samvv/bolt
1
0
Fork 0
Code Issues 24 Pull requests Projects Releases Packages Wiki Activity Actions
bolt/deps/llvm-18.1.8/llvm/test/CodeGen/AArch64/fcvt.ll
Sam Vervaeck 174b6f4d4e
Embed LLVM 18.1.8
2025-02-14 19:21:04 +01:00

3296 lines
118 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-NOFP16
; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-FP16
; RUN: llc -mtriple=aarch64-none-eabi -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-NOFP16
; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-FP16
define double @ceil_f64(double %a) {
; CHECK-LABEL: ceil_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.ceil.f64(double %a)
ret double %c
}
define float @ceil_f32(float %a) {
; CHECK-LABEL: ceil_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.ceil.f32(float %a)
ret float %c
}
define half @ceil_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: ceil_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frintp s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: ceil_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintp h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: ceil_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frintp s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: ceil_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintp h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.ceil.f16(half %a)
ret half %c
}
define <2 x double> @ceil_v2f64(<2 x double> %a) {
; CHECK-LABEL: ceil_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.ceil.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @ceil_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: ceil_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frintp v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frintp v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: ceil_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frintp d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frintp v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.ceil.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @ceil_v4f64(<4 x double> %a) {
; CHECK-LABEL: ceil_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.2d, v0.2d
; CHECK-NEXT: frintp v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.ceil.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @ceil_v2f32(<2 x float> %a) {
; CHECK-LABEL: ceil_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.ceil.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @ceil_v3f32(<3 x float> %a) {
; CHECK-LABEL: ceil_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.ceil.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @ceil_v4f32(<4 x float> %a) {
; CHECK-LABEL: ceil_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.ceil.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @ceil_v8f32(<8 x float> %a) {
; CHECK-LABEL: ceil_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintp v0.4s, v0.4s
; CHECK-NEXT: frintp v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.ceil.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @ceil_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: ceil_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintp s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintp s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintp s2, s3
; CHECK-SD-NOFP16-NEXT: frintp s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintp s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintp s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintp s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintp s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: ceil_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: ceil_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintp v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintp v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: ceil_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.ceil.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @ceil_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: ceil_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintp s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frintp s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frintp s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintp s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: ceil_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintp v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: ceil_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frintp v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: ceil_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintp v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.ceil.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @ceil_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: ceil_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintp s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintp s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintp s2, s3
; CHECK-SD-NOFP16-NEXT: frintp s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintp s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintp s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintp s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintp s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: ceil_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: ceil_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frintp v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frintp v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: ceil_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.ceil.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @ceil_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: ceil_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frintp s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frintp s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frintp s18, s2
; CHECK-SD-NOFP16-NEXT: frintp s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frintp s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frintp s6, s7
; CHECK-SD-NOFP16-NEXT: frintp s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frintp s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frintp s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintp s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frintp s7, s7
; CHECK-SD-NOFP16-NEXT: frintp s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintp s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frintp s7, s17
; CHECK-SD-NOFP16-NEXT: frintp s0, s0
; CHECK-SD-NOFP16-NEXT: frintp s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: ceil_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frintp v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: ceil_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frintp v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frintp v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frintp v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frintp v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: ceil_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintp v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frintp v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.ceil.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @floor_f64(double %a) {
