clang/test/CodeGenOpenCL/amdgpu-abi-struct-arg-byref.cl

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
   2 // RUN: %clang_cc1 %s -emit-llvm -o - -cl-std=CL2.0 -O0 -triple amdgcn | FileCheck -check-prefix=AMDGCN %s
   3
   4 typedef int int2 __attribute__((ext_vector_type(2)));
   5
   6 typedef struct {
   7   int cells[9];
   8 } Mat3X3;
   9
  10 typedef struct {
  11   int cells[16];
  12 } Mat4X4;
  13
  14 typedef struct {
  15   int cells[1024];
  16 } Mat32X32;
  17
  18 typedef struct {
  19   int cells[4096];
  20 } Mat64X64;
  21
  22 struct StructOneMember {
  23   int2 x;
  24 };
  25
  26 struct StructTwoMember {
  27   int2 x;
  28   int2 y;
  29 };
  30
  31 struct LargeStructOneMember {
  32   int2 x[100];
  33 };
  34
  35 struct LargeStructTwoMember {
  36   int2 x[40];
  37   int2 y[20];
  38 };
  39
  40 #if (__OPENCL_C_VERSION__ == 200) || (__OPENCL_C_VERSION__ >= 300 && defined(__opencl_c_program_scope_global_variables))
  41 struct LargeStructOneMember g_s;
  42 #endif
  43
  44
  45 // AMDGCN-LABEL: define dso_local %struct.Mat4X4 @foo
  46 // AMDGCN-SAME: ([9 x i32] [[IN_COERCE:%.*]]) #[[ATTR0:[0-9]+]] {
  47 // AMDGCN-NEXT:  entry:
  48 // AMDGCN-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_MAT4X4:%.*]], align 4, addrspace(5)
  49 // AMDGCN-NEXT:    [[IN:%.*]] = alloca [[STRUCT_MAT3X3:%.*]], align 4, addrspace(5)
  50 // AMDGCN-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_MAT3X3]], ptr addrspace(5) [[IN]], i32 0, i32 0
  51 // AMDGCN-NEXT:    store [9 x i32] [[IN_COERCE]], ptr addrspace(5) [[COERCE_DIVE]], align 4
  52 // AMDGCN-NEXT:    [[TMP0:%.*]] = load [[STRUCT_MAT4X4]], ptr addrspace(5) [[RETVAL]], align 4
  53 // AMDGCN-NEXT:    ret [[STRUCT_MAT4X4]] [[TMP0]]
  54 //
  55 Mat4X4 __attribute__((noinline)) foo(Mat3X3 in) {
  56   Mat4X4 out;
  57   return out;
  58 }
  59
  60 // Expect two mem copies: one for the argument "in", and one for
  61 // the return value.
  62
  63 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @ker
  64 // AMDGCN-SAME: (ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR1:[0-9]+]] !kernel_arg_addr_space !4 !kernel_arg_access_qual !5 !kernel_arg_type !6 !kernel_arg_base_type !6 !kernel_arg_type_qual !7 {
  65 // AMDGCN-NEXT:  entry:
  66 // AMDGCN-NEXT:    [[IN_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
  67 // AMDGCN-NEXT:    [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
  68 // AMDGCN-NEXT:    [[TMP:%.*]] = alloca [[STRUCT_MAT4X4:%.*]], align 4, addrspace(5)
  69 // AMDGCN-NEXT:    store ptr addrspace(1) [[IN]], ptr addrspace(5) [[IN_ADDR]], align 8
  70 // AMDGCN-NEXT:    store ptr addrspace(1) [[OUT]], ptr addrspace(5) [[OUT_ADDR]], align 8
  71 // AMDGCN-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr addrspace(5) [[OUT_ADDR]], align 8
  72 // AMDGCN-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_MAT4X4]], ptr addrspace(1) [[TMP0]], i64 0
  73 // AMDGCN-NEXT:    [[TMP1:%.*]] = load ptr addrspace(1), ptr addrspace(5) [[IN_ADDR]], align 8
  74 // AMDGCN-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds [[STRUCT_MAT3X3:%.*]], ptr addrspace(1) [[TMP1]], i64 1
  75 // AMDGCN-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_MAT3X3]], ptr addrspace(1) [[ARRAYIDX1]], i32 0, i32 0
  76 // AMDGCN-NEXT:    [[TMP3:%.*]] = load [9 x i32], ptr addrspace(1) [[TMP2]], align 4
  77 // AMDGCN-NEXT:    [[CALL:%.*]] = call [[STRUCT_MAT4X4]] @foo([9 x i32] [[TMP3]]) #[[ATTR3:[0-9]+]]
  78 // AMDGCN-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_MAT4X4]], ptr addrspace(5) [[TMP]], i32 0, i32 0
