File: neon-fpextend_f16.ll

package info (click to toggle)
llvm-toolchain-14 1%3A14.0.6-12
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,496,180 kB
  • sloc: cpp: 5,593,972; ansic: 986,872; asm: 585,869; python: 184,223; objc: 72,530; lisp: 31,119; f90: 27,793; javascript: 9,780; pascal: 9,762; sh: 9,482; perl: 7,468; ml: 5,432; awk: 3,523; makefile: 2,538; xml: 953; cs: 573; fortran: 567
file content (26 lines) | stat: -rw-r--r-- 1,038 bytes parent folder | download | duplicates (7) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
 
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon -fp-contract=fast | FileCheck %s

; This is testing that we can scalarize the v1f16 input to fp_extend even
; though the v1f64 result is legal.

define <1 x double> @fpext_v1f16_v1f64(<1 x half>* %a) {
; CHECK-LABEL: fpext_v1f16_v1f64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr h0, [x0]
; CHECK-NEXT:    fcvt d0, h0
; CHECK-NEXT:    ret
  %b = load <1 x half>, <1 x half>* %a
  %c = fpext <1 x half> %b to <1 x double>
  ret <1 x double> %c
}

define <1 x double> @strict_fpext_v1f32_v1f64(<1 x half> %x) #0 {
; CHECK-LABEL: strict_fpext_v1f32_v1f64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    fcvt d0, h0
; CHECK-NEXT:    ret
  %val = call <1 x double> @llvm.experimental.constrained.fpext.v1f64.v1f16(<1 x half> %x, metadata !"fpexcept.strict") strictfp
  ret <1 x double> %val
}
declare <1 x double> @llvm.experimental.constrained.fpext.v1f64.v1f16(<1 x half>, metadata)