1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+avx,+f16c | FileCheck %s --check-prefixes=CHECK,X86
3 ; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c | FileCheck %s --check-prefixes=CHECK,X64
5 ; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/f16c-builtins.c
7 define float @test_cvtsh_ss(i16 %a0) nounwind {
8 ; X86-LABEL: test_cvtsh_ss:
10 ; X86-NEXT: pushl %eax
11 ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
12 ; X86-NEXT: vmovd %eax, %xmm0
13 ; X86-NEXT: vcvtph2ps %xmm0, %xmm0
14 ; X86-NEXT: vmovss %xmm0, (%esp)
15 ; X86-NEXT: flds (%esp)
19 ; X64-LABEL: test_cvtsh_ss:
21 ; X64-NEXT: movzwl %di, %eax
22 ; X64-NEXT: vmovd %eax, %xmm0
23 ; X64-NEXT: vcvtph2ps %xmm0, %xmm0
25 %ins0 = insertelement <8 x i16> undef, i16 %a0, i32 0
26 %ins1 = insertelement <8 x i16> %ins0, i16 0, i32 1
27 %ins2 = insertelement <8 x i16> %ins1, i16 0, i32 2
28 %ins3 = insertelement <8 x i16> %ins2, i16 0, i32 3
29 %ins4 = insertelement <8 x i16> %ins3, i16 0, i32 4
30 %ins5 = insertelement <8 x i16> %ins4, i16 0, i32 5
31 %ins6 = insertelement <8 x i16> %ins5, i16 0, i32 6
32 %ins7 = insertelement <8 x i16> %ins6, i16 0, i32 7
33 %shuffle = shufflevector <8 x i16> %ins7, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
34 %bc = bitcast <4 x i16> %shuffle to <4 x half>
35 %cvt = fpext <4 x half> %bc to <4 x float>
36 %res = extractelement <4 x float> %cvt, i32 0
40 define i16 @test_cvtss_sh(float %a0) nounwind {
41 ; X86-LABEL: test_cvtss_sh:
43 ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
44 ; X86-NEXT: vxorps %xmm1, %xmm1, %xmm1
45 ; X86-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
46 ; X86-NEXT: vcvtps2ph $0, %xmm0, %xmm0
47 ; X86-NEXT: vmovd %xmm0, %eax
48 ; X86-NEXT: # kill: def $ax killed $ax killed $eax
51 ; X64-LABEL: test_cvtss_sh:
53 ; X64-NEXT: vxorps %xmm1, %xmm1, %xmm1
54 ; X64-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
55 ; X64-NEXT: vcvtps2ph $0, %xmm0, %xmm0
56 ; X64-NEXT: vmovd %xmm0, %eax
57 ; X64-NEXT: # kill: def $ax killed $ax killed $eax
59 %ins0 = insertelement <4 x float> undef, float %a0, i32 0
60 %ins1 = insertelement <4 x float> %ins0, float 0.000000e+00, i32 1
61 %ins2 = insertelement <4 x float> %ins1, float 0.000000e+00, i32 2
62 %ins3 = insertelement <4 x float> %ins2, float 0.000000e+00, i32 3
63 %cvt = call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %ins3, i32 0)
64 %res = extractelement <8 x i16> %cvt, i32 0
68 define <4 x float> @test_mm_cvtph_ps(<2 x i64> %a0) nounwind {
69 ; CHECK-LABEL: test_mm_cvtph_ps:
71 ; CHECK-NEXT: vcvtph2ps %xmm0, %xmm0
72 ; CHECK-NEXT: ret{{[l|q]}}
73 %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
74 %shuffle = shufflevector <8 x i16> %arg0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
75 %bc = bitcast <4 x i16> %shuffle to <4 x half>
76 %res = fpext <4 x half> %bc to <4 x float>
80 define <8 x float> @test_mm256_cvtph_ps(<2 x i64> %a0) nounwind {
81 ; CHECK-LABEL: test_mm256_cvtph_ps:
83 ; CHECK-NEXT: vcvtph2ps %xmm0, %ymm0
84 ; CHECK-NEXT: ret{{[l|q]}}
85 %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
86 %bc = bitcast <8 x i16> %arg0 to <8 x half>
87 %res = fpext <8 x half> %bc to <8 x float>
91 define <2 x i64> @test_mm_cvtps_ph(<4 x float> %a0) nounwind {
92 ; CHECK-LABEL: test_mm_cvtps_ph:
94 ; CHECK-NEXT: vcvtps2ph $0, %xmm0, %xmm0
95 ; CHECK-NEXT: ret{{[l|q]}}
96 %cvt = call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a0, i32 0)
97 %res = bitcast <8 x i16> %cvt to <2 x i64>
101 define <2 x i64> @test_mm256_cvtps_ph(<8 x float> %a0) nounwind {
102 ; CHECK-LABEL: test_mm256_cvtps_ph:
104 ; CHECK-NEXT: vcvtps2ph $0, %ymm0, %xmm0
105 ; CHECK-NEXT: vzeroupper
106 ; CHECK-NEXT: ret{{[l|q]}}
107 %cvt = call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a0, i32 0)
108 %res = bitcast <8 x i16> %cvt to <2 x i64>
112 declare <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16>) nounwind readonly
113 declare <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16>) nounwind readonly
115 declare <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float>, i32) nounwind readonly
116 declare <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float>, i32) nounwind readonly