derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
1046 lines
32 KiB
LLVM
1046 lines
32 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc --mtriple=loongarch32 -mattr=-32s,+d < %s | FileCheck %s --check-prefix=LA32R
|
|
; RUN: llc --mtriple=loongarch32 -mattr=+32s,+d < %s | FileCheck %s --check-prefix=LA32S
|
|
; RUN: llc --mtriple=loongarch64 -mattr=+d < %s | FileCheck %s --check-prefix=LA64
|
|
|
|
declare i8 @llvm.ctlz.i8(i8, i1)
|
|
declare i16 @llvm.ctlz.i16(i16, i1)
|
|
declare i32 @llvm.ctlz.i32(i32, i1)
|
|
declare i64 @llvm.ctlz.i64(i64, i1)
|
|
declare i8 @llvm.ctpop.i8(i8)
|
|
declare i16 @llvm.ctpop.i16(i16)
|
|
declare i32 @llvm.ctpop.i32(i32)
|
|
declare i64 @llvm.ctpop.i64(i64)
|
|
declare i8 @llvm.cttz.i8(i8, i1)
|
|
declare i16 @llvm.cttz.i16(i16, i1)
|
|
declare i32 @llvm.cttz.i32(i32, i1)
|
|
declare i64 @llvm.cttz.i64(i64, i1)
|
|
|
|
define i8 @test_ctlz_i8(i8 %a) nounwind {
|
|
; LA32R-LABEL: test_ctlz_i8:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: andi $a1, $a0, 254
|
|
; LA32R-NEXT: srli.w $a1, $a1, 1
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 252
|
|
; LA32R-NEXT: srli.w $a1, $a1, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 240
|
|
; LA32R-NEXT: srli.w $a1, $a1, 4
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: andi $a1, $a1, 85
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 51
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: andi $a0, $a0, 51
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a0, $a0, 15
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctlz_i8:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: andi $a0, $a0, 255
|
|
; LA32S-NEXT: clz.w $a0, $a0
|
|
; LA32S-NEXT: addi.w $a0, $a0, -24
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_ctlz_i8:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: andi $a0, $a0, 255
|
|
; LA64-NEXT: clz.d $a0, $a0
|
|
; LA64-NEXT: addi.d $a0, $a0, -56
|
|
; LA64-NEXT: ret
|
|
%tmp = call i8 @llvm.ctlz.i8(i8 %a, i1 false)
|
|
ret i8 %tmp
|
|
}
|
|
|
|
define i16 @test_ctlz_i16(i16 %a) nounwind {
|
|
; LA32R-LABEL: test_ctlz_i16:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: lu12i.w $a1, 15
|
|
; LA32R-NEXT: ori $a2, $a1, 4094
|
|
; LA32R-NEXT: and $a2, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a2, $a2, 1
|
|
; LA32R-NEXT: or $a0, $a0, $a2
|
|
; LA32R-NEXT: ori $a2, $a1, 4092
|
|
; LA32R-NEXT: and $a2, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a2, $a2, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a2
|
|
; LA32R-NEXT: ori $a2, $a1, 4080
|
|
; LA32R-NEXT: and $a2, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a2, $a2, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a2
|
|
; LA32R-NEXT: ori $a1, $a1, 3840
|
|
; LA32R-NEXT: and $a1, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a1, 8
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 5
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 3
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 15
|
|
; LA32R-NEXT: andi $a0, $a0, 3840
|
|
; LA32R-NEXT: srli.w $a0, $a0, 8
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctlz_i16:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: bstrpick.w $a0, $a0, 15, 0
|
|
; LA32S-NEXT: clz.w $a0, $a0
|
|
; LA32S-NEXT: addi.w $a0, $a0, -16
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_ctlz_i16:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: bstrpick.d $a0, $a0, 15, 0
|
|
; LA64-NEXT: clz.d $a0, $a0
|
|
; LA64-NEXT: addi.d $a0, $a0, -48
