; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple aarch64 -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
; RUN: llc < %s -mtriple aarch64 -o - -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
define <8 x i16> @extadds_v8i8_i16(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-LABEL: extadds_v8i8_i16:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: saddl v0.8h, v0.8b, v1.8b
; CHECK-NEXT: ret
entry:
%s0s = sext <8 x i8> %s0 to <8 x i16>
%s1s = sext <8 x i8> %s1 to <8 x i16>
%m = add <8 x i16> %s0s, %s1s
ret <8 x i16> %m
}
define <8 x i16> @extaddu_v8i8_i16(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-LABEL: extaddu_v8i8_i16:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-NEXT: ret
entry:
%s0s = zext <8 x i8> %s0 to <8 x i16>
%s1s = zext <8 x i8> %s1 to <8 x i16>
%m = add <8 x i16> %s0s, %s1s
ret <8 x i16> %m
}
define <16 x i16> @extadds_v16i8_i16(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extadds_v16i8_i16:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v2.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: saddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v16i8_i16:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: saddl2 v1.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <16 x i8> %s0 to <16 x i16>
%s1s = sext <16 x i8> %s1 to <16 x i16>
%m = add <16 x i16> %s0s, %s1s
ret <16 x i16> %m
}
define <16 x i16> @extaddu_v16i8_i16(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extaddu_v16i8_i16:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v2.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v16i8_i16:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: uaddl2 v1.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i16>
%s1s = zext <16 x i8> %s1 to <16 x i16>
%m = add <16 x i16> %s0s, %s1s
ret <16 x i16> %m
}
define <32 x i16> @extadds_v32i8_i16(<32 x i8> %s0, <32 x i8> %s1) {
; CHECK-SD-LABEL: extadds_v32i8_i16:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v4.8h, v1.16b, v3.16b
; CHECK-SD-NEXT: saddl v5.8h, v0.8b, v2.8b
; CHECK-SD-NEXT: saddl2 v6.8h, v0.16b, v2.16b
; CHECK-SD-NEXT: saddl v2.8h, v1.8b, v3.8b
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v32i8_i16:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v4.8h, v0.8b, v2.8b
; CHECK-GI-NEXT: saddl2 v5.8h, v0.16b, v2.16b
; CHECK-GI-NEXT: saddl v2.8h, v1.8b, v3.8b
; CHECK-GI-NEXT: saddl2 v3.8h, v1.16b, v3.16b
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <32 x i8> %s0 to <32 x i16>
%s1s = sext <32 x i8> %s1 to <32 x i16>
%m = add <32 x i16> %s0s, %s1s
ret <32 x i16> %m
}
define <32 x i16> @extaddu_v32i8_i16(<32 x i8> %s0, <32 x i8> %s1) {
; CHECK-SD-LABEL: extaddu_v32i8_i16:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v4.8h, v1.16b, v3.16b
; CHECK-SD-NEXT: uaddl v5.8h, v0.8b, v2.8b
; CHECK-SD-NEXT: uaddl2 v6.8h, v0.16b, v2.16b
; CHECK-SD-NEXT: uaddl v2.8h, v1.8b, v3.8b
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v32i8_i16:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v4.8h, v0.8b, v2.8b
; CHECK-GI-NEXT: uaddl2 v5.8h, v0.16b, v2.16b
; CHECK-GI-NEXT: uaddl v2.8h, v1.8b, v3.8b
; CHECK-GI-NEXT: uaddl2 v3.8h, v1.16b, v3.16b
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <32 x i8> %s0 to <32 x i16>
%s1s = zext <32 x i8> %s1 to <32 x i16>
%m = add <32 x i16> %s0s, %s1s
ret <32 x i16> %m
}
define <8 x i32> @extadds_v8i8_i32(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extadds_v8i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v8i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v1.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v0.4s, v1.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i8> %s0 to <8 x i32>
%s1s = sext <8 x i8> %s1 to <8 x i32>
%m = add <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <8 x i32> @extaddu_v8i8_i32(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extaddu_v8i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ushll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT: ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v8i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v1.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: ushll v0.4s, v1.4h, #0
; CHECK-GI-NEXT: ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i8> %s0 to <8 x i32>
%s1s = zext <8 x i8> %s1 to <8 x i32>
%m = add <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <8 x i32> @extsubs_v8i8_i32(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extsubs_v8i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ssubl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v8i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ssubl v1.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v0.4s, v1.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i8> %s0 to <8 x i32>
%s1s = sext <8 x i8> %s1 to <8 x i32>
%m = sub <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <8 x i32> @extsubu_v8i8_i32(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extsubu_v8i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: usubl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v8i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: usubl v1.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v0.4s, v1.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i8> %s0 to <8 x i32>
%s1s = zext <8 x i8> %s1 to <8 x i32>
%m = sub <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <16 x i32> @extadds_v16i8_i32(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extadds_v16i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: saddl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v16i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: saddl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <16 x i8> %s0 to <16 x i32>
