1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG %s
4 ; This test just checks that the compiler doesn't crash.
6 define amdgpu_kernel void @i8ptr_v16i8ptr(<16 x i8> addrspace(1)* %out, i8 addrspace(1)* %in) {
7 ; EG-LABEL: i8ptr_v16i8ptr:
8 ; EG: ; %bb.0: ; %entry
9 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
11 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
12 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
15 ; EG-NEXT: Fetch clause starting at 6:
16 ; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1
17 ; EG-NEXT: ALU clause starting at 8:
18 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
19 ; EG-NEXT: ALU clause starting at 9:
20 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
21 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
23 %0 = bitcast i8 addrspace(1)* %in to <16 x i8> addrspace(1)*
24 %1 = load <16 x i8>, <16 x i8> addrspace(1)* %0
25 store <16 x i8> %1, <16 x i8> addrspace(1)* %out
29 define amdgpu_kernel void @f32_to_v2i16(<2 x i16> addrspace(1)* %out, float addrspace(1)* %in) nounwind {
30 ; EG-LABEL: f32_to_v2i16:
32 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
34 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
35 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
38 ; EG-NEXT: Fetch clause starting at 6:
39 ; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
40 ; EG-NEXT: ALU clause starting at 8:
41 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
42 ; EG-NEXT: ALU clause starting at 9:
43 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
44 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
45 %load = load float, float addrspace(1)* %in, align 4
46 %bc = bitcast float %load to <2 x i16>
47 store <2 x i16> %bc, <2 x i16> addrspace(1)* %out, align 4
51 define amdgpu_kernel void @v2i16_to_f32(float addrspace(1)* %out, <2 x i16> addrspace(1)* %in) nounwind {
52 ; EG-LABEL: v2i16_to_f32:
54 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
56 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
57 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
60 ; EG-NEXT: Fetch clause starting at 6:
61 ; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
62 ; EG-NEXT: ALU clause starting at 8:
63 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
64 ; EG-NEXT: ALU clause starting at 9:
65 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
66 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
67 %load = load <2 x i16>, <2 x i16> addrspace(1)* %in, align 4
68 %bc = bitcast <2 x i16> %load to float
69 store float %bc, float addrspace(1)* %out, align 4
73 define amdgpu_kernel void @v4i8_to_i32(i32 addrspace(1)* %out, <4 x i8> addrspace(1)* %in) nounwind {
74 ; EG-LABEL: v4i8_to_i32:
76 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
78 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
79 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
82 ; EG-NEXT: Fetch clause starting at 6:
83 ; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
84 ; EG-NEXT: ALU clause starting at 8:
85 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
86 ; EG-NEXT: ALU clause starting at 9:
87 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
88 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
89 %load = load <4 x i8>, <4 x i8> addrspace(1)* %in, align 4
90 %bc = bitcast <4 x i8> %load to i32
91 store i32 %bc, i32 addrspace(1)* %out, align 4
95 define amdgpu_kernel void @i32_to_v4i8(<4 x i8> addrspace(1)* %out, i32 addrspace(1)* %in) nounwind {
96 ; EG-LABEL: i32_to_v4i8:
98 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
100 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
101 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
104 ; EG-NEXT: Fetch clause starting at 6:
105 ; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
