clang/test/CodeGenSYCL/cuda-address-space-conversions.cpp

   1 // RUN: %clang_cc1 -triple nvptx64-nvidia-cuda -fsycl-is-device -disable-llvm-passes -emit-llvm %s -o - | FileCheck %s
   2 void bar(int &Data) {}
   3 // CHECK: define dso_local void @[[RAW_REF:[a-zA-Z0-9_]+]](ptr noundef nonnull align 4 dereferenceable(4) %
   4 void bar2(int &Data) {}
   5 // CHECK: define dso_local void @[[RAW_REF2:[a-zA-Z0-9_]+]](ptr noundef nonnull align 4 dereferenceable(4) %
   6 void bar(__attribute__((opencl_local)) int &Data) {}
   7 // CHECK: define dso_local void @[[LOCAL_REF:[a-zA-Z0-9_]+]](ptr addrspace(3) noundef align 4 dereferenceable(4) %
   8 void foo(int *Data) {}
   9 // CHECK: define dso_local void @[[RAW_PTR:[a-zA-Z0-9_]+]](ptr noundef %
  10 void foo2(int *Data) {}
  11 // CHECK: define dso_local void @[[RAW_PTR2:[a-zA-Z0-9_]+]](ptr noundef %
  12 void foo(__attribute__((opencl_local)) int *Data) {}
  13 // CHECK: define dso_local void @[[LOC_PTR:[a-zA-Z0-9_]+]](ptr addrspace(3) noundef %
  14
  15 template <typename T>
  16 void tmpl(T t);
  17 // See Check Lines below.
  18
  19 void usages() {
  20   int *NoAS;
  21   // CHECK: [[NoAS:%[a-zA-Z0-9]+]] = alloca ptr, align 8
  22   __attribute__((opencl_global)) int *GLOB;
  23   // CHECK: [[GLOB:%[a-zA-Z0-9]+]] = alloca ptr addrspace(1), align 8
  24   __attribute__((opencl_local)) int *LOC;
  25   // CHECK: [[LOC:%[a-zA-Z0-9]+]] = alloca ptr addrspace(3), align 8
  26   __attribute__((opencl_private)) int *PRIV;
  27   // CHECK: [[PRIV:%[a-zA-Z0-9]+]] = alloca ptr, align 8
  28   __attribute__((opencl_global_device)) int *GLOBDEVICE;
  29   // CHECK: [[GLOB_DEVICE:%[a-zA-Z0-9]+]] = alloca ptr addrspace(1), align 8
  30   __attribute__((opencl_global_host)) int *GLOBHOST;
  31   // CHECK: [[GLOB_HOST:%[a-zA-Z0-9]+]] = alloca ptr addrspace(1), align 8
  32   LOC = nullptr;
  33   // CHECK: store ptr addrspace(3) addrspacecast (ptr null to ptr addrspace(3)), ptr [[LOC]], align 8
  34   GLOB = nullptr;
  35   // CHECK: store ptr addrspace(1) null, ptr [[GLOB]], align 8
  36   NoAS = (int *)GLOB;
  37   // CHECK: [[GLOB_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
  38   // CHECK: [[GLOB_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(1) [[GLOB_LOAD]] to ptr
  39   // CHECK: store ptr [[GLOB_CAST]], ptr [[NoAS]], align 8
  40   NoAS = (int *)LOC;
  41   // CHECK: [[LOC_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
  42   // CHECK: [[LOC_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(3) [[LOC_LOAD]] to ptr
  43   // CHECK: store ptr [[LOC_CAST]], ptr [[NoAS]], align 8
  44   NoAS = (int *)PRIV;
  45   // CHECK: [[LOC_LOAD:%[a-zA-Z0-9]+]] = load ptr, ptr [[PRIV]], align 8
  46   // CHECK: store ptr [[LOC_LOAD]], ptr [[NoAS]], align 8
  47   GLOB = (__attribute__((opencl_global)) int *)NoAS;
  48   // CHECK: [[NoAS_LOAD:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
  49   // CHECK: [[NoAS_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr [[NoAS_LOAD]] to ptr addrspace(1)
  50   // CHECK: store ptr addrspace(1) [[NoAS_CAST]], ptr [[GLOB]], align 8
  51   LOC = (__attribute__((opencl_local)) int *)NoAS;
  52   // CHECK: [[NoAS_LOAD:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
  53   // CHECK: [[NoAS_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr [[NoAS_LOAD]] to ptr addrspace(3)
  54   // CHECK: store ptr addrspace(3) [[NoAS_CAST]], ptr [[LOC]], align 8
  55   PRIV = (__attribute__((opencl_private)) int *)NoAS;
  56   // CHECK: [[NoAS_LOAD:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
  57   // CHECK: store ptr [[NoAS_LOAD]], ptr [[PRIV]], align 8
  58   GLOB = (__attribute__((opencl_global)) int *)GLOBDEVICE;
  59   // CHECK: [[GLOBDEVICE_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB_DEVICE]], align 8
  60   // CHECK: store ptr addrspace(1) [[GLOBDEVICE_LOAD]], ptr %GLOB, align 8
  61   GLOB = (__attribute__((opencl_global)) int *)GLOBHOST;