|
|
; LA64-NEXT: ret
|
|
%tmp = call i16 @llvm.ctlz.i16(i16 %a, i1 false)
|
|
ret i16 %tmp
|
|
}
|
|
|
|
define i32 @test_ctlz_i32(i32 %a) nounwind {
|
|
; LA32R-LABEL: test_ctlz_i32:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 349525
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 209715
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 61680
|
|
; LA32R-NEXT: ori $a1, $a1, 3855
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 4112
|
|
; LA32R-NEXT: ori $a1, $a1, 257
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctlz_i32:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: clz.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_ctlz_i32:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: clz.w $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i32 @llvm.ctlz.i32(i32 %a, i1 false)
|
|
ret i32 %tmp
|
|
}
|
|
|
|
define i64 @test_ctlz_i64(i64 %a) nounwind {
|
|
; LA32R-LABEL: test_ctlz_i64:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: lu12i.w $a2, 349525
|
|
; LA32R-NEXT: ori $a5, $a2, 1365
|
|
; LA32R-NEXT: lu12i.w $a2, 209715
|
|
; LA32R-NEXT: ori $a4, $a2, 819
|
|
; LA32R-NEXT: lu12i.w $a2, 61680
|
|
; LA32R-NEXT: ori $a2, $a2, 3855
|
|
; LA32R-NEXT: lu12i.w $a3, 4112
|
|
; LA32R-NEXT: ori $a3, $a3, 257
|
|
; LA32R-NEXT: bne $a1, $zero, .LBB3_2
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: and $a1, $a1, $a5
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a1, $a0, $a4
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a4
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a0, $a0, $a2
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a3
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: addi.w $a0, $a0, 32
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB3_2:
|
|
; LA32R-NEXT: srli.w $a0, $a1, 1
|
|
; LA32R-NEXT: or $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: and $a1, $a1, $a5
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a1, $a0, $a4
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a4
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a0, $a0, $a2
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a3
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctlz_i64:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: sltu $a2, $zero, $a1
|
|
; LA32S-NEXT: clz.w $a1, $a1
|
|
; LA32S-NEXT: maskeqz $a1, $a1, $a2
|
|
; LA32S-NEXT: clz.w $a0, $a0
|
|
; LA32S-NEXT: addi.w $a0, $a0, 32
|
|
; LA32S-NEXT: masknez $a0, $a0, $a2
|
|
; LA32S-NEXT: or $a0, $a1, $a0
|
|
; LA32S-NEXT: move $a1, $zero
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_ctlz_i64:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: clz.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i64 @llvm.ctlz.i64(i64 %a, i1 false)
|
|
ret i64 %tmp
|
|
}
|
|
|
|
define i8 @test_not_ctlz_i8(i8 %a) nounwind {
|
|
; LA32R-LABEL: test_not_ctlz_i8:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: ori $a1, $zero, 254
|
|
; LA32R-NEXT: andn $a1, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a1, 1
|
|
; LA32R-NEXT: orn $a0, $a1, $a0
|
|
; LA32R-NEXT: andi $a1, $a0, 252
|
|
; LA32R-NEXT: srli.w $a1, $a1, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 240
|
|
; LA32R-NEXT: srli.w $a1, $a1, 4
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: andi $a1, $a1, 85
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 51
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: andi $a0, $a0, 51
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a0, $a0, 15
