; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc --mtriple=loongarch32 -mattr=-32s,+lsx < %s | FileCheck %s --check-prefix=LA32R
; RUN: llc --mtriple=loongarch32 -mattr=+32s,+lsx < %s | FileCheck %s --check-prefix=LA32S
; RUN: llc --mtriple=loongarch64 -mattr=+lsx < %s | FileCheck %s --check-prefix=LA64

declare i8 @llvm.ctpop.i8(i8)
declare i16 @llvm.ctpop.i16(i16)
declare i32 @llvm.ctpop.i32(i32)
declare i64 @llvm.ctpop.i64(i64)

define i8 @test_ctpop_i8(i8 %a) nounwind {
; LA32R-LABEL: test_ctpop_i8:
; LA32R:       # %bb.0:
; LA32R-NEXT:    vreplgr2vr.b $vr0, $a0
; LA32R-NEXT:    vpcnt.b $vr0, $vr0
; LA32R-NEXT:    vpickve2gr.b $a0, $vr0, 0
; LA32R-NEXT:    ret
;
; LA32S-LABEL: test_ctpop_i8:
; LA32S:       # %bb.0:
; LA32S-NEXT:    vreplgr2vr.b $vr0, $a0
; LA32S-NEXT:    vpcnt.b $vr0, $vr0
; LA32S-NEXT:    vpickve2gr.b $a0, $vr0, 0
; LA32S-NEXT:    ret
;
; LA64-LABEL: test_ctpop_i8:
; LA64:       # %bb.0:
; LA64-NEXT:    vreplgr2vr.b $vr0, $a0
; LA64-NEXT:    vpcnt.b $vr0, $vr0
; LA64-NEXT:    vpickve2gr.b $a0, $vr0, 0
; LA64-NEXT:    ret
  %1 = call i8 @llvm.ctpop.i8(i8 %a)
  ret i8 %1
}

define i16 @test_ctpop_i16(i16 %a) nounwind {
; LA32R-LABEL: test_ctpop_i16:
; LA32R:       # %bb.0:
; LA32R-NEXT:    vreplgr2vr.h $vr0, $a0
; LA32R-NEXT:    vpcnt.h $vr0, $vr0
; LA32R-NEXT:    vpickve2gr.h $a0, $vr0, 0
; LA32R-NEXT:    ret
;
; LA32S-LABEL: test_ctpop_i16:
; LA32S:       # %bb.0:
; LA32S-NEXT:    vreplgr2vr.h $vr0, $a0
; LA32S-NEXT:    vpcnt.h $vr0, $vr0
; LA32S-NEXT:    vpickve2gr.h $a0, $vr0, 0
; LA32S-NEXT:    ret
;
; LA64-LABEL: test_ctpop_i16:
; LA64:       # %bb.0:
; LA64-NEXT:    vreplgr2vr.h $vr0, $a0
; LA64-NEXT:    vpcnt.h $vr0, $vr0
; LA64-NEXT:    vpickve2gr.h $a0, $vr0, 0
; LA64-NEXT:    ret
  %1 = call i16 @llvm.ctpop.i16(i16 %a)
  ret i16 %1
}

define i32 @test_ctpop_i32(i32 %a) nounwind {
; LA32R-LABEL: test_ctpop_i32:
; LA32R:       # %bb.0:
; LA32R-NEXT:    vreplgr2vr.w $vr0, $a0
; LA32R-NEXT:    vpcnt.w $vr0, $vr0
; LA32R-NEXT:    vpickve2gr.w $a0, $vr0, 0
; LA32R-NEXT:    ret
;
; LA32S-LABEL: test_ctpop_i32:
; LA32S:       # %bb.0:
; LA32S-NEXT:    vreplgr2vr.w $vr0, $a0
; LA32S-NEXT:    vpcnt.w $vr0, $vr0
; LA32S-NEXT:    vpickve2gr.w $a0, $vr0, 0
; LA32S-NEXT:    ret
;
; LA64-LABEL: test_ctpop_i32:
; LA64:       # %bb.0:
; LA64-NEXT:    vreplgr2vr.w $vr0, $a0
; LA64-NEXT:    vpcnt.w $vr0, $vr0
; LA64-NEXT:    vpickve2gr.w $a0, $vr0, 0
; LA64-NEXT:    ret
  %1 = call i32 @llvm.ctpop.i32(i32 %a)
  ret i32 %1
}

define i64 @test_ctpop_i64(i64 %a) nounwind {
; LA32R-LABEL: test_ctpop_i64:
; LA32R:       # %bb.0:
; LA32R-NEXT:    vreplgr2vr.w $vr0, $a1
; LA32R-NEXT:    vpcnt.w $vr0, $vr0
; LA32R-NEXT:    vpickve2gr.w $a1, $vr0, 0
; LA32R-NEXT:    vreplgr2vr.w $vr0, $a0
; LA32R-NEXT:    vpcnt.w $vr0, $vr0
; LA32R-NEXT:    vpickve2gr.w $a0, $vr0, 0
; LA32R-NEXT:    add.w $a0, $a0, $a1
; LA32R-NEXT:    move $a1, $zero
; LA32R-NEXT:    ret
;
; LA32S-LABEL: test_ctpop_i64:
; LA32S:       # %bb.0:
; LA32S-NEXT:    vreplgr2vr.w $vr0, $a1
; LA32S-NEXT:    vpcnt.w $vr0, $vr0
; LA32S-NEXT:    vpickve2gr.w $a1, $vr0, 0
; LA32S-NEXT:    vreplgr2vr.w $vr0, $a0
; LA32S-NEXT:    vpcnt.w $vr0, $vr0
; LA32S-NEXT:    vpickve2gr.w $a0, $vr0, 0
; LA32S-NEXT:    add.w $a0, $a0, $a1
; LA32S-NEXT:    move $a1, $zero
; LA32S-NEXT:    ret
;
; LA64-LABEL: test_ctpop_i64:
; LA64:       # %bb.0:
; LA64-NEXT:    vreplgr2vr.d $vr0, $a0
; LA64-NEXT:    vpcnt.d $vr0, $vr0
; LA64-NEXT:    vpickve2gr.d $a0, $vr0, 0
; LA64-NEXT:    ret
  %1 = call i64 @llvm.ctpop.i64(i64 %a)
  ret i64 %1
}