From 534e635d1b9c3cc2815a3b38748c1d584dda2630 Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Fri, 13 Jan 2017 23:04:10 +0000
Subject: [PATCH] [InstCombine] use m_APInt to allow lshr folds for vectors
 with splat constants

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291972 91177308-0d34-0410-b5e6-96231b3b80d8
---
 .../InstCombine/InstCombineShifts.cpp         | 31 +++++++++----------
 test/Transforms/InstCombine/lshr.ll           | 14 ++++-----
 2 files changed, 21 insertions(+), 24 deletions(-)

diff --git a/lib/Transforms/InstCombine/InstCombineShifts.cpp b/lib/Transforms/InstCombine/InstCombineShifts.cpp
index 5209dad31b5..078c938ff9a 100644
--- a/lib/Transforms/InstCombine/InstCombineShifts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineShifts.cpp
@@ -774,34 +774,31 @@ Instruction *InstCombiner::visitLShr(BinaryOperator &I) {
   if (Instruction *R = commonShiftTransforms(I))
     return R;
 
-  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
-    unsigned ShAmt = Op1C->getZExtValue();
-
-    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Op0)) {
-      unsigned BitWidth = Op0->getType()->getScalarSizeInBits();
+  const APInt *ShAmtAPInt;
+  if (match(Op1, m_APInt(ShAmtAPInt))) {
+    unsigned ShAmt = ShAmtAPInt->getZExtValue();
+    unsigned BitWidth = Op0->getType()->getScalarSizeInBits();
+    auto *II = dyn_cast<IntrinsicInst>(Op0);
+    if (II && isPowerOf2_32(BitWidth) && Log2_32(BitWidth) == ShAmt &&
+        (II->getIntrinsicID() == Intrinsic::ctlz ||
+         II->getIntrinsicID() == Intrinsic::cttz ||
+         II->getIntrinsicID() == Intrinsic::ctpop)) {
       // ctlz.i32(x)>>5  --> zext(x == 0)
       // cttz.i32(x)>>5  --> zext(x == 0)
       // ctpop.i32(x)>>5 --> zext(x == -1)
-      if ((II->getIntrinsicID() == Intrinsic::ctlz ||
-           II->getIntrinsicID() == Intrinsic::cttz ||
-           II->getIntrinsicID() == Intrinsic::ctpop) &&
-          isPowerOf2_32(BitWidth) && Log2_32(BitWidth) == ShAmt) {
-        bool isCtPop = II->getIntrinsicID() == Intrinsic::ctpop;
-        Constant *RHS = ConstantInt::getSigned(Op0->getType(), isCtPop ? -1:0);
-        Value *Cmp = Builder->CreateICmpEQ(II->getArgOperand(0), RHS);
-        return new ZExtInst(Cmp, II->getType());
-      }
+      bool IsPop = II->getIntrinsicID() == Intrinsic::ctpop;
+      Constant *RHS = ConstantInt::getSigned(Op0->getType(), IsPop ? -1 : 0);
+      Value *Cmp = Builder->CreateICmpEQ(II->getArgOperand(0), RHS);
+      return new ZExtInst(Cmp, II->getType());
     }
 
     // If the shifted-out value is known-zero, then this is an exact shift.
     if (!I.isExact() &&
-        MaskedValueIsZero(Op0, APInt::getLowBitsSet(Op1C->getBitWidth(), ShAmt),
-                          0, &I)){
+        MaskedValueIsZero(Op0, APInt::getLowBitsSet(BitWidth, ShAmt), 0, &I)) {
       I.setIsExact();
       return &I;
     }
   }
-
   return nullptr;
 }
 
diff --git a/test/Transforms/InstCombine/lshr.ll b/test/Transforms/InstCombine/lshr.ll
index 13aa5e7dc09..b81371b0304 100644
--- a/test/Transforms/InstCombine/lshr.ll
+++ b/test/Transforms/InstCombine/lshr.ll
@@ -43,8 +43,8 @@ define i32 @lshr_ctpop(i32 %x) {
 
 define <2 x i8> @lshr_ctlz_zero_is_not_undef_splat_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @lshr_ctlz_zero_is_not_undef_splat_vec(
-; CHECK-NEXT:    [[CT:%.*]] = call <2 x i8> @llvm.ctlz.v2i8(<2 x i8> %x, i1 false)
-; CHECK-NEXT:    [[SH:%.*]] = lshr <2 x i8> [[CT]], <i8 3, i8 3>
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq <2 x i8> %x, zeroinitializer
+; CHECK-NEXT:    [[SH:%.*]] = zext <2 x i1> [[TMP1]] to <2 x i8>
 ; CHECK-NEXT:    ret <2 x i8> [[SH]]
 ;
   %ct = call <2 x i8> @llvm.ctlz.v2i8(<2 x i8> %x, i1 false)
@@ -54,8 +54,8 @@ define <2 x i8> @lshr_ctlz_zero_is_not_undef_splat_vec(<2 x i8> %x) {
 
 define <2 x i8> @lshr_cttz_zero_is_not_undef_splat_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @lshr_cttz_zero_is_not_undef_splat_vec(
-; CHECK-NEXT:    [[CT:%.*]] = call <2 x i8> @llvm.cttz.v2i8(<2 x i8> %x, i1 false)
-; CHECK-NEXT:    [[SH:%.*]] = lshr <2 x i8> [[CT]], <i8 3, i8 3>
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq <2 x i8> %x, zeroinitializer
+; CHECK-NEXT:    [[SH:%.*]] = zext <2 x i1> [[TMP1]] to <2 x i8>
 ; CHECK-NEXT:    ret <2 x i8> [[SH]]
 ;
   %ct = call <2 x i8> @llvm.cttz.v2i8(<2 x i8> %x, i1 false)
@@ -65,8 +65,8 @@ define <2 x i8> @lshr_cttz_zero_is_not_undef_splat_vec(<2 x i8> %x) {
 
 define <2 x i8> @lshr_ctpop_splat_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @lshr_ctpop_splat_vec(
-; CHECK-NEXT:    [[CT:%.*]] = call <2 x i8> @llvm.ctpop.v2i8(<2 x i8> %x)
-; CHECK-NEXT:    [[SH:%.*]] = lshr <2 x i8> [[CT]], <i8 3, i8 3>
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq <2 x i8> %x, <i8 -1, i8 -1>
+; CHECK-NEXT:    [[SH:%.*]] = zext <2 x i1> [[TMP1]] to <2 x i8>
 ; CHECK-NEXT:    ret <2 x i8> [[SH]]
 ;
   %ct = call <2 x i8> @llvm.ctpop.v2i8(<2 x i8> %x)
@@ -91,7 +91,7 @@ define <2 x i8> @lshr_exact_splat_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @lshr_exact_splat_vec(
 ; CHECK-NEXT:    [[SHL:%.*]] = shl <2 x i8> %x, <i8 2, i8 2>
 ; CHECK-NEXT:    [[ADD:%.*]] = add <2 x i8> [[SHL]], <i8 4, i8 4>
-; CHECK-NEXT:    [[LSHR:%.*]] = lshr <2 x i8> [[ADD]], <i8 2, i8 2>
+; CHECK-NEXT:    [[LSHR:%.*]] = lshr exact <2 x i8> [[ADD]], <i8 2, i8 2>
 ; CHECK-NEXT:    ret <2 x i8> [[LSHR]]
 ;
   %shl = shl <2 x i8> %x, <i8 2, i8 2>
-- 
2.40.0