clang/test/CodeGenOpenCL/addr-space-struct-arg.cl

   1 // RUN: %clang_cc1 %s -emit-llvm -o - -O0 -ffake-address-space-map -triple i686-pc-darwin | FileCheck -enable-var-scope -check-prefixes=ALL,X86 %s
   2 // RUN: %clang_cc1 %s -emit-llvm -o - -O0 -triple amdgcn | FileCheck -enable-var-scope -check-prefixes=ALL,AMDGCN %s
   3 // RUN: %clang_cc1 %s -emit-llvm -o - -cl-std=CL2.0 -O0 -triple amdgcn | FileCheck -enable-var-scope -check-prefixes=ALL,AMDGCN,AMDGCN20 %s
   4 // RUN: %clang_cc1 %s -emit-llvm -o - -cl-std=CL1.2 -O0 -triple spir-unknown-unknown-unknown | FileCheck -enable-var-scope -check-prefixes=SPIR %s
   5 // RUN: %clang_cc1 %s -emit-llvm -o - -cl-std=CL3.0 -O0 -triple amdgcn -cl-ext=+__opencl_c_program_scope_global_variables | FileCheck -enable-var-scope -check-prefixes=ALL,AMDGCN,AMDGCN20 %s
   6 // RUN: %clang_cc1 %s -emit-llvm -o - -cl-std=CL3.0 -O0 -triple amdgcn | FileCheck -enable-var-scope -check-prefixes=ALL,AMDGCN %s
   7
   8 typedef int int2 __attribute__((ext_vector_type(2)));
   9
  10 typedef struct {
  11   int cells[9];
  12 } Mat3X3;
  13
  14 typedef struct {
  15   int cells[16];
  16 } Mat4X4;
  17
  18 typedef struct {
  19   int cells[1024];
  20 } Mat32X32;
  21
  22 typedef struct {
  23   int cells[4096];
  24 } Mat64X64;
  25
  26 struct StructOneMember {
  27   int2 x;
  28 };
  29
  30 struct StructTwoMember {
  31   int2 x;
  32   int2 y;
  33 };
  34
  35 struct LargeStructOneMember {
  36   int2 x[100];
  37 };
  38
  39 struct LargeStructTwoMember {
  40   int2 x[40];
  41   int2 y[20];
  42 };
  43
  44 #if (__OPENCL_C_VERSION__ == 200) || (__OPENCL_C_VERSION__ >= 300 && defined(__opencl_c_program_scope_global_variables))
  45 struct LargeStructOneMember g_s;
  46 #endif
  47
  48 // X86-LABEL: define{{.*}} void @foo(ptr noalias sret(%struct.Mat4X4) align 4 %agg.result, ptr noundef byval(%struct.Mat3X3) align 4 %in)
  49 // AMDGCN-LABEL: define{{.*}} %struct.Mat4X4 @foo([9 x i32] %in.coerce)
  50 Mat4X4 __attribute__((noinline)) foo(Mat3X3 in) {
  51   Mat4X4 out;
  52   return out;
  53 }
  54
  55 // ALL-LABEL: define {{.*}} void @ker
  56 // Expect two mem copies: one for the argument "in", and one for
  57 // the return value.
  58 // X86: call void @llvm.memcpy.p0.p1.i32(ptr
  59 // X86: call void @llvm.memcpy.p1.p0.i32(ptr addrspace(1)
  60
  61 // AMDGCN: load [9 x i32], ptr addrspace(1)
  62 // AMDGCN: call %struct.Mat4X4 @foo([9 x i32]
  63 // AMDGCN: call void @llvm.memcpy.p1.p5.i64(ptr addrspace(1)
  64 kernel void ker(global Mat3X3 *in, global Mat4X4 *out) {
  65   out[0] = foo(in[1]);
  66 }
  67
  68 // X86-LABEL: define{{.*}} void @foo_large(ptr noalias sret(%struct.Mat64X64) align 4 %agg.result, ptr noundef byval(%struct.Mat32X32) align 4 %in)
  69 // AMDGCN-LABEL: define{{.*}} void @foo_large(ptr addrspace(5) noalias sret(%struct.Mat64X64) align 4 %agg.result, ptr addrspace(5) noundef byref(%struct.Mat32X32) align 4 %{{.*}}
  70 // AMDGCN:       %in = alloca %struct.Mat32X32, align 4, addrspace(5)
  71 // AMDGCN-NEXT:  call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 4 %in, ptr addrspace(5) align 4 %{{.*}}, i64 4096, i1 false)
  72 Mat64X64 __attribute__((noinline)) foo_large(Mat32X32 in) {
  73   Mat64X64 out;
  74   return out;
  75 }
  76
  77 // ALL-LABEL: define {{.*}} void @ker_large
  78 // Expect two mem copies: one for the argument "in", and one for
  79 // the return value.
