2012-04-05 18:01:12 +08:00
|
|
|
; RUN: llc -mtriple armv7 %s -o - | FileCheck %s
|
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f:
|
2012-04-05 18:01:12 +08:00
|
|
|
define float @f(<4 x i16>* nocapture %in) {
|
2015-03-06 03:37:53 +08:00
|
|
|
; CHECK: vld1
|
2012-04-05 18:01:12 +08:00
|
|
|
; CHECK: vmovl.u16
|
2015-02-28 05:17:42 +08:00
|
|
|
%1 = load <4 x i16>, <4 x i16>* %in
|
2012-04-05 18:01:12 +08:00
|
|
|
; CHECK: vcvt.f32.u32
|
|
|
|
%2 = uitofp <4 x i16> %1 to <4 x float>
|
|
|
|
%3 = extractelement <4 x float> %2, i32 0
|
|
|
|
%4 = extractelement <4 x float> %2, i32 1
|
|
|
|
%5 = extractelement <4 x float> %2, i32 2
|
|
|
|
|
|
|
|
; CHECK: vadd.f32
|
|
|
|
%6 = fadd float %3, %4
|
|
|
|
%7 = fadd float %6, %5
|
|
|
|
|
|
|
|
ret float %7
|
|
|
|
}
|
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: g:
|
2012-04-05 18:01:12 +08:00
|
|
|
define float @g(<4 x i8>* nocapture %in) {
|
2012-04-26 16:46:29 +08:00
|
|
|
; Note: vld1 here is reasonably important. Mixing VFP and NEON
|
|
|
|
; instructions is bad on some cores
|
|
|
|
; CHECK: vld1
|
2012-04-05 18:01:12 +08:00
|
|
|
; CHECK: vmovl.u8
|
|
|
|
; CHECK: vmovl.u16
|
2015-02-28 05:17:42 +08:00
|
|
|
%1 = load <4 x i8>, <4 x i8>* %in
|
2012-04-05 18:01:12 +08:00
|
|
|
; CHECK: vcvt.f32.u32
|
|
|
|
%2 = uitofp <4 x i8> %1 to <4 x float>
|
|
|
|
%3 = extractelement <4 x float> %2, i32 0
|
|
|
|
%4 = extractelement <4 x float> %2, i32 1
|
|
|
|
%5 = extractelement <4 x float> %2, i32 2
|
|
|
|
|
|
|
|
; CHECK: vadd.f32
|
|
|
|
%6 = fadd float %3, %4
|
|
|
|
%7 = fadd float %6, %5
|
|
|
|
|
|
|
|
ret float %7
|
|
|
|
}
|
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: h:
|
2012-04-05 18:01:12 +08:00
|
|
|
define <4 x i8> @h(<4 x float> %v) {
|
|
|
|
; CHECK: vcvt.{{[us]}}32.f32
|
|
|
|
; CHECK: vmovn.i32
|
|
|
|
%1 = fptoui <4 x float> %v to <4 x i8>
|
|
|
|
ret <4 x i8> %1
|
|
|
|
}
|
2012-04-17 16:18:00 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: i:
|
2012-04-17 16:18:00 +08:00
|
|
|
define <4 x i8> @i(<4 x i8>* %x) {
|
2012-04-26 16:46:29 +08:00
|
|
|
; Note: vld1 here is reasonably important. Mixing VFP and NEON
|
|
|
|
; instructions is bad on some cores
|
|
|
|
; CHECK: vld1
|
2012-04-17 16:18:00 +08:00
|
|
|
; CHECK: vmovl.s8
|
|
|
|
; CHECK: vmovl.s16
|
|
|
|
; CHECK: vrecpe
|
|
|
|
; CHECK: vrecps
|
|
|
|
; CHECK: vmul
|
|
|
|
; CHECK: vmovn
|
2015-02-28 05:17:42 +08:00
|
|
|
%1 = load <4 x i8>, <4 x i8>* %x, align 4
|
2012-04-17 16:18:00 +08:00
|
|
|
%2 = sdiv <4 x i8> zeroinitializer, %1
|
|
|
|
ret <4 x i8> %2
|
|
|
|
}
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: j:
|
2012-09-05 16:57:21 +08:00
|
|
|
define <4 x i32> @j(<4 x i8>* %in) nounwind {
|
|
|
|
; CHECK: vld1
|
|
|
|
; CHECK: vmovl.u8
|
|
|
|
; CHECK: vmovl.u16
|
|
|
|
; CHECK-NOT: vand
|
2015-02-28 05:17:42 +08:00
|
|
|
%1 = load <4 x i8>, <4 x i8>* %in, align 4
|
2012-09-05 16:57:21 +08:00
|
|
|
%2 = zext <4 x i8> %1 to <4 x i32>
|
|
|
|
ret <4 x i32> %2
|
|
|
|
}
|
|
|
|
|