[X86] The TEST instruction is eliminated when BSF/TZCNT is used

Summary:
These changes cover the PR#31399.
Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0
and it corresponds to the following llvm code:
  %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
  %tobool = icmp eq i32 %v, 0
  %.op = add nuw nsw i32 %cnt, 1
  %add = select i1 %tobool, i32 0, i32 %.op
and x86 asm code:
  bsfl     %edi, %ecx
  addl     $1, %ecx
  testl    %edi, %edi
  movl     $0, %eax
  cmovnel  %ecx, %eax
In this case the 'test' instruction can't be eliminated because
the 'add' instruction modifies the EFLAGS, namely, ZF flag
that is set by the 'bsf' instruction when 'x' is zero.

We now produce the following code:
  bsfl     %edi, %ecx
  movl     $-1, %eax
  cmovnel  %ecx, %eax
  addl     $1, %eax

Patch by Ivan Kulagin

Reviewers: davide, craig.topper, spatel, RKSimon

Reviewed By: craig.topper

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D48765

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@336768 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/X86/X86ISelLowering.cpp

@@ -33452,6 +33452,36 @@ static SDValue combineCMov(SDNode *N, SelectionDAG &DAG,
     }
   }
 
+  // Handle (CMOV (ADD (CTTZ X), C), C-1, (X != 0)) ->
+  // (ADD (CMOV (CTTZ X), -1, (X != 0)), C) or
+  // (CMOV C-1, (ADD (CTTZ X), C), (X == 0)) ->
+  // (ADD (CMOV C-1, (CTTZ X), (X == 0)), C)
+  if (CC == X86::COND_NE || CC == X86::COND_E) {
+    auto *Cnst = CC == X86::COND_E ? dyn_cast<ConstantSDNode>(TrueOp)
+                                   : dyn_cast<ConstantSDNode>(FalseOp);
+    SDValue Add = CC == X86::COND_E ? FalseOp : TrueOp;
+
+    if (Cnst && Add.getOpcode() == ISD::ADD && Add.hasOneUse()) {
+      auto *AddOp1 = dyn_cast<ConstantSDNode>(Add.getOperand(1));
+      SDValue AddOp2 = Add.getOperand(0);
+      if (AddOp1 && (AddOp2.getOpcode() == ISD::CTTZ_ZERO_UNDEF ||
+                     AddOp2.getOpcode() == ISD::CTTZ)) {
+        APInt Diff = Cnst->getAPIntValue() - AddOp1->getAPIntValue();
+        if (CC == X86::COND_NE) {
+          Add = DAG.getNode(X86ISD::CMOV, DL, Add.getValueType(), AddOp2,
+                            DAG.getConstant(Diff, DL, Add.getValueType()),
+                            DAG.getConstant(CC, DL, MVT::i8), Cond);
+        } else {
+          Add = DAG.getNode(X86ISD::CMOV, DL, Add.getValueType(),
+                            DAG.getConstant(Diff, DL, Add.getValueType()),
+                            AddOp2, DAG.getConstant(CC, DL, MVT::i8), Cond);
+        }
+        return DAG.getNode(X86ISD::ADD, DL, Add.getValueType(), Add,
+                           SDValue(AddOp1, 0));
+      }
+    }
+  }
+
   return SDValue();
 }
 

lib/Target/X86/X86InstrInfo.cpp

@@ -3591,6 +3591,13 @@ static X86::CondCode isUseDefConvertible(MachineInstr &MI) {
   case X86::TZCNT32rr: case X86::TZCNT32rm:
   case X86::TZCNT64rr: case X86::TZCNT64rm:
     return X86::COND_B;
+  case X86::BSF16rr:
+  case X86::BSF16rm:
+  case X86::BSF32rr:
+  case X86::BSF32rm:
+  case X86::BSF64rr:
+  case X86::BSF64rm:
+    return X86::COND_E;
   }
 }
 

test/CodeGen/X86/dagcombine-select.ll

@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-unknown -verify-machineinstrs < %s | FileCheck -enable-var-scope %s
+; RUN: llc -mtriple=x86_64-unknown-unknown -verify-machineinstrs -mattr=+bmi < %s | FileCheck -check-prefix=BMI -enable-var-scope %s
 