%s1s = sext <16 x i8> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @extaddu_v16i8_i32(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extaddu_v16i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: uaddl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: ushll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: ushll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: ushll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: ushll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v16i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: uaddl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: ushll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: ushll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: ushll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: ushll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i32>
%s1s = zext <16 x i8> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @extsubs_v16i8_i32(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extsubs_v16i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ssubl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v16i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: ssubl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <16 x i8> %s0 to <16 x i32>
%s1s = sext <16 x i8> %s1 to <16 x i32>
%m = sub <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @extsubu_v16i8_i32(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: extsubu_v16i8_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: usubl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v16i8_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: usubl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i32>
%s1s = zext <16 x i8> %s1 to <16 x i32>
%m = sub <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <8 x i64> @extadds_v8i8_i64(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extadds_v8i8_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v8i8_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v0.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i8> %s0 to <8 x i64>
%s1s = sext <8 x i8> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extaddu_v8i8_i64(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extaddu_v8i8_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ushll v1.4s, v0.4h, #0
; CHECK-SD-NEXT: ushll2 v2.4s, v0.8h, #0
; CHECK-SD-NEXT: ushll v0.2d, v1.2s, #0
; CHECK-SD-NEXT: ushll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: ushll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT: ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v8i8_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT: ushll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT: ushll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: ushll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i8> %s0 to <8 x i64>
%s1s = zext <8 x i8> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extsubs_v8i8_i64(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extsubs_v8i8_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ssubl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v8i8_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ssubl v0.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i8> %s0 to <8 x i64>
%s1s = sext <8 x i8> %s1 to <8 x i64>
%m = sub <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extsubu_v8i8_i64(<8 x i8> %s0, <8 x i8> %s1) {
; CHECK-SD-LABEL: extsubu_v8i8_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: usubl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v8i8_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: usubl v0.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i8> %s0 to <8 x i64>
%s1s = zext <8 x i8> %s1 to <8 x i64>
%m = sub <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <16 x i64> @extaddu_v16i8_i64(<16 x i8> %a, <16 x i8> %b) {
; CHECK-SD-LABEL: extaddu_v16i8_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: uaddl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: uaddl2 v0.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: ushll v3.4s, v2.4h, #0
; CHECK-SD-NEXT: ushll2 v2.4s, v2.8h, #0
; CHECK-SD-NEXT: ushll v5.4s, v0.4h, #0
; CHECK-SD-NEXT: ushll2 v6.4s, v0.8h, #0
; CHECK-SD-NEXT: ushll2 v1.2d, v3.4s, #0
; CHECK-SD-NEXT: ushll v0.2d, v3.2s, #0
; CHECK-SD-NEXT: ushll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ushll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: ushll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: ushll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v16i8_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: uaddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: uaddl2 v0.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: ushll v1.4s, v2.4h, #0
; CHECK-GI-NEXT: ushll2 v3.4s, v2.8h, #0
; CHECK-GI-NEXT: ushll v5.4s, v0.4h, #0
; CHECK-GI-NEXT: ushll2 v7.4s, v0.8h, #0
; CHECK-GI-NEXT: ushll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: ushll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ushll v4.2d, v5.2s, #0
; CHECK-GI-NEXT: ushll2 v5.2d, v5.4s, #0
; CHECK-GI-NEXT: ushll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: ushll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = zext <16 x i8> %a to <16 x i64>
%d = zext <16 x i8> %b to <16 x i64>
%e = add <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extadds_v16i8_i64(<16 x i8> %a, <16 x i8> %b) {
; CHECK-SD-LABEL: extadds_v16i8_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: saddl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: saddl2 v0.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v3.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v6.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v3.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v3.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v16i8_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: saddl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: saddl2 v0.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v1.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v7.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = sext <16 x i8> %a to <16 x i64>
%d = sext <16 x i8> %b to <16 x i64>
%e = add <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extsubu_v16i8_i64(<16 x i8> %a, <16 x i8> %b) {
