Changeset 07646b6 in sasmodels for sasmodels/kernel_header.c


Ignore:
Timestamp:
Oct 25, 2018 3:43:01 PM (6 years ago)
Author:
Paul Kienzle <pkienzle@…>
Branches:
master, core_shell_microgels, magnetic_model, ticket-1257-vesicle-product, ticket_1156, ticket_1265_superball, ticket_822_more_unit_tests
Children:
149eb53
Parents:
31fc4ad (diff), d5ce7fa (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.
git-author:
Paul Kienzle <pkienzle@…> (10/25/18 14:41:48)
git-committer:
Paul Kienzle <pkienzle@…> (10/25/18 15:43:01)
Message:

Merge branch 'cuda-test' into beta_approx

File:
1 edited

Legend:

Unmodified
Added
Removed
  • sasmodels/kernel_header.c

    r296c52b r07646b6  
    11#ifdef __OPENCL_VERSION__ 
    22# define USE_OPENCL 
     3#elif defined(__CUDACC__) 
     4# define USE_CUDA 
    35#elif defined(_OPENMP) 
    46# define USE_OPENMP 
    57#endif 
     8 
     9// Use SAS_DOUBLE to force the use of double even for float kernels 
     10#define SAS_DOUBLE dou ## ble 
    611 
    712// If opencl is not available, then we are compiling a C function 
    813// Note: if using a C++ compiler, then define kernel as extern "C" 
    914#ifdef USE_OPENCL 
     15 
     16   #define USE_GPU 
     17   #define pglobal global 
     18   #define pconstant constant 
     19 
    1020   typedef int int32_t; 
    11 #  if defined(USE_SINCOS) 
    12 #    define SINCOS(angle,svar,cvar) svar=sincos(angle,&cvar) 
    13 #  else 
    14 #    define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) 
    15 #  endif 
     21 
     22   #if defined(USE_SINCOS) 
     23   #  define SINCOS(angle,svar,cvar) svar=sincos(angle,&cvar) 
     24   #else 
     25   #  define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) 
     26   #endif 
    1627   // Intel CPU on Mac gives strange values for erf(); on the verified 
    1728   // platforms (intel, nvidia, amd), the cephes erf() is significantly 
     
    2435   #  define erfcf erfc 
    2536   #endif 
    26 #else // !USE_OPENCL 
    27 // Use SAS_DOUBLE to force the use of double even for float kernels 
    28 #  define SAS_DOUBLE dou ## ble 
    29 #  ifdef __cplusplus 
     37 
     38#elif defined(USE_CUDA) 
     39 
     40   #define USE_GPU 
     41   #define local __shared__ 
     42   #define pglobal 
     43   #define constant __constant__ 
     44   #define pconstant const 
     45   #define kernel extern "C" __global__ 
     46 
     47   // OpenCL powr(a,b) = C99 pow(a,b), b >= 0 
     48   // OpenCL pown(a,b) = C99 pow(a,b), b integer 
     49   #define powr(a,b) pow(a,b) 
     50   #define pown(a,b) pow(a,b) 
     51   //typedef int int32_t; 
     52   #if defined(USE_SINCOS) 
     53   #  define SINCOS(angle,svar,cvar) sincos(angle,&svar,&cvar) 
     54   #else 
     55   #  define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) 
     56   #endif 
     57 
     58#else // !USE_OPENCL && !USE_CUDA 
     59 
     60   #define local 
     61   #define pglobal 
     62   #define constant const 
     63   #define pconstant const 
     64 
     65   #ifdef __cplusplus 
    3066      #include <cstdio> 
    3167      #include <cmath> 
     
    5187     #endif 
    5288     inline void SINCOS(double angle, double &svar, double &cvar) { svar=sin(angle); cvar=cos(angle); } 
    53 else // !__cplusplus 
     89   #else // !__cplusplus 
    5490     #include <inttypes.h>  // C99 guarantees that int32_t types is here 
    5591     #include <stdio.h> 
     
    77113     #define kernel 
    78114     #define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) 
    79 #  endif  // !__cplusplus 
    80 #  define global 
    81 #  define local 
    82 #  define constant const 
    83 // OpenCL powr(a,b) = C99 pow(a,b), b >= 0 
    84 // OpenCL pown(a,b) = C99 pow(a,b), b integer 
    85 #  define powr(a,b) pow(a,b) 
    86 #  define pown(a,b) pow(a,b) 
     115   #endif  // !__cplusplus 
     116   // OpenCL powr(a,b) = C99 pow(a,b), b >= 0 
     117   // OpenCL pown(a,b) = C99 pow(a,b), b integer 
     118   #define powr(a,b) pow(a,b) 
     119   #define pown(a,b) pow(a,b) 
     120 
    87121#endif // !USE_OPENCL 
    88122 
Note: See TracChangeset for help on using the changeset viewer.