clang/test/CodeGenOpenCL/amdgpu-enqueue-kernel.cl

   1 // RUN: %clang_cc1 -no-opaque-pointers %s -cl-std=CL2.0 -O0 -emit-llvm -o - -triple amdgcn | FileCheck %s --check-prefix=CHECK
   2
   3 typedef struct {int a;} ndrange_t;
   4
   5 void callee(long id, global long *out) {
   6   out[id] = id;
   7 }
   8
   9 // CHECK-LABEL: define{{.*}} amdgpu_kernel void @test
  10 kernel void test(global char *a, char b, global long *c, long d) {
  11   queue_t default_queue;
  12   unsigned flags = 0;
  13   ndrange_t ndrange;
  14
  15   enqueue_kernel(default_queue, flags, ndrange,
  16                  ^(void) {
  17                  a[0] = b;
  18                  });
  19
  20   enqueue_kernel(default_queue, flags, ndrange,
  21                  ^(void) {
  22                  a[0] = b;
  23                  c[0] = d;
  24                  });
  25   enqueue_kernel(default_queue, flags, ndrange,
  26                  ^(local void *lp) {
  27                  a[0] = b;
  28                  c[0] = d;
  29                  ((local int*)lp)[0] = 1;
  30                  }, 100);
  31
  32   void (^block)(void) = ^{
  33     callee(d, c);
  34   };
  35
  36   enqueue_kernel(default_queue, flags, ndrange, block);
  37 }
  38
  39 // CHECK-LABEL: define internal amdgpu_kernel void @__test_block_invoke_kernel(<{ i32, i32, i8*, i8 addrspace(1)*, i8 }> %0)
  40 // CHECK-SAME: #[[ATTR:[0-9]+]] !kernel_arg_addr_space !{{.*}} !kernel_arg_access_qual !{{.*}} !kernel_arg_type !{{.*}} !kernel_arg_base_type !{{.*}} !kernel_arg_type_qual !{{.*}}
  41 // CHECK: entry:
  42 // CHECK:  %1 = alloca <{ i32, i32, i8*, i8 addrspace(1)*, i8 }>, align 8, addrspace(5)
  43 // CHECK:  store <{ i32, i32, i8*, i8 addrspace(1)*, i8 }> %0, <{ i32, i32, i8*, i8 addrspace(1)*, i8 }> addrspace(5)* %1, align 8
  44 // CHECK:  %2 ={{.*}} addrspacecast <{ i32, i32, i8*, i8 addrspace(1)*, i8 }> addrspace(5)* %1 to i8*
  45 // CHECK:  call void @__test_block_invoke(i8* %2)
  46 // CHECK:  ret void
  47 // CHECK:}
  48
  49 // CHECK-LABEL: define internal amdgpu_kernel void @__test_block_invoke_2_kernel(<{ i32, i32, i8*, i8 addrspace(1)*, i64 addrspace(1)*, i64, i8 }> %0)
  50 // CHECK-SAME: #[[ATTR]] !kernel_arg_addr_space !{{.*}} !kernel_arg_access_qual !{{.*}} !kernel_arg_type !{{.*}} !kernel_arg_base_type !{{.*}} !kernel_arg_type_qual !{{.*}}
  51
  52 // CHECK-LABEL: define internal amdgpu_kernel void @__test_block_invoke_3_kernel(<{ i32, i32, i8*, i8 addrspace(1)*, i64 addrspace(1)*, i64, i8 }> %0, i8 addrspace(3)* %1)
  53 // CHECK-SAME: #[[ATTR]] !kernel_arg_addr_space !{{.*}} !kernel_arg_access_qual !{{.*}} !kernel_arg_type !{{.*}} !kernel_arg_base_type !{{.*}} !kernel_arg_type_qual !{{.*}}
  54
  55 // CHECK-LABEL: define internal amdgpu_kernel void @__test_block_invoke_4_kernel(<{ i32, i32, i8*, i64, i64 addrspace(1)* }> %0)
  56 // CHECK-SAME: #[[ATTR]] !kernel_arg_addr_space !{{.*}} !kernel_arg_access_qual !{{.*}} !kernel_arg_type !{{.*}} !kernel_arg_base_type !{{.*}} !kernel_arg_type_qual !{{.*}}
  57
  58 // CHECK: attributes #[[ATTR]] = { nounwind "enqueued-block" }