  80 // X86: call void @llvm.memcpy.p0.p1.i32(ptr
  81 // X86: call void @llvm.memcpy.p1.p0.i32(ptr addrspace(1)
  82 // AMDGCN: call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5)
  83 // AMDGCN: call void @llvm.memcpy.p1.p5.i64(ptr addrspace(1)
  84 kernel void ker_large(global Mat32X32 *in, global Mat64X64 *out) {
  85   out[0] = foo_large(in[1]);
  86 }
  87
  88 // AMDGCN-LABEL: define{{.*}} void @FuncOneMember(<2 x i32> %u.coerce)
  89 void FuncOneMember(struct StructOneMember u) {
  90   u.x = (int2)(0, 0);
  91 }
  92
  93 // AMDGCN-LABEL: define{{.*}} void @FuncOneLargeMember(ptr addrspace(5) noundef byref(%struct.LargeStructOneMember) align 8 %{{.*}}
  94 // AMDGCN:  %u = alloca %struct.LargeStructOneMember, align 8, addrspace(5)
  95 // AMDGCN:  call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 8 %u, ptr addrspace(5) align 8 %{{.*}}, i64 800, i1 false)
  96 // AMDGCN-NOT: addrspacecast
  97 // AMDGCN:   store <2 x i32> %{{.*}}, ptr addrspace(5)
  98 void FuncOneLargeMember(struct LargeStructOneMember u) {
  99   u.x[0] = (int2)(0, 0);
 100 }
 101
 102 // AMDGCN20-LABEL: define{{.*}} void @test_indirect_arg_globl()
 103 // AMDGCN20:  %[[byval_temp:.*]] = alloca %struct.LargeStructOneMember, align 8, addrspace(5)
 104 // AMDGCN20:  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)
 105 // AMDGCN20:  call void @FuncOneLargeMember(ptr addrspace(5) noundef byref(%struct.LargeStructOneMember) align 8 %[[byval_temp]])
 106 #if (__OPENCL_C_VERSION__ == 200) || (__OPENCL_C_VERSION__ >= 300 && defined(__opencl_c_program_scope_global_variables))
 107 void test_indirect_arg_globl(void) {
 108   FuncOneLargeMember(g_s);
 109 }
 110 #endif
 111
 112 // AMDGCN-LABEL: define{{.*}} amdgpu_kernel void @test_indirect_arg_local()
 113 // AMDGCN: %[[byval_temp:.*]] = alloca %struct.LargeStructOneMember, align 8, addrspace(5)
 114 // AMDGCN: 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)
 115 // AMDGCN: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref(%struct.LargeStructOneMember) align 8 %[[byval_temp]])
 116 kernel void test_indirect_arg_local(void) {
 117   local struct LargeStructOneMember l_s;
 118   FuncOneLargeMember(l_s);
 119 }
 120
 121 // AMDGCN-LABEL: define{{.*}} void @test_indirect_arg_private()
 122 // AMDGCN: %[[p_s:.*]] = alloca %struct.LargeStructOneMember, align 8, addrspace(5)
 123 // AMDGCN-NOT: @llvm.memcpy
 124 // AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref(%struct.LargeStructOneMember) align 8 %[[p_s]])
 125 void test_indirect_arg_private(void) {
 126   struct LargeStructOneMember p_s;
 127   FuncOneLargeMember(p_s);
 128 }
 129
 130 // AMDGCN-LABEL: define{{.*}} amdgpu_kernel void @KernelOneMember
 131 // AMDGCN-SAME:  (<2 x i32> %[[u_coerce:.*]])
 132 // AMDGCN:  %[[u:.*]] = alloca %struct.StructOneMember, align 8, addrspace(5)
