63 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mattr=+sme-fa64 -force-streaming-compatible < %s | FileCheck %s -check-prefix=FA643; RUN: llc -mattr=+sve -force-streaming-compatible < %s | FileCheck %s -check-prefix=NO-FA644; RUN: llc -force-streaming-compatible < %s | FileCheck %s --check-prefix=NONEON-NOSVE5 6 7target triple = "aarch64-unknown-linux-gnu"8 9define half @fadda_v4f16(half %start, <4 x half> %a) {10; FA64-LABEL: fadda_v4f16:11; FA64: // %bb.0:12; FA64-NEXT: ptrue p0.h, vl413; FA64-NEXT: // kill: def $h0 killed $h0 def $z014; FA64-NEXT: // kill: def $d1 killed $d1 def $z115; FA64-NEXT: fadda h0, p0, h0, z1.h16; FA64-NEXT: // kill: def $h0 killed $h0 killed $z017; FA64-NEXT: ret18;19; NO-FA64-LABEL: fadda_v4f16:20; NO-FA64: // %bb.0:21; NO-FA64-NEXT: // kill: def $d1 killed $d1 def $z122; NO-FA64-NEXT: fadd h0, h0, h123; NO-FA64-NEXT: mov z2.h, z1.h[1]24; NO-FA64-NEXT: fadd h0, h0, h225; NO-FA64-NEXT: mov z2.h, z1.h[2]26; NO-FA64-NEXT: mov z1.h, z1.h[3]27; NO-FA64-NEXT: fadd h0, h0, h228; NO-FA64-NEXT: fadd h0, h0, h129; NO-FA64-NEXT: ret30;31; NONEON-NOSVE-LABEL: fadda_v4f16:32; NONEON-NOSVE: // %bb.0:33; NONEON-NOSVE-NEXT: sub sp, sp, #1634; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 1635; NONEON-NOSVE-NEXT: str d1, [sp, #8]36; NONEON-NOSVE-NEXT: fcvt s0, h037; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]38; NONEON-NOSVE-NEXT: fcvt s1, h139; NONEON-NOSVE-NEXT: fadd s0, s0, s140; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]41; NONEON-NOSVE-NEXT: fcvt s1, h142; NONEON-NOSVE-NEXT: fcvt h0, s043; NONEON-NOSVE-NEXT: fcvt s0, h044; NONEON-NOSVE-NEXT: fadd s0, s0, s145; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]46; NONEON-NOSVE-NEXT: fcvt s1, h147; NONEON-NOSVE-NEXT: fcvt h0, s048; NONEON-NOSVE-NEXT: fcvt s0, h049; NONEON-NOSVE-NEXT: fadd s0, s0, s150; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]51; NONEON-NOSVE-NEXT: fcvt s1, h152; NONEON-NOSVE-NEXT: fcvt h0, s053; NONEON-NOSVE-NEXT: fcvt s0, h054; NONEON-NOSVE-NEXT: fadd s0, s0, s155; NONEON-NOSVE-NEXT: fcvt h0, s056; NONEON-NOSVE-NEXT: add sp, sp, #1657; NONEON-NOSVE-NEXT: ret58 %res = call half @llvm.vector.reduce.fadd.v4f16(half %start, <4 x half> %a)59 ret half %res60}61 62declare half @llvm.vector.reduce.fadd.v4f16(half, <4 x half>)63