  62   // CHECK: [[GLOB_HOST_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB_HOST]], align 8
  63   // CHECK: store ptr addrspace(1) [[GLOB_HOST_LOAD]], ptr [[GLOB]], align 8
  64   bar(*GLOB);
  65   // CHECK: [[GLOB_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
  66   // CHECK: [[GLOB_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(1) [[GLOB_LOAD]] to ptr
  67   // CHECK: call void @[[RAW_REF]](ptr noundef nonnull align 4 dereferenceable(4) [[GLOB_CAST]])
  68   bar2(*GLOB);
  69   // CHECK: [[GLOB_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
  70   // CHECK: [[GLOB_CAST:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(1) [[GLOB_LOAD]] to ptr
  71   // CHECK: call void @[[RAW_REF2]](ptr noundef nonnull align 4 dereferenceable(4) [[GLOB_CAST]])
  72   bar(*LOC);
  73   // CHECK: [[LOC_LOAD:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
  74   // CHECK: call void @[[LOCAL_REF]](ptr addrspace(3) noundef align 4 dereferenceable(4) [[LOC_LOAD]])
  75   bar2(*LOC);
  76   // CHECK: [[LOC_LOAD2:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
  77   // CHECK: [[LOC_CAST2:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(3) [[LOC_LOAD2]] to ptr
  78   // CHECK: call void @[[RAW_REF2]](ptr noundef nonnull align 4 dereferenceable(4) [[LOC_CAST2]])
  79   bar(*NoAS);
  80   // CHECK: [[NoAS_LOAD:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
  81   // CHECK: call void @[[RAW_REF]](ptr noundef nonnull align 4 dereferenceable(4) [[NoAS_LOAD]])
  82   bar2(*NoAS);
  83   // CHECK: [[NoAS_LOAD2:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
  84   // CHECK: call void @[[RAW_REF2]](ptr noundef nonnull align 4 dereferenceable(4) [[NoAS_LOAD2]])
  85   foo(GLOB);
  86   // CHECK: [[GLOB_LOAD3:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
  87   // CHECK: [[GLOB_CAST3:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(1) [[GLOB_LOAD3]] to ptr
  88   // CHECK: call void @[[RAW_PTR]](ptr noundef [[GLOB_CAST3]])
  89   foo2(GLOB);
  90   // CHECK: [[GLOB_LOAD4:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
  91   // CHECK: [[GLOB_CAST4:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(1) [[GLOB_LOAD4]] to ptr
  92   // CHECK: call void @[[RAW_PTR2]](ptr noundef [[GLOB_CAST4]])
  93   foo(LOC);
  94   // CHECK: [[LOC_LOAD3:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
  95   // CHECK: call void @[[LOC_PTR]](ptr addrspace(3) noundef [[LOC_LOAD3]])
  96   foo2(LOC);
  97   // CHECK: [[LOC_LOAD4:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
  98   // CHECK: [[LOC_CAST4:%[a-zA-Z0-9]+]] = addrspacecast ptr addrspace(3) [[LOC_LOAD4]] to ptr
  99   // CHECK: call void @[[RAW_PTR2]](ptr noundef [[LOC_CAST4]])
 100   foo(NoAS);
 101   // CHECK: [[NoAS_LOAD3:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
 102   // CHECK: call void @[[RAW_PTR]](ptr noundef [[NoAS_LOAD3]])
 103   foo2(NoAS);
 104   // CHECK: [[NoAS_LOAD4:%[a-zA-Z0-9]+]] = load ptr, ptr [[NoAS]], align 8
 105   // CHECK: call void @[[RAW_PTR2]](ptr noundef [[NoAS_LOAD4]])
 106   tmpl(GLOB);
 107   // CHECK: [[GLOB_LOAD4:%[a-zA-Z0-9]+]] = load ptr addrspace(1), ptr [[GLOB]], align 8
 108   // CHECK: call void @_Z4tmplIPU3AS1iEvT_(ptr addrspace(1) noundef [[GLOB_LOAD4]])
 109   tmpl(LOC);
 110   // CHECK: [[LOC_LOAD5:%[a-zA-Z0-9]+]] = load ptr addrspace(3), ptr [[LOC]], align 8
 111   // CHECK: call void @_Z4tmplIPU3AS3iEvT_(ptr addrspace(3) noundef [[LOC_LOAD5]])
 112   tmpl(PRIV);
 113   // CHECK: [[PRIV_LOAD5:%[a-zA-Z0-9]+]] = load ptr, ptr [[PRIV]], align 8
 114   // CHECK: call void @_Z4tmplIPiEvT_(ptr noundef [[PRIV_LOAD5]])
 115   tmpl(NoAS);
 116 // CHECK: %33 = load ptr, ptr %NoAS, align 8
 117 // CHECK: call void @_Z4tmplIPiEvT_(ptr noundef %33)
 118 }
 119
 120 // CHECK: declare void @_Z4tmplIPU3AS1iEvT_(ptr addrspace(1) noundef)
 121 // CHECK: declare void @_Z4tmplIPU3AS3iEvT_(ptr addrspace(3) noundef)
 122 // CHECK: declare void @_Z4tmplIPiEvT_(ptr noundef)