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_ctlz_i8:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: slli.w $a0, $a0, 24
|
|
; LA32S-NEXT: clo.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_ctlz_i8:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: slli.d $a0, $a0, 56
|
|
; LA64-NEXT: clo.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i8 %a, -1
|
|
%tmp = call i8 @llvm.ctlz.i8(i8 %neg, i1 false)
|
|
ret i8 %tmp
|
|
}
|
|
|
|
define i16 @test_not_ctlz_i16(i16 %a) nounwind {
|
|
; LA32R-LABEL: test_not_ctlz_i16:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: lu12i.w $a1, 15
|
|
; LA32R-NEXT: ori $a2, $a1, 4094
|
|
; LA32R-NEXT: andn $a2, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a2, $a2, 1
|
|
; LA32R-NEXT: orn $a0, $a2, $a0
|
|
; LA32R-NEXT: ori $a2, $a1, 4092
|
|
; LA32R-NEXT: and $a2, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a2, $a2, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a2
|
|
; LA32R-NEXT: ori $a2, $a1, 4080
|
|
; LA32R-NEXT: and $a2, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a2, $a2, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a2
|
|
; LA32R-NEXT: ori $a1, $a1, 3840
|
|
; LA32R-NEXT: and $a1, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a1, 8
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 5
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 3
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 15
|
|
; LA32R-NEXT: andi $a0, $a0, 3840
|
|
; LA32R-NEXT: srli.w $a0, $a0, 8
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_ctlz_i16:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: slli.w $a0, $a0, 16
|
|
; LA32S-NEXT: clo.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_ctlz_i16:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: slli.d $a0, $a0, 48
|
|
; LA64-NEXT: clo.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i16 %a, -1
|
|
%tmp = call i16 @llvm.ctlz.i16(i16 %neg, i1 false)
|
|
ret i16 %tmp
|
|
}
|
|
|
|
define i32 @test_not_ctlz_i32(i32 %a) nounwind {
|
|
; LA32R-LABEL: test_not_ctlz_i32:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: nor $a1, $a0, $zero
|
|
; LA32R-NEXT: srli.w $a1, $a1, 1
|
|
; LA32R-NEXT: orn $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 349525
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 209715
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 61680
|
|
; LA32R-NEXT: ori $a1, $a1, 3855
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 4112
|
|
; LA32R-NEXT: ori $a1, $a1, 257
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_ctlz_i32:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: clo.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_ctlz_i32:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: clo.w $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i32 %a, -1
|
|
%tmp = call i32 @llvm.ctlz.i32(i32 %neg, i1 false)
|
|
ret i32 %tmp
|
|
}
|
|
|
|
define i64 @test_not_ctlz_i64(i64 %a) nounwind {
|
|
; LA32R-LABEL: test_not_ctlz_i64:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: nor $a6, $a1, $zero
|
|
; LA32R-NEXT: lu12i.w $a2, 349525
|
|
; LA32R-NEXT: ori $a5, $a2, 1365
|
|
; LA32R-NEXT: lu12i.w $a2, 209715
|
|
; LA32R-NEXT: ori $a4, $a2, 819
|
|
; LA32R-NEXT: lu12i.w $a2, 61680
|
|
; LA32R-NEXT: ori $a2, $a2, 3855
|
|
; LA32R-NEXT: lu12i.w $a3, 4112
|
|
; LA32R-NEXT: ori $a3, $a3, 257
|
|
; LA32R-NEXT: bne $a6, $zero, .LBB7_2
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: nor $a1, $a0, $zero
