Files
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
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

308 lines
9.4 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc --mtriple=loongarch32 -mattr=-32s,+d --verify-machineinstrs < %s \
; RUN: | FileCheck %s --check-prefix=LA32R
; RUN: llc --mtriple=loongarch32 -mattr=+32s,+d --verify-machineinstrs < %s \
; RUN: | FileCheck %s --check-prefix=LA32S
; RUN: llc --mtriple=loongarch64 -mattr=+d --verify-machineinstrs < %s \
; RUN: | FileCheck %s --check-prefix=LA64
declare i16 @llvm.bswap.i16(i16)
declare i32 @llvm.bswap.i32(i32)
declare i48 @llvm.bswap.i48(i48)
declare i64 @llvm.bswap.i64(i64)
declare i80 @llvm.bswap.i80(i80)
declare i128 @llvm.bswap.i128(i128)
define i16 @test_bswap_i16(i16 %a) nounwind {
; LA32R-LABEL: test_bswap_i16:
; LA32R: # %bb.0:
; LA32R-NEXT: lu12i.w $a1, 15
; LA32R-NEXT: ori $a1, $a1, 3840
; LA32R-NEXT: and $a1, $a0, $a1
; LA32R-NEXT: srli.w $a1, $a1, 8
; LA32R-NEXT: slli.w $a0, $a0, 8
; LA32R-NEXT: or $a0, $a0, $a1
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i16:
; LA32S: # %bb.0:
; LA32S-NEXT: revb.2h $a0, $a0
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i16:
; LA64: # %bb.0:
; LA64-NEXT: revb.2h $a0, $a0
; LA64-NEXT: ret
%tmp = call i16 @llvm.bswap.i16(i16 %a)
ret i16 %tmp
}
define i32 @test_bswap_i32(i32 %a) nounwind {
; LA32R-LABEL: test_bswap_i32:
; LA32R: # %bb.0:
; LA32R-NEXT: srli.w $a1, $a0, 8
; LA32R-NEXT: lu12i.w $a2, 15
; LA32R-NEXT: ori $a2, $a2, 3840
; LA32R-NEXT: and $a1, $a1, $a2
; LA32R-NEXT: srli.w $a3, $a0, 24
; LA32R-NEXT: or $a1, $a1, $a3
; LA32R-NEXT: and $a2, $a0, $a2
; LA32R-NEXT: slli.w $a2, $a2, 8
; LA32R-NEXT: slli.w $a0, $a0, 24
; LA32R-NEXT: or $a0, $a0, $a2
; LA32R-NEXT: or $a0, $a0, $a1
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i32:
; LA32S: # %bb.0:
; LA32S-NEXT: revb.2h $a0, $a0
; LA32S-NEXT: rotri.w $a0, $a0, 16
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i32:
; LA64: # %bb.0:
; LA64-NEXT: revb.2w $a0, $a0
; LA64-NEXT: ret
%tmp = call i32 @llvm.bswap.i32(i32 %a)
ret i32 %tmp
}
define i64 @test_bswap_i64(i64 %a) nounwind {
; LA32R-LABEL: test_bswap_i64:
; LA32R: # %bb.0:
; LA32R-NEXT: srli.w $a2, $a1, 8
; LA32R-NEXT: lu12i.w $a3, 15
; LA32R-NEXT: ori $a3, $a3, 3840
; LA32R-NEXT: and $a2, $a2, $a3
; LA32R-NEXT: srli.w $a4, $a1, 24
; LA32R-NEXT: or $a2, $a2, $a4
; LA32R-NEXT: and $a4, $a1, $a3
; LA32R-NEXT: slli.w $a4, $a4, 8
; LA32R-NEXT: slli.w $a1, $a1, 24
; LA32R-NEXT: or $a1, $a1, $a4
; LA32R-NEXT: or $a2, $a1, $a2
; LA32R-NEXT: srli.w $a1, $a0, 8
; LA32R-NEXT: and $a1, $a1, $a3
; LA32R-NEXT: srli.w $a4, $a0, 24
; LA32R-NEXT: or $a1, $a1, $a4
; LA32R-NEXT: and $a3, $a0, $a3
; LA32R-NEXT: slli.w $a3, $a3, 8
; LA32R-NEXT: slli.w $a0, $a0, 24
; LA32R-NEXT: or $a0, $a0, $a3
; LA32R-NEXT: or $a1, $a0, $a1
; LA32R-NEXT: move $a0, $a2
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i64:
; LA32S: # %bb.0:
; LA32S-NEXT: revb.2h $a1, $a1
; LA32S-NEXT: rotri.w $a2, $a1, 16
; LA32S-NEXT: revb.2h $a0, $a0
; LA32S-NEXT: rotri.w $a1, $a0, 16
; LA32S-NEXT: move $a0, $a2
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i64:
; LA64: # %bb.0:
; LA64-NEXT: revb.d $a0, $a0
; LA64-NEXT: ret
%tmp = call i64 @llvm.bswap.i64(i64 %a)
ret i64 %tmp
}
;; Bswap on non-native integer widths.
