1 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=R600-CHECK %s 2 ; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK %s 3 4 ; R600-CHECK: @ngroups_x 5 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 6 ; R600-CHECK: MOV [[VAL]], KC0[0].X 7 ; SI-CHECK: @ngroups_x 8 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0 9 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 10 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 11 define void @ngroups_x (i32 addrspace(1)* %out) { 12 entry: 13 %0 = call i32 @llvm.r600.read.ngroups.x() #0 14 store i32 %0, i32 addrspace(1)* %out 15 ret void 16 } 17 18 ; R600-CHECK: @ngroups_y 19 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 20 ; R600-CHECK: MOV [[VAL]], KC0[0].Y 21 ; SI-CHECK: @ngroups_y 22 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x1 23 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 24 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 25 define void @ngroups_y (i32 addrspace(1)* %out) { 26 entry: 27 %0 = call i32 @llvm.r600.read.ngroups.y() #0 28 store i32 %0, i32 addrspace(1)* %out 29 ret void 30 } 31 32 ; R600-CHECK: @ngroups_z 33 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 34 ; R600-CHECK: MOV [[VAL]], KC0[0].Z 35 ; SI-CHECK: @ngroups_z 36 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x2 37 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 38 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 39 define void @ngroups_z (i32 addrspace(1)* %out) { 40 entry: 41 %0 = call i32 @llvm.r600.read.ngroups.z() #0 42 store i32 %0, i32 addrspace(1)* %out 43 ret void 44 } 45 46 ; R600-CHECK: @global_size_x 47 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 48 ; R600-CHECK: MOV [[VAL]], KC0[0].W 49 ; SI-CHECK: @global_size_x 50 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x3 51 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 52 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 53 define void @global_size_x (i32 addrspace(1)* %out) { 54 entry: 55 %0 = call i32 @llvm.r600.read.global.size.x() #0 56 store i32 %0, i32 addrspace(1)* %out 57 ret void 58 } 59 60 ; R600-CHECK: @global_size_y 61 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 62 ; R600-CHECK: MOV [[VAL]], KC0[1].X 63 ; SI-CHECK: @global_size_y 64 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x4 65 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 66 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 67 define void @global_size_y (i32 addrspace(1)* %out) { 68 entry: 69 %0 = call i32 @llvm.r600.read.global.size.y() #0 70 store i32 %0, i32 addrspace(1)* %out 71 ret void 72 } 73 74 ; R600-CHECK: @global_size_z 75 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 76 ; R600-CHECK: MOV [[VAL]], KC0[1].Y 77 ; SI-CHECK: @global_size_z 78 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x5 79 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 80 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 81 define void @global_size_z (i32 addrspace(1)* %out) { 82 entry: 83 %0 = call i32 @llvm.r600.read.global.size.z() #0 84 store i32 %0, i32 addrspace(1)* %out 85 ret void 86 } 87 88 ; R600-CHECK: @local_size_x 89 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 90 ; R600-CHECK: MOV [[VAL]], KC0[1].Z 91 ; SI-CHECK: @local_size_x 92 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x6 93 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 94 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 95 define void @local_size_x (i32 addrspace(1)* %out) { 96 entry: 97 %0 = call i32 @llvm.r600.read.local.size.x() #0 98 store i32 %0, i32 addrspace(1)* %out 99 ret void 100 } 101 102 ; R600-CHECK: @local_size_y 103 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 104 ; R600-CHECK: MOV [[VAL]], KC0[1].W 105 ; SI-CHECK: @local_size_y 106 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x7 107 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 108 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 109 define void @local_size_y (i32 addrspace(1)* %out) { 110 entry: 111 %0 = call i32 @llvm.r600.read.local.size.y() #0 112 store i32 %0, i32 addrspace(1)* %out 113 ret void 114 } 115 116 ; R600-CHECK: @local_size_z 117 ; R600-CHECK: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]] 118 ; R600-CHECK: MOV [[VAL]], KC0[2].X 119 ; SI-CHECK: @local_size_z 120 ; SI-CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]], s[0:1], 0x8 121 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], [[VAL]] 122 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 123 define void @local_size_z (i32 addrspace(1)* %out) { 124 entry: 125 %0 = call i32 @llvm.r600.read.local.size.z() #0 126 store i32 %0, i32 addrspace(1)* %out 127 ret void 128 } 129 130 ; The tgid values are stored in ss offset by the number of user ss. 131 ; Currently we always use exactly 2 user ss for the pointer to the 132 ; kernel arguments, but this may change in the future. 133 134 ; SI-CHECK: @tgid_x 135 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], s2 136 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 137 define void @tgid_x (i32 addrspace(1)* %out) { 138 entry: 139 %0 = call i32 @llvm.r600.read.tgid.x() #0 140 store i32 %0, i32 addrspace(1)* %out 141 ret void 142 } 143 144 ; SI-CHECK: @tgid_y 145 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], s3 146 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 147 define void @tgid_y (i32 addrspace(1)* %out) { 148 entry: 149 %0 = call i32 @llvm.r600.read.tgid.y() #0 150 store i32 %0, i32 addrspace(1)* %out 151 ret void 152 } 153 154 ; SI-CHECK: @tgid_z 155 ; SI-CHECK: V_MOV_B32_e32 [[VVAL:v[0-9]+]], s4 156 ; SI-CHECK: BUFFER_STORE_DWORD [[VVAL]] 157 define void @tgid_z (i32 addrspace(1)* %out) { 158 entry: 159 %0 = call i32 @llvm.r600.read.tgid.z() #0 160 store i32 %0, i32 addrspace(1)* %out 161 ret void 162 } 163 164 ; SI-CHECK: @tidig_x 165 ; SI-CHECK: BUFFER_STORE_DWORD v0 166 define void @tidig_x (i32 addrspace(1)* %out) { 167 entry: 168 %0 = call i32 @llvm.r600.read.tidig.x() #0 169 store i32 %0, i32 addrspace(1)* %out 170 ret void 171 } 172 173 ; SI-CHECK: @tidig_y 174 ; SI-CHECK: BUFFER_STORE_DWORD v1 175 define void @tidig_y (i32 addrspace(1)* %out) { 176 entry: 177 %0 = call i32 @llvm.r600.read.tidig.y() #0 178 store i32 %0, i32 addrspace(1)* %out 179 ret void 180 } 181 182 ; SI-CHECK: @tidig_z 183 ; SI-CHECK: BUFFER_STORE_DWORD v2 184 define void @tidig_z (i32 addrspace(1)* %out) { 185 entry: 186 %0 = call i32 @llvm.r600.read.tidig.z() #0 187 store i32 %0, i32 addrspace(1)* %out 188 ret void 189 } 190 191 declare i32 @llvm.r600.read.ngroups.x() #0 192 declare i32 @llvm.r600.read.ngroups.y() #0 193 declare i32 @llvm.r600.read.ngroups.z() #0 194 195 declare i32 @llvm.r600.read.global.size.x() #0 196 declare i32 @llvm.r600.read.global.size.y() #0 197 declare i32 @llvm.r600.read.global.size.z() #0 198 199 declare i32 @llvm.r600.read.local.size.x() #0 200 declare i32 @llvm.r600.read.local.size.y() #0 201 declare i32 @llvm.r600.read.local.size.z() #0 202 203 declare i32 @llvm.r600.read.tgid.x() #0 204 declare i32 @llvm.r600.read.tgid.y() #0 205 declare i32 @llvm.r600.read.tgid.z() #0 206 207 declare i32 @llvm.r600.read.tidig.x() #0 208 declare i32 @llvm.r600.read.tidig.y() #0 209 declare i32 @llvm.r600.read.tidig.z() #0 210 211 attributes #0 = { readnone } 212