|
|
; LA32R-NEXT: srli.w $a1, $a1, 1
|
|
; LA32R-NEXT: orn $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: and $a1, $a1, $a5
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a1, $a0, $a4
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a4
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a0, $a0, $a2
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a3
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: addi.w $a0, $a0, 32
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB7_2:
|
|
; LA32R-NEXT: srli.w $a0, $a6, 1
|
|
; LA32R-NEXT: orn $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 2
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 8
|
|
; LA32R-NEXT: or $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 16
|
|
; LA32R-NEXT: nor $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: and $a1, $a1, $a5
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a1, $a0, $a4
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a4
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: and $a0, $a0, $a2
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a3
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_ctlz_i64:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: nor $a2, $a1, $zero
|
|
; LA32S-NEXT: sltu $a2, $zero, $a2
|
|
; LA32S-NEXT: clo.w $a0, $a0
|
|
; LA32S-NEXT: addi.w $a0, $a0, 32
|
|
; LA32S-NEXT: masknez $a0, $a0, $a2
|
|
; LA32S-NEXT: clo.w $a1, $a1
|
|
; LA32S-NEXT: maskeqz $a1, $a1, $a2
|
|
; LA32S-NEXT: or $a0, $a1, $a0
|
|
; LA32S-NEXT: move $a1, $zero
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_ctlz_i64:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: clo.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i64 %a, -1
|
|
%tmp = call i64 @llvm.ctlz.i64(i64 %neg, i1 false)
|
|
ret i64 %tmp
|
|
}
|
|
|
|
define i8 @test_ctpop_i8(i8 %a) nounwind {
|
|
; LA32R-LABEL: test_ctpop_i8:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: andi $a1, $a1, 85
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 51
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: andi $a0, $a0, 51
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a0, $a0, 15
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctpop_i8:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: srli.w $a1, $a0, 1
|
|
; LA32S-NEXT: andi $a1, $a1, 85
|
|
; LA32S-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32S-NEXT: andi $a1, $a0, 51
|
|
; LA32S-NEXT: srli.w $a0, $a0, 2
|
|
; LA32S-NEXT: andi $a0, $a0, 51
|
|
; LA32S-NEXT: add.w $a0, $a1, $a0
|
|
; LA32S-NEXT: srli.w $a1, $a0, 4
|
|
; LA32S-NEXT: add.w $a0, $a0, $a1
|
|
; LA32S-NEXT: andi $a0, $a0, 15
|
|
; 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: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 5
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 3
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 15
|
|
; LA32R-NEXT: andi $a0, $a0, 3840
|
|
; LA32R-NEXT: srli.w $a0, $a0, 8
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctpop_i16:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: srli.w $a1, $a0, 1
|
|
; LA32S-NEXT: lu12i.w $a2, 5
|
|
; LA32S-NEXT: ori $a2, $a2, 1365
|
|
; LA32S-NEXT: and $a1, $a1, $a2
|
|
; LA32S-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32S-NEXT: lu12i.w $a1, 3
|
|
; LA32S-NEXT: ori $a1, $a1, 819
|
|
; LA32S-NEXT: and $a2, $a0, $a1
|
|
; LA32S-NEXT: srli.w $a0, $a0, 2
|
|
; LA32S-NEXT: and $a0, $a0, $a1
|
|
; LA32S-NEXT: add.w $a0, $a2, $a0
|
|
