1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=avx512fp16,avx512vl -O3 | FileCheck %s --check-prefixes=CHECK,X86
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512fp16,avx512vl -O3 | FileCheck %s --check-prefixes=CHECK,X64
5 declare <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i1(<8 x i1>, metadata, metadata)
6 declare <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i1(<8 x i1>, metadata, metadata)
7 declare <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i8(<8 x i8>, metadata, metadata)
8 declare <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i8(<8 x i8>, metadata, metadata)
9 declare <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i16(<8 x i16>, metadata, metadata)
10 declare <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i16(<8 x i16>, metadata, metadata)
11 declare <4 x half> @llvm.experimental.constrained.sitofp.v4f16.v4i32(<4 x i32>, metadata, metadata)
12 declare <4 x half> @llvm.experimental.constrained.uitofp.v4f16.v4i32(<4 x i32>, metadata, metadata)
13 declare <2 x half> @llvm.experimental.constrained.sitofp.v2f16.v2i64(<2 x i64>, metadata, metadata)
14 declare <2 x half> @llvm.experimental.constrained.uitofp.v2f16.v2i64(<2 x i64>, metadata, metadata)
16 define <4 x half> @sitofp_v4i32_v4f16(<4 x i32> %x) #0 {
17 ; CHECK-LABEL: sitofp_v4i32_v4f16:
19 ; CHECK-NEXT: vcvtdq2ph %xmm0, %xmm0
20 ; CHECK-NEXT: ret{{[l|q]}}
21 %result = call <4 x half> @llvm.experimental.constrained.sitofp.v4f16.v4i32(<4 x i32> %x,
22 metadata !"round.dynamic",
23 metadata !"fpexcept.strict") #0
24 ret <4 x half> %result
27 define <4 x half> @uitofp_v4i32_v4f16(<4 x i32> %x) #0 {
28 ; CHECK-LABEL: uitofp_v4i32_v4f16:
30 ; CHECK-NEXT: vcvtudq2ph %xmm0, %xmm0
31 ; CHECK-NEXT: ret{{[l|q]}}
32 %result = call <4 x half> @llvm.experimental.constrained.uitofp.v4f16.v4i32(<4 x i32> %x,
33 metadata !"round.dynamic",
34 metadata !"fpexcept.strict") #0
35 ret <4 x half> %result
38 define <2 x half> @sitofp_v2i64_v2f16(<2 x i64> %x) #0 {
39 ; CHECK-LABEL: sitofp_v2i64_v2f16:
41 ; CHECK-NEXT: vcvtqq2ph %xmm0, %xmm0
42 ; CHECK-NEXT: ret{{[l|q]}}
43 %result = call <2 x half> @llvm.experimental.constrained.sitofp.v2f16.v2i64(<2 x i64> %x,
44 metadata !"round.dynamic",
45 metadata !"fpexcept.strict") #0
46 ret <2 x half> %result
49 define <2 x half> @uitofp_v2i64_v2f16(<2 x i64> %x) #0 {
50 ; CHECK-LABEL: uitofp_v2i64_v2f16:
52 ; CHECK-NEXT: vcvtuqq2ph %xmm0, %xmm0
53 ; CHECK-NEXT: ret{{[l|q]}}
54 %result = call <2 x half> @llvm.experimental.constrained.uitofp.v2f16.v2i64(<2 x i64> %x,
55 metadata !"round.dynamic",
56 metadata !"fpexcept.strict") #0
57 ret <2 x half> %result
60 define <8 x half> @sitofp_v8i1_v8f16(<8 x i1> %x) #0 {
61 ; CHECK-LABEL: sitofp_v8i1_v8f16:
63 ; CHECK-NEXT: vpsllw $15, %xmm0, %xmm0
64 ; CHECK-NEXT: vpsraw $15, %xmm0, %xmm0
65 ; CHECK-NEXT: vcvtw2ph %xmm0, %xmm0
66 ; CHECK-NEXT: ret{{[l|q]}}
67 %result = call <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i1(<8 x i1> %x,
68 metadata !"round.dynamic",
69 metadata !"fpexcept.strict") #0
70 ret <8 x half> %result
73 define <8 x half> @uitofp_v8i1_v8f16(<8 x i1> %x) #0 {
74 ; X86-LABEL: uitofp_v8i1_v8f16:
76 ; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}{1to4}, %xmm0, %xmm0
77 ; X86-NEXT: vcvtuw2ph %xmm0, %xmm0
80 ; X64-LABEL: uitofp_v8i1_v8f16:
82 ; X64-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
83 ; X64-NEXT: vcvtuw2ph %xmm0, %xmm0
85 %result = call <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i1(<8 x i1> %x,
86 metadata !"round.dynamic",
87 metadata !"fpexcept.strict") #0
88 ret <8 x half> %result
91 define <8 x half> @sitofp_v8i8_v8f16(<8 x i8> %x) #0 {
92 ; CHECK-LABEL: sitofp_v8i8_v8f16:
94 ; CHECK-NEXT: vpmovsxbw %xmm0, %xmm0
95 ; CHECK-NEXT: vcvtw2ph %xmm0, %xmm0
96 ; CHECK-NEXT: ret{{[l|q]}}
97 %result = call <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i8(<8 x i8> %x,
98 metadata !"round.dynamic",
99 metadata !"fpexcept.strict") #0
100 ret <8 x half> %result
103 define <8 x half> @uitofp_v8i8_v8f16(<8 x i8> %x) #0 {
104 ; CHECK-LABEL: uitofp_v8i8_v8f16:
106 ; CHECK-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
107 ; CHECK-NEXT: vcvtuw2ph %xmm0, %xmm0
108 ; CHECK-NEXT: ret{{[l|q]}}
109 %result = call <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i8(<8 x i8> %x,
110 metadata !"round.dynamic",
111 metadata !"fpexcept.strict") #0
112 ret <8 x half> %result
115 define <8 x half> @sitofp_v8i16_v8f16(<8 x i16> %x) #0 {
116 ; CHECK-LABEL: sitofp_v8i16_v8f16:
118 ; CHECK-NEXT: vcvtw2ph %xmm0, %xmm0
119 ; CHECK-NEXT: ret{{[l|q]}}
120 %result = call <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i16(<8 x i16> %x,
121 metadata !"round.dynamic",
122 metadata !"fpexcept.strict") #0
123 ret <8 x half> %result
126 define <8 x half> @uitofp_v8i16_v8f16(<8 x i16> %x) #0 {
127 ; CHECK-LABEL: uitofp_v8i16_v8f16:
129 ; CHECK-NEXT: vcvtuw2ph %xmm0, %xmm0
130 ; CHECK-NEXT: ret{{[l|q]}}
131 %result = call <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i16(<8 x i16> %x,
132 metadata !"round.dynamic",
133 metadata !"fpexcept.strict") #0
134 ret <8 x half> %result
137 attributes #0 = { strictfp }