; CHECK-LABEL: floor_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.floor.f64(double %a)
ret double %c
}
define float @floor_f32(float %a) {
; CHECK-LABEL: floor_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.floor.f32(float %a)
ret float %c
}
define half @floor_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: floor_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frintm s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: floor_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintm h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: floor_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frintm s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: floor_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintm h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.floor.f16(half %a)
ret half %c
}
define <2 x double> @floor_v2f64(<2 x double> %a) {
; CHECK-LABEL: floor_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.floor.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @floor_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: floor_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frintm v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frintm v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: floor_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frintm d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frintm v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.floor.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @floor_v4f64(<4 x double> %a) {
; CHECK-LABEL: floor_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.2d, v0.2d
; CHECK-NEXT: frintm v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.floor.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @floor_v2f32(<2 x float> %a) {
; CHECK-LABEL: floor_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.floor.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @floor_v3f32(<3 x float> %a) {
; CHECK-LABEL: floor_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.floor.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @floor_v4f32(<4 x float> %a) {
; CHECK-LABEL: floor_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.floor.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @floor_v8f32(<8 x float> %a) {
; CHECK-LABEL: floor_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintm v0.4s, v0.4s
; CHECK-NEXT: frintm v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.floor.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @floor_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: floor_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintm s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintm s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintm s2, s3
; CHECK-SD-NOFP16-NEXT: frintm s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintm s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintm s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintm s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintm s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: floor_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: floor_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintm v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintm v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: floor_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.floor.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @floor_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: floor_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintm s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frintm s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frintm s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintm s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: floor_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintm v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: floor_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frintm v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: floor_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintm v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.floor.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @floor_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: floor_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintm s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintm s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintm s2, s3
; CHECK-SD-NOFP16-NEXT: frintm s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintm s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintm s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintm s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintm s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: floor_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: floor_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frintm v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frintm v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: floor_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.floor.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @floor_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: floor_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frintm s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frintm s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frintm s18, s2
; CHECK-SD-NOFP16-NEXT: frintm s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frintm s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frintm s6, s7
; CHECK-SD-NOFP16-NEXT: frintm s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frintm s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frintm s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintm s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frintm s7, s7
; CHECK-SD-NOFP16-NEXT: frintm s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintm s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frintm s7, s17
; CHECK-SD-NOFP16-NEXT: frintm s0, s0
; CHECK-SD-NOFP16-NEXT: frintm s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: floor_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frintm v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: floor_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frintm v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frintm v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frintm v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frintm v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: floor_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintm v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frintm v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.floor.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @nearbyint_f64(double %a) {
; CHECK-LABEL: nearbyint_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.nearbyint.f64(double %a)
ret double %c
}
define float @nearbyint_f32(float %a) {
; CHECK-LABEL: nearbyint_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.nearbyint.f32(float %a)
ret float %c
}