; CHECK-SD-LABEL: extsubu_v16i8_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: usubl2 v0.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v3.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v6.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v3.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v3.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v16i8_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: usubl2 v0.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v1.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v7.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = zext <16 x i8> %a to <16 x i64>
%d = zext <16 x i8> %b to <16 x i64>
%e = sub <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extsubs_v16i8_i64(<16 x i8> %a, <16 x i8> %b) {
; CHECK-SD-LABEL: extsubs_v16i8_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ssubl2 v0.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v3.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v2.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-SD-NEXT: sshll2 v6.4s, v0.8h, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v3.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v3.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v16i8_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: ssubl2 v0.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v1.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v5.4s, v0.4h, #0
; CHECK-GI-NEXT: sshll2 v7.4s, v0.8h, #0
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = sext <16 x i8> %a to <16 x i64>
%d = sext <16 x i8> %b to <16 x i64>
%e = sub <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extaddu_v16i16_i64(<16 x i16> %a, <16 x i16> %b) {
; CHECK-SD-LABEL: extaddu_v16i16_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: uaddl v5.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: uaddl v4.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: uaddl2 v2.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: uaddl2 v6.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: ushll2 v1.2d, v4.4s, #0
; CHECK-SD-NEXT: ushll v0.2d, v4.2s, #0
; CHECK-SD-NEXT: ushll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: ushll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: ushll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: ushll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v16i16_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: uaddl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: uaddl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: uaddl v6.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: uaddl2 v7.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: ushll v0.2d, v4.2s, #0
; CHECK-GI-NEXT: ushll2 v1.2d, v4.4s, #0
; CHECK-GI-NEXT: ushll v2.2d, v5.2s, #0
; CHECK-GI-NEXT: ushll2 v3.2d, v5.4s, #0
; CHECK-GI-NEXT: ushll v4.2d, v6.2s, #0
; CHECK-GI-NEXT: ushll2 v5.2d, v6.4s, #0
; CHECK-GI-NEXT: ushll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: ushll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = zext <16 x i16> %a to <16 x i64>
%d = zext <16 x i16> %b to <16 x i64>
%e = add <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extadds_v16i16_i64(<16 x i16> %a, <16 x i16> %b) {
; CHECK-SD-LABEL: extadds_v16i16_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: saddl v5.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: saddl v4.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: saddl2 v2.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: saddl2 v6.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v16i16_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: saddl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: saddl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: saddl v6.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: saddl2 v7.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v6.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v6.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = sext <16 x i16> %a to <16 x i64>
%d = sext <16 x i16> %b to <16 x i64>
%e = add <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extsubu_v16i16_i64(<16 x i16> %a, <16 x i16> %b) {
; CHECK-SD-LABEL: extsubu_v16i16_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: usubl v5.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: usubl v4.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: usubl2 v2.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: usubl2 v6.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v16i16_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: usubl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: usubl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: usubl v6.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: usubl2 v7.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v6.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v6.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = zext <16 x i16> %a to <16 x i64>
%d = zext <16 x i16> %b to <16 x i64>
%e = sub <16 x i64> %c, %d
ret <16 x i64> %e
}
define <16 x i64> @extsubs_v16i16_i64(<16 x i16> %a, <16 x i16> %b) {
; CHECK-SD-LABEL: extsubs_v16i16_i64:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ssubl v5.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: ssubl v4.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: ssubl2 v2.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: ssubl2 v6.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll v4.2d, v5.2s, #0
; CHECK-SD-NEXT: sshll2 v7.2d, v6.4s, #0
; CHECK-SD-NEXT: sshll2 v5.2d, v5.4s, #0
; CHECK-SD-NEXT: sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v16i16_i64:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: ssubl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: ssubl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: ssubl v6.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: ssubl2 v7.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: sshll v0.2d, v4.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v4.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v5.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v5.4s, #0
