OpenCL Memory Transfer Issue (Err. code -6) - opencl

Question

I'm dying a little inside. I've been working on this all day to no avail. I was having issues running some code that previously ran fine, so I wrote a short "toy" OpenCL program to try and figure out what was going on, but my toy program has me baffled and incredibly frustrated.
I'm working with an Nvidia 780i with 3Gb of global memory. It has a maximum allocation of ~780 Mb. At first, it wouldn't error out when I was intentionally over-allocating. Solved that (it was typographical, but the compiler/analyzer wasn't catching it). Now, even when trying to allocate WAY below what the device should be able to handle, I get an error code of -6 (CL_OUT_OF_HOST_MEMORY) on the second big buffer allocation.
I've been researching this error and I just can't track how it applies in this case. I have 32 gb of ram in the machine I'm using so there's certainly not a shortage there. I figure there's something that I just don't understand going on here.
It will allocate the first buffer alright, but then choke on the second. I basically just cannot allocate nearly the amount of global memory I both want and need to.
Any help is much appreciated. If you can help me with this and you're near LA I'll take you out for drinks. That's how frustrated I am.
Code and output is below for my machine.
Thanks,
John
Main program:
#define _CRT_SECURE_NO_WARNINGS
#define PROGRAM_FILE "kernels.cl"
#define KERNEL_NAME "test"
#include <CL/cl.h>
#include <stdlib.h>
#include <stdio.h>
#include <sys/types.h>
#define N_PROJ 4000
#define N_CHANNELS 736
#define N_ROWS 32
/* Find a GPU or CPU associated with the first available platform */
cl_device_id create_device() {
cl_platform_id platform;
cl_device_id dev;
int err;
/* Identify a platform */
err = clGetPlatformIDs(1, &platform, NULL);
if(err < 0) {
perror("Couldn't identify a platform");
exit(1);
}
/* Access a device */
err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &dev, NULL);
if(err == CL_DEVICE_NOT_FOUND) {
perror("Just a heads up: I'm not going to run on the GPU");
err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, 1, &dev, NULL);
}
if(err < 0) {
perror("Couldn't access any devices");
exit(1);
}
cl_ulong16 alloc_size,mem_size;
char name[40];
clGetDeviceInfo(dev,CL_DEVICE_MAX_MEM_ALLOC_SIZE,sizeof(cl_ulong16),&alloc_size,NULL);
clGetDeviceInfo(dev,CL_DEVICE_NAME,sizeof(name),name,NULL);
clGetDeviceInfo(dev,CL_DEVICE_GLOBAL_MEM_SIZE,sizeof(cl_ulong16),&mem_size,NULL);
printf("Using device: %s\n",name);
printf("Global memory size: %lu\n",mem_size);
printf("Max. allocation: %lu\n",alloc_size);
return dev;
}
/* Create program from a file and compile it */
cl_program build_program(cl_context ctx, cl_device_id dev, const char* filename) {
cl_program program;
FILE *program_handle;
char *program_buffer, *program_log;
size_t program_size, log_size;
int err;
/* Read program file and place content into buffer */
program_handle = fopen(filename, "r");
if(program_handle == NULL) {
perror("Couldn't find the program file");
exit(1);
}
fseek(program_handle, 0, SEEK_END);
program_size = ftell(program_handle)-13;
rewind(program_handle);
program_buffer = (char*)malloc(program_size + 1);
program_buffer[program_size] = '\0';
fread(program_buffer, sizeof(char), program_size, program_handle);
fclose(program_handle);
/* Create program from file */
program = clCreateProgramWithSource(ctx, 1,
(const char**)&program_buffer, &program_size, &err);
if(err < 0) {
perror("Couldn't create the program");