; LA32S-NEXT: srli.w $a1, $a0, 4
|
|
; LA32S-NEXT: add.w $a0, $a0, $a1
|
|
; LA32S-NEXT: bstrpick.w $a1, $a0, 11, 8
|
|
; LA32S-NEXT: andi $a0, $a0, 15
|
|
; LA32S-NEXT: add.w $a0, $a0, $a1
|
|
; 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: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 349525
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 209715
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 61680
|
|
; LA32R-NEXT: ori $a1, $a1, 3855
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 4112
|
|
; LA32R-NEXT: ori $a1, $a1, 257
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_ctpop_i32:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: srli.w $a1, $a0, 1
|
|
; LA32S-NEXT: lu12i.w $a2, 349525
|
|
; LA32S-NEXT: ori $a2, $a2, 1365
|
|
; LA32S-NEXT: and $a1, $a1, $a2
|
|
; LA32S-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32S-NEXT: lu12i.w $a1, 209715
|
|
; LA32S-NEXT: ori $a1, $a1, 819
|
|
; LA32S-NEXT: and $a2, $a0, $a1
|
|
; LA32S-NEXT: srli.w $a0, $a0, 2
|
|
; LA32S-NEXT: and $a0, $a0, $a1
|
|
; LA32S-NEXT: add.w $a0, $a2, $a0
|
|
; LA32S-NEXT: srli.w $a1, $a0, 4
|
|
; LA32S-NEXT: add.w $a0, $a0, $a1
|
|
; LA32S-NEXT: lu12i.w $a1, 61680
|
|
; LA32S-NEXT: ori $a1, $a1, 3855
|
|
; LA32S-NEXT: and $a0, $a0, $a1
|
|
; LA32S-NEXT: lu12i.w $a1, 4112
|
|
; LA32S-NEXT: ori $a1, $a1, 257
|
|
; LA32S-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32S-NEXT: srli.w $a0, $a0, 24
|
|
; 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: srli.w $a2, $a1, 1
|
|
; LA32R-NEXT: lu12i.w $a3, 349525
|
|
; LA32R-NEXT: ori $a3, $a3, 1365
|
|
; LA32R-NEXT: and $a2, $a2, $a3
|
|
; LA32R-NEXT: sub.w $a1, $a1, $a2
|
|
; LA32R-NEXT: lu12i.w $a2, 209715
|
|
; LA32R-NEXT: ori $a2, $a2, 819
|
|
; LA32R-NEXT: and $a4, $a1, $a2
|
|
; LA32R-NEXT: srli.w $a1, $a1, 2
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: add.w $a1, $a4, $a1
|
|
; LA32R-NEXT: srli.w $a4, $a1, 4
|
|
; LA32R-NEXT: add.w $a1, $a1, $a4
|
|
; LA32R-NEXT: lu12i.w $a4, 61680
|
|
; LA32R-NEXT: ori $a4, $a4, 3855
|
|
; LA32R-NEXT: and $a1, $a1, $a4
|
|
; LA32R-NEXT: lu12i.w $a5, 4112
|
|
; LA32R-NEXT: ori $a5, $a5, 257
|
|
; LA32R-NEXT: mul.w $a1, $a1, $a5
|
|
; LA32R-NEXT: srli.w $a1, $a1, 24
|
|
; LA32R-NEXT: srli.w $a6, $a0, 1
|
|
; LA32R-NEXT: and $a3, $a6, $a3
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a3
|
|
; LA32R-NEXT: and $a3, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a2
|
|
; LA32R-NEXT: add.w $a0, $a3, $a0
|
|
; LA32R-NEXT: srli.w $a2, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a2
|
|
; LA32R-NEXT: and $a0, $a0, $a4
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a5
|
|
; LA32R-NEXT: srli.w $a0, $a0, 24
|
|
; 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: srli.w $a2, $a1, 1
|
|
; LA32S-NEXT: lu12i.w $a3, 349525
|
|
; LA32S-NEXT: ori $a3, $a3, 1365
|
|
; LA32S-NEXT: and $a2, $a2, $a3
|
|
; LA32S-NEXT: sub.w $a1, $a1, $a2
|
|
; LA32S-NEXT: lu12i.w $a2, 209715
|
|
; LA32S-NEXT: ori $a2, $a2, 819
|
|
; LA32S-NEXT: and $a4, $a1, $a2
|
|
; LA32S-NEXT: srli.w $a1, $a1, 2
|
|
; LA32S-NEXT: and $a1, $a1, $a2
|
|
; LA32S-NEXT: add.w $a1, $a4, $a1
|
|
; LA32S-NEXT: srli.w $a4, $a1, 4
|
|
; LA32S-NEXT: add.w $a1, $a1, $a4
|
|
; LA32S-NEXT: lu12i.w $a4, 61680
|
|
; LA32S-NEXT: ori $a4, $a4, 3855
|
|
; LA32S-NEXT: and $a1, $a1, $a4
|
|
; LA32S-NEXT: lu12i.w $a5, 4112
|
|
; LA32S-NEXT: ori $a5, $a5, 257
|
|
; LA32S-NEXT: mul.w $a1, $a1, $a5
|
|
; LA32S-NEXT: srli.w $a1, $a1, 24
|
|
; LA32S-NEXT: srli.w $a6, $a0, 1
