1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
llvm-mirror/test/CodeGen/AArch64/bf16-convert-intrinsics.ll
Mikhail Maltsev 14bad468ca [BFloat] Add convert/copy instrinsic support
This patch is part of a series implementing the Bfloat16 extension of the Armv8.6-a architecture, as detailed here:

https://community.arm.com/developer/ip-products/processors/b/processors-ip-blog/posts/arm-architecture-developments-armv8-6-a

Specifically it adds intrinsic support in clang and llvm for Arm and AArch64.

The bfloat type, and its properties are specified in the Arm Architecture Reference Manual:

https://developer.arm.com/docs/ddi0487/latest/arm-architecture-reference-manual-armv8-for-armv8-a-architecture-profile

The following people contributed to this patch:
  - Alexandros Lamprineas
  - Luke Cheeseman
  - Mikhail Maltsev
  - Momchil Velikov
  - Luke Geeson

Differential Revision: https://reviews.llvm.org/D80928
2020-06-23 14:27:05 +00:00

35 lines
1.1 KiB
LLVM

; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-arm-none-eabi -mattr=+neon -mattr=+bf16 | FileCheck %s
declare bfloat @llvm.aarch64.neon.bfcvt(float)
declare <8 x bfloat> @llvm.aarch64.neon.bfcvtn(<4 x float>)
declare <8 x bfloat> @llvm.aarch64.neon.bfcvtn2(<8 x bfloat>, <4 x float>)
; CHECK-LABEL: test_vcvth_bf16_f32
; CHECK: bfcvt h0, s0
; CHECK-NEXT: ret
define bfloat @test_vcvth_bf16_f32(float %a) {
entry:
%vcvth_bf16_f32 = call bfloat @llvm.aarch64.neon.bfcvt(float %a)
ret bfloat %vcvth_bf16_f32
}
; CHECK-LABEL: test_vcvtq_low_bf16_f32
; CHECK: bfcvtn v0.4h, v0.4s
; CHECK-NEXT: ret
define <8 x bfloat> @test_vcvtq_low_bf16_f32(<4 x float> %a) {
entry:
%cvt = call <8 x bfloat> @llvm.aarch64.neon.bfcvtn(<4 x float> %a)
ret <8 x bfloat> %cvt
}
; CHECK-LABEL: test_vcvtq_high_bf16_f32
; CHECK: bfcvtn2 v1.8h, v0.4s
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: ret
define <8 x bfloat> @test_vcvtq_high_bf16_f32(<4 x float> %a, <8 x bfloat> %inactive) {
entry:
%cvt = call <8 x bfloat> @llvm.aarch64.neon.bfcvtn2(<8 x bfloat> %inactive, <4 x float> %a)
ret <8 x bfloat> %cvt
}