define i48 @test_bswap_i48(i48 %a) nounwind {
; LA32R-LABEL: test_bswap_i48:
; LA32R: # %bb.0:
; LA32R-NEXT: srli.w $a2, $a0, 8
; LA32R-NEXT: lu12i.w $a3, 15
; LA32R-NEXT: ori $a3, $a3, 3840
; LA32R-NEXT: and $a2, $a2, $a3
; LA32R-NEXT: srli.w $a4, $a0, 24
; LA32R-NEXT: or $a2, $a2, $a4
; LA32R-NEXT: slli.w $a2, $a2, 16
; LA32R-NEXT: and $a4, $a1, $a3
; LA32R-NEXT: slli.w $a4, $a4, 8
; LA32R-NEXT: slli.w $a1, $a1, 24
; LA32R-NEXT: or $a1, $a1, $a4
; LA32R-NEXT: srli.w $a1, $a1, 16
; LA32R-NEXT: or $a2, $a1, $a2
; LA32R-NEXT: and $a1, $a0, $a3
; LA32R-NEXT: slli.w $a1, $a1, 8
; LA32R-NEXT: slli.w $a0, $a0, 24
; LA32R-NEXT: or $a0, $a0, $a1
; LA32R-NEXT: srli.w $a1, $a0, 16
; LA32R-NEXT: move $a0, $a2
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i48:
; LA32S: # %bb.0:
; LA32S-NEXT: revb.2h $a0, $a0
; LA32S-NEXT: rotri.w $a2, $a0, 16
; LA32S-NEXT: revb.2h $a0, $a1
; LA32S-NEXT: rotri.w $a0, $a0, 16
; LA32S-NEXT: bytepick.w $a0, $a0, $a2, 2
; LA32S-NEXT: srli.w $a1, $a2, 16
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i48:
; LA64: # %bb.0:
; LA64-NEXT: revb.d $a0, $a0
; LA64-NEXT: srli.d $a0, $a0, 16
; LA64-NEXT: ret
%tmp = call i48 @llvm.bswap.i48(i48 %a)
ret i48 %tmp
}
define i80 @test_bswap_i80(i80 %a) nounwind {
; LA32R-LABEL: test_bswap_i80:
; LA32R: # %bb.0:
; LA32R-NEXT: ld.w $a2, $a1, 4
; LA32R-NEXT: ld.w $a3, $a1, 0
; LA32R-NEXT: ld.w $a1, $a1, 8
; LA32R-NEXT: srli.w $a4, $a2, 8
; LA32R-NEXT: lu12i.w $a5, 15
; LA32R-NEXT: ori $a5, $a5, 3840
; LA32R-NEXT: and $a4, $a4, $a5
; LA32R-NEXT: srli.w $a6, $a2, 24
; LA32R-NEXT: or $a4, $a4, $a6
; LA32R-NEXT: slli.w $a4, $a4, 16
; LA32R-NEXT: and $a6, $a1, $a5
; LA32R-NEXT: slli.w $a6, $a6, 8
; LA32R-NEXT: slli.w $a1, $a1, 24
; LA32R-NEXT: or $a1, $a1, $a6
; LA32R-NEXT: srli.w $a1, $a1, 16
; LA32R-NEXT: or $a1, $a1, $a4
; LA32R-NEXT: and $a4, $a2, $a5
; LA32R-NEXT: slli.w $a4, $a4, 8