  79 // AMDGCN-NEXT:    [[TMP5:%.*]] = extractvalue [[STRUCT_MAT4X4]] [[CALL]], 0
  80 // AMDGCN-NEXT:    store [16 x i32] [[TMP5]], ptr addrspace(5) [[TMP4]], align 4
  81 // AMDGCN-NEXT:    call void @llvm.memcpy.p1.p5.i64(ptr addrspace(1) align 4 [[ARRAYIDX]], ptr addrspace(5) align 4 [[TMP]], i64 64, i1 false)
  82 // AMDGCN-NEXT:    ret void
  83 //
  84 kernel void ker(global Mat3X3 *in, global Mat4X4 *out) {
  85   out[0] = foo(in[1]);
  86 }
  87
  88 // AMDGCN-LABEL: define dso_local void @foo_large
  89 // AMDGCN-SAME: (ptr addrspace(5) noalias sret([[STRUCT_MAT64X64:%.*]]) align 4 [[AGG_RESULT:%.*]], ptr addrspace(5) noundef byref([[STRUCT_MAT32X32:%.*]]) align 4 [[TMP0:%.*]]) #[[ATTR0]] {
  90 // AMDGCN-NEXT:  entry:
  91 // AMDGCN-NEXT:    [[IN:%.*]] = alloca [[STRUCT_MAT32X32]], align 4, addrspace(5)
  92 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 4 [[IN]], ptr addrspace(5) align 4 [[TMP0]], i64 4096, i1 false)
  93 // AMDGCN-NEXT:    ret void
  94 //
  95 Mat64X64 __attribute__((noinline)) foo_large(Mat32X32 in) {
  96   Mat64X64 out;
  97   return out;
  98 }
  99
 100 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @ker_large
 101 // AMDGCN-SAME: (ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !4 !kernel_arg_access_qual !5 !kernel_arg_type !8 !kernel_arg_base_type !8 !kernel_arg_type_qual !7 {
 102 // AMDGCN-NEXT:  entry:
 103 // AMDGCN-NEXT:    [[IN_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
 104 // AMDGCN-NEXT:    [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
 105 // AMDGCN-NEXT:    [[TMP:%.*]] = alloca [[STRUCT_MAT64X64:%.*]], align 4, addrspace(5)
 106 // AMDGCN-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_MAT32X32:%.*]], align 4, addrspace(5)
 107 // AMDGCN-NEXT:    store ptr addrspace(1) [[IN]], ptr addrspace(5) [[IN_ADDR]], align 8
 108 // AMDGCN-NEXT:    store ptr addrspace(1) [[OUT]], ptr addrspace(5) [[OUT_ADDR]], align 8
 109 // AMDGCN-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr addrspace(5) [[OUT_ADDR]], align 8
 110 // AMDGCN-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_MAT64X64]], ptr addrspace(1) [[TMP0]], i64 0
 111 // AMDGCN-NEXT:    [[TMP1:%.*]] = load ptr addrspace(1), ptr addrspace(5) [[IN_ADDR]], align 8
 112 // AMDGCN-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds [[STRUCT_MAT32X32]], ptr addrspace(1) [[TMP1]], i64 1
 113 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) align 4 [[BYVAL_TEMP]], ptr addrspace(1) align 4 [[ARRAYIDX1]], i64 4096, i1 false)
 114 // AMDGCN-NEXT:    call void @foo_large(ptr addrspace(5) sret([[STRUCT_MAT64X64]]) align 4 [[TMP]], ptr addrspace(5) noundef byref([[STRUCT_MAT32X32]]) align 4 [[BYVAL_TEMP]]) #[[ATTR3]]
 115 // AMDGCN-NEXT:    call void @llvm.memcpy.p1.p5.i64(ptr addrspace(1) align 4 [[ARRAYIDX]], ptr addrspace(5) align 4 [[TMP]], i64 16384, i1 false)
 116 // AMDGCN-NEXT:    ret void
 117 //
 118 kernel void ker_large(global Mat32X32 *in, global Mat64X64 *out) {
 119   out[0] = foo_large(in[1]);
 120 }
 121
 122 // AMDGCN-LABEL: define dso_local void @FuncOneMember
 123 // AMDGCN-SAME: (<2 x i32> [[U_COERCE:%.*]]) #[[ATTR0]] {
 124 // AMDGCN-NEXT:  entry:
 125 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_STRUCTONEMEMBER:%.*]], align 8, addrspace(5)
 126 // AMDGCN-NEXT:    [[DOTCOMPOUNDLITERAL:%.*]] = alloca <2 x i32>, align 8, addrspace(5)
 127 // AMDGCN-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 128 // AMDGCN-NEXT:    store <2 x i32> [[U_COERCE]], ptr addrspace(5) [[COERCE_DIVE]], align 8