; CHECK-GI-NEXT: sshll v4.2d, v6.2s, #0
; CHECK-GI-NEXT: sshll2 v5.2d, v6.4s, #0
; CHECK-GI-NEXT: sshll v6.2d, v7.2s, #0
; CHECK-GI-NEXT: sshll2 v7.2d, v7.4s, #0
; CHECK-GI-NEXT: ret
%c = sext <16 x i16> %a to <16 x i64>
%d = sext <16 x i16> %b to <16 x i64>
%e = sub <16 x i64> %c, %d
ret <16 x i64> %e
}
define <4 x i32> @extadds_v4i16_i32(<4 x i16> %s0, <4 x i16> %s1) {
; CHECK-LABEL: extadds_v4i16_i32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: saddl v0.4s, v0.4h, v1.4h
; CHECK-NEXT: ret
entry:
%s0s = sext <4 x i16> %s0 to <4 x i32>
%s1s = sext <4 x i16> %s1 to <4 x i32>
%m = add <4 x i32> %s0s, %s1s
ret <4 x i32> %m
}
define <4 x i32> @extaddu_v4i16_i32(<4 x i16> %s0, <4 x i16> %s1) {
; CHECK-LABEL: extaddu_v4i16_i32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: uaddl v0.4s, v0.4h, v1.4h
; CHECK-NEXT: ret
entry:
%s0s = zext <4 x i16> %s0 to <4 x i32>
%s1s = zext <4 x i16> %s1 to <4 x i32>
%m = add <4 x i32> %s0s, %s1s
ret <4 x i32> %m
}
define <8 x i32> @extadds_v8i16_i32(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extadds_v8i16_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v2.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: saddl v0.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v8i16_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: saddl2 v1.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i16> %s0 to <8 x i32>
%s1s = sext <8 x i16> %s1 to <8 x i32>
%m = add <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <8 x i32> @extaddu_v8i16_i32(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extaddu_v8i16_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v2.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: uaddl v0.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v8i16_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: uaddl2 v1.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i16> %s0 to <8 x i32>
%s1s = zext <8 x i16> %s1 to <8 x i32>
%m = add <8 x i32> %s0s, %s1s
ret <8 x i32> %m
}
define <16 x i32> @extadds_v16i16_i32(<16 x i16> %s0, <16 x i16> %s1) {
; CHECK-SD-LABEL: extadds_v16i16_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v4.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: saddl v5.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: saddl2 v6.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: saddl v2.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v16i16_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: saddl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: saddl v2.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: saddl2 v3.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <16 x i16> %s0 to <16 x i32>
%s1s = sext <16 x i16> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @extaddu_v16i16_i32(<16 x i16> %s0, <16 x i16> %s1) {
; CHECK-SD-LABEL: extaddu_v16i16_i32:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v4.4s, v1.8h, v3.8h
; CHECK-SD-NEXT: uaddl v5.4s, v0.4h, v2.4h
; CHECK-SD-NEXT: uaddl2 v6.4s, v0.8h, v2.8h
; CHECK-SD-NEXT: uaddl v2.4s, v1.4h, v3.4h
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v16i16_i32:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v4.4s, v0.4h, v2.4h
; CHECK-GI-NEXT: uaddl2 v5.4s, v0.8h, v2.8h
; CHECK-GI-NEXT: uaddl v2.4s, v1.4h, v3.4h
; CHECK-GI-NEXT: uaddl2 v3.4s, v1.8h, v3.8h
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i16> %s0 to <16 x i32>
%s1s = zext <16 x i16> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <4 x i64> @extadds_v4i16_i64(<4 x i16> %s0, <4 x i16> %s1) {
; CHECK-SD-LABEL: extadds_v4i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl v0.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: sshll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT: sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v4i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <4 x i16> %s0 to <4 x i64>
%s1s = sext <4 x i16> %s1 to <4 x i64>
%m = add <4 x i64> %s0s, %s1s
ret <4 x i64> %m
}
define <4 x i64> @extaddu_v4i16_i64(<4 x i16> %s0, <4 x i16> %s1) {
; CHECK-SD-LABEL: extaddu_v4i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl v0.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: ushll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT: ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v4i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: ushll v0.2d, v1.2s, #0
; CHECK-GI-NEXT: ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <4 x i16> %s0 to <4 x i64>
%s1s = zext <4 x i16> %s1 to <4 x i64>
%m = add <4 x i64> %s0s, %s1s
ret <4 x i64> %m
}
define <8 x i64> @extadds_v8i16_i64(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extadds_v8i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl v2.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: saddl2 v4.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v4.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v8i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: saddl2 v3.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i16> %s0 to <8 x i64>
%s1s = sext <8 x i16> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extaddu_v8i16_i64(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extaddu_v8i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl v2.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: uaddl2 v4.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: ushll v0.2d, v2.2s, #0
; CHECK-SD-NEXT: ushll2 v3.2d, v4.4s, #0
; CHECK-SD-NEXT: ushll2 v1.2d, v2.4s, #0
; CHECK-SD-NEXT: ushll v2.2d, v4.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v8i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: uaddl2 v3.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: ushll v0.2d, v2.2s, #0
; CHECK-GI-NEXT: ushll2 v1.2d, v2.4s, #0
; CHECK-GI-NEXT: ushll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i16> %s0 to <8 x i64>