define half @nearbyint_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: nearbyint_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frinti s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: nearbyint_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinti h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: nearbyint_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frinti s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: nearbyint_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinti h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.nearbyint.f16(half %a)
ret half %c
}
define <2 x double> @nearbyint_v2f64(<2 x double> %a) {
; CHECK-LABEL: nearbyint_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @nearbyint_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: nearbyint_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frinti v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frinti v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: nearbyint_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frinti d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frinti v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.nearbyint.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @nearbyint_v4f64(<4 x double> %a) {
; CHECK-LABEL: nearbyint_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.2d, v0.2d
; CHECK-NEXT: frinti v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @nearbyint_v2f32(<2 x float> %a) {
; CHECK-LABEL: nearbyint_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.nearbyint.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @nearbyint_v3f32(<3 x float> %a) {
; CHECK-LABEL: nearbyint_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.nearbyint.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @nearbyint_v4f32(<4 x float> %a) {
; CHECK-LABEL: nearbyint_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @nearbyint_v8f32(<8 x float> %a) {
; CHECK-LABEL: nearbyint_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinti v0.4s, v0.4s
; CHECK-NEXT: frinti v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @nearbyint_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: nearbyint_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinti s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frinti s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinti s2, s3
; CHECK-SD-NOFP16-NEXT: frinti s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frinti s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frinti s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinti s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frinti s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: nearbyint_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: nearbyint_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frinti v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frinti v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: nearbyint_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.nearbyint.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @nearbyint_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: nearbyint_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinti s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frinti s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frinti s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinti s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: nearbyint_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinti v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: nearbyint_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frinti v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: nearbyint_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinti v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.nearbyint.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @nearbyint_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: nearbyint_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinti s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frinti s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinti s2, s3
; CHECK-SD-NOFP16-NEXT: frinti s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frinti s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frinti s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinti s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frinti s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: nearbyint_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: nearbyint_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frinti v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frinti v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: nearbyint_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.nearbyint.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @nearbyint_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: nearbyint_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frinti s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frinti s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frinti s18, s2
; CHECK-SD-NOFP16-NEXT: frinti s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frinti s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frinti s6, s7
; CHECK-SD-NOFP16-NEXT: frinti s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frinti s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frinti s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinti s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frinti s7, s7
; CHECK-SD-NOFP16-NEXT: frinti s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinti s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frinti s7, s17
; CHECK-SD-NOFP16-NEXT: frinti s0, s0
; CHECK-SD-NOFP16-NEXT: frinti s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: nearbyint_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frinti v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: nearbyint_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frinti v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frinti v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frinti v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frinti v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: nearbyint_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinti v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frinti v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.nearbyint.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @roundeven_f64(double %a) {
; CHECK-LABEL: roundeven_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.roundeven.f64(double %a)
ret double %c
}
define float @roundeven_f32(float %a) {
; CHECK-LABEL: roundeven_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.roundeven.f32(float %a)
ret float %c
}
define half @roundeven_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: roundeven_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frintn s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: roundeven_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintn h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: roundeven_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frintn s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: roundeven_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintn h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.roundeven.f16(half %a)
ret half %c
}
define <2 x double> @roundeven_v2f64(<2 x double> %a) {
