1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve < %s | FileCheck %s
4 ; fpext <vscale x 2 x half> -> <vscale x 2 x double>
5 define <vscale x 2 x double> @ext2_f16_f64(<vscale x 2 x half> *%ptr, i64 %index) {
6 ; CHECK-LABEL: ext2_f16_f64:
8 ; CHECK-NEXT: ptrue p0.d
9 ; CHECK-NEXT: ld1h { z0.d }, p0/z, [x0]
10 ; CHECK-NEXT: fcvt z0.d, p0/m, z0.h
12 %load = load <vscale x 2 x half>, <vscale x 2 x half>* %ptr, align 4
13 %load.ext = fpext <vscale x 2 x half> %load to <vscale x 2 x double>
14 ret <vscale x 2 x double> %load.ext
17 ; fpext <vscale x 4 x half> -> <vscale x 4 x double>
18 define <vscale x 4 x double> @ext4_f16_f64(<vscale x 4 x half> *%ptr, i64 %index) {
19 ; CHECK-LABEL: ext4_f16_f64:
21 ; CHECK-NEXT: ptrue p0.s
22 ; CHECK-NEXT: ld1h { z0.s }, p0/z, [x0]
23 ; CHECK-NEXT: ptrue p0.d
24 ; CHECK-NEXT: uunpklo z1.d, z0.s
25 ; CHECK-NEXT: uunpkhi z2.d, z0.s
26 ; CHECK-NEXT: movprfx z0, z1
27 ; CHECK-NEXT: fcvt z0.d, p0/m, z1.h
28 ; CHECK-NEXT: movprfx z1, z2
29 ; CHECK-NEXT: fcvt z1.d, p0/m, z2.h
31 %load = load <vscale x 4 x half>, <vscale x 4 x half>* %ptr, align 4
32 %load.ext = fpext <vscale x 4 x half> %load to <vscale x 4 x double>
33 ret <vscale x 4 x double> %load.ext
36 ; fpext <vscale x 8 x half> -> <vscale x 8 x double>
37 define <vscale x 8 x double> @ext8_f16_f64(<vscale x 8 x half> *%ptr, i64 %index) {
38 ; CHECK-LABEL: ext8_f16_f64:
40 ; CHECK-NEXT: ptrue p0.h
41 ; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
42 ; CHECK-NEXT: ptrue p0.d
43 ; CHECK-NEXT: uunpklo z1.s, z0.h
44 ; CHECK-NEXT: uunpkhi z0.s, z0.h
45 ; CHECK-NEXT: uunpklo z2.d, z1.s
46 ; CHECK-NEXT: uunpkhi z1.d, z1.s
47 ; CHECK-NEXT: uunpklo z3.d, z0.s
48 ; CHECK-NEXT: uunpkhi z4.d, z0.s
49 ; CHECK-NEXT: movprfx z0, z2
50 ; CHECK-NEXT: fcvt z0.d, p0/m, z2.h
51 ; CHECK-NEXT: movprfx z2, z3
52 ; CHECK-NEXT: fcvt z2.d, p0/m, z3.h
53 ; CHECK-NEXT: movprfx z3, z4
54 ; CHECK-NEXT: fcvt z3.d, p0/m, z4.h
55 ; CHECK-NEXT: fcvt z1.d, p0/m, z1.h
57 %load = load <vscale x 8 x half>, <vscale x 8 x half>* %ptr, align 4
58 %load.ext = fpext <vscale x 8 x half> %load to <vscale x 8 x double>
59 ret <vscale x 8 x double> %load.ext
62 ; fpext <vscale x 2 x float> -> <vscale x 2 x double>
63 define <vscale x 2 x double> @ext2_f32_f64(<vscale x 2 x float> *%ptr, i64 %index) {
64 ; CHECK-LABEL: ext2_f32_f64:
66 ; CHECK-NEXT: ptrue p0.d
67 ; CHECK-NEXT: ld1w { z0.d }, p0/z, [x0]
68 ; CHECK-NEXT: fcvt z0.d, p0/m, z0.s
70 %load = load <vscale x 2 x float>, <vscale x 2 x float>* %ptr, align 4
71 %load.ext = fpext <vscale x 2 x float> %load to <vscale x 2 x double>
72 ret <vscale x 2 x double> %load.ext
75 ; fpext <vscale x 4 x float> -> <vscale x 4 x double>
76 define <vscale x 4 x double> @ext4_f32_f64(<vscale x 4 x float> *%ptr, i64 %index) {
77 ; CHECK-LABEL: ext4_f32_f64:
79 ; CHECK-NEXT: ptrue p0.s
80 ; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
81 ; CHECK-NEXT: ptrue p0.d
82 ; CHECK-NEXT: uunpklo z1.d, z0.s
83 ; CHECK-NEXT: uunpkhi z2.d, z0.s
84 ; CHECK-NEXT: movprfx z0, z1
85 ; CHECK-NEXT: fcvt z0.d, p0/m, z1.s
86 ; CHECK-NEXT: movprfx z1, z2
87 ; CHECK-NEXT: fcvt z1.d, p0/m, z2.s
89 %load = load <vscale x 4 x float>, <vscale x 4 x float>* %ptr, align 4
90 %load.ext = fpext <vscale x 4 x float> %load to <vscale x 4 x double>
91 ret <vscale x 4 x double> %load.ext