1; RUN: llc -march=amdgcn -mcpu=SI -show-mc-encoding -verify-machineinstrs < %s | FileCheck %s
2
3declare i32 @llvm.r600.read.tidig.x() readnone
4
5;;;==========================================================================;;;
6;;; MUBUF LOAD TESTS
7;;;==========================================================================;;;
8
9; MUBUF load with an immediate byte offset that fits into 12-bits
10; CHECK-LABEL: {{^}}mubuf_load0:
11; CHECK: buffer_load_dword v{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0 offset:4 ; encoding: [0x04,0x00,0x30,0xe0
12define void @mubuf_load0(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
13entry:
14  %0 = getelementptr i32, i32 addrspace(1)* %in, i64 1
15  %1 = load i32, i32 addrspace(1)* %0
16  store i32 %1, i32 addrspace(1)* %out
17  ret void
18}
19
20; MUBUF load with the largest possible immediate offset
21; CHECK-LABEL: {{^}}mubuf_load1:
22; CHECK: buffer_load_ubyte v{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0 offset:4095 ; encoding: [0xff,0x0f,0x20,0xe0
23define void @mubuf_load1(i8 addrspace(1)* %out, i8 addrspace(1)* %in) {
24entry:
25  %0 = getelementptr i8, i8 addrspace(1)* %in, i64 4095
26  %1 = load i8, i8 addrspace(1)* %0
27  store i8 %1, i8 addrspace(1)* %out
28  ret void
29}
30
31; MUBUF load with an immediate byte offset that doesn't fit into 12-bits
32; CHECK-LABEL: {{^}}mubuf_load2:
33; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
34; CHECK: buffer_load_dword v{{[0-9]}}, s[{{[0-9]+:[0-9]+}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x30,0xe0
35define void @mubuf_load2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
36entry:
37  %0 = getelementptr i32, i32 addrspace(1)* %in, i64 1024
38  %1 = load i32, i32 addrspace(1)* %0
39  store i32 %1, i32 addrspace(1)* %out
40  ret void
41}
42
43; MUBUF load with a 12-bit immediate offset and a register offset
44; CHECK-LABEL: {{^}}mubuf_load3:
45; CHECK-NOT: ADD
46; CHECK: buffer_load_dword v{{[0-9]}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], 0 addr64 offset:4 ; encoding: [0x04,0x80,0x30,0xe0
47define void @mubuf_load3(i32 addrspace(1)* %out, i32 addrspace(1)* %in, i64 %offset) {
48entry:
49  %0 = getelementptr i32, i32 addrspace(1)* %in, i64 %offset
50  %1 = getelementptr i32, i32 addrspace(1)* %0, i64 1
51  %2 = load i32, i32 addrspace(1)* %1
52  store i32 %2, i32 addrspace(1)* %out
53  ret void
54}
55
56; CHECK-LABEL: {{^}}soffset_max_imm:
57; CHECK: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 64 offen glc
58define void @soffset_max_imm([6 x <16 x i8>] addrspace(2)* byval, [17 x <16 x i8>] addrspace(2)* byval, [16 x <4 x i32>] addrspace(2)* byval, [32 x <8 x i32>] addrspace(2)* byval, i32 inreg, i32 inreg, i32, i32, i32, i32, i32, i32, i32, i32) #1 {
59main_body:
60  %tmp0 = getelementptr [6 x <16 x i8>], [6 x <16 x i8>] addrspace(2)* %0, i32 0, i32 0
61  %tmp1 = load <16 x i8>, <16 x i8> addrspace(2)* %tmp0
62  %tmp2 = shl i32 %6, 2
63  %tmp3 = call i32 @llvm.SI.buffer.load.dword.i32.i32(<16 x i8> %tmp1, i32 %tmp2, i32 64, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0)
64  %tmp4 = add i32 %6, 16
65  %tmp5 = bitcast float 0.0 to i32
66  call void @llvm.SI.tbuffer.store.i32(<16 x i8> %tmp1, i32 %tmp5, i32 1, i32 %tmp4, i32 %4, i32 0, i32 4, i32 4, i32 1, i32 0, i32 1, i32 1, i32 0)
67  ret void
68}
69
70; Make sure immediates that aren't inline constants don't get folded into
71; the soffset operand.
72; FIXME: for this test we should be smart enough to shift the immediate into
73; the offset field.