|
|
; LA32S-NEXT: and $a3, $a6, $a3
|
|
; LA32S-NEXT: sub.w $a0, $a0, $a3
|
|
; LA32S-NEXT: and $a3, $a0, $a2
|
|
; LA32S-NEXT: srli.w $a0, $a0, 2
|
|
; LA32S-NEXT: and $a0, $a0, $a2
|
|
; LA32S-NEXT: add.w $a0, $a3, $a0
|
|
; LA32S-NEXT: srli.w $a2, $a0, 4
|
|
; LA32S-NEXT: add.w $a0, $a0, $a2
|
|
; LA32S-NEXT: and $a0, $a0, $a4
|
|
; LA32S-NEXT: mul.w $a0, $a0, $a5
|
|
; LA32S-NEXT: srli.w $a0, $a0, 24
|
|
; 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
|
|
}
|
|
|
|
define i8 @test_cttz_i8(i8 %a) nounwind {
|
|
; LA32R-LABEL: test_cttz_i8:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: addi.w $a1, $a0, -1
|
|
; LA32R-NEXT: andn $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: andi $a1, $a1, 85
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 51
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: andi $a0, $a0, 51
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a0, $a0, 15
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_cttz_i8:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: ori $a0, $a0, 256
|
|
; LA32S-NEXT: ctz.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_cttz_i8:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: ori $a0, $a0, 256
|
|
; LA64-NEXT: ctz.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i8 @llvm.cttz.i8(i8 %a, i1 false)
|
|
ret i8 %tmp
|
|
}
|
|
|
|
define i16 @test_cttz_i16(i16 %a) nounwind {
|
|
; LA32R-LABEL: test_cttz_i16:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: addi.w $a1, $a0, -1
|
|
; LA32R-NEXT: andn $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 5
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 3
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 15
|
|
; LA32R-NEXT: andi $a0, $a0, 3840
|
|
; LA32R-NEXT: srli.w $a0, $a0, 8
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_cttz_i16:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: lu12i.w $a1, 16
|
|
; LA32S-NEXT: or $a0, $a0, $a1
|
|
; LA32S-NEXT: ctz.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_cttz_i16:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: lu12i.w $a1, 16
|
|
; LA64-NEXT: or $a0, $a0, $a1
|
|
; LA64-NEXT: ctz.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i16 @llvm.cttz.i16(i16 %a, i1 false)
|
|
ret i16 %tmp
|
|
}
|
|
|
|
define i32 @test_cttz_i32(i32 %a) nounwind {
|
|
; LA32R-LABEL: test_cttz_i32:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: bne $a0, $zero, .LBB14_2
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: ori $a0, $zero, 32
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB14_2:
|
|
; LA32R-NEXT: sub.w $a1, $zero, $a0
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 30667
|
|
; LA32R-NEXT: ori $a1, $a1, 1329
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 27
|
|
; LA32R-NEXT: .Lpcadd_hi0:
|
|
; LA32R-NEXT: pcaddu12i $a1, %pcadd_hi20(.LCPI14_0)
|
|
; LA32R-NEXT: addi.w $a1, $a1, %pcadd_lo12(.Lpcadd_hi0)
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ld.bu $a0, $a0, 0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_cttz_i32:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: ctz.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_cttz_i32:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: ctz.w $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i32 @llvm.cttz.i32(i32 %a, i1 false)
|
|
ret i32 %tmp
|
|
}
|
|
|
|
define i64 @test_cttz_i64(i64 %a) nounwind {
|
|
; LA32R-LABEL: test_cttz_i64:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: lu12i.w $a2, 30667
