[InstCombine] Simplify ctpop with bitreverse/bswap

[xbolva00]

Summary

Fixes PR41337

[spatel]

LGTM

lib/Transforms/InstCombine/InstCombineCalls.cpp

static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombiner &IC) {
 
   return nullptr;
 }
 
 static Instruction *foldCtpop(IntrinsicInst &II, InstCombiner &IC) {
   assert(II.getIntrinsicID() == Intrinsic::ctpop &&
          "Expected ctpop intrinsic");
   Value *Op0 = II.getArgOperand(0);
+  Value *X;
+  // ctpop(bitreverse(x)) -> ctpop(x)
+  // ctpop(bswap(x)) -> ctpop(x)
+  if (match(Op0, m_BitReverse(m_Value(X))) || match(Op0, m_BSwap(m_Value(X)))) {
+    II.setOperand(0, X);
+    return ⅈ
+  }
+
   // FIXME: Try to simplify vectors of integers.
   auto *IT = dyn_cast<IntegerType>(Op0->getType());
   if (!IT)
     return nullptr;
 
   unsigned BitWidth = IT->getBitWidth();
   KnownBits Known(BitWidth);
   IC.computeKnownBits(Op0, Known, 0, &II);

test/Transforms/InstCombine/ctpop-bswap-bitreverse.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -instcombine -S < %s | FileCheck %s
 
 define i32 @ctpop_bitreverse(i32 %x) {
 ; CHECK-LABEL: @ctpop_bitreverse(
-; CHECK-NEXT:    [[A:%.*]] = tail call i32 @llvm.bitreverse.i32(i32 [[X:%.*]])
-; CHECK-NEXT:    [[B:%.*]] = tail call i32 @llvm.ctpop.i32(i32 [[A]]), !range !0
+; CHECK-NEXT:    [[B:%.*]] = tail call i32 @llvm.ctpop.i32(i32 [[X:%.*]]), !range !0
 ; CHECK-NEXT:    ret i32 [[B]]
 ;
   %a = tail call i32 @llvm.bitreverse.i32(i32 %x)
   %b = tail call i32 @llvm.ctpop.i32(i32 %a)
   ret i32 %b
 }
 
 define <2 x i64> @ctpop_bitreverse_vec(<2 x i64> %x) {
 ; CHECK-LABEL: @ctpop_bitreverse_vec(
-; CHECK-NEXT:    [[A:%.*]] = tail call <2 x i64> @llvm.bitreverse.v2i64(<2 x i64> [[X:%.*]])
-; CHECK-NEXT:    [[B:%.*]] = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> [[A]])
+; CHECK-NEXT:    [[B:%.*]] = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> [[X:%.*]])
 ; CHECK-NEXT:    ret <2 x i64> [[B]]
 ;
   %a = tail call <2 x i64> @llvm.bitreverse.v2i64(<2 x i64> %x)
   %b = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
   ret <2 x i64> %b
 }
 
 define i32 @ctpop_bswap(i32 %x) {
 ; CHECK-LABEL: @ctpop_bswap(
-; CHECK-NEXT:    [[A:%.*]] = tail call i32 @llvm.bswap.i32(i32 [[X:%.*]])
-; CHECK-NEXT:    [[B:%.*]] = tail call i32 @llvm.ctpop.i32(i32 [[A]]), !range !0
+; CHECK-NEXT:    [[B:%.*]] = tail call i32 @llvm.ctpop.i32(i32 [[X:%.*]]), !range !0
 ; CHECK-NEXT:    ret i32 [[B]]
 ;
   %a = tail call i32 @llvm.bswap.i32(i32 %x)
   %b = tail call i32 @llvm.ctpop.i32(i32 %a)
   ret i32 %b
 }
 
 define <2 x i64> @ctpop_bswap_vec(<2 x i64> %x) {
 ; CHECK-LABEL: @ctpop_bswap_vec(
-; CHECK-NEXT:    [[A:%.*]] = tail call <2 x i64> @llvm.bswap.v2i64(<2 x i64> [[X:%.*]])
-; CHECK-NEXT:    [[B:%.*]] = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> [[A]])
+; CHECK-NEXT:    [[B:%.*]] = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> [[X:%.*]])
 ; CHECK-NEXT:    ret <2 x i64> [[B]]
 ;
   %a = tail call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %x)
   %b = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
   ret <2 x i64> %b
 }
 
 declare i32 @llvm.bitreverse.i32(i32)
 declare <2 x i64> @llvm.bitreverse.v2i64(<2 x i64>)
 declare i32 @llvm.bswap.i32(i32)
 declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
 declare i32 @llvm.ctpop.i32(i32)
 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)