; CHECK-LABEL: roundeven_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.roundeven.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @roundeven_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: roundeven_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frintn v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frintn v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: roundeven_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frintn d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frintn v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.roundeven.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @roundeven_v4f64(<4 x double> %a) {
; CHECK-LABEL: roundeven_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.2d, v0.2d
; CHECK-NEXT: frintn v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.roundeven.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @roundeven_v2f32(<2 x float> %a) {
; CHECK-LABEL: roundeven_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.roundeven.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @roundeven_v3f32(<3 x float> %a) {
; CHECK-LABEL: roundeven_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.roundeven.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @roundeven_v4f32(<4 x float> %a) {
; CHECK-LABEL: roundeven_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.roundeven.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @roundeven_v8f32(<8 x float> %a) {
; CHECK-LABEL: roundeven_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintn v0.4s, v0.4s
; CHECK-NEXT: frintn v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.roundeven.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @roundeven_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: roundeven_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintn s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintn s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintn s2, s3
; CHECK-SD-NOFP16-NEXT: frintn s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintn s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintn s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintn s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintn s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: roundeven_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: roundeven_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintn v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintn v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: roundeven_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.roundeven.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @roundeven_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: roundeven_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintn s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frintn s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frintn s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintn s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: roundeven_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintn v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: roundeven_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frintn v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: roundeven_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintn v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.roundeven.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @roundeven_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: roundeven_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintn s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintn s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintn s2, s3
; CHECK-SD-NOFP16-NEXT: frintn s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintn s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintn s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintn s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintn s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: roundeven_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: roundeven_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frintn v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frintn v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: roundeven_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.roundeven.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @roundeven_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: roundeven_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frintn s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frintn s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frintn s18, s2
; CHECK-SD-NOFP16-NEXT: frintn s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frintn s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frintn s6, s7
; CHECK-SD-NOFP16-NEXT: frintn s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frintn s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frintn s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintn s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frintn s7, s7
; CHECK-SD-NOFP16-NEXT: frintn s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintn s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frintn s7, s17
; CHECK-SD-NOFP16-NEXT: frintn s0, s0
; CHECK-SD-NOFP16-NEXT: frintn s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: roundeven_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frintn v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: roundeven_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frintn v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frintn v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frintn v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frintn v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: roundeven_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintn v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frintn v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.roundeven.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @rint_f64(double %a) {
; CHECK-LABEL: rint_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.rint.f64(double %a)
ret double %c
}
define float @rint_f32(float %a) {
; CHECK-LABEL: rint_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.rint.f32(float %a)
ret float %c
}
define half @rint_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: rint_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frintx s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: rint_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintx h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: rint_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frintx s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: rint_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintx h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.rint.f16(half %a)
ret half %c
}
define <2 x double> @rint_v2f64(<2 x double> %a) {
; CHECK-LABEL: rint_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.rint.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @rint_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: rint_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frintx v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frintx v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: rint_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frintx d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frintx v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.rint.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @rint_v4f64(<4 x double> %a) {