 129 // AMDGCN-NEXT:    store <2 x i32> zeroinitializer, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 130 // AMDGCN-NEXT:    [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 131 // AMDGCN-NEXT:    [[X:%.*]] = getelementptr inbounds [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 132 // AMDGCN-NEXT:    store <2 x i32> [[TMP0]], ptr addrspace(5) [[X]], align 8
 133 // AMDGCN-NEXT:    ret void
 134 //
 135 void FuncOneMember(struct StructOneMember u) {
 136   u.x = (int2)(0, 0);
 137 }
 138
 139 // AMDGCN-LABEL: define dso_local void @FuncOneLargeMember
 140 // AMDGCN-SAME: (ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR0]] {
 141 // AMDGCN-NEXT:  entry:
 142 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
 143 // AMDGCN-NEXT:    [[DOTCOMPOUNDLITERAL:%.*]] = alloca <2 x i32>, align 8, addrspace(5)
 144 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 800, i1 false)
 145 // AMDGCN-NEXT:    store <2 x i32> zeroinitializer, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 146 // AMDGCN-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 147 // AMDGCN-NEXT:    [[X:%.*]] = getelementptr inbounds [[STRUCT_LARGESTRUCTONEMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 148 // AMDGCN-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x <2 x i32>], ptr addrspace(5) [[X]], i64 0, i64 0
 149 // AMDGCN-NEXT:    store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
 150 // AMDGCN-NEXT:    ret void
 151 //
 152 void FuncOneLargeMember(struct LargeStructOneMember u) {
 153   u.x[0] = (int2)(0, 0);
 154 }
 155
 156 #if (__OPENCL_C_VERSION__ == 200) || (__OPENCL_C_VERSION__ >= 300 && defined(__opencl_c_program_scope_global_variables))
 157 // AMDGCN-LABEL: define dso_local void @test_indirect_arg_globl
 158 // AMDGCN-SAME: () #[[ATTR0]] {
 159 // AMDGCN-NEXT:  entry:
 160 // AMDGCN-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
 161 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr addrspace(1) align 8 @g_s, i64 800, i1 false)
 162 // AMDGCN-NEXT:    call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
 163 // AMDGCN-NEXT:    ret void
 164 //
 165 void test_indirect_arg_globl(void) {
 166   FuncOneLargeMember(g_s);
 167 }
 168 #endif
 169
 170 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @test_indirect_arg_local
 171 // AMDGCN-SAME: () #[[ATTR1]] !kernel_arg_addr_space !9 !kernel_arg_access_qual !9 !kernel_arg_type !9 !kernel_arg_base_type !9 !kernel_arg_type_qual !9 {
 172 // AMDGCN-NEXT:  entry:
 173 // AMDGCN-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
 174 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p3.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr addrspace(3) align 8 @test_indirect_arg_local.l_s, i64 800, i1 false)
 175 // AMDGCN-NEXT:    call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
 176 // AMDGCN-NEXT:    ret void
 177 //
 178 kernel void test_indirect_arg_local(void) {
 179   local struct LargeStructOneMember l_s;
 180   FuncOneLargeMember(l_s);
 181 }
 182
 183 // AMDGCN-LABEL: define dso_local void @test_indirect_arg_private
 184 // AMDGCN-SAME: () #[[ATTR0]] {
 185 // AMDGCN-NEXT:  entry:
 186 // AMDGCN-NEXT:    [[P_S:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
 187 // AMDGCN-NEXT:    call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[P_S]]) #[[ATTR3]]
 188 // AMDGCN-NEXT:    ret void
 189 //
 190 void test_indirect_arg_private(void) {
 191   struct LargeStructOneMember p_s;
 192   FuncOneLargeMember(p_s);