|
|
; LA32R-NEXT: ori $a2, $a2, 1329
|
|
; LA32R-NEXT: .Lpcadd_hi1:
|
|
; LA32R-NEXT: pcaddu12i $a3, %pcadd_hi20(.LCPI15_0)
|
|
; LA32R-NEXT: addi.w $a3, $a3, %pcadd_lo12(.Lpcadd_hi1)
|
|
; LA32R-NEXT: bne $a1, $zero, .LBB15_3
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: ori $a1, $zero, 32
|
|
; LA32R-NEXT: beq $a0, $zero, .LBB15_4
|
|
; LA32R-NEXT: .LBB15_2:
|
|
; LA32R-NEXT: sub.w $a1, $zero, $a0
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a2
|
|
; LA32R-NEXT: srli.w $a0, $a0, 27
|
|
; LA32R-NEXT: add.w $a0, $a3, $a0
|
|
; LA32R-NEXT: ld.bu $a0, $a0, 0
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB15_3:
|
|
; LA32R-NEXT: sub.w $a4, $zero, $a1
|
|
; LA32R-NEXT: and $a1, $a1, $a4
|
|
; LA32R-NEXT: mul.w $a1, $a1, $a2
|
|
; LA32R-NEXT: srli.w $a1, $a1, 27
|
|
; LA32R-NEXT: add.w $a1, $a3, $a1
|
|
; LA32R-NEXT: ld.bu $a1, $a1, 0
|
|
; LA32R-NEXT: bne $a0, $zero, .LBB15_2
|
|
; LA32R-NEXT: .LBB15_4:
|
|
; LA32R-NEXT: addi.w $a0, $a1, 32
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_cttz_i64:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: sltu $a2, $zero, $a0
|
|
; LA32S-NEXT: ctz.w $a0, $a0
|
|
; LA32S-NEXT: maskeqz $a0, $a0, $a2
|
|
; LA32S-NEXT: ctz.w $a1, $a1
|
|
; LA32S-NEXT: addi.w $a1, $a1, 32
|
|
; LA32S-NEXT: masknez $a1, $a1, $a2
|
|
; LA32S-NEXT: or $a0, $a0, $a1
|
|
; LA32S-NEXT: move $a1, $zero
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_cttz_i64:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: ctz.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%tmp = call i64 @llvm.cttz.i64(i64 %a, i1 false)
|
|
ret i64 %tmp
|
|
}
|
|
|
|
define i8 @test_not_cttz_i8(i8 %a) nounwind {
|
|
; LA32R-LABEL: test_not_cttz_i8:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: addi.w $a1, $a0, 1
|
|
; LA32R-NEXT: andn $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: andi $a1, $a1, 85
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 51
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: andi $a0, $a0, 51
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a0, $a0, 15
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_cttz_i8:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: andi $a0, $a0, 255
|
|
; LA32S-NEXT: cto.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_cttz_i8:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: andi $a0, $a0, 255
|
|
; LA64-NEXT: cto.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i8 %a, -1
|
|
%tmp = call i8 @llvm.cttz.i8(i8 %neg, i1 false)
|
|
ret i8 %tmp
|
|
}
|
|
|
|
define i16 @test_not_cttz_i16(i16 %a) nounwind {
|
|
; LA32R-LABEL: test_not_cttz_i16:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: addi.w $a1, $a0, 1
|
|
; LA32R-NEXT: andn $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a1, $a0, 1
|
|
; LA32R-NEXT: lu12i.w $a2, 5
|
|
; LA32R-NEXT: ori $a2, $a2, 1365
|
|
; LA32R-NEXT: and $a1, $a1, $a2
|
|
; LA32R-NEXT: sub.w $a0, $a0, $a1
|
|
; LA32R-NEXT: lu12i.w $a1, 3
|
|
; LA32R-NEXT: ori $a1, $a1, 819
|
|
; LA32R-NEXT: and $a2, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 2
|
|
; LA32R-NEXT: and $a0, $a0, $a1
|
|
; LA32R-NEXT: add.w $a0, $a2, $a0
|
|
; LA32R-NEXT: srli.w $a1, $a0, 4
|
|
; LA32R-NEXT: add.w $a0, $a0, $a1
|
|
; LA32R-NEXT: andi $a1, $a0, 15
|
|
; LA32R-NEXT: andi $a0, $a0, 3840
|
|
; LA32R-NEXT: srli.w $a0, $a0, 8
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_cttz_i16:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: bstrpick.w $a0, $a0, 15, 0
