xref: /llvm-project/llvm/test/CodeGen/AArch64/fp16_intrinsic_scalar_2op.ll (revision 3d18c8cd265c0c0bf1d85226c4770a2dd0f86e8f)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=aarch64 -mattr=+v8.2a,+fullfp16  | FileCheck %s
3
4declare half @llvm.aarch64.sisd.fabd.f16(half, half)
5declare half @llvm.aarch64.neon.fmax.f16(half, half)
6declare half @llvm.aarch64.neon.fmin.f16(half, half)
7declare half @llvm.aarch64.neon.frsqrts.f16(half, half)
8declare half @llvm.aarch64.neon.frecps.f16(half, half)
9declare half @llvm.aarch64.neon.fmulx.f16(half, half)
10declare half @llvm.fabs.f16(half)
11declare i32 @llvm.aarch64.neon.facge.i32.f16(half, half)
12declare i32 @llvm.aarch64.neon.facgt.i32.f16(half, half)
13
14define dso_local half @t_vabdh_f16(half %a, half %b) {
15; CHECK-LABEL: t_vabdh_f16:
16; CHECK:       // %bb.0: // %entry
17; CHECK-NEXT:    fabd h0, h0, h1
18; CHECK-NEXT:    ret
19entry:
20  %vabdh_f16 = tail call half @llvm.aarch64.sisd.fabd.f16(half %a, half %b)
21  ret half %vabdh_f16
22}
23
24define dso_local half @t_vabdh_f16_from_fsub_fabs(half %a, half %b) {
25; CHECK-LABEL: t_vabdh_f16_from_fsub_fabs:
26; CHECK:       // %bb.0: // %entry
27; CHECK-NEXT:    fabd h0, h0, h1
28; CHECK-NEXT:    ret
29entry:
30  %sub = fsub half %a, %b
31  %abs = tail call half @llvm.fabs.f16(half %sub)
32  ret half %abs
33}
34
35define dso_local i16 @t_vceqh_f16(half %a, half %b) {
36; CHECK-LABEL: t_vceqh_f16:
37; CHECK:       // %bb.0: // %entry
38; CHECK-NEXT:    fcmp h0, h1
39; CHECK-NEXT:    csetm w0, eq
40; CHECK-NEXT:    ret
41entry:
42  %0 = fcmp oeq half %a, %b
43  %vcmpd = sext i1 %0 to i16
44  ret i16 %vcmpd
45}
46
47define dso_local i16 @t_vcgeh_f16(half %a, half %b) {
48; CHECK-LABEL: t_vcgeh_f16:
49; CHECK:       // %bb.0: // %entry
50; CHECK-NEXT:    fcmp h0, h1
51; CHECK-NEXT:    csetm w0, ge
52; CHECK-NEXT:    ret
53entry:
54  %0 = fcmp oge half %a, %b
55  %vcmpd = sext i1 %0 to i16
56  ret i16 %vcmpd
57}
58
59define dso_local i16 @t_vcgth_f16(half %a, half %b) {
60; CHECK-LABEL: t_vcgth_f16:
61; CHECK:       // %bb.0: // %entry
62; CHECK-NEXT:    fcmp h0, h1
63; CHECK-NEXT:    csetm w0, gt
64; CHECK-NEXT:    ret
65entry:
66  %0 = fcmp ogt half %a, %b
67  %vcmpd = sext i1 %0 to i16
68  ret i16 %vcmpd
69}
70
71define dso_local i16 @t_vcleh_f16(half %a, half %b) {
72; CHECK-LABEL: t_vcleh_f16:
73; CHECK:       // %bb.0: // %entry
74; CHECK-NEXT:    fcmp h0, h1
75; CHECK-NEXT:    csetm w0, ls
76; CHECK-NEXT:    ret
77entry:
78  %0 = fcmp ole half %a, %b
79  %vcmpd = sext i1 %0 to i16
80  ret i16 %vcmpd
81}
82