%s1s = zext <8 x i16> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extsubs_v8i16_i64(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extsubs_v8i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ssubl v2.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: ssubl2 v4.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v4.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubs_v8i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ssubl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: ssubl2 v3.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i16> %s0 to <8 x i64>
%s1s = sext <8 x i16> %s1 to <8 x i64>
%m = sub <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extsubu_v8i16_i64(<8 x i16> %s0, <8 x i16> %s1) {
; CHECK-SD-LABEL: extsubu_v8i16_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: usubl v2.4s, v0.4h, v1.4h
; CHECK-SD-NEXT: usubl2 v4.4s, v0.8h, v1.8h
; CHECK-SD-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-SD-NEXT: sshll2 v3.2d, v4.4s, #0
; CHECK-SD-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-SD-NEXT: sshll v2.2d, v4.2s, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extsubu_v8i16_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: usubl v2.4s, v0.4h, v1.4h
; CHECK-GI-NEXT: usubl2 v3.4s, v0.8h, v1.8h
; CHECK-GI-NEXT: sshll v0.2d, v2.2s, #0
; CHECK-GI-NEXT: sshll2 v1.2d, v2.4s, #0
; CHECK-GI-NEXT: sshll v2.2d, v3.2s, #0
; CHECK-GI-NEXT: sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i16> %s0 to <8 x i64>
%s1s = zext <8 x i16> %s1 to <8 x i64>
%m = sub <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <2 x i64> @extadds_v2i32_i64(<2 x i32> %s0, <2 x i32> %s1) {
; CHECK-LABEL: extadds_v2i32_i64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: saddl v0.2d, v0.2s, v1.2s
; CHECK-NEXT: ret
entry:
%s0s = sext <2 x i32> %s0 to <2 x i64>
%s1s = sext <2 x i32> %s1 to <2 x i64>
%m = add <2 x i64> %s0s, %s1s
ret <2 x i64> %m
}
define <2 x i64> @extaddu_v2i32_i64(<2 x i32> %s0, <2 x i32> %s1) {
; CHECK-LABEL: extaddu_v2i32_i64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: uaddl v0.2d, v0.2s, v1.2s
; CHECK-NEXT: ret
entry:
%s0s = zext <2 x i32> %s0 to <2 x i64>
%s1s = zext <2 x i32> %s1 to <2 x i64>
%m = add <2 x i64> %s0s, %s1s
ret <2 x i64> %m
}
define <4 x i64> @extadds_v4i32_i64(<4 x i32> %s0, <4 x i32> %s1) {
; CHECK-SD-LABEL: extadds_v4i32_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v2.2d, v0.4s, v1.4s
; CHECK-SD-NEXT: saddl v0.2d, v0.2s, v1.2s
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v4i32_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v2.2d, v0.2s, v1.2s
; CHECK-GI-NEXT: saddl2 v1.2d, v0.4s, v1.4s
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <4 x i32> %s0 to <4 x i64>
%s1s = sext <4 x i32> %s1 to <4 x i64>
%m = add <4 x i64> %s0s, %s1s
ret <4 x i64> %m
}
define <4 x i64> @extaddu_v4i32_i64(<4 x i32> %s0, <4 x i32> %s1) {
; CHECK-SD-LABEL: extaddu_v4i32_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v2.2d, v0.4s, v1.4s
; CHECK-SD-NEXT: uaddl v0.2d, v0.2s, v1.2s
; CHECK-SD-NEXT: mov v1.16b, v2.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v4i32_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v2.2d, v0.2s, v1.2s
; CHECK-GI-NEXT: uaddl2 v1.2d, v0.4s, v1.4s
; CHECK-GI-NEXT: mov v0.16b, v2.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <4 x i32> %s0 to <4 x i64>
%s1s = zext <4 x i32> %s1 to <4 x i64>
%m = add <4 x i64> %s0s, %s1s
ret <4 x i64> %m
}
define <8 x i64> @extadds_v8i32_i64(<8 x i32> %s0, <8 x i32> %s1) {
; CHECK-SD-LABEL: extadds_v8i32_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: saddl2 v4.2d, v1.4s, v3.4s
; CHECK-SD-NEXT: saddl v5.2d, v0.2s, v2.2s
; CHECK-SD-NEXT: saddl2 v6.2d, v0.4s, v2.4s
; CHECK-SD-NEXT: saddl v2.2d, v1.2s, v3.2s
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extadds_v8i32_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: saddl v4.2d, v0.2s, v2.2s
; CHECK-GI-NEXT: saddl2 v5.2d, v0.4s, v2.4s
; CHECK-GI-NEXT: saddl v2.2d, v1.2s, v3.2s
; CHECK-GI-NEXT: saddl2 v3.2d, v1.4s, v3.4s
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <8 x i32> %s0 to <8 x i64>
%s1s = sext <8 x i32> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <8 x i64> @extaddu_v8i32_i64(<8 x i32> %s0, <8 x i32> %s1) {
; CHECK-SD-LABEL: extaddu_v8i32_i64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: uaddl2 v4.2d, v1.4s, v3.4s
; CHECK-SD-NEXT: uaddl v5.2d, v0.2s, v2.2s
; CHECK-SD-NEXT: uaddl2 v6.2d, v0.4s, v2.4s
; CHECK-SD-NEXT: uaddl v2.2d, v1.2s, v3.2s
; CHECK-SD-NEXT: mov v0.16b, v5.16b
; CHECK-SD-NEXT: mov v1.16b, v6.16b
; CHECK-SD-NEXT: mov v3.16b, v4.16b
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: extaddu_v8i32_i64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: uaddl v4.2d, v0.2s, v2.2s
; CHECK-GI-NEXT: uaddl2 v5.2d, v0.4s, v2.4s
; CHECK-GI-NEXT: uaddl v2.2d, v1.2s, v3.2s
; CHECK-GI-NEXT: uaddl2 v3.2d, v1.4s, v3.4s
; CHECK-GI-NEXT: mov v0.16b, v4.16b
; CHECK-GI-NEXT: mov v1.16b, v5.16b
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <8 x i32> %s0 to <8 x i64>
%s1s = zext <8 x i32> %s1 to <8 x i64>
%m = add <8 x i64> %s0s, %s1s
ret <8 x i64> %m
}
define <16 x i32> @add_zs(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: add_zs:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ushll v2.8h, v0.8b, #0
; CHECK-SD-NEXT: ushll2 v0.8h, v0.16b, #0
; CHECK-SD-NEXT: saddw v2.8h, v2.8h, v1.8b
; CHECK-SD-NEXT: saddw2 v4.8h, v0.8h, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: add_zs:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: sshll v2.8h, v1.8b, #0
; CHECK-GI-NEXT: sshll2 v1.8h, v1.16b, #0
; CHECK-GI-NEXT: ushll v3.8h, v0.8b, #0
; CHECK-GI-NEXT: ushll2 v4.8h, v0.16b, #0
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v5.4s, v1.4h, #0
; CHECK-GI-NEXT: sshll2 v6.4s, v1.8h, #0
; CHECK-GI-NEXT: uaddw v0.4s, v0.4s, v3.4h
; CHECK-GI-NEXT: uaddw2 v1.4s, v2.4s, v3.8h
; CHECK-GI-NEXT: uaddw v2.4s, v5.4s, v4.4h
; CHECK-GI-NEXT: uaddw2 v3.4s, v6.4s, v4.8h