; CHECK-LABEL: rint_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.2d, v0.2d
; CHECK-NEXT: frintx v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.rint.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @rint_v2f32(<2 x float> %a) {
; CHECK-LABEL: rint_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.rint.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @rint_v3f32(<3 x float> %a) {
; CHECK-LABEL: rint_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.rint.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @rint_v4f32(<4 x float> %a) {
; CHECK-LABEL: rint_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.rint.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @rint_v8f32(<8 x float> %a) {
; CHECK-LABEL: rint_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintx v0.4s, v0.4s
; CHECK-NEXT: frintx v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.rint.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @rint_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: rint_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintx s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintx s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintx s2, s3
; CHECK-SD-NOFP16-NEXT: frintx s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintx s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintx s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintx s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintx s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: rint_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: rint_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintx v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintx v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: rint_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.rint.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @rint_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: rint_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintx s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frintx s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frintx s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintx s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: rint_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintx v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: rint_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frintx v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: rint_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintx v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.rint.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @rint_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: rint_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintx s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintx s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintx s2, s3
; CHECK-SD-NOFP16-NEXT: frintx s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintx s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintx s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintx s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintx s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: rint_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: rint_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frintx v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frintx v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: rint_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.rint.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @rint_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: rint_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frintx s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frintx s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frintx s18, s2
; CHECK-SD-NOFP16-NEXT: frintx s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frintx s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frintx s6, s7
; CHECK-SD-NOFP16-NEXT: frintx s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frintx s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frintx s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintx s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frintx s7, s7
; CHECK-SD-NOFP16-NEXT: frintx s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintx s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frintx s7, s17
; CHECK-SD-NOFP16-NEXT: frintx s0, s0
; CHECK-SD-NOFP16-NEXT: frintx s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: rint_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frintx v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: rint_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frintx v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frintx v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frintx v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frintx v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: rint_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintx v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frintx v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.rint.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @round_f64(double %a) {
; CHECK-LABEL: round_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.round.f64(double %a)
ret double %c
}
define float @round_f32(float %a) {
; CHECK-LABEL: round_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.round.f32(float %a)
ret float %c
}
define half @round_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: round_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frinta s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: round_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinta h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: round_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frinta s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: round_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinta h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.round.f16(half %a)
ret half %c
}
define <2 x double> @round_v2f64(<2 x double> %a) {
; CHECK-LABEL: round_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.round.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @round_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: round_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frinta v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frinta v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: round_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frinta d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frinta v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.round.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @round_v4f64(<4 x double> %a) {
; CHECK-LABEL: round_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.2d, v0.2d
; CHECK-NEXT: frinta v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.round.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @round_v2f32(<2 x float> %a) {
; CHECK-LABEL: round_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.round.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @round_v3f32(<3 x float> %a) {