exit(1);
}
free(program_buffer);
/* Build program */
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if(err < 0) {
/* Find size of log and print to std output */
clGetProgramBuildInfo(program, dev, CL_PROGRAM_BUILD_LOG,
0, NULL, &log_size);
program_log = (char*) malloc(log_size + 1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(program, dev, CL_PROGRAM_BUILD_LOG,
log_size + 1, program_log, NULL);
printf("%s\n", program_log);
free(program_log);
exit(1);
}
return program;
}
int main(int argc, const char * argv[])
{
/* This file serves as a backbone for OpenCL programs. */
/* All the user needs to do is enter their OpenCL data */
/* structures, set kernel args, and kernel dispatches. */
/* Standard OCL structures */
cl_device_id device;
cl_context context;
cl_program program;
cl_kernel kernel;
cl_command_queue queue;
device=create_device();
context = clCreateContext(NULL, 1, &device, NULL, NULL, NULL);
program = build_program(context, device, PROGRAM_FILE);
queue = clCreateCommandQueue(context, device,0, NULL);
kernel = clCreateKernel(program, KERNEL_NAME, NULL);
/* User code goes here */
cl_int err;
/* Declare and set data */
int a[]={1,2,3};
float *rebin;
rebin =(float*) calloc(N_PROJ*N_CHANNELS*N_ROWS,sizeof(float));
float *mat;
mat =(float*) calloc(N_PROJ*N_CHANNELS*N_ROWS,sizeof(float));
printf("\nAllocation size: %lu\n",N_PROJ*N_CHANNELS*N_ROWS*sizeof(float));
/* Declare and set buffer objects */
cl_mem a_buff,rebin_buff,mat_buff;
printf("Total memory to be allocated: %lu\n",2*N_PROJ*N_CHANNELS*N_ROWS*sizeof(float)+sizeof(a) );
a_buff =clCreateBuffer(context,CL_MEM_COPY_HOST_PTR|CL_MEM_READ_WRITE,sizeof(a),a,NULL);
rebin_buff =clCreateBuffer(context,CL_MEM_COPY_HOST_PTR|CL_MEM_READ_WRITE,N_PROJ*N_CHANNELS*N_ROWS*sizeof(float),rebin,&err);
if (err<0){
printf("Error: %i\n",err);
perror("Couldn't create buffer 1");
exit(1);
}
mat_buff =clCreateBuffer(context,CL_MEM_COPY_HOST_PTR|CL_MEM_READ_WRITE,N_PROJ*N_CHANNELS*N_ROWS*sizeof(float),mat ,&err);
if (err<0){
printf("Error: %i\n",err);
perror("Couldn't create buffer 2");
exit(1);
}
/* Copy data over to the device */
err=clSetKernelArg(kernel,0,sizeof(cl_mem),&mat_buff);
if (err<0){
perror("Couldn't set kernel argument");
exit(1);
}
err=clSetKernelArg(kernel,1,sizeof(cl_mem),&rebin_buff);
err=clSetKernelArg(kernel,2,sizeof(cl_mem),&a_buff);
clEnqueueTask(queue,kernel,0,NULL,NULL);
clEnqueueReadBuffer(queue,mat_buff ,CL_TRUE,0,N_PROJ*N_CHANNELS*N_ROWS*sizeof(float),mat ,0,NULL,NULL);
clEnqueueReadBuffer(queue,rebin_buff,CL_TRUE,0,N_PROJ*N_CHANNELS*N_ROWS*sizeof(float),rebin,0,NULL,NULL);
clEnqueueReadBuffer(queue,a_buff, CL_TRUE,0,sizeof(a),a,0,NULL,NULL);
printf("%f %f %f\n",mat[1],mat[2],mat[3]);
printf("%f %f %f\n",rebin[1],rebin[2],rebin[3]);
printf("%i %i %i",a[0],a[1],a[2]);
/***********************/
clReleaseKernel(kernel);
clReleaseCommandQueue(queue);
clReleaseProgram(program);
clReleaseContext(context);
//clReleaseDevice(device);
printf("\n\nProgram apparently executed fully. \n");
return 0;
}
Kernel:
__kernel void test(__global float *mat,__global float *rebin,__global int *a){
a[0]=3;
a[1]=2;
a[2]=1;
rebin[1]=1.0f;
rebin[2]=2.0f;
rebin[3]=3.0f;
mat[1]=3.0f;
mat[2]=2.0f;
mat[3]=1.0f;
}
Console output for my machine:
Using device: GeForce GTX 780
Global memory size: 3221225472
Max. allocation: 805306368
Allocation size: 376832000
Total memory to be allocated: 753664012
Error: -6
Couldn't create buffer 2: No error
Process returned 1 (0x1) execution time : 0.435 s
Press any key to continue.