Mercurial > hg > CbC > CbC_llvm

annotate test/CodeGen/AMDGPU/uint_to_fp.f64.ll @ 146:3fc4d5c3e21e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
set tail call flag for code segment in CGCAll
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Sun, 23 Dec 2018 19:23:36 +0900
parents 803732b1fca8
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
2
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
3 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5 ; SI-LABEL: {{^}}v_uint_to_fp_i64_to_f64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6 ; SI: buffer_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 ; SI-DAG: v_cvt_f64_u32_e32 [[HI_CONV:v\[[0-9]+:[0-9]+\]]], v[[HI]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 ; SI-DAG: v_cvt_f64_u32_e32 [[LO_CONV:v\[[0-9]+:[0-9]+\]]], v[[LO]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 ; SI-DAG: v_ldexp_f64 [[LDEXP:v\[[0-9]+:[0-9]+\]]], [[HI_CONV]], 32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 ; SI: v_add_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[LDEXP]], [[LO_CONV]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11 ; SI: buffer_store_dwordx2 [[RESULT]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
12 define amdgpu_kernel void @v_uint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 addrspace(1)* %in) {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
13 %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
14 %gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
15 %val = load i64, i64 addrspace(1)* %gep, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
16 %result = uitofp i64 %val to double
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17 store double %result, double addrspace(1)* %out
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
21 ; SI-LABEL: {{^}}s_uint_to_fp_i64_to_f64
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
22 define amdgpu_kernel void @s_uint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 %cast = uitofp i64 %in to double
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 store double %cast, double addrspace(1)* %out, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28 ; SI-LABEL: {{^}}s_uint_to_fp_v2i64_to_v2f64
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
29 define amdgpu_kernel void @s_uint_to_fp_v2i64_to_v2f64(<2 x double> addrspace(1)* %out, <2 x i64> %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 %cast = uitofp <2 x i64> %in to <2 x double>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 store <2 x double> %cast, <2 x double> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
32 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
33 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
34
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 ; SI-LABEL: {{^}}s_uint_to_fp_v4i64_to_v4f64
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
36 define amdgpu_kernel void @s_uint_to_fp_v4i64_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i64> %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 %cast = uitofp <4 x i64> %in to <4 x double>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
38 store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
39 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 ; SI-LABEL: {{^}}s_uint_to_fp_i32_to_f64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
44 ; SI: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
45 define amdgpu_kernel void @s_uint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 %cast = uitofp i32 %in to double
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
47 store double %cast, double addrspace(1)* %out, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
48 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
49 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
50
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
51 ; SI-LABEL: {{^}}s_uint_to_fp_v2i32_to_v2f64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
52 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
53 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
54 ; SI: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
55 define amdgpu_kernel void @s_uint_to_fp_v2i32_to_v2f64(<2 x double> addrspace(1)* %out, <2 x i32> %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
56 %cast = uitofp <2 x i32> %in to <2 x double>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
57 store <2 x double> %cast, <2 x double> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
58 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
59 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
61 ; SI-LABEL: {{^}}s_uint_to_fp_v4i32_to_v4f64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
62 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
63 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
64 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 ; SI: v_cvt_f64_u32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 ; SI: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
67 define amdgpu_kernel void @s_uint_to_fp_v4i32_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i32> %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
68 %cast = uitofp <4 x i32> %in to <4 x double>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
69 store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
73 ; We can't fold the SGPRs into v_cndmask_b32_e32, because it already
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
74 ; uses an SGPR (implicit vcc).
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
75
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
76 ; SI-LABEL: {{^}}uint_to_fp_i1_to_f64:
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
77 ; SI-DAG: v_cmp_eq_u32_e64 vcc
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
78 ; SI-DAG: v_cndmask_b32_e32 v[[SEL:[0-9]+]], 0, v{{[0-9]+}}
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
79 ; SI-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
80 ; SI: buffer_store_dwordx2 v{{\[}}[[ZERO]]:[[SEL]]{{\]}}
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81 ; SI: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
82 define amdgpu_kernel void @uint_to_fp_i1_to_f64(double addrspace(1)* %out, i32 %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83 %cmp = icmp eq i32 %in, 0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 %fp = uitofp i1 %cmp to double
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 store double %fp, double addrspace(1)* %out, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
86 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
87 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89 ; SI-LABEL: {{^}}uint_to_fp_i1_to_f64_load:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90 ; SI: v_cndmask_b32_e64 [[IRESULT:v[0-9]]], 0, 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
91 ; SI-NEXT: v_cvt_f64_u32_e32 [[RESULT:v\[[0-9]+:[0-9]\]]], [[IRESULT]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92 ; SI: buffer_store_dwordx2 [[RESULT]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93 ; SI: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
94 define amdgpu_kernel void @uint_to_fp_i1_to_f64_load(double addrspace(1)* %out, i1 %in) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
95 %fp = uitofp i1 %in to double
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
96 store double %fp, double addrspace(1)* %out, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
97 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
98 }