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i32>
%s1s = sext <16 x i8> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <20 x i32> @v20(<20 x i8> %s0, <20 x i8> %s1) {
; CHECK-SD-LABEL: v20:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: fmov s0, w0
; CHECK-SD-NEXT: ldr b2, [sp, #160]
; CHECK-SD-NEXT: add x10, sp, #168
; CHECK-SD-NEXT: ldr b3, [sp]
; CHECK-SD-NEXT: add x11, sp, #8
; CHECK-SD-NEXT: ldr b1, [sp, #96]
; CHECK-SD-NEXT: ld1 { v2.b }[1], [x10]
; CHECK-SD-NEXT: add x9, sp, #104
; CHECK-SD-NEXT: add x10, sp, #176
; CHECK-SD-NEXT: mov v0.b[1], w1
; CHECK-SD-NEXT: ld1 { v3.b }[1], [x11]
; CHECK-SD-NEXT: ld1 { v1.b }[1], [x9]
; CHECK-SD-NEXT: add x11, sp, #16
; CHECK-SD-NEXT: add x9, sp, #112
; CHECK-SD-NEXT: add x13, sp, #184
; CHECK-SD-NEXT: ld1 { v2.b }[2], [x10]
; CHECK-SD-NEXT: add x12, sp, #120
; CHECK-SD-NEXT: add x14, sp, #32
; CHECK-SD-NEXT: ld1 { v3.b }[2], [x11]
; CHECK-SD-NEXT: ld1 { v1.b }[2], [x9]
; CHECK-SD-NEXT: ldr b5, [sp, #64]
; CHECK-SD-NEXT: mov v0.b[2], w2
; CHECK-SD-NEXT: ldr b4, [sp, #224]
; CHECK-SD-NEXT: add x11, sp, #128
; CHECK-SD-NEXT: ld1 { v2.b }[3], [x13]
; CHECK-SD-NEXT: add x13, sp, #24
; CHECK-SD-NEXT: add x10, sp, #136
; CHECK-SD-NEXT: ld1 { v3.b }[3], [x13]
; CHECK-SD-NEXT: ld1 { v1.b }[3], [x12]
; CHECK-SD-NEXT: add x12, sp, #192
; CHECK-SD-NEXT: add x13, sp, #200
; CHECK-SD-NEXT: add x15, sp, #80
; CHECK-SD-NEXT: add x9, sp, #144
; CHECK-SD-NEXT: mov v0.b[3], w3
; CHECK-SD-NEXT: ld1 { v2.b }[4], [x12]
; CHECK-SD-NEXT: add x12, sp, #232
; CHECK-SD-NEXT: ld1 { v3.b }[4], [x14]
; CHECK-SD-NEXT: add x14, sp, #72
; CHECK-SD-NEXT: ld1 { v4.b }[1], [x12]
; CHECK-SD-NEXT: ld1 { v5.b }[1], [x14]
; CHECK-SD-NEXT: add x14, sp, #40
; CHECK-SD-NEXT: ld1 { v1.b }[4], [x11]
; CHECK-SD-NEXT: ld1 { v2.b }[5], [x13]
; CHECK-SD-NEXT: add x12, sp, #208
; CHECK-SD-NEXT: add x13, sp, #48
; CHECK-SD-NEXT: mov v0.b[4], w4
; CHECK-SD-NEXT: ld1 { v3.b }[5], [x14]
; CHECK-SD-NEXT: add x14, sp, #240
; CHECK-SD-NEXT: ld1 { v4.b }[2], [x14]
; CHECK-SD-NEXT: ld1 { v5.b }[2], [x15]
; CHECK-SD-NEXT: ld1 { v1.b }[5], [x10]
; CHECK-SD-NEXT: ld1 { v2.b }[6], [x12]
; CHECK-SD-NEXT: add x11, sp, #216
; CHECK-SD-NEXT: add x10, sp, #56
; CHECK-SD-NEXT: ld1 { v3.b }[6], [x13]
; CHECK-SD-NEXT: add x12, sp, #248
; CHECK-SD-NEXT: add x13, sp, #88
; CHECK-SD-NEXT: mov v0.b[5], w5
; CHECK-SD-NEXT: ld1 { v4.b }[3], [x12]
; CHECK-SD-NEXT: ld1 { v5.b }[3], [x13]
; CHECK-SD-NEXT: ld1 { v1.b }[6], [x9]
; CHECK-SD-NEXT: ld1 { v2.b }[7], [x11]
; CHECK-SD-NEXT: add x9, sp, #152
; CHECK-SD-NEXT: ld1 { v3.b }[7], [x10]
; CHECK-SD-NEXT: uaddl v4.8h, v5.8b, v4.8b
; CHECK-SD-NEXT: mov v0.b[6], w6
; CHECK-SD-NEXT: ld1 { v1.b }[7], [x9]
; CHECK-SD-NEXT: uaddl v2.8h, v3.8b, v2.8b
; CHECK-SD-NEXT: ushll v3.4s, v4.4h, #0
; CHECK-SD-NEXT: mov v0.b[7], w7
; CHECK-SD-NEXT: uaddl v0.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ushll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: ushll v2.4s, v2.4h, #0
; CHECK-SD-NEXT: stp q1, q3, [x8, #48]
; CHECK-SD-NEXT: ushll2 v3.4s, v0.8h, #0
; CHECK-SD-NEXT: ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: stp q3, q2, [x8, #16]
; CHECK-SD-NEXT: str q0, [x8]
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: v20:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ldr w9, [sp, #64]
; CHECK-GI-NEXT: ldr w10, [sp, #72]
; CHECK-GI-NEXT: and w13, w2, #0xff
; CHECK-GI-NEXT: ldr w11, [sp, #80]
; CHECK-GI-NEXT: ldr w12, [sp, #88]
; CHECK-GI-NEXT: fmov s19, w13
; CHECK-GI-NEXT: fmov s0, w9
; CHECK-GI-NEXT: ldr w9, [sp, #224]
; CHECK-GI-NEXT: fmov s16, w10
; CHECK-GI-NEXT: ldr w10, [sp, #232]
; CHECK-GI-NEXT: fmov s3, w11
; CHECK-GI-NEXT: ldr w11, [sp, #240]
; CHECK-GI-NEXT: fmov s2, w9
; CHECK-GI-NEXT: ldr w9, [sp, #248]
; CHECK-GI-NEXT: fmov s1, w12
; CHECK-GI-NEXT: fmov s7, w10
; CHECK-GI-NEXT: and w10, w1, #0xff
; CHECK-GI-NEXT: fmov s5, w11
; CHECK-GI-NEXT: fmov s4, w9
; CHECK-GI-NEXT: and w9, w0, #0xff
; CHECK-GI-NEXT: ldrb w11, [sp]
; CHECK-GI-NEXT: ldrb w12, [sp, #8]
; CHECK-GI-NEXT: fmov s6, w9
; CHECK-GI-NEXT: fmov s20, w10
; CHECK-GI-NEXT: ldrb w9, [sp, #96]
; CHECK-GI-NEXT: ldrb w10, [sp, #104]
; CHECK-GI-NEXT: fmov s17, w11
; CHECK-GI-NEXT: fmov s21, w12
; CHECK-GI-NEXT: ldrb w11, [sp, #160]
; CHECK-GI-NEXT: mov v0.b[1], v16.b[0]
; CHECK-GI-NEXT: fmov s18, w9
; CHECK-GI-NEXT: fmov s22, w10
; CHECK-GI-NEXT: ldrb w9, [sp, #168]
; CHECK-GI-NEXT: mov v6.h[1], v20.h[0]
; CHECK-GI-NEXT: fmov s20, w11
; CHECK-GI-NEXT: ldrb w10, [sp, #16]
; CHECK-GI-NEXT: mov v17.h[1], v21.h[0]
; CHECK-GI-NEXT: fmov s21, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #112]
; CHECK-GI-NEXT: mov v18.h[1], v22.h[0]
; CHECK-GI-NEXT: fmov s23, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #176]
; CHECK-GI-NEXT: and w11, w3, #0xff
; CHECK-GI-NEXT: mov v2.b[1], v7.b[0]
; CHECK-GI-NEXT: mov v0.b[2], v3.b[0]
; CHECK-GI-NEXT: mov v6.h[2], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w9
; CHECK-GI-NEXT: mov v20.h[1], v21.h[0]
; CHECK-GI-NEXT: ldrb w9, [sp, #24]
; CHECK-GI-NEXT: fmov s22, w11
; CHECK-GI-NEXT: mov v17.h[2], v23.h[0]
; CHECK-GI-NEXT: and w11, w4, #0xff
; CHECK-GI-NEXT: mov v18.h[2], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #120]
; CHECK-GI-NEXT: fmov s23, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #184]
; CHECK-GI-NEXT: mov v6.h[3], v22.h[0]
; CHECK-GI-NEXT: fmov s21, w11
; CHECK-GI-NEXT: and w11, w6, #0xff
; CHECK-GI-NEXT: mov v2.b[2], v5.b[0]
; CHECK-GI-NEXT: mov v20.h[2], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w10
; CHECK-GI-NEXT: fmov s16, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #128]
; CHECK-GI-NEXT: and w10, w5, #0xff
; CHECK-GI-NEXT: mov v17.h[3], v23.h[0]
; CHECK-GI-NEXT: mov v6.h[4], v21.h[0]
; CHECK-GI-NEXT: mov v0.b[3], v1.b[0]
; CHECK-GI-NEXT: mov v18.h[3], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #192]
; CHECK-GI-NEXT: mov v20.h[3], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #32]
; CHECK-GI-NEXT: mov v2.b[3], v4.b[0]
; CHECK-GI-NEXT: ushll v0.8h, v0.8b, #0
