arno/hashcat
1
0
Fork 0
mirror of https://github.com/hashcat/hashcat.git synced 2025-07-29 18:08:46 +00:00
Code Issues Releases Wiki Activity

Add runtime OpenCL version detection and discrete Intel GPU tuning

Browse Source 
Detect the highest supported OpenCL version at runtime and use the
appropriate -cl-std= flag when compiling kernels. This improves
compatibility with the Intel NEO driver. Note: behavior is untested
on other platforms (NVIDIA, AMD, Apple, etc.). Feedback will be
monitored.

Add tuningdb entries for discrete Intel GPUs. Copy over hash-mode
patterns that benefit from vectorizing on scalar compute units, based
on existing AMD and NVIDIA entries. This change also removes the
artificial thread limit previously enforced for discrete Intel GPUs.

Disable automatic vector width detection from the OpenCL runtime
except on CPU, where it remains in use.
This commit is contained in:
Jens Steube 2025-07-24 22:19:58 +02:00
parent 3344558396
commit b2c308d403
4 changed files with 113 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
include/types.h
View File

@ -1518,9 +1518,10 @@ typedef struct hc_device_param
// Some more attributes
bool use_opencl11;
bool use_opencl12;
bool use_opencl20;
bool use_opencl21;
bool use_opencl30;
// AMD
bool has_vadd;

73
src/backend.c
View File

@ -5634,9 +5634,10 @@ static void backend_ctx_devices_init_cuda (hashcat_ctx_t *hashcat_ctx, int *virt
device_param->is_metal = false;
device_param->is_opencl = false;
device_param->use_opencl11 = false;
device_param->use_opencl12 = false;
device_param->use_opencl20 = false;
device_param->use_opencl21 = false;
device_param->use_opencl30 = false;
// device_name
@ -6111,9 +6112,10 @@ static void backend_ctx_devices_init_hip (hashcat_ctx_t *hashcat_ctx, int *virth
device_param->is_metal = false;
device_param->is_opencl = false;
device_param->use_opencl11 = false;
device_param->use_opencl12 = false;
device_param->use_opencl20 = false;
device_param->use_opencl21 = false;
device_param->use_opencl30 = false;
// device_name
@ -6621,9 +6623,10 @@ static void backend_ctx_devices_init_metal (hashcat_ctx_t *hashcat_ctx, MAYBE_UN
device_param->is_metal = true;
device_param->is_opencl = false;
device_param->use_opencl11 = false;
device_param->use_opencl12 = false;
device_param->use_opencl20 = false;
device_param->use_opencl21 = false;
device_param->use_opencl30 = false;
device_param->is_apple_silicon = is_apple_silicon ();
@ -7029,16 +7032,21 @@ static void backend_ctx_devices_init_opencl (hashcat_ctx_t *hashcat_ctx, int *vi
// check OpenCL version
device_param->use_opencl11 = false;
device_param->use_opencl12 = false;
device_param->use_opencl20 = false;
device_param->use_opencl21 = false;
device_param->use_opencl30 = false;
int opencl_version_min = 0;
int opencl_version_maj = 0;
if (sscanf (opencl_platform_version, "OpenCL %d.%d", &opencl_version_min, &opencl_version_maj) == 2)
{
if ((opencl_version_min == 1) && (opencl_version_maj == 2))
if ((opencl_version_min == 1) && (opencl_version_maj == 1))
{
device_param->use_opencl11 = true;
}
else if ((opencl_version_min == 1) && (opencl_version_maj == 2))
{
device_param->use_opencl12 = true;
}
@ -7046,9 +7054,9 @@ static void backend_ctx_devices_init_opencl (hashcat_ctx_t *hashcat_ctx, int *vi
{
device_param->use_opencl20 = true;
}
else if ((opencl_version_min == 2) && (opencl_version_maj == 1))
else if ((opencl_version_min == 3) && (opencl_version_maj == 0))
{
device_param->use_opencl21 = true;
device_param->use_opencl30 = true;
}
}
@ -10433,10 +10441,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
if (device_param->is_opencl == true)
{
if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG, sizeof (vector_width), &vector_width, NULL) == -1)
// For CPU we can ask the runtime
// For GPUs we want to be more selective and we will use the tuning db
vector_width = 1;
if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
{
device_param->skipped = true;
continue;
if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG, sizeof (vector_width), &vector_width, NULL) == -1)
{
device_param->skipped = true;
continue;
}
}
}
}
@ -10467,11 +10484,19 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
if (device_param->is_opencl == true)
{
if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_NATIVE_VECTOR_WIDTH_INT, sizeof (vector_width), &vector_width, NULL) == -1)
{
device_param->skipped = true;
// For CPU we can ask the runtime
// For GPUs we want to be more selective and we will use the tuning db
continue;
vector_width = 1;
if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
{
if (hc_clGetDeviceInfo (hashcat_ctx, device_param->opencl_device, CL_DEVICE_NATIVE_VECTOR_WIDTH_INT, sizeof (vector_width), &vector_width, NULL) == -1)
{
device_param->skipped = true;
continue;
}
}
}
}
@ -11124,10 +11149,16 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
/* currently disabled, hangs NEO drivers since 20.09.
was required for NEO driver 20.08 to workaround the same issue!
we go with the latest version
v7 re-enabled
*/
if (device_param->is_opencl == true)
{
if (device_param->use_opencl12 == true)
if (device_param->use_opencl11 == true)
{
build_options_len += snprintf (build_options_buf + build_options_len, build_options_sz - build_options_len, "-cl-std=CL1.1 ");
}
else if (device_param->use_opencl12 == true)
{
build_options_len += snprintf (build_options_buf + build_options_len, build_options_sz - build_options_len, "-cl-std=CL1.2 ");
}
@ -11135,12 +11166,11 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
{
build_options_len += snprintf (build_options_buf + build_options_len, build_options_sz - build_options_len, "-cl-std=CL2.0 ");
}
else if (device_param->use_opencl21 == true)
else if (device_param->use_opencl30 == true)
{
build_options_len += snprintf (build_options_buf + build_options_len, build_options_sz - build_options_len, "-cl-std=CL2.1 ");
build_options_len += snprintf (build_options_buf + build_options_len, build_options_sz - build_options_len, "-cl-std=CL3.0 ");
}
}
*/
// we don't have sm_* on vendors not NV but it doesn't matter
@ -16200,13 +16230,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
threads_per_block = opencl_query_threads_per_block (hashcat_ctx, device_param, kernel);
if ((device_param->opencl_device_type & CL_DEVICE_TYPE_GPU) && (device_param->device_host_unified_memory == 0) && (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK))
{
// Intel is highly inaccurate here: https://github.com/hashcat/hashcat/issues/4356
threads_per_block = MIN (threads_per_block, device_param->kernel_preferred_wgs_multiple);
}
// num_regs check should be included in opencl's CL_KERNEL_WORK_GROUP_SIZE
}
else if (device_param->is_metal == true)

