Try to evaluate available OpenCL device memory and use this information instead of total available OpenCL device memory for autotune

pull/1656/head
jsteube 6 years ago
parent af2626e67a
commit e67195aa11

@ -1,5 +1,11 @@
* changes v4.2.0 -> xxx
##
## Improvements
##
- Try to evaluate available OpenCL device memory and use this information instead of total available OpenCL device memory for autotune
* changes v4.1.0 -> v4.2.0
##

@ -953,6 +953,7 @@ typedef struct hc_device_param
u32 device_processors;
u64 device_maxmem_alloc;
u64 device_global_mem;
u64 device_available_mem;
u32 device_maxclock_frequency;
size_t device_maxworkgroup_size;
u64 device_local_mem_size;

@ -3225,6 +3225,8 @@ int opencl_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
device_param->device_global_mem = device_global_mem;
device_param->device_available_mem = 0;
// device_maxmem_alloc
cl_ulong device_maxmem_alloc;
@ -4259,6 +4261,57 @@ int opencl_session_begin (hashcat_ctx_t *hashcat_ctx)
if (CL_rc == -1) return -1;
// OK, so the problem here is the following:
// There's just CL_DEVICE_GLOBAL_MEM_SIZE to ask OpenCL about the total memory on the device,
// but there's no way to ask for available memory on the device.
// In combination, most OpenCL runtimes implementation of clCreateBuffer()
// are doing so called lazy memory allocation on the device.
// Now, if the user has X11 (or a game or anything that takes a lot of GPU memory)
// running on the host we end up with an error type of this:
// clEnqueueNDRangeKernel(): CL_MEM_OBJECT_ALLOCATION_FAILURE
// The clEnqueueNDRangeKernel() is because of the lazy allocation
// The best way to workaround this problem is if we would be able to ask for available memory,
// The idea here is to try to evaluate available memory by allocating it till it errors
#define MAX_ALLOC_CHECKS_CNT 8192
#define MAX_ALLOC_CHECKS_SIZE (32 * 1024 * 1024)
cl_mem *tmp_device = (cl_mem *) hccalloc (MAX_ALLOC_CHECKS_CNT, sizeof (cl_mem));
char *tmp_host = (char *) hcmalloc (MAX_ALLOC_CHECKS_SIZE);
int c;
for (c = 0; c < MAX_ALLOC_CHECKS_CNT; c++)
{
cl_int CL_err;
OCL_PTR *ocl = opencl_ctx->ocl;
tmp_device[c] = ocl->clCreateBuffer (device_param->context, CL_MEM_READ_ONLY, MAX_ALLOC_CHECKS_SIZE, NULL, &CL_err);
if (CL_err != CL_SUCCESS) break;
CL_err = ocl->clEnqueueReadBuffer (device_param->command_queue, tmp_device[c], CL_TRUE, 0, MAX_ALLOC_CHECKS_SIZE, tmp_host, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) break;
}
int r;
for (r = 0; r < c; r++)
{
CL_rc = hc_clReleaseMemObject (hashcat_ctx, tmp_device[r]);
if (CL_rc == -1) return -1;
}
hcfree (tmp_host);
hcfree (tmp_device);
device_param->device_available_mem = c * MAX_ALLOC_CHECKS_SIZE;
/**
* create input buffers on device : calculate size of fixed memory buffers
*/
@ -4436,7 +4489,7 @@ int opencl_session_begin (hashcat_ctx_t *hashcat_ctx)
continue;
}
if ((size_scrypt + scrypt_extra_space) > device_param->device_global_mem)
if ((size_scrypt + scrypt_extra_space) > device_param->device_available_mem)
{
if (user_options->quiet == false) event_log_warning (hashcat_ctx, "Increasing total device memory allocatable for --scrypt-tmto %u.", tmto);
@ -6095,7 +6148,7 @@ int opencl_session_begin (hashcat_ctx_t *hashcat_ctx)
#define PWS_SPACE (1024 * 1024 * 1024)
// sometimes device_global_mem and device_maxmem_alloc reported back from the opencl runtime are a bit inaccurate.
// sometimes device_available_mem and device_maxmem_alloc reported back from the opencl runtime are a bit inaccurate.
// let's add some extra space just to be sure.
#define EXTRA_SPACE (64 * 1024 * 1024)
@ -6173,7 +6226,7 @@ int opencl_session_begin (hashcat_ctx_t *hashcat_ctx)
+ size_st_salts
+ size_st_esalts;
if ((size_total + EXTRA_SPACE) > device_param->device_global_mem) memory_limit_hit = 1;
if ((size_total + EXTRA_SPACE) > device_param->device_available_mem) memory_limit_hit = 1;
if (memory_limit_hit == 1)
{

Loading…
Cancel
Save