; CHECK-GI-NEXT: mov v18.h[4], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #136]
; CHECK-GI-NEXT: mov v6.h[5], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #48]
; CHECK-GI-NEXT: ushll v2.8h, v2.8b, #0
; CHECK-GI-NEXT: mov v17.h[4], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #40]
; CHECK-GI-NEXT: mov v18.h[5], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w9
; CHECK-GI-NEXT: ldrb w9, [sp, #144]
; CHECK-GI-NEXT: mov v20.h[4], v19.h[0]
; CHECK-GI-NEXT: fmov s19, w11
; CHECK-GI-NEXT: ldrb w11, [sp, #200]
; CHECK-GI-NEXT: add v0.4h, v0.4h, v2.4h
; CHECK-GI-NEXT: fmov s7, w11
; CHECK-GI-NEXT: mov v17.h[5], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w9
; CHECK-GI-NEXT: ldrb w11, [sp, #208]
; CHECK-GI-NEXT: mov v6.h[6], v19.h[0]
; CHECK-GI-NEXT: ldrb w9, [sp, #56]
; CHECK-GI-NEXT: ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT: mov v20.h[5], v7.h[0]
; CHECK-GI-NEXT: fmov s7, w10
; CHECK-GI-NEXT: mov v18.h[6], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w11
; CHECK-GI-NEXT: ldrb w10, [sp, #152]
; CHECK-GI-NEXT: and w11, w7, #0xff
; CHECK-GI-NEXT: fmov s3, w11
; CHECK-GI-NEXT: str q0, [x8, #64]
; CHECK-GI-NEXT: fmov s5, w10
; CHECK-GI-NEXT: ldrb w10, [sp, #216]
; CHECK-GI-NEXT: mov v17.h[6], v7.h[0]
; CHECK-GI-NEXT: mov v20.h[6], v16.h[0]
; CHECK-GI-NEXT: fmov s7, w9
; CHECK-GI-NEXT: mov v6.h[7], v3.h[0]
; CHECK-GI-NEXT: fmov s3, w10
; CHECK-GI-NEXT: mov v18.h[7], v5.h[0]
; CHECK-GI-NEXT: mov v17.h[7], v7.h[0]
; CHECK-GI-NEXT: mov v20.h[7], v3.h[0]
; CHECK-GI-NEXT: add v1.8h, v6.8h, v18.8h
; CHECK-GI-NEXT: add v3.8h, v17.8h, v20.8h
; CHECK-GI-NEXT: ushll v4.4s, v1.4h, #0
; CHECK-GI-NEXT: ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT: ushll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: ushll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: stp q4, q1, [x8]
; CHECK-GI-NEXT: stp q2, q3, [x8, #32]
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <20 x i8> %s0 to <20 x i32>
%s1s = zext <20 x i8> %s1 to <20 x i32>
%m = add <20 x i32> %s0s, %s1s
ret <20 x i32> %m
}
define <16 x i32> @i12(<16 x i12> %s0, <16 x i12> %s1) {
; CHECK-SD-LABEL: i12:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: str x23, [sp, #-48]! // 8-byte Folded Spill
; CHECK-SD-NEXT: stp x22, x21, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT: stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT: .cfi_def_cfa_offset 48
; CHECK-SD-NEXT: .cfi_offset w19, -8
; CHECK-SD-NEXT: .cfi_offset w20, -16
; CHECK-SD-NEXT: .cfi_offset w21, -24
; CHECK-SD-NEXT: .cfi_offset w22, -32
; CHECK-SD-NEXT: .cfi_offset w23, -48
; CHECK-SD-NEXT: ldr w13, [sp, #112]
; CHECK-SD-NEXT: ldr w14, [sp, #144]
; CHECK-SD-NEXT: fmov s2, w4
; CHECK-SD-NEXT: ldr w17, [sp, #176]
; CHECK-SD-NEXT: ldr w19, [sp, #208]
; CHECK-SD-NEXT: fmov s3, w0
; CHECK-SD-NEXT: ldr w20, [sp, #80]
; CHECK-SD-NEXT: ldr w21, [sp, #48]
; CHECK-SD-NEXT: fmov s5, w13
; CHECK-SD-NEXT: fmov s4, w19
; CHECK-SD-NEXT: fmov s6, w17
; CHECK-SD-NEXT: fmov s7, w14
; CHECK-SD-NEXT: fmov s0, w20
; CHECK-SD-NEXT: fmov s1, w21
; CHECK-SD-NEXT: ldr w10, [sp, #120]
; CHECK-SD-NEXT: ldr w11, [sp, #152]
; CHECK-SD-NEXT: ldr w12, [sp, #184]
; CHECK-SD-NEXT: ldr w15, [sp, #216]
; CHECK-SD-NEXT: ldr w22, [sp, #88]
; CHECK-SD-NEXT: ldr w23, [sp, #56]
; CHECK-SD-NEXT: mov v2.h[1], w5
; CHECK-SD-NEXT: mov v3.h[1], w1
; CHECK-SD-NEXT: mov v5.h[1], w10
; CHECK-SD-NEXT: mov v4.h[1], w15
; CHECK-SD-NEXT: mov v0.h[1], w22
; CHECK-SD-NEXT: mov v1.h[1], w23
; CHECK-SD-NEXT: mov v6.h[1], w12
; CHECK-SD-NEXT: mov v7.h[1], w11
; CHECK-SD-NEXT: ldr w8, [sp, #128]
; CHECK-SD-NEXT: ldr w9, [sp, #160]
; CHECK-SD-NEXT: ldr w16, [sp, #64]
; CHECK-SD-NEXT: ldr w18, [sp, #96]
; CHECK-SD-NEXT: ldr w10, [sp, #192]
; CHECK-SD-NEXT: ldr w11, [sp, #224]
; CHECK-SD-NEXT: mov v2.h[2], w6
; CHECK-SD-NEXT: mov v3.h[2], w2
; CHECK-SD-NEXT: mov v0.h[2], w18
; CHECK-SD-NEXT: mov v1.h[2], w16
; CHECK-SD-NEXT: mov v5.h[2], w8
; CHECK-SD-NEXT: mov v4.h[2], w11
; CHECK-SD-NEXT: mov v6.h[2], w10
; CHECK-SD-NEXT: mov v7.h[2], w9
; CHECK-SD-NEXT: ldr w12, [sp, #72]
; CHECK-SD-NEXT: ldr w13, [sp, #104]
; CHECK-SD-NEXT: ldr w8, [sp, #136]
; CHECK-SD-NEXT: ldr w9, [sp, #168]
; CHECK-SD-NEXT: ldr w10, [sp, #200]
; CHECK-SD-NEXT: ldr w11, [sp, #232]
; CHECK-SD-NEXT: mov v0.h[3], w13
; CHECK-SD-NEXT: mov v1.h[3], w12
; CHECK-SD-NEXT: mov v2.h[3], w7
; CHECK-SD-NEXT: mov v3.h[3], w3
; CHECK-SD-NEXT: mov v5.h[3], w8
; CHECK-SD-NEXT: mov v4.h[3], w11
; CHECK-SD-NEXT: mov v6.h[3], w10
; CHECK-SD-NEXT: mov v7.h[3], w9
; CHECK-SD-NEXT: movi v16.4s, #15, msl #8
; CHECK-SD-NEXT: ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT: ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT: ushll v1.4s, v1.4h, #0
; CHECK-SD-NEXT: ushll v2.4s, v2.4h, #0
; CHECK-SD-NEXT: ushll v3.4s, v3.4h, #0
; CHECK-SD-NEXT: ushll v5.4s, v5.4h, #0
; CHECK-SD-NEXT: ushll v4.4s, v4.4h, #0
; CHECK-SD-NEXT: ushll v6.4s, v6.4h, #0
; CHECK-SD-NEXT: ushll v7.4s, v7.4h, #0
; CHECK-SD-NEXT: and v17.16b, v0.16b, v16.16b
; CHECK-SD-NEXT: and v18.16b, v1.16b, v16.16b
; CHECK-SD-NEXT: and v1.16b, v2.16b, v16.16b
; CHECK-SD-NEXT: and v0.16b, v3.16b, v16.16b
; CHECK-SD-NEXT: and v2.16b, v5.16b, v16.16b
; CHECK-SD-NEXT: and v3.16b, v4.16b, v16.16b
; CHECK-SD-NEXT: and v4.16b, v6.16b, v16.16b
; CHECK-SD-NEXT: and v5.16b, v7.16b, v16.16b
; CHECK-SD-NEXT: ldp x22, x21, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NEXT: add v0.4s, v0.4s, v2.4s
; CHECK-SD-NEXT: add v3.4s, v17.4s, v3.4s
; CHECK-SD-NEXT: add v1.4s, v1.4s, v5.4s
; CHECK-SD-NEXT: add v2.4s, v18.4s, v4.4s
; CHECK-SD-NEXT: ldr x23, [sp], #48 // 8-byte Folded Reload
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: i12:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: fmov s0, w0
; CHECK-GI-NEXT: fmov s1, w1
; CHECK-GI-NEXT: ldr w8, [sp]
; CHECK-GI-NEXT: fmov s2, w5
; CHECK-GI-NEXT: ldr w9, [sp, #8]
; CHECK-GI-NEXT: ldr w11, [sp, #32]
; CHECK-GI-NEXT: ldr w12, [sp, #40]
; CHECK-GI-NEXT: fmov s5, w7
; CHECK-GI-NEXT: ldr w10, [sp, #16]