; CHECK-LABEL: round_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.round.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @round_v4f32(<4 x float> %a) {
; CHECK-LABEL: round_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.round.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @round_v8f32(<8 x float> %a) {
; CHECK-LABEL: round_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frinta v0.4s, v0.4s
; CHECK-NEXT: frinta v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.round.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @round_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: round_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinta s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frinta s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinta s2, s3
; CHECK-SD-NOFP16-NEXT: frinta s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frinta s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frinta s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinta s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frinta s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: round_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: round_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frinta v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frinta v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: round_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.round.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @round_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: round_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinta s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frinta s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frinta s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinta s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: round_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinta v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: round_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frinta v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: round_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinta v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.round.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @round_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: round_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frinta s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frinta s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frinta s2, s3
; CHECK-SD-NOFP16-NEXT: frinta s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frinta s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frinta s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinta s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frinta s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: round_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: round_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frinta v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frinta v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: round_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.round.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @round_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: round_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frinta s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frinta s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frinta s18, s2
; CHECK-SD-NOFP16-NEXT: frinta s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frinta s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frinta s6, s7
; CHECK-SD-NOFP16-NEXT: frinta s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frinta s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frinta s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinta s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frinta s7, s7
; CHECK-SD-NOFP16-NEXT: frinta s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frinta s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frinta s7, s17
; CHECK-SD-NOFP16-NEXT: frinta s0, s0
; CHECK-SD-NOFP16-NEXT: frinta s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: round_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frinta v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: round_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frinta v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frinta v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frinta v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frinta v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: round_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frinta v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frinta v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.round.v16f16(<16 x half> %a)
ret <16 x half> %c
}
define double @trunc_f64(double %a) {
; CHECK-LABEL: trunc_f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz d0, d0
; CHECK-NEXT: ret
entry:
%c = call double @llvm.trunc.f64(double %a)
ret double %c
}
define float @trunc_f32(float %a) {
; CHECK-LABEL: trunc_f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz s0, s0
; CHECK-NEXT: ret
entry:
%c = call float @llvm.trunc.f32(float %a)
ret float %c
}
define half @trunc_f16(half %a) {
; CHECK-SD-NOFP16-LABEL: trunc_f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: frintz s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: trunc_f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintz h0, h0
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: trunc_f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvt s0, h0
; CHECK-GI-NOFP16-NEXT: frintz s0, s0
; CHECK-GI-NOFP16-NEXT: fcvt h0, s0
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: trunc_f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintz h0, h0
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call half @llvm.trunc.f16(half %a)
ret half %c
}
define <2 x double> @trunc_v2f64(<2 x double> %a) {
; CHECK-LABEL: trunc_v2f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.2d, v0.2d
; CHECK-NEXT: ret
entry:
%c = call <2 x double> @llvm.trunc.v2f64(<2 x double> %a)
ret <2 x double> %c
}
define <3 x double> @trunc_v3f64(<3 x double> %a) {
; CHECK-SD-LABEL: trunc_v3f64:
; CHECK-SD: // %bb.0: // %entry
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT: frintz v2.2d, v2.2d
; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT: frintz v0.2d, v0.2d
; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: trunc_v3f64:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT: frintz d2, d2
; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT: frintz v0.2d, v0.2d
; CHECK-GI-NEXT: mov d1, v0.d[1]
; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT: ret
entry:
%c = call <3 x double> @llvm.trunc.v3f64(<3 x double> %a)
ret <3 x double> %c
}
define <4 x double> @trunc_v4f64(<4 x double> %a) {
; CHECK-LABEL: trunc_v4f64:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.2d, v0.2d
; CHECK-NEXT: frintz v1.2d, v1.2d
; CHECK-NEXT: ret
entry:
%c = call <4 x double> @llvm.trunc.v4f64(<4 x double> %a)
ret <4 x double> %c
}
define <2 x float> @trunc_v2f32(<2 x float> %a) {
; CHECK-LABEL: trunc_v2f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.2s, v0.2s
; CHECK-NEXT: ret
entry:
%c = call <2 x float> @llvm.trunc.v2f32(<2 x float> %a)
ret <2 x float> %c
}
define <3 x float> @trunc_v3f32(<3 x float> %a) {