83define dso_local i16 @t_vclth_f16(half %a, half %b) {
84; CHECK-LABEL: t_vclth_f16:
85; CHECK:       // %bb.0: // %entry
86; CHECK-NEXT:    fcmp h0, h1
87; CHECK-NEXT:    csetm w0, mi
88; CHECK-NEXT:    ret
89entry:
90  %0 = fcmp olt half %a, %b
91  %vcmpd = sext i1 %0 to i16
92  ret i16 %vcmpd
93}
94
95define dso_local half @t_vmaxh_f16(half %a, half %b) {
96; CHECK-LABEL: t_vmaxh_f16:
97; CHECK:       // %bb.0: // %entry
98; CHECK-NEXT:    fmax h0, h0, h1
99; CHECK-NEXT:    ret
100entry:
101  %vmax = tail call half @llvm.aarch64.neon.fmax.f16(half %a, half %b)
102  ret half %vmax
103}
104
105define dso_local half @t_vminh_f16(half %a, half %b) {
106; CHECK-LABEL: t_vminh_f16:
107; CHECK:       // %bb.0: // %entry
108; CHECK-NEXT:    fmin h0, h0, h1
109; CHECK-NEXT:    ret
110entry:
111  %vmin = tail call half @llvm.aarch64.neon.fmin.f16(half %a, half %b)
112  ret half %vmin
113}
114
115define dso_local half @t_vmulxh_f16(half %a, half %b) {
116; CHECK-LABEL: t_vmulxh_f16:
117; CHECK:       // %bb.0: // %entry
118; CHECK-NEXT:    fmulx h0, h0, h1
119; CHECK-NEXT:    ret
120entry:
121  %vmulxh_f16 = tail call half @llvm.aarch64.neon.fmulx.f16(half %a, half %b)
122  ret half %vmulxh_f16
123}
124
125define dso_local half @t_vrecpsh_f16(half %a, half %b) {
126; CHECK-LABEL: t_vrecpsh_f16:
127; CHECK:       // %bb.0: // %entry
128; CHECK-NEXT:    frecps h0, h0, h1
129; CHECK-NEXT:    ret
130entry:
131  %vrecps = tail call half @llvm.aarch64.neon.frecps.f16(half %a, half %b)
132  ret half %vrecps
133}
134
135define dso_local half @t_vrsqrtsh_f16(half %a, half %b) {
136; CHECK-LABEL: t_vrsqrtsh_f16:
137; CHECK:       // %bb.0: // %entry
138; CHECK-NEXT:    frsqrts h0, h0, h1
139; CHECK-NEXT:    ret
140entry:
141  %vrsqrtsh_f16 = tail call half @llvm.aarch64.neon.frsqrts.f16(half %a, half %b)
142  ret half %vrsqrtsh_f16
143}
144
145declare half @llvm.aarch64.neon.vcvtfxs2fp.f16.i32(i32, i32) #1
146declare half @llvm.aarch64.neon.vcvtfxs2fp.f16.i64(i64, i32) #1
147declare i32 @llvm.aarch64.neon.vcvtfp2fxs.i32.f16(half, i32) #1
148declare i64 @llvm.aarch64.neon.vcvtfp2fxs.i64.f16(half, i32) #1
149declare half @llvm.aarch64.neon.vcvtfxu2fp.f16.i32(i32, i32) #1
150declare i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half, i32) #1
151
152define dso_local half @test_vcvth_n_f16_s16_1(i16 %a) {
153; CHECK-LABEL: test_vcvth_n_f16_s16_1:
154; CHECK:       // %bb.0: // %entry
155; CHECK-NEXT:    fmov s0, w0
156; CHECK-NEXT:    scvtf h0, h0, #1
157; CHECK-NEXT:    ret
158entry:
159  %sext = sext i16 %a to i32
160  %fcvth_n = tail call half @llvm.aarch64.neon.vcvtfxs2fp.f16.i32(i32 %sext, i32 1)