; CHECK-GI-NEXT: mov v0.h[1], v1.h[0]
; CHECK-GI-NEXT: fmov s1, w4
; CHECK-GI-NEXT: fmov s3, w9
; CHECK-GI-NEXT: fmov s4, w12
; CHECK-GI-NEXT: ldr w12, [sp, #96]
; CHECK-GI-NEXT: ldr w13, [sp, #104]
; CHECK-GI-NEXT: ldr w14, [sp, #128]
; CHECK-GI-NEXT: ldr w15, [sp, #136]
; CHECK-GI-NEXT: ldr w16, [sp, #160]
; CHECK-GI-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NEXT: fmov s2, w2
; CHECK-GI-NEXT: fmov s7, w13
; CHECK-GI-NEXT: fmov s16, w15
; CHECK-GI-NEXT: ldr w17, [sp, #168]
; CHECK-GI-NEXT: ldr w9, [sp, #24]
; CHECK-GI-NEXT: ldr w13, [sp, #176]
; CHECK-GI-NEXT: mov v0.h[2], v2.h[0]
; CHECK-GI-NEXT: fmov s2, w6
; CHECK-GI-NEXT: fmov s17, w17
; CHECK-GI-NEXT: mov v1.h[2], v2.h[0]
; CHECK-GI-NEXT: fmov s2, w8
; CHECK-GI-NEXT: ldr w8, [sp, #56]
; CHECK-GI-NEXT: mov v2.h[1], v3.h[0]
; CHECK-GI-NEXT: fmov s3, w11
; CHECK-GI-NEXT: ldr w11, [sp, #48]
; CHECK-GI-NEXT: mov v1.h[3], v5.h[0]
; CHECK-GI-NEXT: fmov s5, w10
; CHECK-GI-NEXT: ldr w10, [sp, #64]
; CHECK-GI-NEXT: mov v3.h[1], v4.h[0]
; CHECK-GI-NEXT: fmov s4, w3
; CHECK-GI-NEXT: mov v2.h[2], v5.h[0]
; CHECK-GI-NEXT: fmov s5, w11
; CHECK-GI-NEXT: ldr w11, [sp, #72]
; CHECK-GI-NEXT: ushll v1.4s, v1.4h, #0
; CHECK-GI-NEXT: fmov s6, w11
; CHECK-GI-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NEXT: fmov s4, w9
; CHECK-GI-NEXT: mov v3.h[2], v5.h[0]
; CHECK-GI-NEXT: fmov s5, w10
; CHECK-GI-NEXT: ldr w9, [sp, #80]
; CHECK-GI-NEXT: ldr w10, [sp, #112]
; CHECK-GI-NEXT: ldr w11, [sp, #144]
; CHECK-GI-NEXT: mov v2.h[3], v4.h[0]
; CHECK-GI-NEXT: mov v5.h[1], v6.h[0]
; CHECK-GI-NEXT: fmov s6, w12
; CHECK-GI-NEXT: fmov s18, w11
; CHECK-GI-NEXT: ldr w12, [sp, #88]
; CHECK-GI-NEXT: ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT: mov v6.h[1], v7.h[0]
; CHECK-GI-NEXT: fmov s7, w14
; CHECK-GI-NEXT: ushll v2.4s, v2.4h, #0
; CHECK-GI-NEXT: mov v7.h[1], v16.h[0]
; CHECK-GI-NEXT: fmov s16, w16
; CHECK-GI-NEXT: mov v16.h[1], v17.h[0]
; CHECK-GI-NEXT: fmov s17, w9
; CHECK-GI-NEXT: ldr w9, [sp, #152]
; CHECK-GI-NEXT: mov v7.h[2], v18.h[0]
; CHECK-GI-NEXT: fmov s18, w8
; CHECK-GI-NEXT: ldr w8, [sp, #120]
; CHECK-GI-NEXT: mov v5.h[2], v17.h[0]
; CHECK-GI-NEXT: fmov s17, w10
; CHECK-GI-NEXT: ldr w10, [sp, #184]
; CHECK-GI-NEXT: mov v3.h[3], v18.h[0]
; CHECK-GI-NEXT: fmov s4, w8
; CHECK-GI-NEXT: fmov s18, w10
; CHECK-GI-NEXT: mov v6.h[2], v17.h[0]
; CHECK-GI-NEXT: fmov s17, w13
; CHECK-GI-NEXT: ushll v3.4s, v3.4h, #0
; CHECK-GI-NEXT: mov v16.h[2], v17.h[0]
; CHECK-GI-NEXT: fmov s17, w12
; CHECK-GI-NEXT: mov v6.h[3], v4.h[0]
; CHECK-GI-NEXT: movi v4.4s, #15, msl #8
; CHECK-GI-NEXT: mov v5.h[3], v17.h[0]
; CHECK-GI-NEXT: fmov s17, w9
; CHECK-GI-NEXT: mov v16.h[3], v18.h[0]
; CHECK-GI-NEXT: ushll v6.4s, v6.4h, #0
; CHECK-GI-NEXT: and v0.16b, v0.16b, v4.16b
; CHECK-GI-NEXT: and v1.16b, v1.16b, v4.16b
; CHECK-GI-NEXT: mov v7.h[3], v17.h[0]
; CHECK-GI-NEXT: and v2.16b, v2.16b, v4.16b
; CHECK-GI-NEXT: and v3.16b, v3.16b, v4.16b
; CHECK-GI-NEXT: ushll v5.4s, v5.4h, #0
; CHECK-GI-NEXT: ushll v16.4s, v16.4h, #0
; CHECK-GI-NEXT: and v6.16b, v6.16b, v4.16b
; CHECK-GI-NEXT: ushll v7.4s, v7.4h, #0
; CHECK-GI-NEXT: and v5.16b, v5.16b, v4.16b
; CHECK-GI-NEXT: add v1.4s, v1.4s, v6.4s
; CHECK-GI-NEXT: and v7.16b, v7.16b, v4.16b
; CHECK-GI-NEXT: and v4.16b, v16.16b, v4.16b
; CHECK-GI-NEXT: add v0.4s, v0.4s, v5.4s
; CHECK-GI-NEXT: add v2.4s, v2.4s, v7.4s
; CHECK-GI-NEXT: add v3.4s, v3.4s, v4.4s
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i12> %s0 to <16 x i32>
%s1s = zext <16 x i12> %s1 to <16 x i32>
%m = add <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @sub_zz(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: sub_zz:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: usubl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sub_zz:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: usubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: usubl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i32>
%s1s = zext <16 x i8> %s1 to <16 x i32>
%m = sub <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @sub_ss(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: sub_ss:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-SD-NEXT: ssubl2 v4.8h, v0.16b, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sub_ss:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ssubl v2.8h, v0.8b, v1.8b
; CHECK-GI-NEXT: ssubl2 v3.8h, v0.16b, v1.16b
; CHECK-GI-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-GI-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-GI-NEXT: sshll v2.4s, v3.4h, #0
; CHECK-GI-NEXT: sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT: ret
entry:
%s0s = sext <16 x i8> %s0 to <16 x i32>
%s1s = sext <16 x i8> %s1 to <16 x i32>
%m = sub <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
define <16 x i32> @sub_zs(<16 x i8> %s0, <16 x i8> %s1) {
; CHECK-SD-LABEL: sub_zs:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: ushll v2.8h, v0.8b, #0
; CHECK-SD-NEXT: ushll2 v0.8h, v0.16b, #0
; CHECK-SD-NEXT: ssubw v2.8h, v2.8h, v1.8b
; CHECK-SD-NEXT: ssubw2 v4.8h, v0.8h, v1.16b
; CHECK-SD-NEXT: sshll v0.4s, v2.4h, #0
; CHECK-SD-NEXT: sshll2 v3.4s, v4.8h, #0
; CHECK-SD-NEXT: sshll2 v1.4s, v2.8h, #0
; CHECK-SD-NEXT: sshll v2.4s, v4.4h, #0
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sub_zs:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: ushll v2.8h, v0.8b, #0
; CHECK-GI-NEXT: ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT: sshll v3.8h, v1.8b, #0
; CHECK-GI-NEXT: sshll2 v4.8h, v1.16b, #0
; CHECK-GI-NEXT: ushll v1.4s, v2.4h, #0
; CHECK-GI-NEXT: ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT: ushll v5.4s, v0.4h, #0
; CHECK-GI-NEXT: ushll2 v6.4s, v0.8h, #0
; CHECK-GI-NEXT: ssubw v0.4s, v1.4s, v3.4h
; CHECK-GI-NEXT: ssubw2 v1.4s, v2.4s, v3.8h
; CHECK-GI-NEXT: ssubw v2.4s, v5.4s, v4.4h
; CHECK-GI-NEXT: ssubw2 v3.4s, v6.4s, v4.8h
; CHECK-GI-NEXT: ret
entry:
%s0s = zext <16 x i8> %s0 to <16 x i32>
%s1s = sext <16 x i8> %s1 to <16 x i32>
%m = sub <16 x i32> %s0s, %s1s
ret <16 x i32> %m
}
Codebase Browser
llvm