1 ; RUN: llc < %s -mtriple=arm64-apple-ios -asm-verbose=false | FileCheck %s
3 define float @load0(i16* nocapture readonly %a) nounwind {
5 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0]
6 ; CHECK-NEXT: fcvt s0, [[HREG]]
9 %tmp = load i16* %a, align 2
10 %tmp1 = tail call float @llvm.convert.from.fp16.f32(i16 %tmp)
14 define double @load1(i16* nocapture readonly %a) nounwind {
16 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0]
17 ; CHECK-NEXT: fcvt d0, [[HREG]]
20 %tmp = load i16* %a, align 2
21 %conv = tail call double @llvm.convert.from.fp16.f64(i16 %tmp)
25 define float @load2(i16* nocapture readonly %a, i32 %i) nounwind {
27 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, w1, sxtw #1]
28 ; CHECK-NEXT: fcvt s0, [[HREG]]
31 %idxprom = sext i32 %i to i64
32 %arrayidx = getelementptr inbounds i16* %a, i64 %idxprom
33 %tmp = load i16* %arrayidx, align 2
34 %tmp1 = tail call float @llvm.convert.from.fp16.f32(i16 %tmp)
38 define double @load3(i16* nocapture readonly %a, i32 %i) nounwind {
40 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, w1, sxtw #1]
41 ; CHECK-NEXT: fcvt d0, [[HREG]]
44 %idxprom = sext i32 %i to i64
45 %arrayidx = getelementptr inbounds i16* %a, i64 %idxprom
46 %tmp = load i16* %arrayidx, align 2
47 %conv = tail call double @llvm.convert.from.fp16.f64(i16 %tmp)
51 define float @load4(i16* nocapture readonly %a, i64 %i) nounwind {
53 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, x1, lsl #1]
54 ; CHECK-NEXT: fcvt s0, [[HREG]]
57 %arrayidx = getelementptr inbounds i16* %a, i64 %i
58 %tmp = load i16* %arrayidx, align 2
59 %tmp1 = tail call float @llvm.convert.from.fp16.f32(i16 %tmp)
63 define double @load5(i16* nocapture readonly %a, i64 %i) nounwind {
65 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, x1, lsl #1]
66 ; CHECK-NEXT: fcvt d0, [[HREG]]
69 %arrayidx = getelementptr inbounds i16* %a, i64 %i
70 %tmp = load i16* %arrayidx, align 2
71 %conv = tail call double @llvm.convert.from.fp16.f64(i16 %tmp)
75 define float @load6(i16* nocapture readonly %a) nounwind {
77 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, #20]
78 ; CHECK-NEXT: fcvt s0, [[HREG]]
81 %arrayidx = getelementptr inbounds i16* %a, i64 10
82 %tmp = load i16* %arrayidx, align 2
83 %tmp1 = tail call float @llvm.convert.from.fp16.f32(i16 %tmp)
87 define double @load7(i16* nocapture readonly %a) nounwind {
89 ; CHECK-NEXT: ldr [[HREG:h[0-9]+]], [x0, #20]
90 ; CHECK-NEXT: fcvt d0, [[HREG]]
93 %arrayidx = getelementptr inbounds i16* %a, i64 10
94 %tmp = load i16* %arrayidx, align 2
95 %conv = tail call double @llvm.convert.from.fp16.f64(i16 %tmp)
99 define float @load8(i16* nocapture readonly %a) nounwind {
100 ; CHECK-LABEL: load8:
101 ; CHECK-NEXT: ldur [[HREG:h[0-9]+]], [x0, #-20]
102 ; CHECK-NEXT: fcvt s0, [[HREG]]
105 %arrayidx = getelementptr inbounds i16* %a, i64 -10
106 %tmp = load i16* %arrayidx, align 2
107 %tmp1 = tail call float @llvm.convert.from.fp16.f32(i16 %tmp)
111 define double @load9(i16* nocapture readonly %a) nounwind {
112 ; CHECK-LABEL: load9:
113 ; CHECK-NEXT: ldur [[HREG:h[0-9]+]], [x0, #-20]
114 ; CHECK-NEXT: fcvt d0, [[HREG]]
117 %arrayidx = getelementptr inbounds i16* %a, i64 -10
118 %tmp = load i16* %arrayidx, align 2
119 %conv = tail call double @llvm.convert.from.fp16.f64(i16 %tmp)
123 define void @store0(i16* nocapture %a, float %val) nounwind {
124 ; CHECK-LABEL: store0:
125 ; CHECK-NEXT: fcvt h0, s0
126 ; CHECK-NEXT: str h0, [x0]
129 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %val)