 193 }
 194
 195 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @KernelOneMember
 196 // AMDGCN-SAME: (<2 x i32> [[U_COERCE:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !10 !kernel_arg_access_qual !11 !kernel_arg_type !12 !kernel_arg_base_type !12 !kernel_arg_type_qual !13 {
 197 // AMDGCN-NEXT:  entry:
 198 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_STRUCTONEMEMBER:%.*]], align 8, addrspace(5)
 199 // AMDGCN-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 200 // AMDGCN-NEXT:    store <2 x i32> [[U_COERCE]], ptr addrspace(5) [[COERCE_DIVE]], align 8
 201 // AMDGCN-NEXT:    [[COERCE_DIVE1:%.*]] = getelementptr inbounds [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 202 // AMDGCN-NEXT:    [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[COERCE_DIVE1]], align 8
 203 // AMDGCN-NEXT:    call void @FuncOneMember(<2 x i32> [[TMP0]]) #[[ATTR3]]
 204 // AMDGCN-NEXT:    ret void
 205 //
 206 kernel void KernelOneMember(struct StructOneMember u) {
 207   FuncOneMember(u);
 208 }
 209
 210 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @KernelOneMemberSpir
 211 // AMDGCN-SAME: (ptr addrspace(1) noundef align 8 [[U:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !14 !kernel_arg_access_qual !11 !kernel_arg_type !15 !kernel_arg_base_type !15 !kernel_arg_type_qual !13 {
 212 // AMDGCN-NEXT:  entry:
 213 // AMDGCN-NEXT:    [[U_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
 214 // AMDGCN-NEXT:    store ptr addrspace(1) [[U]], ptr addrspace(5) [[U_ADDR]], align 8
 215 // AMDGCN-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr addrspace(5) [[U_ADDR]], align 8
 216 // AMDGCN-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_STRUCTONEMEMBER:%.*]], ptr addrspace(1) [[TMP0]], i32 0, i32 0
 217 // AMDGCN-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr addrspace(1) [[COERCE_DIVE]], align 8
 218 // AMDGCN-NEXT:    call void @FuncOneMember(<2 x i32> [[TMP1]]) #[[ATTR3]]
 219 // AMDGCN-NEXT:    ret void
 220 //
 221 kernel void KernelOneMemberSpir(global struct StructOneMember* u) {
 222   FuncOneMember(*u);
 223 }
 224
 225 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @KernelLargeOneMember
 226 // AMDGCN-SAME: ([[STRUCT_LARGESTRUCTONEMEMBER:%.*]] [[U_COERCE:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !10 !kernel_arg_access_qual !11 !kernel_arg_type !16 !kernel_arg_base_type !16 !kernel_arg_type_qual !13 {
 227 // AMDGCN-NEXT:  entry:
 228 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
 229 // AMDGCN-NEXT:    store [[STRUCT_LARGESTRUCTONEMEMBER]] [[U_COERCE]], ptr addrspace(5) [[U]], align 8
 230 // AMDGCN-NEXT:    call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[U]]) #[[ATTR3]]
 231 // AMDGCN-NEXT:    ret void
 232 //
 233 kernel void KernelLargeOneMember(struct LargeStructOneMember u) {
 234   FuncOneLargeMember(u);
 235 }
 236
 237 // AMDGCN-LABEL: define dso_local void @FuncTwoMember
 238 // AMDGCN-SAME: (<2 x i32> [[U_COERCE0:%.*]], <2 x i32> [[U_COERCE1:%.*]]) #[[ATTR0]] {
 239 // AMDGCN-NEXT:  entry:
 240 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_STRUCTTWOMEMBER:%.*]], align 8, addrspace(5)
 241 // AMDGCN-NEXT:    [[DOTCOMPOUNDLITERAL:%.*]] = alloca <2 x i32>, align 8, addrspace(5)
 242 // AMDGCN-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 243 // AMDGCN-NEXT:    store <2 x i32> [[U_COERCE0]], ptr addrspace(5) [[TMP0]], align 8
 244 // AMDGCN-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 1
 245 // AMDGCN-NEXT:    store <2 x i32> [[U_COERCE1]], ptr addrspace(5) [[TMP1]], align 8
