1; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-promote-alloca < %s | FileCheck --check-prefix=OPT %s
2
3; Make sure that array alloca loaded and stored as multi-element aggregates are handled correctly
4; Strictly the promote-alloca pass shouldn't have to deal with this case as it is non-canonical, but
5; the pass should handle it gracefully if it is
6; The checks look for lines that previously caused issues in PromoteAlloca (non-canonical). Opt
7; should now leave these unchanged
8
9; OPT-LABEL: @promote_1d_aggr(
10; OPT: store [1 x float] %tmp3, [1 x float]* %f1
11
12%Block = type { [1 x float], i32 }
13%gl_PerVertex = type { <4 x float>, float, [1 x float], [1 x float] }
14%struct = type { i32, i32 }
15
16@block = external addrspace(1) global %Block
17@pv = external addrspace(1) global %gl_PerVertex
18
19define amdgpu_vs void @promote_1d_aggr() #0 {
20  %i = alloca i32
21  %f1 = alloca [1 x float]
22  %tmp = getelementptr %Block, %Block addrspace(1)* @block, i32 0, i32 1
23  %tmp1 = load i32, i32 addrspace(1)* %tmp
24  store i32 %tmp1, i32* %i
25  %tmp2 = getelementptr %Block, %Block addrspace(1)* @block, i32 0, i32 0
26  %tmp3 = load [1 x float], [1 x float] addrspace(1)* %tmp2
27  store [1 x float] %tmp3, [1 x float]* %f1
28  %tmp4 = load i32, i32* %i
29  %tmp5 = getelementptr [1 x float], [1 x float]* %f1, i32 0, i32 %tmp4
30  %tmp6 = load float, float* %tmp5
31  %tmp7 = alloca <4 x float>
32  %tmp8 = load <4 x float>, <4 x float>* %tmp7
33  %tmp9 = insertelement <4 x float> %tmp8, float %tmp6, i32 0
34  %tmp10 = insertelement <4 x float> %tmp9, float %tmp6, i32 1
35  %tmp11 = insertelement <4 x float> %tmp10, float %tmp6, i32 2
36  %tmp12 = insertelement <4 x float> %tmp11, float %tmp6, i32 3
37  %tmp13 = getelementptr %gl_PerVertex, %gl_PerVertex addrspace(1)* @pv, i32 0, i32 0
38  store <4 x float> %tmp12, <4 x float> addrspace(1)* %tmp13
39  ret void
40}
41
42
43; OPT-LABEL: @promote_store_aggr(
44; OPT: %tmp6 = load [2 x float], [2 x float]* %f1
45
46%Block2 = type { i32, [2 x float] }
47@block2 = external addrspace(1) global %Block2
48
49define amdgpu_vs void @promote_store_aggr() #0 {
50  %i = alloca i32
51  %f1 = alloca [2 x float]
52  %tmp = getelementptr %Block2, %Block2 addrspace(1)* @block2, i32 0, i32 0
53  %tmp1 = load i32, i32 addrspace(1)* %tmp
54  store i32 %tmp1, i32* %i
55  %tmp2 = load i32, i32* %i
56  %tmp3 = sitofp i32 %tmp2 to float
57  %tmp4 = getelementptr [2 x float], [2 x float]* %f1, i32 0, i32 0
58  store float %tmp3, float* %tmp4
59  %tmp5 = getelementptr [2 x float], [2 x float]* %f1, i32 0, i32 1
60  store float 2.000000e+00, float* %tmp5
61  %tmp6 = load [2 x float], [2 x float]* %f1
62  %tmp7 = getelementptr %Block2, %Block2 addrspace(1)* @block2, i32 0, i32 1
63  store [2 x float] %tmp6, [2 x float] addrspace(1)* %tmp7
64  %tmp8 = getelementptr %gl_PerVertex, %gl_PerVertex addrspace(1)* @pv, i32 0, i32 0
65  store <4 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, <4 x float> addrspace(1)* %tmp8
66  ret void
67}
68
69; OPT-LABEL: @promote_load_from_store_aggr(
70; OPT: store [2 x float] %tmp3, [2 x float]* %f1
71
72%Block3 = type { [2 x float], i32 }
73@block3 = external addrspace(1) global %Block3
74
75define amdgpu_vs void @promote_load_from_store_aggr() #0 {
76  %i = alloca i32
