Subversion Repositories QNX 8.QNX8 LLVM/Clang compiler suite

/*===---- cuda_builtin_vars.h - CUDA built-in variables ---------------------===

 *

 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

 * See https://llvm.org/LICENSE.txt for license information.

 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

 *

 *===-----------------------------------------------------------------------===

 */

#ifndef __CUDA_BUILTIN_VARS_H

#define __CUDA_BUILTIN_VARS_H

// Forward declares from vector_types.h.

struct uint3;

struct dim3;

// The file implements built-in CUDA variables using __declspec(property).

// https://msdn.microsoft.com/en-us/library/yhfk0thd.aspx

// All read accesses of built-in variable fields get converted into calls to a

// getter function which in turn calls the appropriate builtin to fetch the

// value.

//

// Example:

//    int x = threadIdx.x;

// IR output:

//  %0 = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() #3

// PTX output:

//  mov.u32     %r2, %tid.x;

#define __CUDA_DEVICE_BUILTIN(FIELD, INTRINSIC)                                \

  __declspec(property(get = __fetch_builtin_##FIELD)) unsigned int FIELD;      \

  static inline __attribute__((always_inline))                                 \

      __attribute__((device)) unsigned int __fetch_builtin_##FIELD(void) {     \

    return INTRINSIC;                                                          \

  }

#if __cplusplus >= 201103L

#define __DELETE =delete

#else

#define __DELETE

#endif

// Make sure nobody can create instances of the special variable types.  nvcc

// also disallows taking address of special variables, so we disable address-of

// operator as well.

#define __CUDA_DISALLOW_BUILTINVAR_ACCESS(TypeName)                            \

  __attribute__((device)) TypeName() __DELETE;                                 \

  __attribute__((device)) TypeName(const TypeName &) __DELETE;                 \

  __attribute__((device)) void operator=(const TypeName &) const __DELETE;     \

  __attribute__((device)) TypeName *operator&() const __DELETE

struct __cuda_builtin_threadIdx_t {

  __CUDA_DEVICE_BUILTIN(x,__nvvm_read_ptx_sreg_tid_x());

  __CUDA_DEVICE_BUILTIN(y,__nvvm_read_ptx_sreg_tid_y());

  __CUDA_DEVICE_BUILTIN(z,__nvvm_read_ptx_sreg_tid_z());

  // threadIdx should be convertible to uint3 (in fact in nvcc, it *is* a

  // uint3).  This function is defined after we pull in vector_types.h.

  __attribute__((device)) operator dim3() const;

  __attribute__((device)) operator uint3() const;

private:

  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_threadIdx_t);

};

struct __cuda_builtin_blockIdx_t {

  __CUDA_DEVICE_BUILTIN(x,__nvvm_read_ptx_sreg_ctaid_x());

  __CUDA_DEVICE_BUILTIN(y,__nvvm_read_ptx_sreg_ctaid_y());

  __CUDA_DEVICE_BUILTIN(z,__nvvm_read_ptx_sreg_ctaid_z());

  // blockIdx should be convertible to uint3 (in fact in nvcc, it *is* a

  // uint3).  This function is defined after we pull in vector_types.h.

  __attribute__((device)) operator dim3() const;

  __attribute__((device)) operator uint3() const;

private:

  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockIdx_t);

};

struct __cuda_builtin_blockDim_t {

  __CUDA_DEVICE_BUILTIN(x,__nvvm_read_ptx_sreg_ntid_x());

  __CUDA_DEVICE_BUILTIN(y,__nvvm_read_ptx_sreg_ntid_y());

  __CUDA_DEVICE_BUILTIN(z,__nvvm_read_ptx_sreg_ntid_z());

  // blockDim should be convertible to dim3 (in fact in nvcc, it *is* a

  // dim3).  This function is defined after we pull in vector_types.h.

  __attribute__((device)) operator dim3() const;

  __attribute__((device)) operator uint3() const;

private:

  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockDim_t);

};

struct __cuda_builtin_gridDim_t {

  __CUDA_DEVICE_BUILTIN(x,__nvvm_read_ptx_sreg_nctaid_x());

  __CUDA_DEVICE_BUILTIN(y,__nvvm_read_ptx_sreg_nctaid_y());

  __CUDA_DEVICE_BUILTIN(z,__nvvm_read_ptx_sreg_nctaid_z());

  // gridDim should be convertible to dim3 (in fact in nvcc, it *is* a

  // dim3).  This function is defined after we pull in vector_types.h.

  __attribute__((device)) operator dim3() const;

  __attribute__((device)) operator uint3() const;

private:

  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_gridDim_t);

};

#define __CUDA_BUILTIN_VAR                                                     \

  extern const __attribute__((device)) __attribute__((weak))

__CUDA_BUILTIN_VAR __cuda_builtin_threadIdx_t threadIdx;

__CUDA_BUILTIN_VAR __cuda_builtin_blockIdx_t blockIdx;

__CUDA_BUILTIN_VAR __cuda_builtin_blockDim_t blockDim;

__CUDA_BUILTIN_VAR __cuda_builtin_gridDim_t gridDim;

// warpSize should translate to read of %WARP_SZ but there's currently no

// builtin to do so. According to PTX v4.2 docs 'to date, all target

// architectures have a WARP_SZ value of 32'.

__attribute__((device)) const int warpSize = 32;

#undef __CUDA_DEVICE_BUILTIN

#undef __CUDA_BUILTIN_VAR

#undef __CUDA_DISALLOW_BUILTINVAR_ACCESS

#undef __DELETE

#endif /* __CUDA_BUILTIN_VARS_H */