; LA32R-NEXT: slli.w $a2, $a2, 24
; LA32R-NEXT: or $a2, $a2, $a4
; LA32R-NEXT: srli.w $a2, $a2, 16
; LA32R-NEXT: srli.w $a4, $a3, 8
; LA32R-NEXT: and $a4, $a4, $a5
; LA32R-NEXT: srli.w $a6, $a3, 24
; LA32R-NEXT: or $a4, $a4, $a6
; LA32R-NEXT: slli.w $a4, $a4, 16
; LA32R-NEXT: or $a2, $a4, $a2
; LA32R-NEXT: and $a4, $a3, $a5
; LA32R-NEXT: slli.w $a4, $a4, 8
; LA32R-NEXT: slli.w $a3, $a3, 24
; LA32R-NEXT: or $a3, $a3, $a4
; LA32R-NEXT: srli.w $a3, $a3, 16
; LA32R-NEXT: st.h $a3, $a0, 8
; LA32R-NEXT: st.w $a2, $a0, 4
; LA32R-NEXT: st.w $a1, $a0, 0
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i80:
; LA32S: # %bb.0:
; LA32S-NEXT: ld.w $a2, $a1, 4
; LA32S-NEXT: ld.w $a3, $a1, 8
; LA32S-NEXT: ld.w $a1, $a1, 0
; LA32S-NEXT: revb.2h $a2, $a2
; LA32S-NEXT: rotri.w $a2, $a2, 16
; LA32S-NEXT: revb.2h $a3, $a3
; LA32S-NEXT: rotri.w $a3, $a3, 16
; LA32S-NEXT: bytepick.w $a3, $a3, $a2, 2
; LA32S-NEXT: revb.2h $a1, $a1
; LA32S-NEXT: rotri.w $a1, $a1, 16
; LA32S-NEXT: bytepick.w $a2, $a2, $a1, 2
; LA32S-NEXT: srli.w $a1, $a1, 16
; LA32S-NEXT: st.w $a2, $a0, 4
; LA32S-NEXT: st.w $a3, $a0, 0
; LA32S-NEXT: st.h $a1, $a0, 8
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i80:
; LA64: # %bb.0:
; LA64-NEXT: revb.d $a2, $a0
; LA64-NEXT: revb.d $a0, $a1
; LA64-NEXT: bytepick.d $a0, $a0, $a2, 2
; LA64-NEXT: srli.d $a1, $a2, 48
; LA64-NEXT: ret
%tmp = call i80 @llvm.bswap.i80(i80 %a)
ret i80 %tmp
}
define i128 @test_bswap_i128(i128 %a) nounwind {
; LA32R-LABEL: test_bswap_i128:
; LA32R: # %bb.0:
; LA32R-NEXT: ld.w $a2, $a1, 12
; LA32R-NEXT: ld.w $a3, $a1, 0
; LA32R-NEXT: ld.w $a4, $a1, 4
; LA32R-NEXT: ld.w $a1, $a1, 8
; LA32R-NEXT: srli.w $a5, $a2, 8
; LA32R-NEXT: lu12i.w $a6, 15
; LA32R-NEXT: ori $a6, $a6, 3840
; LA32R-NEXT: and $a5, $a5, $a6
; LA32R-NEXT: srli.w $a7, $a2, 24
; LA32R-NEXT: or $a5, $a5, $a7
; LA32R-NEXT: and $a7, $a2, $a6
; LA32R-NEXT: slli.w $a7, $a7, 8