; CHECK-LABEL: trunc_v3f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <3 x float> @llvm.trunc.v3f32(<3 x float> %a)
ret <3 x float> %c
}
define <4 x float> @trunc_v4f32(<4 x float> %a) {
; CHECK-LABEL: trunc_v4f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.4s, v0.4s
; CHECK-NEXT: ret
entry:
%c = call <4 x float> @llvm.trunc.v4f32(<4 x float> %a)
ret <4 x float> %c
}
define <8 x float> @trunc_v8f32(<8 x float> %a) {
; CHECK-LABEL: trunc_v8f32:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: frintz v0.4s, v0.4s
; CHECK-NEXT: frintz v1.4s, v1.4s
; CHECK-NEXT: ret
entry:
%c = call <8 x float> @llvm.trunc.v8f32(<8 x float> %a)
ret <8 x float> %c
}
define <7 x half> @trunc_v7f16(<7 x half> %a) {
; CHECK-SD-NOFP16-LABEL: trunc_v7f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintz s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintz s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintz s2, s3
; CHECK-SD-NOFP16-NEXT: frintz s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintz s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintz s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintz s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintz s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: trunc_v7f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: trunc_v7f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[4]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[5]
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[6]
; CHECK-GI-NOFP16-NEXT: mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintz v2.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: mov v1.h[3], v0.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT: frintz v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v3.h[0]
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v4.h[0]
; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1]
; CHECK-GI-NOFP16-NEXT: mov v0.h[4], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[2]
; CHECK-GI-NOFP16-NEXT: mov v0.h[5], v2.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[6], v1.h[0]
; CHECK-GI-NOFP16-NEXT: mov v0.h[7], v0.h[0]
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: trunc_v7f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <7 x half> @llvm.trunc.v7f16(<7 x half> %a)
ret <7 x half> %c
}
define <4 x half> @trunc_v4f16(<4 x half> %a) {
; CHECK-SD-NOFP16-LABEL: trunc_v4f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintz s0, s2
; CHECK-SD-NOFP16-NEXT: fcvt s2, h3
; CHECK-SD-NOFP16-NEXT: fcvt s3, h4
; CHECK-SD-NOFP16-NEXT: frintz s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: frintz s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintz s2, s3
; CHECK-SD-NOFP16-NEXT: mov v0.h[2], v1.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: mov v0.h[3], v1.h[0]
; CHECK-SD-NOFP16-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: trunc_v4f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintz v0.4h, v0.4h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: trunc_v4f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v0.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: frintz v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: trunc_v4f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintz v0.4h, v0.4h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <4 x half> @llvm.trunc.v4f16(<4 x half> %a)
ret <4 x half> %c
}
define <8 x half> @trunc_v8f16(<8 x half> %a) {
; CHECK-SD-NOFP16-LABEL: trunc_v8f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h1, v0.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h0
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2]
; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: frintz s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: fcvt s4, h4
; CHECK-SD-NOFP16-NEXT: frintz s5, s1
; CHECK-SD-NOFP16-NEXT: fcvt h1, s2
; CHECK-SD-NOFP16-NEXT: frintz s2, s3
; CHECK-SD-NOFP16-NEXT: frintz s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h3, s5
; CHECK-SD-NOFP16-NEXT: fcvt s5, h6
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v1.h[1], v3.h[0]
; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[5]
; CHECK-SD-NOFP16-NEXT: frintz s5, s5
; CHECK-SD-NOFP16-NEXT: mov v1.h[2], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov v1.h[3], v4.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s5
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: frintz s3, s3
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: mov v1.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintz s2, s2
; CHECK-SD-NOFP16-NEXT: fcvt h3, s3
; CHECK-SD-NOFP16-NEXT: frintz s0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h2, s2
; CHECK-SD-NOFP16-NEXT: mov v1.h[5], v3.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: mov v1.h[6], v2.h[0]
; CHECK-SD-NOFP16-NEXT: mov v1.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: trunc_v8f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: trunc_v8f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v1.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: frintz v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: frintz v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: trunc_v8f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <8 x half> @llvm.trunc.v8f16(<8 x half> %a)
ret <8 x half> %c
}
define <16 x half> @trunc_v16f16(<16 x half> %a) {
; CHECK-SD-NOFP16-LABEL: trunc_v16f16:
; CHECK-SD-NOFP16: // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT: mov h2, v0.h[1]
; CHECK-SD-NOFP16-NEXT: mov h3, v1.h[1]
; CHECK-SD-NOFP16-NEXT: fcvt s4, h0
; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2]
; CHECK-SD-NOFP16-NEXT: fcvt s6, h1
; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[2]
; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[3]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3]
; CHECK-SD-NOFP16-NEXT: mov h20, v0.h[4]
; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4]
; CHECK-SD-NOFP16-NEXT: fcvt s2, h2
; CHECK-SD-NOFP16-NEXT: fcvt s3, h3
; CHECK-SD-NOFP16-NEXT: frintz s4, s4
; CHECK-SD-NOFP16-NEXT: fcvt s5, h5
; CHECK-SD-NOFP16-NEXT: frintz s6, s6
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s16, h16
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: frintz s18, s2
; CHECK-SD-NOFP16-NEXT: frintz s19, s3
; CHECK-SD-NOFP16-NEXT: fcvt h2, s4
; CHECK-SD-NOFP16-NEXT: frintz s4, s5
; CHECK-SD-NOFP16-NEXT: fcvt h3, s6
; CHECK-SD-NOFP16-NEXT: frintz s6, s7
; CHECK-SD-NOFP16-NEXT: frintz s16, s16
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt h7, s19
; CHECK-SD-NOFP16-NEXT: fcvt s18, h20
; CHECK-SD-NOFP16-NEXT: fcvt s19, h21
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: fcvt h6, s6
; CHECK-SD-NOFP16-NEXT: fcvt h16, s16
; CHECK-SD-NOFP16-NEXT: mov v2.h[1], v5.h[0]
; CHECK-SD-NOFP16-NEXT: frintz s5, s17
; CHECK-SD-NOFP16-NEXT: mov v3.h[1], v7.h[0]
; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5]
; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5]
; CHECK-SD-NOFP16-NEXT: frintz s18, s18
; CHECK-SD-NOFP16-NEXT: mov v2.h[2], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintz s4, s19
; CHECK-SD-NOFP16-NEXT: fcvt h5, s5
; CHECK-SD-NOFP16-NEXT: mov v3.h[2], v6.h[0]
; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[6]
; CHECK-SD-NOFP16-NEXT: fcvt s7, h7
; CHECK-SD-NOFP16-NEXT: fcvt s17, h17
; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[6]
; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7]
; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7]
; CHECK-SD-NOFP16-NEXT: mov v2.h[3], v16.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v3.h[3], v5.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s18
; CHECK-SD-NOFP16-NEXT: fcvt s6, h6
; CHECK-SD-NOFP16-NEXT: frintz s7, s7
; CHECK-SD-NOFP16-NEXT: frintz s16, s17
; CHECK-SD-NOFP16-NEXT: fcvt s17, h19
; CHECK-SD-NOFP16-NEXT: fcvt s0, h0
; CHECK-SD-NOFP16-NEXT: fcvt s1, h1
; CHECK-SD-NOFP16-NEXT: mov v2.h[4], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[4], v4.h[0]
; CHECK-SD-NOFP16-NEXT: frintz s4, s6
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h6, s16
; CHECK-SD-NOFP16-NEXT: frintz s7, s17
; CHECK-SD-NOFP16-NEXT: frintz s0, s0
; CHECK-SD-NOFP16-NEXT: frintz s1, s1
; CHECK-SD-NOFP16-NEXT: fcvt h4, s4
; CHECK-SD-NOFP16-NEXT: mov v2.h[5], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[5], v6.h[0]
; CHECK-SD-NOFP16-NEXT: fcvt h5, s7
; CHECK-SD-NOFP16-NEXT: fcvt h0, s0
; CHECK-SD-NOFP16-NEXT: fcvt h1, s1
; CHECK-SD-NOFP16-NEXT: mov v2.h[6], v4.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[6], v5.h[0]
; CHECK-SD-NOFP16-NEXT: mov v2.h[7], v0.h[0]
; CHECK-SD-NOFP16-NEXT: mov v3.h[7], v1.h[0]
; CHECK-SD-NOFP16-NEXT: mov v0.16b, v2.16b
; CHECK-SD-NOFP16-NEXT: mov v1.16b, v3.16b
; CHECK-SD-NOFP16-NEXT: ret
;
; CHECK-SD-FP16-LABEL: trunc_v16f16:
; CHECK-SD-FP16: // %bb.0: // %entry
; CHECK-SD-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-SD-FP16-NEXT: frintz v1.8h, v1.8h
; CHECK-SD-FP16-NEXT: ret
;
; CHECK-GI-NOFP16-LABEL: trunc_v16f16:
; CHECK-GI-NOFP16: // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT: fcvtl v2.4s, v0.4h
; CHECK-GI-NOFP16-NEXT: fcvtl v3.4s, v1.4h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v0.4s, v0.8h
; CHECK-GI-NOFP16-NEXT: fcvtl2 v1.4s, v1.8h
; CHECK-GI-NOFP16-NEXT: frintz v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT: frintz v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT: frintz v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT: frintz v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT: fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT: fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT: ret
;
; CHECK-GI-FP16-LABEL: trunc_v16f16:
; CHECK-GI-FP16: // %bb.0: // %entry
; CHECK-GI-FP16-NEXT: frintz v0.8h, v0.8h
; CHECK-GI-FP16-NEXT: frintz v1.8h, v1.8h
; CHECK-GI-FP16-NEXT: ret
entry:
%c = call <16 x half> @llvm.trunc.v16f16(<16 x half> %a)
ret <16 x half> %c
}
declare <16 x half> @llvm.ceil.v16f16(<16 x half>)
declare <16 x half> @llvm.floor.v16f16(<16 x half>)
declare <16 x half> @llvm.nearbyint.v16f16(<16 x half>)
declare <16 x half> @llvm.rint.v16f16(<16 x half>)
declare <16 x half> @llvm.round.v16f16(<16 x half>)
declare <16 x half> @llvm.roundeven.v16f16(<16 x half>)
declare <16 x half> @llvm.trunc.v16f16(<16 x half>)
declare <2 x double> @llvm.ceil.v2f64(<2 x double>)
declare <2 x double> @llvm.floor.v2f64(<2 x double>)
declare <2 x double> @llvm.nearbyint.v2f64(<2 x double>)
declare <2 x double> @llvm.rint.v2f64(<2 x double>)
declare <2 x double> @llvm.round.v2f64(<2 x double>)
declare <2 x double> @llvm.roundeven.v2f64(<2 x double>)
declare <2 x double> @llvm.trunc.v2f64(<2 x double>)
declare <2 x float> @llvm.ceil.v2f32(<2 x float>)
declare <2 x float> @llvm.floor.v2f32(<2 x float>)
declare <2 x float> @llvm.nearbyint.v2f32(<2 x float>)
declare <2 x float> @llvm.rint.v2f32(<2 x float>)
declare <2 x float> @llvm.round.v2f32(<2 x float>)
declare <2 x float> @llvm.roundeven.v2f32(<2 x float>)
declare <2 x float> @llvm.trunc.v2f32(<2 x float>)
declare <3 x double> @llvm.ceil.v3f64(<3 x double>)
declare <3 x double> @llvm.floor.v3f64(<3 x double>)
declare <3 x double> @llvm.nearbyint.v3f64(<3 x double>)
declare <3 x double> @llvm.rint.v3f64(<3 x double>)
declare <3 x double> @llvm.round.v3f64(<3 x double>)
declare <3 x double> @llvm.roundeven.v3f64(<3 x double>)
declare <3 x double> @llvm.trunc.v3f64(<3 x double>)
declare <3 x float> @llvm.ceil.v3f32(<3 x float>)
declare <3 x float> @llvm.floor.v3f32(<3 x float>)
declare <3 x float> @llvm.nearbyint.v3f32(<3 x float>)
declare <3 x float> @llvm.rint.v3f32(<3 x float>)
declare <3 x float> @llvm.round.v3f32(<3 x float>)
declare <3 x float> @llvm.roundeven.v3f32(<3 x float>)
declare <3 x float> @llvm.trunc.v3f32(<3 x float>)
declare <4 x double> @llvm.ceil.v4f64(<4 x double>)
declare <4 x double> @llvm.floor.v4f64(<4 x double>)
declare <4 x double> @llvm.nearbyint.v4f64(<4 x double>)
declare <4 x double> @llvm.rint.v4f64(<4 x double>)
declare <4 x double> @llvm.round.v4f64(<4 x double>)
declare <4 x double> @llvm.roundeven.v4f64(<4 x double>)
declare <4 x double> @llvm.trunc.v4f64(<4 x double>)
declare <4 x float> @llvm.ceil.v4f32(<4 x float>)
declare <4 x float> @llvm.floor.v4f32(<4 x float>)
declare <4 x float> @llvm.nearbyint.v4f32(<4 x float>)
declare <4 x float> @llvm.rint.v4f32(<4 x float>)
declare <4 x float> @llvm.round.v4f32(<4 x float>)
declare <4 x float> @llvm.roundeven.v4f32(<4 x float>)
declare <4 x float> @llvm.trunc.v4f32(<4 x float>)
declare <4 x half> @llvm.ceil.v4f16(<4 x half>)
declare <4 x half> @llvm.floor.v4f16(<4 x half>)
declare <4 x half> @llvm.nearbyint.v4f16(<4 x half>)
declare <4 x half> @llvm.rint.v4f16(<4 x half>)
declare <4 x half> @llvm.round.v4f16(<4 x half>)
declare <4 x half> @llvm.roundeven.v4f16(<4 x half>)
declare <4 x half> @llvm.trunc.v4f16(<4 x half>)
declare <7 x half> @llvm.ceil.v7f16(<7 x half>)
declare <7 x half> @llvm.floor.v7f16(<7 x half>)
declare <7 x half> @llvm.nearbyint.v7f16(<7 x half>)
declare <7 x half> @llvm.rint.v7f16(<7 x half>)
declare <7 x half> @llvm.round.v7f16(<7 x half>)
declare <7 x half> @llvm.roundeven.v7f16(<7 x half>)
declare <7 x half> @llvm.trunc.v7f16(<7 x half>)
declare <8 x float> @llvm.ceil.v8f32(<8 x float>)
declare <8 x float> @llvm.floor.v8f32(<8 x float>)
declare <8 x float> @llvm.nearbyint.v8f32(<8 x float>)
declare <8 x float> @llvm.rint.v8f32(<8 x float>)
declare <8 x float> @llvm.round.v8f32(<8 x float>)
declare <8 x float> @llvm.roundeven.v8f32(<8 x float>)
declare <8 x float> @llvm.trunc.v8f32(<8 x float>)
declare <8 x half> @llvm.ceil.v8f16(<8 x half>)
declare <8 x half> @llvm.floor.v8f16(<8 x half>)
declare <8 x half> @llvm.nearbyint.v8f16(<8 x half>)
declare <8 x half> @llvm.rint.v8f16(<8 x half>)
declare <8 x half> @llvm.round.v8f16(<8 x half>)
declare <8 x half> @llvm.roundeven.v8f16(<8 x half>)
declare <8 x half> @llvm.trunc.v8f16(<8 x half>)
declare double @llvm.ceil.f64(double)
declare double @llvm.floor.f64(double)
declare double @llvm.nearbyint.f64(double)
declare double @llvm.rint.f64(double)
declare double @llvm.round.f64(double)
declare double @llvm.roundeven.f64(double)
declare double @llvm.trunc.f64(double)
declare float @llvm.ceil.f32(float)
declare float @llvm.floor.f32(float)
declare float @llvm.nearbyint.f32(float)
declare float @llvm.rint.f32(float)
declare float @llvm.round.f32(float)
declare float @llvm.roundeven.f32(float)
declare float @llvm.trunc.f32(float)
declare half @llvm.ceil.f16(half)
declare half @llvm.floor.f16(half)
declare half @llvm.nearbyint.f16(half)
declare half @llvm.rint.f16(half)
declare half @llvm.round.f16(half)
declare half @llvm.roundeven.f16(half)
declare half @llvm.trunc.f16(half)
Reference in a new issue View git blame Copy permalink