130 store i16 %tmp, i16* %a, align 2
134 define void @store1(i16* nocapture %a, double %val) nounwind {
135 ; CHECK-LABEL: store1:
136 ; CHECK-NEXT: fcvt h0, d0
137 ; CHECK-NEXT: str h0, [x0]
140 %conv = fptrunc double %val to float
141 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %conv)
142 store i16 %tmp, i16* %a, align 2
146 define void @store2(i16* nocapture %a, i32 %i, float %val) nounwind {
147 ; CHECK-LABEL: store2:
148 ; CHECK-NEXT: fcvt h0, s0
149 ; CHECK-NEXT: str h0, [x0, w1, sxtw #1]
152 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %val)
153 %idxprom = sext i32 %i to i64
154 %arrayidx = getelementptr inbounds i16* %a, i64 %idxprom
155 store i16 %tmp, i16* %arrayidx, align 2
159 define void @store3(i16* nocapture %a, i32 %i, double %val) nounwind {
160 ; CHECK-LABEL: store3:
161 ; CHECK-NEXT: fcvt h0, d0
162 ; CHECK-NEXT: str h0, [x0, w1, sxtw #1]
165 %conv = fptrunc double %val to float
166 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %conv)
167 %idxprom = sext i32 %i to i64
168 %arrayidx = getelementptr inbounds i16* %a, i64 %idxprom
169 store i16 %tmp, i16* %arrayidx, align 2
173 define void @store4(i16* nocapture %a, i64 %i, float %val) nounwind {
174 ; CHECK-LABEL: store4:
175 ; CHECK-NEXT: fcvt h0, s0
176 ; CHECK-NEXT: str h0, [x0, x1, lsl #1]
179 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %val)
180 %arrayidx = getelementptr inbounds i16* %a, i64 %i
181 store i16 %tmp, i16* %arrayidx, align 2
185 define void @store5(i16* nocapture %a, i64 %i, double %val) nounwind {
186 ; CHECK-LABEL: store5:
187 ; CHECK-NEXT: fcvt h0, d0
188 ; CHECK-NEXT: str h0, [x0, x1, lsl #1]
191 %conv = fptrunc double %val to float
192 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %conv)
193 %arrayidx = getelementptr inbounds i16* %a, i64 %i
194 store i16 %tmp, i16* %arrayidx, align 2
198 define void @store6(i16* nocapture %a, float %val) nounwind {
199 ; CHECK-LABEL: store6:
200 ; CHECK-NEXT: fcvt h0, s0
201 ; CHECK-NEXT: str h0, [x0, #20]
204 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %val)
205 %arrayidx = getelementptr inbounds i16* %a, i64 10
206 store i16 %tmp, i16* %arrayidx, align 2
210 define void @store7(i16* nocapture %a, double %val) nounwind {
211 ; CHECK-LABEL: store7:
212 ; CHECK-NEXT: fcvt h0, d0
213 ; CHECK-NEXT: str h0, [x0, #20]
216 %conv = fptrunc double %val to float
217 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %conv)
218 %arrayidx = getelementptr inbounds i16* %a, i64 10
219 store i16 %tmp, i16* %arrayidx, align 2
223 define void @store8(i16* nocapture %a, float %val) nounwind {
224 ; CHECK-LABEL: store8:
225 ; CHECK-NEXT: fcvt h0, s0
226 ; CHECK-NEXT: stur h0, [x0, #-20]
229 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %val)
230 %arrayidx = getelementptr inbounds i16* %a, i64 -10
231 store i16 %tmp, i16* %arrayidx, align 2
235 define void @store9(i16* nocapture %a, double %val) nounwind {
236 ; CHECK-LABEL: store9:
237 ; CHECK-NEXT: fcvt h0, d0
238 ; CHECK-NEXT: stur h0, [x0, #-20]
241 %conv = fptrunc double %val to float
242 %tmp = tail call i16 @llvm.convert.to.fp16.f32(float %conv)
243 %arrayidx = getelementptr inbounds i16* %a, i64 -10
244 store i16 %tmp, i16* %arrayidx, align 2
248 declare i16 @llvm.convert.to.fp16.f32(float) nounwind readnone
249 declare float @llvm.convert.from.fp16.f32(i16) nounwind readnone
250 declare i16 @llvm.convert.to.fp16.f64(double) nounwind readnone
251 declare double @llvm.convert.from.fp16.f64(i16) nounwind readnone