20
tunings/Alias.hctune
View File

@ -321,3 +321,23 @@ AMD_Instinct_MI300X ALIAS_AMD
AMD_Instinct_MI325X ALIAS_AMD
AMD_Instinct_MI350X ALIAS_AMD
AMD_Instinct_MI355X ALIAS_AMD
Intel(R)_Arc(TM)_A310_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_A380_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_A580_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_A750_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_A770_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_B570_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_B580_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_B770_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Cxxx_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Dxxx_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_A30M_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_A40_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_A50_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_A60_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_A60M_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_B50_Graphics ALIAS_INTEL
Intel(R)_Arc(TM)_Pro_B60_Graphics ALIAS_INTEL
Intel(R)_Data_Center_GPU_Max_1100 ALIAS_INTEL

43
tunings/Modules_default.hctune
View File

@ -147,6 +147,49 @@ ALIAS_AMD 3 23700 2 A
ALIAS_AMD 3 24700 2 A A
ALIAS_AMD 3 99999 8 A A
ALIAS_INTEL 3 0 4 A A
ALIAS_INTEL 3 10 4 A A
ALIAS_INTEL 3 11 4 A A
ALIAS_INTEL 3 12 4 A A
ALIAS_INTEL 3 20 4 A A
ALIAS_INTEL 3 21 4 A A
ALIAS_INTEL 3 22 4 A A
ALIAS_INTEL 3 23 4 A A
ALIAS_INTEL 3 24 4 A A
ALIAS_INTEL 3 200 4 A A
ALIAS_INTEL 3 500 2 A A
ALIAS_INTEL 3 900 4 A A
ALIAS_INTEL 3 1000 4 A A
ALIAS_INTEL 3 1100 4 A A
ALIAS_INTEL 3 1600 2 A A
ALIAS_INTEL 3 2400 4 A A
ALIAS_INTEL 3 2410 4 A A
ALIAS_INTEL 3 2611 2 A A
ALIAS_INTEL 3 2711 2 A A
ALIAS_INTEL 3 2811 2 A A
ALIAS_INTEL 3 3710 4 A A
ALIAS_INTEL 3 3711 4 A A
ALIAS_INTEL 3 3800 2 A A
ALIAS_INTEL 3 4010 4 A A
ALIAS_INTEL 3 4110 2 A A
ALIAS_INTEL 3 4300 2 A A
ALIAS_INTEL 3 4800 4 A A
ALIAS_INTEL 3 4900 2 A A
ALIAS_INTEL 3 5300 2 A A
ALIAS_INTEL 3 5400 2 A A
ALIAS_INTEL 3 5500 4 A A
ALIAS_INTEL 3 9900 4 A A
ALIAS_INTEL 3 11100 2 A A
ALIAS_INTEL 3 13200 2 A A
ALIAS_INTEL 3 16400 4 A A
ALIAS_INTEL 3 18700 4 A A
ALIAS_INTEL 3 21200 4 A A
ALIAS_INTEL 3 21700 2 A A
ALIAS_INTEL 3 23600 2 A A
ALIAS_INTEL 3 23700 2 A A
ALIAS_INTEL 3 24700 2 A A
ALIAS_INTEL 3 99999 4 A A
##
## CryptoAPI
##