.smul.with.overflow.v4i8(<4 x i8> [[X:%.*]], <4 x i8> <i8 -1, i8 -1, i8 poison, i8 -1>)
-; CHECK-NEXT:    [[R:%.*]] = sub <4 x i8> zeroinitializer, [[X]]
-; CHECK-NEXT:    [[OV:%.*]] = extractvalue { <4 x i8>, <4 x i1> } [[M]], 1
+; CHECK-NEXT:    [[R:%.*]] = sub <4 x i8> zeroinitializer, [[X:%.*]]
+; CHECK-NEXT:    [[OV:%.*]] = icmp eq <4 x i8> [[X]], <i8 -128, i8 -128, i8 -128, i8 -128>
 ; CHECK-NEXT:    store <4 x i1> [[OV]], <4 x i1>* [[P:%.*]], align 1
 ; CHECK-NEXT:    ret <4 x i8> [[R]]
 ;
@@ -853,13 +850,10 @@
   ret <4 x i8> %r
 }
 
-; TODO: partly failed to match vector constant with poison element
-
 define <4 x i8> @umul_neg1_vec_poison(<4 x i8> %x, <4 x i1>* %p) {
 ; CHECK-LABEL: @umul_neg1_vec_poison(
-; CHECK-NEXT:    [[M:%.*]] = tail call { <4 x i8>, <4 x i1> } @llvm.umul.with.overflow.v4i8(<4 x i8> [[X:%.*]], <4 x i8> <i8 poison, i8 -1, i8 -1, i8 poison>)
-; CHECK-NEXT:    [[R:%.*]] = sub <4 x i8> zeroinitializer, [[X]]
-; CHECK-NEXT:    [[OV:%.*]] = extractvalue { <4 x i8>, <4 x i1> } [[M]], 1
+; CHECK-NEXT:    [[R:%.*]] = sub <4 x i8> zeroinitializer, [[X:%.*]]
+; CHECK-NEXT:    [[OV:%.*]] = icmp ugt <4 x i8> [[X]], <i8 1, i8 1, i8 1, i8 1>
 ; CHECK-NEXT:    store <4 x i1> [[OV]], <4 x i1>* [[P:%.*]], align 1
 ; CHECK-NEXT:    ret <4 x i8> [[R]]
 ;
Index: llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -3251,11 +3251,14 @@
 
   // extractvalue (any_mul_with_overflow X, -1), 0 --> -X
   Intrinsic::ID OvID = WO->getIntrinsicID();
-  if (*EV.idx_begin() == 0 &&
-      (OvID == Intrinsic::smul_with_overflow ||
-       OvID == Intrinsic::umul_with_overflow) &&
-      match(WO->getArgOperand(1), m_AllOnes())) {
-    return BinaryOperator::CreateNeg(WO->getArgOperand(0));
+  const APInt *C = nullptr;
+  if (match(WO->getRHS(), m_APIntAllowUndef(C))) {
+    if (*EV.idx_begin() == 0 &&
+        (OvID == Intrinsic::smul_with_overflow ||
+         OvID == Intrinsic::umul_with_overflow) &&
+        C->isAllOnes()) {
+      return BinaryOperator::CreateNeg(WO->getArgOperand(0));
+    }
   }
 
   // We're extracting from an overflow intrinsic. See if we're the only user.
@@ -3284,8 +3287,7 @@
   // If only the overflow result is used, and the right hand side is a
   // constant (or constant splat), we can remove the intrinsic by directly
   // checking for overflow.
-  const APInt *C;
-  if (match(WO->getRHS(), m_APInt(C))) {
+  if (C) {
     // Compute the no-wrap range for LHS given RHS=C, then construct an
     // equivalent icmp, potentially using an offset.
     ConstantRange NWR = ConstantRange::makeExactNoWrapRegion(


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D132996.457162.patch
Type: text/x-patch
Size: 3301 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220901/e98d5448/attachment.bin>