Legends never die: GPUPI

Zitat von clauskadrnoschka

Heut abend wird ausprobiert; trotzdem schon eine Frage @mat: wie hast Du geprüft ob Dein Ergebnis richtig ist?

Würde mich auch interessieren Bzg generell wie der Kernel grob implementiert ist, wenn das kein Geheimniss ist. Source gibts ja keinen oder?

du brauchst anscheinend die visual c++ 2013 redist x86 am 64bit OS. damit startets bei mir.
http://www.microsoft.com/en-us/down...n.aspx?id=40784


app startet, aber:

Code:

Selected device: NVS 5100M
OpenCL 1.1 CUDA 4.2.1 is ready.

Compiling OpenCL kernels 
ERROR: Error happened during the build of OpenCL program.
Build log:
:11:10: fatal error: 'uint128.h' file not found
#include "uint128.h"
         ^

auch bei 1.0.1

ERROR: No connected OpenCL devices found on this system!

R9 290, catalyst 14.9, Win7 x64

Nein, das sind prinzipiell keine simplen Bugs, sondern Kompatibilitätsprobleme von OpenCL. Das liegt daran, dass OpenCL einen JIT-Compiler hat, der die Kernels on-the-fly auf dem System berechnet. Daher braucht man sehr viel Erfahrung, um die Kernel für alle System korrekt anzupassen. Dasselbe gilt leider auch für die Auswahl der Devices, die schon im Code sehr komplex ist.

Aber deshalb machen wir das ja gemeinsam. Wir schaffen das!

Same here:

Code:

Selected device: GeForce GTX 770
OpenCL 1.1 CUDA 6.5.21 is ready.

Compiling OpenCL kernels 
ERROR: Error happened during the build of OpenCL program.
Build log:
:11:10: fatal error: 'uint128.h' file not found
#include "uint128.h"
         ^

Wenn ich für jeden Absturz von Visual Studio 2013 einen Euro bekommen würde, hätte ich mittlerweile einen Butler daheim.

Btw, I'm on it. Die Header-Files werden in der Non-Debug-Version der exe künftig implizit in den Kernel-Source geladen. Dann sollte es problemlos funktionieren (für NVIDIA-Karten). AMD-Support kommt gleich danach.

Wird womöglich an der HD 4000 liegen, oder?

Code:

Selected device: Intel(R) HD Graphics P4000
OpenCL 1.2 is ready.

Compiling OpenCL kernels 
ERROR: Error happened during the build of OpenCL program.
Build log:
:13:26: error: OpenCL extension 'cl_khr_fp64' is unsupported
#pragma OPENCL EXTENSION cl_khr_fp64: enable
                         ^
:15:8: error: unknown type name 'double2'; did you mean 'double'?
inline double2 make_double2(const double head, const double tail)
       ^~~~~~~
       double
:17:2: error: use of undeclared identifier 'double2'; did you mean 'double'?
        double2 z;
        ^~~~~~~
        double
:19:3: error: member reference base type 'double' is not a structure or union
        z.x = tail;
        ~^~
:20:3: error: member reference base type 'double' is not a structure or union
        z.y = head;
        ~^~
:32:1: error: unknown type name 'double2'; did you mean 'double'?
double2 add_double2(const double2 a, const double2 b)
^~~~~~~
double
:32:27: error: unknown type name 'double2'; did you mean 'double'?
double2 add_double2(const double2 a, const double2 b)
                          ^~~~~~~
                          double
:32:44: error: unknown type name 'double2'; did you mean 'double'?
double2 add_double2(const double2 a, const double2 b)
                                           ^~~~~~~
                                           double
:34:2: error: use of undeclared identifier 'double2'; did you mean 'double'?
        double2 z;
        ^~~~~~~
        double
:37:8: error: member reference base type 'const double' is not a structure or union
        t1 = a.y + b.y;
             ~^~
:37:14: error: member reference base type 'const double' is not a structure or union
        t1 = a.y + b.y;
                   ~^~
:38:13: error: member reference base type 'const double' is not a structure or union
        t2 = t1 - a.y;
                  ~^~
:39:9: error: member reference base type 'const double' is not a structure or union
        t3 = (a.y + (t2 - t1)) + (b.y - t2);
              ~^~
:39:29: error: member reference base type 'const double' is not a structure or union
        t3 = (a.y + (t2 - t1)) + (b.y - t2);
                                  ~^~
:40:8: error: member reference base type 'const double' is not a structure or union
        t4 = a.x + b.x;
             ~^~
:40:14: error: member reference base type 'const double' is not a structure or union
        t4 = a.x + b.x;
                   ~^~
:41:13: error: member reference base type 'const double' is not a structure or union
        t2 = t4 - a.x;
                  ~^~
:42:9: error: member reference base type 'const double' is not a structure or union
        t5 = (a.x + (t2 - t4)) + (b.x - t2);
              ~^~
:42:29: error: member reference base type 'const double' is not a structure or union
        t5 = (a.x + (t2 - t4)) + (b.x - t2);
                                  ~^~
:47:3: error: member reference base type 'double' is not a structure or union
        z.y = e = t4 + t3;
        ~^~
:48:3: error: member reference base type 'double' is not a structure or union
        z.x = (t4 - e) + t3;
        ~^~
:53:8: error: unknown type name 'double2'; did you mean 'double'?
inline double2 div_double2_nofma(const double a, const double b)
       ^~~~~~~
       double
:55:2: error: use of undeclared identifier 'double2'; did you mean 'double'?
        double2 t;
        ^~~~~~~
        double
:59:3: error: member reference base type 'double' is not a structure or union
        t.y = a * r;
        ~^~
:60:12: error: member reference base type 'double' is not a structure or union
        e = b * -t.y + a;
                 ~^~
:61:3: error: member reference base type 'double' is not a structure or union
        t.y = r * e + t.y;
        ~^~
:61:17: error: member reference base type 'double' is not a structure or union
        t.y = r * e + t.y;
                      ~^~
:62:3: error: member reference base type 'double' is not a structure or union
        t.x = b * -t.y + a;
        ~^~
:62:14: error: member reference base type 'double' is not a structure or union
        t.x = b * -t.y + a;
                   ~^~
:63:11: error: member reference base type 'double' is not 

Läuft bei mir nur bei 1M und 32M, darüber

Code:

Selected device: GeForce GT 630
OpenCL 1.1 CUDA 6.5.19 is ready.

Compiling OpenCL kernels ... done.

Calculating 100.000.000nth digit of PI. 20 iterations.

Allocated device memory : 671088640 Bytes
Batch Size              : 20M
Reduction Size          : 64


ERROR: OpenCL error: CL_MEM_OBJECT_ALLOCATION_FAILURE happened in file pi.cpp at line 553.