 define i32 @select_and1(i32 %x, i32 %y) {
 ; CHECK-LABEL: select_and1:
@@ -279,3 +280,97 @@ define double @frem_constant_sel_constants(i1 %cond) {
   %bo = frem double 5.1, %sel
   ret double %bo
 }
+
+declare i64 @llvm.cttz.i64(i64, i1)
+define i64 @cttz_64_eq_select(i64 %v) nounwind {
+; CHECK-LABEL: cttz_64_eq_select:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    bsfq %rdi, %rcx
+; CHECK-NEXT:    movq $-1, %rax
+; CHECK-NEXT:    cmoveq %rcx, %rax
+; CHECK-NEXT:    addq $6, %rax
+; CHECK-NEXT:    retq
+
+; BMI-LABEL: cttz_64_eq_select:
+; BMI:       # %bb.0:
+; BMI-NEXT:    tzcntq %rdi, %rcx
+; BMI-NEXT:    movq $-1, %rax
+; BMI-NEXT:    cmovbq %rcx, %rax
+; BMI-NEXT:    addq $6, %rax
+; BMI-NEXT:    retq
+  %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+  %tobool = icmp eq i64 %v, 0
+  %.op = add nuw nsw i64 %cnt, 6
+  %add = select i1 %tobool, i64 5, i64 %.op
+  ret i64 %add
+}
+
+define i64 @cttz_64_ne_select(i64 %v) nounwind {
+; CHECK-LABEL: cttz_64_ne_select:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    bsfq %rdi, %rcx
+; CHECK-NEXT:    movq $-1, %rax
+; CHECK-NEXT:    cmoveq %rcx, %rax
+; CHECK-NEXT:    addq $6, %rax
+; CHECK-NEXT:    retq
+
+; BMI-LABEL: cttz_64_ne_select:
+; BMI:       # %bb.0:
+; BMI-NEXT:    tzcntq %rdi, %rcx
+; BMI-NEXT:    movq $-1, %rax
+; BMI-NEXT:    cmovbq %rcx, %rax
+; BMI-NEXT:    addq $6, %rax
+; BMI-NEXT:    retq
+  %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+  %tobool = icmp ne i64 %v, 0
+  %.op = add nuw nsw i64 %cnt, 6
+  %add = select i1 %tobool, i64 %.op, i64 5
+  ret i64 %add
+}
+
+declare i32 @llvm.cttz.i32(i32, i1)
+define i32 @cttz_32_eq_select(i32 %v) nounwind {
+; CHECK-LABEL: cttz_32_eq_select:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    bsfl %edi, %ecx
+; CHECK-NEXT:    movl $-1, %eax
+; CHECK-NEXT:    cmovel %ecx, %eax
+; CHECK-NEXT:    addl $6, %eax
+; CHECK-NEXT:    retq
+
+; BMI-LABEL: cttz_32_eq_select:
+; BMI:       # %bb.0:
+; BMI-NEXT:    tzcntl %edi, %ecx
+; BMI-NEXT:    movl $-1, %eax
+; BMI-NEXT:    cmovbl %ecx, %eax
+; BMI-NEXT:    addl $6, %eax
+; BMI-NEXT:    retq
+  %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+  %tobool = icmp eq i32 %v, 0
+  %.op = add nuw nsw i32 %cnt, 6
+  %add = select i1 %tobool, i32 5, i32 %.op
+  ret i32 %add
+}
+
+define i32 @cttz_32_ne_select(i32 %v) nounwind {
+; CHECK-LABEL: cttz_32_ne_select:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    bsfl %edi, %ecx
+; CHECK-NEXT:    movl $-1, %eax
+; CHECK-NEXT:    cmovel %ecx, %eax
+; CHECK-NEXT:    addl $6, %eax
+; CHECK-NEXT:    retq
+
+; BMI-LABEL: cttz_32_ne_select:
+; BMI:       # %bb.0:
+; BMI-NEXT:    tzcntl %edi, %ecx
+; BMI-NEXT:    movl $-1, %eax
+; BMI-NEXT:    cmovbl %ecx, %eax
+; BMI-NEXT:    addl $6, %eax
+; BMI-NEXT:    retq
+  %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+  %tobool = icmp ne i32 %v, 0
+  %.op = add nuw nsw i32 %cnt, 6
+  %add = select i1 %tobool, i32 %.op, i32 5
+  ret i32 %add
+}