 246 // AMDGCN-NEXT:    store <2 x i32> zeroinitializer, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 247 // AMDGCN-NEXT:    [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 248 // AMDGCN-NEXT:    [[Y:%.*]] = getelementptr inbounds [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 1
 249 // AMDGCN-NEXT:    store <2 x i32> [[TMP2]], ptr addrspace(5) [[Y]], align 8
 250 // AMDGCN-NEXT:    ret void
 251 //
 252 void FuncTwoMember(struct StructTwoMember u) {
 253   u.y = (int2)(0, 0);
 254 }
 255
 256 // AMDGCN-LABEL: define dso_local void @FuncLargeTwoMember
 257 // AMDGCN-SAME: (ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR0]] {
 258 // AMDGCN-NEXT:  entry:
 259 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
 260 // AMDGCN-NEXT:    [[DOTCOMPOUNDLITERAL:%.*]] = alloca <2 x i32>, align 8, addrspace(5)
 261 // AMDGCN-NEXT:    call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
 262 // AMDGCN-NEXT:    store <2 x i32> zeroinitializer, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 263 // AMDGCN-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr addrspace(5) [[DOTCOMPOUNDLITERAL]], align 8
 264 // AMDGCN-NEXT:    [[Y:%.*]] = getelementptr inbounds [[STRUCT_LARGESTRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 1
 265 // AMDGCN-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [20 x <2 x i32>], ptr addrspace(5) [[Y]], i64 0, i64 0
 266 // AMDGCN-NEXT:    store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
 267 // AMDGCN-NEXT:    ret void
 268 //
 269 void FuncLargeTwoMember(struct LargeStructTwoMember u) {
 270   u.y[0] = (int2)(0, 0);
 271 }
 272
 273 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @KernelTwoMember
 274 // AMDGCN-SAME: ([[STRUCT_STRUCTTWOMEMBER:%.*]] [[U_COERCE:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !10 !kernel_arg_access_qual !11 !kernel_arg_type !17 !kernel_arg_base_type !17 !kernel_arg_type_qual !13 {
 275 // AMDGCN-NEXT:  entry:
 276 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_STRUCTTWOMEMBER]], align 8, addrspace(5)
 277 // AMDGCN-NEXT:    store [[STRUCT_STRUCTTWOMEMBER]] [[U_COERCE]], ptr addrspace(5) [[U]], align 8
 278 // AMDGCN-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 0
 279 // AMDGCN-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP0]], align 8
 280 // AMDGCN-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U]], i32 0, i32 1
 281 // AMDGCN-NEXT:    [[TMP3:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP2]], align 8
 282 // AMDGCN-NEXT:    call void @FuncTwoMember(<2 x i32> [[TMP1]], <2 x i32> [[TMP3]]) #[[ATTR3]]
 283 // AMDGCN-NEXT:    ret void
 284 //
 285 kernel void KernelTwoMember(struct StructTwoMember u) {
 286   FuncTwoMember(u);
 287 }
 288
 289 // AMDGCN-LABEL: define dso_local amdgpu_kernel void @KernelLargeTwoMember
 290 // AMDGCN-SAME: ([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]] [[U_COERCE:%.*]]) #[[ATTR1]] !kernel_arg_addr_space !10 !kernel_arg_access_qual !11 !kernel_arg_type !18 !kernel_arg_base_type !18 !kernel_arg_type_qual !13 {
 291 // AMDGCN-NEXT:  entry:
 292 // AMDGCN-NEXT:    [[U:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
 293 // AMDGCN-NEXT:    store [[STRUCT_LARGESTRUCTTWOMEMBER]] [[U_COERCE]], ptr addrspace(5) [[U]], align 8
 294 // AMDGCN-NEXT:    call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[U]]) #[[ATTR3]]
 295 // AMDGCN-NEXT:    ret void
 296 //
 297 kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
 298   FuncLargeTwoMember(u);
 299 }