[llvm] r306243 - [InstCombine] add (sext i1 X), 1 --> zext (not X)

Sanjay Patel via llvm-commits llvm-commits at lists.llvm.org
Sun Jun 25 07:15:29 PDT 2017


Author: spatel
Date: Sun Jun 25 07:15:28 2017
New Revision: 306243

URL: http://llvm.org/viewvc/llvm-project?rev=306243&view=rev
Log:
[InstCombine] add (sext i1 X), 1 --> zext (not X)

http://rise4fun.com/Alive/i8Q

A narrow bitwise logic op is obviously better than math for value tracking, 
and zext is better than sext. Typically, the 'not' will be folded into an 
icmp predicate.

The IR difference would even survive through codegen for x86, so we would see 
worse code:

https://godbolt.org/g/C14HMF

one_or_zero(int, int):                      # @one_or_zero(int, int)
        xorl    %eax, %eax
        cmpl    %esi, %edi
        setle   %al
        retq

one_or_zero_alt(int, int):                  # @one_or_zero_alt(int, int)
        xorl    %ecx, %ecx
        cmpl    %esi, %edi
        setg    %cl
        movl    $1, %eax
        subl    %ecx, %eax
        retq



Modified:
    llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp
    llvm/trunk/test/Transforms/InstCombine/add.ll

Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp?rev=306243&r1=306242&r2=306243&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp Sun Jun 25 07:15:28 2017
@@ -988,15 +988,24 @@ static Instruction *foldAddWithConstant(
     return new ZExtInst(Builder.CreateNUWAdd(X, NewC), Ty);
   }
 
-  // Shifts and add used to flip and mask off the low bit:
-  // add (ashr (shl i32 X, 31), 31), 1 --> and (not X), 1
-  const APInt *C3;
-  if (C->isOneValue() &&
-      match(Op0,
-            m_OneUse(m_AShr(m_Shl(m_Value(X), m_APInt(C2)), m_APInt(C3)))) &&
-      C2 == C3 && *C2 == Ty->getScalarSizeInBits() - 1) {
-    Value *NotX = Builder.CreateNot(X);
-    return BinaryOperator::CreateAnd(NotX, ConstantInt::get(Ty, 1));
+  if (C->isOneValue() && Op0->hasOneUse()) {
+    // add (sext i1 X), 1 --> zext (not X)
+    // TODO: The smallest IR representation is (select X, 0, 1), and that would
+    // not require the one-use check. But we need to remove a transform in
+    // visitSelect and make sure that IR value tracking for select is equal or
+    // better than for these ops.
+    if (match(Op0, m_SExt(m_Value(X))) &&
+        X->getType()->getScalarSizeInBits() == 1)
+      return new ZExtInst(Builder.CreateNot(X), Ty);
+
+    // Shifts and add used to flip and mask off the low bit:
+    // add (ashr (shl i32 X, 31), 31), 1 --> and (not X), 1
+    const APInt *C3;
+    if (match(Op0, m_AShr(m_Shl(m_Value(X), m_APInt(C2)), m_APInt(C3))) &&
+        C2 == C3 && *C2 == Ty->getScalarSizeInBits() - 1) {
+      Value *NotX = Builder.CreateNot(X);
+      return BinaryOperator::CreateAnd(NotX, ConstantInt::get(Ty, 1));
+    }
   }
 
   return nullptr;

Modified: llvm/trunk/test/Transforms/InstCombine/add.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/add.ll?rev=306243&r1=306242&r2=306243&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/add.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/add.ll Sun Jun 25 07:15:28 2017
@@ -1,12 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-; TODO: This should be canonicalized to either a select or xor+zext.
-
 define i32 @select_0_or_1_from_bool(i1 %x) {
 ; CHECK-LABEL: @select_0_or_1_from_bool(
-; CHECK-NEXT:    [[EXT:%.*]] = sext i1 %x to i32
-; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[EXT]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = xor i1 %x, true
+; CHECK-NEXT:    [[ADD:%.*]] = zext i1 [[TMP1]] to i32
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
   %ext = sext i1 %x to i32
@@ -14,12 +12,10 @@ define i32 @select_0_or_1_from_bool(i1 %
   ret i32 %add
 }
 
-; TODO: This should be canonicalized to either a select or xor+zext.
-
 define <2 x i32> @select_0_or_1_from_bool_vec(<2 x i1> %x) {
 ; CHECK-LABEL: @select_0_or_1_from_bool_vec(
-; CHECK-NEXT:    [[EXT:%.*]] = sext <2 x i1> %x to <2 x i32>
-; CHECK-NEXT:    [[ADD:%.*]] = add nsw <2 x i32> [[EXT]], <i32 1, i32 1>
+; CHECK-NEXT:    [[TMP1:%.*]] = xor <2 x i1> %x, <i1 true, i1 true>
+; CHECK-NEXT:    [[ADD:%.*]] = zext <2 x i1> [[TMP1]] to <2 x i32>
 ; CHECK-NEXT:    ret <2 x i32> [[ADD]]
 ;
   %ext = sext <2 x i1> %x to <2 x i32>




More information about the llvm-commits mailing list