106 ; EG-NEXT: ALU clause starting at 8:
107 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
108 ; EG-NEXT: ALU clause starting at 9:
109 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
110 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
111 %load = load i32, i32 addrspace(1)* %in, align 4
112 %bc = bitcast i32 %load to <4 x i8>
113 store <4 x i8> %bc, <4 x i8> addrspace(1)* %out, align 4
117 define amdgpu_kernel void @v2i16_to_v4i8(<4 x i8> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) nounwind {
118 ; EG-LABEL: v2i16_to_v4i8:
120 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
122 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
123 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
126 ; EG-NEXT: Fetch clause starting at 6:
127 ; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
128 ; EG-NEXT: ALU clause starting at 8:
129 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
130 ; EG-NEXT: ALU clause starting at 9:
131 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
132 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
133 %load = load <2 x i16>, <2 x i16> addrspace(1)* %in, align 4
134 %bc = bitcast <2 x i16> %load to <4 x i8>
135 store <4 x i8> %bc, <4 x i8> addrspace(1)* %out, align 4
139 ; This just checks for crash in BUILD_VECTOR/EXTRACT_ELEMENT combine
140 ; the stack manipulation is tricky to follow
141 ; TODO: This should only use one load
142 define amdgpu_kernel void @v4i16_extract_i8(i8 addrspace(1)* %out, <4 x i16> addrspace(1)* %in) nounwind {
143 ; EG-LABEL: v4i16_extract_i8:
145 ; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[]
147 ; EG-NEXT: ALU 17, @11, KC0[CB0:0-32], KC1[]
148 ; EG-NEXT: MEM_RAT MSKOR T5.XW, T6.X
151 ; EG-NEXT: Fetch clause starting at 6:
152 ; EG-NEXT: VTX_READ_16 T6.X, T5.X, 6, #1
153 ; EG-NEXT: VTX_READ_16 T5.X, T5.X, 4, #1
154 ; EG-NEXT: ALU clause starting at 10:
155 ; EG-NEXT: MOV * T5.X, KC0[2].Z,
156 ; EG-NEXT: ALU clause starting at 11:
157 ; EG-NEXT: LSHL * T0.W, T6.X, literal.x,
158 ; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
159 ; EG-NEXT: OR_INT * T0.W, PV.W, T5.X,
160 ; EG-NEXT: MOV * T3.X, PV.W,
161 ; EG-NEXT: MOV T0.Y, PV.X,
162 ; EG-NEXT: AND_INT T0.W, KC0[2].Y, literal.x,
163 ; EG-NEXT: MOV * T1.W, literal.y,
164 ; EG-NEXT: 3(4.203895e-45), 8(1.121039e-44)
165 ; EG-NEXT: BFE_UINT T1.W, PV.Y, literal.x, PS,
166 ; EG-NEXT: LSHL * T0.W, PV.W, literal.y,
167 ; EG-NEXT: 8(1.121039e-44), 3(4.203895e-45)
168 ; EG-NEXT: LSHL T5.X, PV.W, PS,
169 ; EG-NEXT: LSHL * T5.W, literal.x, PS,
170 ; EG-NEXT: 255(3.573311e-43), 0(0.000000e+00)
171 ; EG-NEXT: MOV T5.Y, 0.0,
172 ; EG-NEXT: MOV * T5.Z, 0.0,
173 ; EG-NEXT: LSHR * T6.X, KC0[2].Y, literal.x,
174 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
175 %load = load <4 x i16>, <4 x i16> addrspace(1)* %in, align 2
176 %bc = bitcast <4 x i16> %load to <8 x i8>
177 %element = extractelement <8 x i8> %bc, i32 5
178 store i8 %element, i8 addrspace(1)* %out
182 define amdgpu_kernel void @bitcast_v2i32_to_f64(double addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
183 ; EG-LABEL: bitcast_v2i32_to_f64:
185 ; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
187 ; EG-NEXT: ALU 1, @9, KC0[CB0:0-32], KC1[]
188 ; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
191 ; EG-NEXT: Fetch clause starting at 6:
192 ; EG-NEXT: VTX_READ_64 T0.XY, T0.X, 0, #1
193 ; EG-NEXT: ALU clause starting at 8:
194 ; EG-NEXT: MOV * T0.X, KC0[2].Z,
195 ; EG-NEXT: ALU clause starting at 9:
196 ; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
197 ; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
198 %val = load <2 x i32>, <2 x i32> addrspace(1)* %in, align 8
199 %bc = bitcast <2 x i32> %val to double
200 store double %bc, double addrspace(1)* %out, align 8