mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-22 07:32:48 +00:00
b0abb4dc42
the overloaded vector types allowed floating-point or integer vector elements. Most of these operations actually depend on the element type, so bitcasting was not an option. If you include the vpadd intrinsics that I updated earlier, this gets rid of 20 intrinsics. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78646 91177308-0d34-0410-b5e6-96231b3b80d8
34 lines
1.2 KiB
LLVM
34 lines
1.2 KiB
LLVM
; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
|
|
; RUN: grep {vrsqrte\\.u32} %t | count 2
|
|
; RUN: grep {vrsqrte\\.f32} %t | count 2
|
|
|
|
define <2 x i32> @vrsqrtei32(<2 x i32>* %A) nounwind {
|
|
%tmp1 = load <2 x i32>* %A
|
|
%tmp2 = call <2 x i32> @llvm.arm.neon.vrsqrte.v2i32(<2 x i32> %tmp1)
|
|
ret <2 x i32> %tmp2
|
|
}
|
|
|
|
define <4 x i32> @vrsqrteQi32(<4 x i32>* %A) nounwind {
|
|
%tmp1 = load <4 x i32>* %A
|
|
%tmp2 = call <4 x i32> @llvm.arm.neon.vrsqrte.v4i32(<4 x i32> %tmp1)
|
|
ret <4 x i32> %tmp2
|
|
}
|
|
|
|
define <2 x float> @vrsqrtef32(<2 x float>* %A) nounwind {
|
|
%tmp1 = load <2 x float>* %A
|
|
%tmp2 = call <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float> %tmp1)
|
|
ret <2 x float> %tmp2
|
|
}
|
|
|
|
define <4 x float> @vrsqrteQf32(<4 x float>* %A) nounwind {
|
|
%tmp1 = load <4 x float>* %A
|
|
%tmp2 = call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %tmp1)
|
|
ret <4 x float> %tmp2
|
|
}
|
|
|
|
declare <2 x i32> @llvm.arm.neon.vrsqrte.v2i32(<2 x i32>) nounwind readnone
|
|
declare <4 x i32> @llvm.arm.neon.vrsqrte.v4i32(<4 x i32>) nounwind readnone
|
|
|
|
declare <2 x float> @llvm.arm.neon.vrsqrte.v2f32(<2 x float>) nounwind readnone
|
|
declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
|