161  ret half %fcvth_n
162}
163
164define dso_local half @test_vcvth_n_f16_s16_16(i16 %a) {
165; CHECK-LABEL: test_vcvth_n_f16_s16_16:
166; CHECK:       // %bb.0: // %entry
167; CHECK-NEXT:    fmov s0, w0
168; CHECK-NEXT:    scvtf h0, h0, #16
169; CHECK-NEXT:    ret
170entry:
171  %sext = sext i16 %a to i32
172  %fcvth_n = tail call half @llvm.aarch64.neon.vcvtfxs2fp.f16.i32(i32 %sext, i32 16)
173  ret half %fcvth_n
174}
175
176define dso_local half @test_vcvth_n_f16_s32_1(i32 %a) {
177; CHECK-LABEL: test_vcvth_n_f16_s32_1:
178; CHECK:       // %bb.0: // %entry
179; CHECK-NEXT:    fmov s0, w0
180; CHECK-NEXT:    scvtf h0, h0, #1
181; CHECK-NEXT:    ret
182entry:
183  %vcvth_n_f16_s32 = tail call half @llvm.aarch64.neon.vcvtfxs2fp.f16.i32(i32 %a, i32 1)
184  ret half %vcvth_n_f16_s32
185}
186
187define dso_local half @test_vcvth_n_f16_s32_16(i32 %a) {
188; CHECK-LABEL: test_vcvth_n_f16_s32_16:
189; CHECK:       // %bb.0: // %entry
190; CHECK-NEXT:    fmov s0, w0
191; CHECK-NEXT:    scvtf h0, h0, #16
192; CHECK-NEXT:    ret
193entry:
194  %vcvth_n_f16_s32 = tail call half @llvm.aarch64.neon.vcvtfxs2fp.f16.i32(i32 %a, i32 16)
195  ret half %vcvth_n_f16_s32
196}
197
198define dso_local i16 @test_vcvth_n_s16_f16_1(half %a) {
199; CHECK-LABEL: test_vcvth_n_s16_f16_1:
200; CHECK:       // %bb.0: // %entry
201; CHECK-NEXT:    fcvtzs h0, h0, #1
202; CHECK-NEXT:    fmov w0, s0
203; CHECK-NEXT:    ret
204entry:
205  %fcvth_n = tail call i32 @llvm.aarch64.neon.vcvtfp2fxs.i32.f16(half %a, i32 1)
206  %0 = trunc i32 %fcvth_n to i16
207  ret i16 %0
208}
209
210define dso_local i16 @test_vcvth_n_s16_f16_16(half %a) {
211; CHECK-LABEL: test_vcvth_n_s16_f16_16:
212; CHECK:       // %bb.0: // %entry
213; CHECK-NEXT:    fcvtzs h0, h0, #16
214; CHECK-NEXT:    fmov w0, s0
215; CHECK-NEXT:    ret
216entry:
217  %fcvth_n = tail call i32 @llvm.aarch64.neon.vcvtfp2fxs.i32.f16(half %a, i32 16)
218  %0 = trunc i32 %fcvth_n to i16
219  ret i16 %0
220}
221
222define dso_local i32 @test_vcvth_n_s32_f16_1(half %a) {
223; CHECK-LABEL: test_vcvth_n_s32_f16_1:
224; CHECK:       // %bb.0: // %entry
225; CHECK-NEXT:    fcvtzs h0, h0, #1
226; CHECK-NEXT:    fmov w0, s0
227; CHECK-NEXT:    ret
228entry:
229  %vcvth_n_s32_f16 = tail call i32 @llvm.aarch64.neon.vcvtfp2fxs.i32.f16(half %a, i32 1)
230  ret i32 %vcvth_n_s32_f16
231}
232
233define dso_local i32 @test_vcvth_n_s32_f16_16(half %a) {
234; CHECK-LABEL: test_vcvth_n_s32_f16_16:
235; CHECK:       // %bb.0: // %entry
236; CHECK-NEXT:    fcvtzs h0, h0, #16
237; CHECK-NEXT:    fmov w0, s0