77  %f1 = alloca [2 x float]
78  %tmp = getelementptr %Block3, %Block3 addrspace(1)* @block3, i32 0, i32 1
79  %tmp1 = load i32, i32 addrspace(1)* %tmp
80  store i32 %tmp1, i32* %i
81  %tmp2 = getelementptr %Block3, %Block3 addrspace(1)* @block3, i32 0, i32 0
82  %tmp3 = load [2 x float], [2 x float] addrspace(1)* %tmp2
83  store [2 x float] %tmp3, [2 x float]* %f1
84  %tmp4 = load i32, i32* %i
85  %tmp5 = getelementptr [2 x float], [2 x float]* %f1, i32 0, i32 %tmp4
86  %tmp6 = load float, float* %tmp5
87  %tmp7 = alloca <4 x float>
88  %tmp8 = load <4 x float>, <4 x float>* %tmp7
89  %tmp9 = insertelement <4 x float> %tmp8, float %tmp6, i32 0
90  %tmp10 = insertelement <4 x float> %tmp9, float %tmp6, i32 1
91  %tmp11 = insertelement <4 x float> %tmp10, float %tmp6, i32 2
92  %tmp12 = insertelement <4 x float> %tmp11, float %tmp6, i32 3
93  %tmp13 = getelementptr %gl_PerVertex, %gl_PerVertex addrspace(1)* @pv, i32 0, i32 0
94  store <4 x float> %tmp12, <4 x float> addrspace(1)* %tmp13
95  ret void
96}
97
98; OPT-LABEL: @promote_double_aggr(
99; OPT: store [2 x double] %tmp5, [2 x double]* %s
100
101@tmp_g = external addrspace(1) global { [4 x double], <2 x double>, <3 x double>, <4 x double> }
102@frag_color = external addrspace(1) global <4 x float>
103
104define amdgpu_ps void @promote_double_aggr() #0 {
105  %s = alloca [2 x double]
106  %tmp = getelementptr { [4 x double], <2 x double>, <3 x double>, <4 x double> }, { [4 x double], <2 x double>, <3 x double>, <4 x double> } addrspace(1)* @tmp_g, i32 0, i32 0, i32 0
107  %tmp1 = load double, double addrspace(1)* %tmp
108  %tmp2 = getelementptr { [4 x double], <2 x double>, <3 x double>, <4 x double> }, { [4 x double], <2 x double>, <3 x double>, <4 x double> } addrspace(1)* @tmp_g, i32 0, i32 0, i32 1
109  %tmp3 = load double, double addrspace(1)* %tmp2
110  %tmp4 = insertvalue [2 x double] undef, double %tmp1, 0
111  %tmp5 = insertvalue [2 x double] %tmp4, double %tmp3, 1
112  store [2 x double] %tmp5, [2 x double]* %s
113  %tmp6 = getelementptr [2 x double], [2 x double]* %s, i32 0, i32 1
114  %tmp7 = load double, double* %tmp6
115  %tmp8 = getelementptr [2 x double], [2 x double]* %s, i32 0, i32 1
116  %tmp9 = load double, double* %tmp8
117  %tmp10 = fadd double %tmp7, %tmp9
118  %tmp11 = getelementptr [2 x double], [2 x double]* %s, i32 0, i32 0
119  store double %tmp10, double* %tmp11
120  %tmp12 = getelementptr [2 x double], [2 x double]* %s, i32 0, i32 0
121  %tmp13 = load double, double* %tmp12
122  %tmp14 = getelementptr [2 x double], [2 x double]* %s, i32 0, i32 1
123  %tmp15 = load double, double* %tmp14
124  %tmp16 = fadd double %tmp13, %tmp15
125  %tmp17 = fptrunc double %tmp16 to float
126  %tmp18 = insertelement <4 x float> undef, float %tmp17, i32 0
127  %tmp19 = insertelement <4 x float> %tmp18, float %tmp17, i32 1
128  %tmp20 = insertelement <4 x float> %tmp19, float %tmp17, i32 2
129  %tmp21 = insertelement <4 x float> %tmp20, float %tmp17, i32 3
130  store <4 x float> %tmp21, <4 x float> addrspace(1)* @frag_color
131  ret void
132}
133
134; Don't crash on a type that isn't a valid vector element.
135; OPT-LABEL: @alloca_struct(
136define amdgpu_kernel void @alloca_struct() #0 {
137entry:
138  %alloca = alloca [2 x %struct], align 4
139  ret void
140}
141