mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-24 11:42:57 +01:00
2f13163a84
This adds a second implementation of the AArch64 architecture to LLVM, accessible in parallel via the "arm64" triple. The plan over the coming weeks & months is to merge the two into a single backend, during which time thorough code review should naturally occur. Everything will be easier with the target in-tree though, hence this commit. llvm-svn: 205090
50 lines
2.0 KiB
LLVM
50 lines
2.0 KiB
LLVM
; RUN: llc < %s -march=arm64 -arm64-neon-syntax=apple | FileCheck %s
|
|
|
|
define <2 x float> @cvtf32fxpu(<2 x i32> %a) nounwind readnone ssp {
|
|
; CHECK-LABEL: cvtf32fxpu:
|
|
; CHECK: ucvtf.2s v0, v0, #9
|
|
; CHECK: ret
|
|
%vcvt_n1 = tail call <2 x float> @llvm.arm64.neon.vcvtfxu2fp.v2f32.v2i32(<2 x i32> %a, i32 9)
|
|
ret <2 x float> %vcvt_n1
|
|
}
|
|
|
|
define <2 x float> @cvtf32fxps(<2 x i32> %a) nounwind readnone ssp {
|
|
; CHECK-LABEL: cvtf32fxps:
|
|
; CHECK: scvtf.2s v0, v0, #12
|
|
; CHECK: ret
|
|
%vcvt_n1 = tail call <2 x float> @llvm.arm64.neon.vcvtfxs2fp.v2f32.v2i32(<2 x i32> %a, i32 12)
|
|
ret <2 x float> %vcvt_n1
|
|
}
|
|
|
|
define <4 x float> @cvtqf32fxpu(<4 x i32> %a) nounwind readnone ssp {
|
|
; CHECK-LABEL: cvtqf32fxpu:
|
|
; CHECK: ucvtf.4s v0, v0, #18
|
|
; CHECK: ret
|
|
%vcvt_n1 = tail call <4 x float> @llvm.arm64.neon.vcvtfxu2fp.v4f32.v4i32(<4 x i32> %a, i32 18)
|
|
ret <4 x float> %vcvt_n1
|
|
}
|
|
|
|
define <4 x float> @cvtqf32fxps(<4 x i32> %a) nounwind readnone ssp {
|
|
; CHECK-LABEL: cvtqf32fxps:
|
|
; CHECK: scvtf.4s v0, v0, #30
|
|
; CHECK: ret
|
|
%vcvt_n1 = tail call <4 x float> @llvm.arm64.neon.vcvtfxs2fp.v4f32.v4i32(<4 x i32> %a, i32 30)
|
|
ret <4 x float> %vcvt_n1
|
|
}
|
|
define <2 x double> @f1(<2 x i64> %a) nounwind readnone ssp {
|
|
%vcvt_n1 = tail call <2 x double> @llvm.arm64.neon.vcvtfxu2fp.v2f64.v2i64(<2 x i64> %a, i32 12)
|
|
ret <2 x double> %vcvt_n1
|
|
}
|
|
|
|
define <2 x double> @f2(<2 x i64> %a) nounwind readnone ssp {
|
|
%vcvt_n1 = tail call <2 x double> @llvm.arm64.neon.vcvtfxs2fp.v2f64.v2i64(<2 x i64> %a, i32 9)
|
|
ret <2 x double> %vcvt_n1
|
|
}
|
|
|
|
declare <4 x float> @llvm.arm64.neon.vcvtfxu2fp.v4f32.v4i32(<4 x i32>, i32) nounwind readnone
|
|
declare <4 x float> @llvm.arm64.neon.vcvtfxs2fp.v4f32.v4i32(<4 x i32>, i32) nounwind readnone
|
|
declare <2 x float> @llvm.arm64.neon.vcvtfxu2fp.v2f32.v2i32(<2 x i32>, i32) nounwind readnone
|
|
declare <2 x float> @llvm.arm64.neon.vcvtfxs2fp.v2f32.v2i32(<2 x i32>, i32) nounwind readnone
|
|
declare <2 x double> @llvm.arm64.neon.vcvtfxu2fp.v2f64.v2i64(<2 x i64>, i32) nounwind readnone
|
|
declare <2 x double> @llvm.arm64.neon.vcvtfxs2fp.v2f64.v2i64(<2 x i64>, i32) nounwind readnone
|