238; CHECK-NEXT:    ret
239entry:
240  %vcvth_n_s32_f16 = tail call i32 @llvm.aarch64.neon.vcvtfp2fxs.i32.f16(half %a, i32 16)
241  ret i32 %vcvth_n_s32_f16
242}
243
244define dso_local i64 @test_vcvth_n_s64_f16_1(half %a) {
245; CHECK-LABEL: test_vcvth_n_s64_f16_1:
246; CHECK:       // %bb.0: // %entry
247; CHECK-NEXT:    fcvtzs h0, h0, #1
248; CHECK-NEXT:    fmov x0, d0
249; CHECK-NEXT:    ret
250entry:
251  %vcvth_n_s64_f16 = tail call i64 @llvm.aarch64.neon.vcvtfp2fxs.i64.f16(half %a, i32 1)
252  ret i64 %vcvth_n_s64_f16
253}
254
255define dso_local i64 @test_vcvth_n_s64_f16_32(half %a) {
256; CHECK-LABEL: test_vcvth_n_s64_f16_32:
257; CHECK:       // %bb.0: // %entry
258; CHECK-NEXT:    fcvtzs h0, h0, #32
259; CHECK-NEXT:    fmov x0, d0
260; CHECK-NEXT:    ret
261entry:
262  %vcvth_n_s64_f16 = tail call i64 @llvm.aarch64.neon.vcvtfp2fxs.i64.f16(half %a, i32 32)
263  ret i64 %vcvth_n_s64_f16
264}
265
266define dso_local half @test_vcvth_n_f16_u16_1(i16 %a) {
267; CHECK-LABEL: test_vcvth_n_f16_u16_1:
268; CHECK:       // %bb.0: // %entry
269; CHECK-NEXT:    fmov s0, w0
270; CHECK-NEXT:    ucvtf h0, h0, #1
271; CHECK-NEXT:    ret
272entry:
273  %0 = zext i16 %a to i32
274  %fcvth_n = tail call half @llvm.aarch64.neon.vcvtfxu2fp.f16.i32(i32 %0, i32 1)
275  ret half %fcvth_n
276}
277
278define dso_local half @test_vcvth_n_f16_u16_16(i16 %a) {
279; CHECK-LABEL: test_vcvth_n_f16_u16_16:
280; CHECK:       // %bb.0: // %entry
281; CHECK-NEXT:    fmov s0, w0
282; CHECK-NEXT:    ucvtf h0, h0, #16
283; CHECK-NEXT:    ret
284entry:
285  %0 = zext i16 %a to i32
286  %fcvth_n = tail call half @llvm.aarch64.neon.vcvtfxu2fp.f16.i32(i32 %0, i32 16)
287  ret half %fcvth_n
288}
289
290define dso_local half @test_vcvth_n_f16_u32_1(i32 %a) {
291; CHECK-LABEL: test_vcvth_n_f16_u32_1:
292; CHECK:       // %bb.0: // %entry
293; CHECK-NEXT:    fmov s0, w0
294; CHECK-NEXT:    ucvtf h0, h0, #1
295; CHECK-NEXT:    ret
296entry:
297  %vcvth_n_f16_u32 = tail call half @llvm.aarch64.neon.vcvtfxu2fp.f16.i32(i32 %a, i32 1)
298  ret half %vcvth_n_f16_u32
299}
300
301define dso_local half @test_vcvth_n_f16_u32_16(i32 %a) {
302; CHECK-LABEL: test_vcvth_n_f16_u32_16:
303; CHECK:       // %bb.0: // %entry
304; CHECK-NEXT:    fmov s0, w0
305; CHECK-NEXT:    ucvtf h0, h0, #16
306; CHECK-NEXT:    ret
307entry:
308  %vcvth_n_f16_u32 = tail call half @llvm.aarch64.neon.vcvtfxu2fp.f16.i32(i32 %a, i32 16)
309  ret half %vcvth_n_f16_u32
310}
311
312define dso_local i16 @test_vcvth_n_u16_f16_1(half %a) {
313; CHECK-LABEL: test_vcvth_n_u16_f16_1:
