From f245111149741e9c4e155f0fbf4a5e809f8231ed Mon Sep 17 00:00:00 2001
From: goldsteinn <35538541+goldsteinn@users.noreply.github.com>
Date: Sat, 11 Jan 2025 15:10:42 -0600
Subject: [PATCH] [InstSimpify] Simplifying `(xor (sub C_Mask, X), C_Mask)` ->
 `X` (#122552)

- **[InstSimpify] Add tests for simplifying `(xor (sub C_Mask, X),
C_Mask)`; NFC**
- **[InstSimpify] Simplifying `(xor (sub C_Mask, X), C_Mask)` -> `X`**

Helps address regressions with folding `clz(Pow2)`.

Proof: https://alive2.llvm.org/ce/z/zGwUBp
---
 llvm/lib/Analysis/InstructionSimplify.cpp     |  16 +++
 .../InstSimplify/subnuw-with-xor.ll           | 118 ++++++++++++++++++
 2 files changed, 134 insertions(+)
 create mode 100644 llvm/test/Transforms/InstSimplify/subnuw-with-xor.ll

diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 999386c0a04917e..d69747e30f884d7 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -871,6 +871,14 @@ static Value *simplifySubInst(Value *Op0, Value *Op1, bool IsNSW, bool IsNUW,
   if (Value *V = simplifyByDomEq(Instruction::Sub, Op0, Op1, Q, MaxRecurse))
     return V;
 
+  // (sub nuw C_Mask, (xor X, C_Mask)) -> X
+  if (IsNUW) {
+    Value *X;
+    if (match(Op1, m_Xor(m_Value(X), m_Specific(Op0))) &&
+        match(Op0, m_LowBitMask()))
+      return X;
+  }
+
   return nullptr;
 }
 
@@ -2540,6 +2548,14 @@ static Value *simplifyXorInst(Value *Op0, Value *Op1, const SimplifyQuery &Q,
   if (Value *V = simplifyByDomEq(Instruction::Xor, Op0, Op1, Q, MaxRecurse))
     return V;
 
+  // (xor (sub nuw C_Mask, X), C_Mask) -> X
+  {
+    Value *X;
+    if (match(Op0, m_NUWSub(m_Specific(Op1), m_Value(X))) &&
+        match(Op1, m_LowBitMask()))
+      return X;
+  }
+
   return nullptr;
 }
 
diff --git a/llvm/test/Transforms/InstSimplify/subnuw-with-xor.ll b/llvm/test/Transforms/InstSimplify/subnuw-with-xor.ll
new file mode 100644
index 000000000000000..a990e2f2ae394e8
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/subnuw-with-xor.ll
@@ -0,0 +1,118 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+
+define i8 @xor_w_sub_fail_missing_nuw(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @xor_w_sub_fail_missing_nuw(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[XOR:%.*]] = xor i8 [[X]], 15
+; CHECK-NEXT:    [[R:%.*]] = sub nsw i8 15, [[XOR]]
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %xor = xor i8 %x, 15
+  %r = sub nsw i8 15, %xor
+  ret i8 %r
+}
+
+define i8 @xor_w_sub_fail_diff_values(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @xor_w_sub_fail_diff_values(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[XOR:%.*]] = xor i8 [[X]], 15
+; CHECK-NEXT:    [[R:%.*]] = sub nuw nsw i8 31, [[XOR]]
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %xor = xor i8 %x, 15
+  %r = sub nsw nuw i8 31, %xor
+  ret i8 %r
+}
+
+define i8 @xor_w_sub_fail_diff_values2(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @xor_w_sub_fail_diff_values2(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[XOR:%.*]] = xor i8 [[X]], 31
+; CHECK-NEXT:    [[R:%.*]] = sub nuw nsw i8 15, [[XOR]]
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %xor = xor i8 %x, 31
+  %r = sub nsw nuw i8 15, %xor
+  ret i8 %r
+}
+
+define i8 @xor_w_sub_fail_not_mask(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @xor_w_sub_fail_not_mask(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[XOR:%.*]] = xor i8 [[X]], 30
+; CHECK-NEXT:    [[R:%.*]] = sub nuw nsw i8 30, [[XOR]]
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %xor = xor i8 %x, 30
+  %r = sub nsw nuw i8 30, %xor
+  ret i8 %r
+}
+
+define i8 @xor_w_sub_okay(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @xor_w_sub_okay(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    ret i8 [[X]]
+;
+  %xor = xor i8 %x, 31
+  %r = sub nsw nuw i8 31, %xor
+  ret i8 %r
+}
+
+define i8 @sub_w_xor_fail_missing_nuw(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @sub_w_xor_fail_missing_nuw(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[SUB:%.*]] = sub nsw i8 15, [[X]]
+; CHECK-NEXT:    [[R:%.*]] = xor i8 [[SUB]], 15
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %sub = sub nsw i8 15, %x
+  %r = xor i8 %sub, 15
+  ret i8 %r
+}
+
+define i8 @sub_w_xor_fail_diff_values(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @sub_w_xor_fail_diff_values(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[SUB:%.*]] = sub nuw nsw i8 15, [[X]]
+; CHECK-NEXT:    [[R:%.*]] = xor i8 [[SUB]], 31
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %sub = sub nsw nuw i8 15, %x
+  %r = xor i8 %sub, 31
+  ret i8 %r
+}
+
+define i8 @sub_w_sub_fail_diff_values2(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @sub_w_sub_fail_diff_values2(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[SUB:%.*]] = sub nuw nsw i8 31, [[X]]
+; CHECK-NEXT:    [[R:%.*]] = xor i8 [[SUB]], 15
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %sub = sub nsw nuw i8 31, %x
+  %r = xor i8 %sub, 15
+  ret i8 %r
+}
+
+define i8 @sub_w_sub_fail_not_mask(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @sub_w_sub_fail_not_mask(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    [[SUB:%.*]] = sub nuw nsw i8 30, [[X]]
+; CHECK-NEXT:    [[R:%.*]] = xor i8 [[SUB]], 30
+; CHECK-NEXT:    ret i8 [[R]]
+;
+  %sub = sub nsw nuw i8 30, %x
+  %r = xor i8 %sub, 30
+  ret i8 %r
+}
+
+define i8 @sub_w_sub_okay(i8 range(i8 0, 16) %x) {
+; CHECK-LABEL: define i8 @sub_w_sub_okay(
+; CHECK-SAME: i8 range(i8 0, 16) [[X:%.*]]) {
+; CHECK-NEXT:    ret i8 [[X]]
+;
+  %sub = sub nsw nuw i8 31, %x
+  %r = xor i8 %sub, 31
+  ret i8 %r
+}