 133 // AMDGCN:  %[[coerce_dive:.*]] = getelementptr inbounds %struct.StructOneMember, ptr addrspace(5) %[[u]], i32 0, i32 0
 134 // AMDGCN:  store <2 x i32> %[[u_coerce]], ptr addrspace(5) %[[coerce_dive]]
 135 // AMDGCN:  call void @FuncOneMember(<2 x i32>
 136 kernel void KernelOneMember(struct StructOneMember u) {
 137   FuncOneMember(u);
 138 }
 139
 140 // SPIR: call void @llvm.memcpy.p0.p1.i32
 141 // SPIR-NOT: addrspacecast
 142 kernel void KernelOneMemberSpir(global struct StructOneMember* u) {
 143   FuncOneMember(*u);
 144 }
 145
 146 // AMDGCN-LABEL: define{{.*}} amdgpu_kernel void @KernelLargeOneMember(
 147 // AMDGCN:  %[[U:.*]] = alloca %struct.LargeStructOneMember, align 8, addrspace(5)
 148 // AMDGCN:  store %struct.LargeStructOneMember %u.coerce, ptr addrspace(5) %[[U]], align 8
 149 // AMDGCN:  call void @FuncOneLargeMember(ptr addrspace(5) noundef byref(%struct.LargeStructOneMember) align 8 %[[U]])
 150 kernel void KernelLargeOneMember(struct LargeStructOneMember u) {
 151   FuncOneLargeMember(u);
 152 }
 153
 154 // AMDGCN-LABEL: define{{.*}} void @FuncTwoMember(<2 x i32> %u.coerce0, <2 x i32> %u.coerce1)
 155 void FuncTwoMember(struct StructTwoMember u) {
 156   u.y = (int2)(0, 0);
 157 }
 158
 159 // AMDGCN-LABEL: define dso_local void @FuncLargeTwoMember
 160 // AMDGCN-SAME: (ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]]) align 8 [[TMP0:%.*]])
 161 // AMDGCN: %[[U:.*]] = alloca %struct.LargeStructTwoMember, align 8, addrspace(5)
 162 // AMDGCN: call void @llvm.memcpy.p5.p5.i64(ptr addrspace(5) align 8 %[[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
 163 void FuncLargeTwoMember(struct LargeStructTwoMember u) {
 164   u.y[0] = (int2)(0, 0);
 165 }
 166
 167 // AMDGCN-LABEL: define{{.*}} amdgpu_kernel void @KernelTwoMember
 168 // AMDGCN-SAME:  (%struct.StructTwoMember %[[u_coerce:.*]])
 169 // AMDGCN:  %[[u:.*]] = alloca %struct.StructTwoMember, align 8, addrspace(5)
 170 // AMDGCN: %[[LD0:.*]] = load <2 x i32>, ptr addrspace(5)
 171 // AMDGCN: %[[LD1:.*]] = load <2 x i32>, ptr addrspace(5)
 172 // AMDGCN: call void @FuncTwoMember(<2 x i32> %[[LD0]], <2 x i32> %[[LD1]])
 173 kernel void KernelTwoMember(struct StructTwoMember u) {
 174   FuncTwoMember(u);
 175 }
 176
 177 // AMDGCN-LABEL: define{{.*}} amdgpu_kernel void @KernelLargeTwoMember
 178 // AMDGCN-SAME:  (%struct.LargeStructTwoMember %[[u_coerce:.*]])
 179 // AMDGCN:  %[[u:.*]] = alloca %struct.LargeStructTwoMember, align 8, addrspace(5)
 180 // AMDGCN:  store %struct.LargeStructTwoMember %[[u_coerce]], ptr addrspace(5) %[[u]]
 181 // AMDGCN:  call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref(%struct.LargeStructTwoMember) align 8 %[[u]])
 182 kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
 183   FuncLargeTwoMember(u);
 184 }