314; CHECK:       // %bb.0: // %entry
315; CHECK-NEXT:    fcvtzu h0, h0, #1
316; CHECK-NEXT:    fmov w0, s0
317; CHECK-NEXT:    ret
318entry:
319  %fcvth_n = tail call i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half %a, i32 1)
320  %0 = trunc i32 %fcvth_n to i16
321  ret i16 %0
322}
323
324define dso_local i16 @test_vcvth_n_u16_f16_16(half %a) {
325; CHECK-LABEL: test_vcvth_n_u16_f16_16:
326; CHECK:       // %bb.0: // %entry
327; CHECK-NEXT:    fcvtzu h0, h0, #16
328; CHECK-NEXT:    fmov w0, s0
329; CHECK-NEXT:    ret
330entry:
331  %fcvth_n = tail call i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half %a, i32 16)
332  %0 = trunc i32 %fcvth_n to i16
333  ret i16 %0
334}
335
336define dso_local i32 @test_vcvth_n_u32_f16_1(half %a) {
337; CHECK-LABEL: test_vcvth_n_u32_f16_1:
338; CHECK:       // %bb.0: // %entry
339; CHECK-NEXT:    fcvtzu h0, h0, #1
340; CHECK-NEXT:    fmov w0, s0
341; CHECK-NEXT:    ret
342entry:
343  %vcvth_n_u32_f16 = tail call i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half %a, i32 1)
344  ret i32 %vcvth_n_u32_f16
345}
346
347define dso_local i32 @test_vcvth_n_u32_f16_16(half %a) {
348; CHECK-LABEL: test_vcvth_n_u32_f16_16:
349; CHECK:       // %bb.0: // %entry
350; CHECK-NEXT:    fcvtzu h0, h0, #16
351; CHECK-NEXT:    fmov w0, s0
352; CHECK-NEXT:    ret
353entry:
354  %vcvth_n_u32_f16 = tail call i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half %a, i32 16)
355  ret i32 %vcvth_n_u32_f16
356}
357
358define dso_local i16 @vcageh_f16_test(half %a, half %b) {
359; CHECK-LABEL: vcageh_f16_test:
360; CHECK:       // %bb.0: // %entry
361; CHECK-NEXT:    facge h0, h0, h1
362; CHECK-NEXT:    fmov w0, s0
363; CHECK-NEXT:    ret
364entry:
365  %facg = tail call i32 @llvm.aarch64.neon.facge.i32.f16(half %a, half %b)
366  %0 = trunc i32 %facg to i16
367  ret i16 %0
368}
369
370define dso_local i16 @vcagth_f16_test(half %a, half %b) {
371; CHECK-LABEL: vcagth_f16_test:
372; CHECK:       // %bb.0: // %entry
373; CHECK-NEXT:    facgt h0, h0, h1
374; CHECK-NEXT:    fmov w0, s0
375; CHECK-NEXT:    ret
376entry:
377  %facg = tail call i32 @llvm.aarch64.neon.facgt.i32.f16(half %a, half %b)
378  %0 = trunc i32 %facg to i16
379  ret i16 %0
380}
381
382define dso_local half @vcvth_n_f16_s64_test(i64 %a) {
383; CHECK-LABEL: vcvth_n_f16_s64_test:
384; CHECK:       // %bb.0: // %entry
385; CHECK-NEXT:    fmov d0, x0
386; CHECK-NEXT:    scvtf h0, h0, #16
387; CHECK-NEXT:    ret
388entry:
389  %vcvth_n_f16_s64 = tail call half @llvm.aarch64.neon.vcvtfxs2fp.f16.i64(i64 %a, i32 16)
390  ret half %vcvth_n_f16_s64
391}
392