|
|
; LA32S-NEXT: cto.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_cttz_i16:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: bstrpick.d $a0, $a0, 15, 0
|
|
; LA64-NEXT: cto.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i16 %a, -1
|
|
%tmp = call i16 @llvm.cttz.i16(i16 %neg, i1 false)
|
|
ret i16 %tmp
|
|
}
|
|
|
|
define i32 @test_not_cttz_i32(i32 %a) nounwind {
|
|
; LA32R-LABEL: test_not_cttz_i32:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: nor $a1, $a0, $zero
|
|
; LA32R-NEXT: bne $a1, $zero, .LBB18_2
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: ori $a0, $zero, 32
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB18_2:
|
|
; LA32R-NEXT: sub.w $a1, $zero, $a1
|
|
; LA32R-NEXT: andn $a0, $a1, $a0
|
|
; LA32R-NEXT: lu12i.w $a1, 30667
|
|
; LA32R-NEXT: ori $a1, $a1, 1329
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a1
|
|
; LA32R-NEXT: srli.w $a0, $a0, 27
|
|
; LA32R-NEXT: .Lpcadd_hi2:
|
|
; LA32R-NEXT: pcaddu12i $a1, %pcadd_hi20(.LCPI18_0)
|
|
; LA32R-NEXT: addi.w $a1, $a1, %pcadd_lo12(.Lpcadd_hi2)
|
|
; LA32R-NEXT: add.w $a0, $a1, $a0
|
|
; LA32R-NEXT: ld.bu $a0, $a0, 0
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_cttz_i32:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: cto.w $a0, $a0
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_cttz_i32:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: cto.w $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i32 %a, -1
|
|
%tmp = call i32 @llvm.cttz.i32(i32 %neg, i1 false)
|
|
ret i32 %tmp
|
|
}
|
|
|
|
define i64 @test_not_cttz_i64(i64 %a) nounwind {
|
|
; LA32R-LABEL: test_not_cttz_i64:
|
|
; LA32R: # %bb.0:
|
|
; LA32R-NEXT: nor $a5, $a1, $zero
|
|
; LA32R-NEXT: nor $a2, $a0, $zero
|
|
; LA32R-NEXT: lu12i.w $a3, 30667
|
|
; LA32R-NEXT: ori $a3, $a3, 1329
|
|
; LA32R-NEXT: .Lpcadd_hi3:
|
|
; LA32R-NEXT: pcaddu12i $a4, %pcadd_hi20(.LCPI19_0)
|
|
; LA32R-NEXT: addi.w $a4, $a4, %pcadd_lo12(.Lpcadd_hi3)
|
|
; LA32R-NEXT: bne $a5, $zero, .LBB19_3
|
|
; LA32R-NEXT: # %bb.1:
|
|
; LA32R-NEXT: ori $a1, $zero, 32
|
|
; LA32R-NEXT: beq $a2, $zero, .LBB19_4
|
|
; LA32R-NEXT: .LBB19_2:
|
|
; LA32R-NEXT: sub.w $a1, $zero, $a2
|
|
; LA32R-NEXT: andn $a0, $a1, $a0
|
|
; LA32R-NEXT: mul.w $a0, $a0, $a3
|
|
; LA32R-NEXT: srli.w $a0, $a0, 27
|
|
; LA32R-NEXT: add.w $a0, $a4, $a0
|
|
; LA32R-NEXT: ld.bu $a0, $a0, 0
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
; LA32R-NEXT: .LBB19_3:
|
|
; LA32R-NEXT: sub.w $a5, $zero, $a5
|
|
; LA32R-NEXT: andn $a1, $a5, $a1
|
|
; LA32R-NEXT: mul.w $a1, $a1, $a3
|
|
; LA32R-NEXT: srli.w $a1, $a1, 27
|
|
; LA32R-NEXT: add.w $a1, $a4, $a1
|
|
; LA32R-NEXT: ld.bu $a1, $a1, 0
|
|
; LA32R-NEXT: bne $a2, $zero, .LBB19_2
|
|
; LA32R-NEXT: .LBB19_4:
|
|
; LA32R-NEXT: addi.w $a0, $a1, 32
|
|
; LA32R-NEXT: move $a1, $zero
|
|
; LA32R-NEXT: ret
|
|
;
|
|
; LA32S-LABEL: test_not_cttz_i64:
|
|
; LA32S: # %bb.0:
|
|
; LA32S-NEXT: nor $a2, $a0, $zero
|
|
; LA32S-NEXT: sltu $a2, $zero, $a2
|
|
; LA32S-NEXT: cto.w $a1, $a1
|
|
; LA32S-NEXT: addi.w $a1, $a1, 32
|
|
; LA32S-NEXT: masknez $a1, $a1, $a2
|
|
; LA32S-NEXT: cto.w $a0, $a0
|
|
; LA32S-NEXT: maskeqz $a0, $a0, $a2
|
|
; LA32S-NEXT: or $a0, $a0, $a1
|
|
; LA32S-NEXT: move $a1, $zero
|
|
; LA32S-NEXT: ret
|
|
;
|
|
; LA64-LABEL: test_not_cttz_i64:
|
|
; LA64: # %bb.0:
|
|
; LA64-NEXT: cto.d $a0, $a0
|
|
; LA64-NEXT: ret
|
|
%neg = xor i64 %a, -1
|
|
%tmp = call i64 @llvm.cttz.i64(i64 %neg, i1 false)
|
|
ret i64 %tmp
|
|
}
|