; LA32R-NEXT: slli.w $a2, $a2, 24
; LA32R-NEXT: or $a2, $a2, $a7
; LA32R-NEXT: or $a2, $a2, $a5
; LA32R-NEXT: srli.w $a5, $a1, 8
; LA32R-NEXT: and $a5, $a5, $a6
; LA32R-NEXT: srli.w $a7, $a1, 24
; LA32R-NEXT: or $a5, $a5, $a7
; LA32R-NEXT: and $a7, $a1, $a6
; LA32R-NEXT: slli.w $a7, $a7, 8
; LA32R-NEXT: slli.w $a1, $a1, 24
; LA32R-NEXT: or $a1, $a1, $a7
; LA32R-NEXT: or $a1, $a1, $a5
; LA32R-NEXT: srli.w $a5, $a4, 8
; LA32R-NEXT: and $a5, $a5, $a6
; LA32R-NEXT: srli.w $a7, $a4, 24
; LA32R-NEXT: or $a5, $a5, $a7
; LA32R-NEXT: and $a7, $a4, $a6
; LA32R-NEXT: slli.w $a7, $a7, 8
; LA32R-NEXT: slli.w $a4, $a4, 24
; LA32R-NEXT: or $a4, $a4, $a7
; LA32R-NEXT: or $a4, $a4, $a5
; LA32R-NEXT: srli.w $a5, $a3, 8
; LA32R-NEXT: and $a5, $a5, $a6
; LA32R-NEXT: srli.w $a7, $a3, 24
; LA32R-NEXT: or $a5, $a5, $a7
; LA32R-NEXT: and $a6, $a3, $a6
; LA32R-NEXT: slli.w $a6, $a6, 8
; LA32R-NEXT: slli.w $a3, $a3, 24
; LA32R-NEXT: or $a3, $a3, $a6
; LA32R-NEXT: or $a3, $a3, $a5
; LA32R-NEXT: st.w $a3, $a0, 12
; LA32R-NEXT: st.w $a4, $a0, 8
; LA32R-NEXT: st.w $a1, $a0, 4
; LA32R-NEXT: st.w $a2, $a0, 0
; LA32R-NEXT: ret
;
; LA32S-LABEL: test_bswap_i128:
; LA32S: # %bb.0:
; LA32S-NEXT: ld.w $a2, $a1, 12
; LA32S-NEXT: ld.w $a3, $a1, 0
; LA32S-NEXT: ld.w $a4, $a1, 8
; LA32S-NEXT: ld.w $a1, $a1, 4
; LA32S-NEXT: revb.2h $a2, $a2
; LA32S-NEXT: rotri.w $a2, $a2, 16
; LA32S-NEXT: revb.2h $a4, $a4
; LA32S-NEXT: rotri.w $a4, $a4, 16
; LA32S-NEXT: revb.2h $a1, $a1
; LA32S-NEXT: rotri.w $a1, $a1, 16
; LA32S-NEXT: revb.2h $a3, $a3
; LA32S-NEXT: rotri.w $a3, $a3, 16
; LA32S-NEXT: st.w $a3, $a0, 12
; LA32S-NEXT: st.w $a1, $a0, 8
; LA32S-NEXT: st.w $a4, $a0, 4
; LA32S-NEXT: st.w $a2, $a0, 0
; LA32S-NEXT: ret
;
; LA64-LABEL: test_bswap_i128:
; LA64: # %bb.0:
; LA64-NEXT: revb.d $a2, $a1
; LA64-NEXT: revb.d $a1, $a0
; LA64-NEXT: move $a0, $a2
; LA64-NEXT: ret
%tmp = call i128 @llvm.bswap.i128(i128 %a)
ret i128 %tmp
}