74; CHECK-LABEL: {{^}}soffset_no_fold:
75; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x41
76; CHECK: buffer_load_dword v{{[0-9+]}}, v{{[0-9+]}}, s[{{[0-9]+}}:{{[0-9]+}}], [[SOFFSET]] offen glc
77define void @soffset_no_fold([6 x <16 x i8>] addrspace(2)* byval, [17 x <16 x i8>] addrspace(2)* byval, [16 x <4 x i32>] addrspace(2)* byval, [32 x <8 x i32>] addrspace(2)* byval, i32 inreg, i32 inreg, i32, i32, i32, i32, i32, i32, i32, i32) #1 {
78main_body:
79  %tmp0 = getelementptr [6 x <16 x i8>], [6 x <16 x i8>] addrspace(2)* %0, i32 0, i32 0
80  %tmp1 = load <16 x i8>, <16 x i8> addrspace(2)* %tmp0
81  %tmp2 = shl i32 %6, 2
82  %tmp3 = call i32 @llvm.SI.buffer.load.dword.i32.i32(<16 x i8> %tmp1, i32 %tmp2, i32 65, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0)
83  %tmp4 = add i32 %6, 16
84  %tmp5 = bitcast float 0.0 to i32
85  call void @llvm.SI.tbuffer.store.i32(<16 x i8> %tmp1, i32 %tmp5, i32 1, i32 %tmp4, i32 %4, i32 0, i32 4, i32 4, i32 1, i32 0, i32 1, i32 1, i32 0)
86  ret void
87}
88
89;;;==========================================================================;;;
90;;; MUBUF STORE TESTS
91;;;==========================================================================;;;
92
93; MUBUF store with an immediate byte offset that fits into 12-bits
94; CHECK-LABEL: {{^}}mubuf_store0:
95; CHECK: buffer_store_dword v{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0 offset:4 ; encoding: [0x04,0x00,0x70,0xe0
96define void @mubuf_store0(i32 addrspace(1)* %out) {
97entry:
98  %0 = getelementptr i32, i32 addrspace(1)* %out, i64 1
99  store i32 0, i32 addrspace(1)* %0
100  ret void
101}
102
103; MUBUF store with the largest possible immediate offset
104; CHECK-LABEL: {{^}}mubuf_store1:
105; CHECK: buffer_store_byte v{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0 offset:4095 ; encoding: [0xff,0x0f,0x60,0xe0
106
107define void @mubuf_store1(i8 addrspace(1)* %out) {
108entry:
109  %0 = getelementptr i8, i8 addrspace(1)* %out, i64 4095
110  store i8 0, i8 addrspace(1)* %0
111  ret void
112}
113
114; MUBUF store with an immediate byte offset that doesn't fit into 12-bits
115; CHECK-LABEL: {{^}}mubuf_store2:
116; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
117; CHECK: buffer_store_dword v{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x70,0xe0
118define void @mubuf_store2(i32 addrspace(1)* %out) {
119entry:
120  %0 = getelementptr i32, i32 addrspace(1)* %out, i64 1024
121  store i32 0, i32 addrspace(1)* %0
122  ret void
123}
124
125; MUBUF store with a 12-bit immediate offset and a register offset
126; CHECK-LABEL: {{^}}mubuf_store3:
127; CHECK-NOT: ADD
128; CHECK: buffer_store_dword v{{[0-9]}}, v[{{[0-9]:[0-9]}}], s[{{[0-9]:[0-9]}}], 0 addr64 offset:4 ; encoding: [0x04,0x80,0x70,0xe0
129define void @mubuf_store3(i32 addrspace(1)* %out, i64 %offset) {
130entry:
131  %0 = getelementptr i32, i32 addrspace(1)* %out, i64 %offset
132  %1 = getelementptr i32, i32 addrspace(1)* %0, i64 1
133  store i32 0, i32 addrspace(1)* %1
134  ret void
135}
136
137; CHECK-LABEL: {{^}}store_sgpr_ptr:
138; CHECK: buffer_store_dword v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, 0
139define void @store_sgpr_ptr(i32 addrspace(1)* %out) #0 {
140  store i32 99, i32 addrspace(1)* %out, align 4
141  ret void
142}
143
144; CHECK-LABEL: {{^}}store_sgpr_ptr_offset:
145; CHECK: buffer_store_dword v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, 0 offset:40
146define void @store_sgpr_ptr_offset(i32 addrspace(1)* %out) #0 {
147  %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 10
148  store i32 99, i32 addrspace(1)* %out.gep, align 4
149  ret void
150}
151
152; CHECK-LABEL: {{^}}store_sgpr_ptr_large_offset:
153; CHECK: s_mov_b32 [[SOFFSET:s[0-9]+]], 0x20000
154; CHECK: buffer_store_dword v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, [[SOFFSET]]
155define void @store_sgpr_ptr_large_offset(i32 addrspace(1)* %out) #0 {
156  %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 32768
157  store i32 99, i32 addrspace(1)* %out.gep, align 4
158  ret void
159}
160
161; CHECK-LABEL: {{^}}store_sgpr_ptr_large_offset_atomic:
162; CHECK: s_mov_b32 [[SOFFSET:s[0-9]+]], 0x20000
163; CHECK: buffer_atomic_add v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, [[SOFFSET]]
164define void @store_sgpr_ptr_large_offset_atomic(i32 addrspace(1)* %out) #0 {
165  %gep = getelementptr i32, i32 addrspace(1)* %out, i32 32768
166  %val = atomicrmw volatile add i32 addrspace(1)* %gep, i32 5 seq_cst
167  ret void
168}
169
170; CHECK-LABEL: {{^}}store_vgpr_ptr:
171; CHECK: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64
172define void @store_vgpr_ptr(i32 addrspace(1)* %out) #0 {
173  %tid = call i32 @llvm.r600.read.tidig.x() readnone
174  %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
175  store i32 99, i32 addrspace(1)* %out.gep, align 4
176  ret void
177}
178
179declare i32 @llvm.SI.buffer.load.dword.i32.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32) #3
180declare void @llvm.SI.tbuffer.store.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32)
181
182attributes #1 = { "ShaderType"="2" "unsafe-fp-math"="true" }
183attributes #3 = { nounwind readonly }
184