diff --git a/docs/changes.txt b/docs/changes.txt
index ccd6f2ee2..2ba1ed9f2 100644
--- a/docs/changes.txt
+++ b/docs/changes.txt
@@ -52,6 +52,7 @@
- CUDA Backend: Give detailed warning if either the NVIDIA CUDA or the NVIDIA RTC library cannot be initialized
- CUDA Backend: Do not warn about missing CUDA SDK installation if --backend-ignore-cuda is used
- CUDA Backend: Use blocking events to avoid 100% CPU core usage (per GPU)
+- OpenCL Runtime: Workaround JiT compiler deadlock on NVIDIA driver >= 465.89
- RAR3 Kernels: Improved loop code, improving performance by 23%
- Startup time: Improved the startup time by avoiding some time intensive operations for skipped devices
diff --git a/include/backend.h b/include/backend.h
index 517e18bba..e9b7a9d4f 100644
--- a/include/backend.h
+++ b/include/backend.h
@@ -80,6 +80,7 @@ int hc_cuLinkDestroy (hashcat_ctx_t *hashcat_ctx, CUlinkState state)
int hc_cuLinkComplete (hashcat_ctx_t *hashcat_ctx, CUlinkState state, void **cubinOut, size_t *sizeOut);
int hc_clBuildProgram (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data);
+int hc_clCompileProgram (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, cl_uint num_input_headers, const cl_program *input_headers, const char **header_include_names, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data);
int hc_clCreateBuffer (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_mem_flags flags, size_t size, void *host_ptr, cl_mem *mem);
int hc_clCreateCommandQueue (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_device_id device, cl_command_queue_properties properties, cl_command_queue *command_queue);
int hc_clCreateContext (hashcat_ctx_t *hashcat_ctx, const cl_context_properties *properties, cl_uint num_devices, const cl_device_id *devices, void (CL_CALLBACK *pfn_notify) (const char *errinfo, const void *private_info, size_t cb, void *user_data), void *user_data, cl_context *context);
@@ -103,6 +104,7 @@ int hc_clGetPlatformIDs (hashcat_ctx_t *hashcat_ctx, cl_uint num_entrie
int hc_clGetPlatformInfo (hashcat_ctx_t *hashcat_ctx, cl_platform_id platform, cl_platform_info param_name, size_t param_value_size, void *param_value, size_t *param_value_size_ret);
int hc_clGetProgramBuildInfo (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_device_id device, cl_program_build_info param_name, size_t param_value_size, void *param_value, size_t *param_value_size_ret);
int hc_clGetProgramInfo (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_program_info param_name, size_t param_value_size, void *param_value, size_t * param_value_size_ret);
+int hc_clLinkProgram (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint num_devices, const cl_device_id *device_list, const char *options, cl_uint num_input_programs, const cl_program *input_programs, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data, cl_program *program);
int hc_clReleaseCommandQueue (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue);
int hc_clReleaseContext (hashcat_ctx_t *hashcat_ctx, cl_context context);
int hc_clReleaseEvent (hashcat_ctx_t *hashcat_ctx, cl_event event);
diff --git a/include/ext_OpenCL.h b/include/ext_OpenCL.h
index 0fc0d01b9..54f3b4182 100644
--- a/include/ext_OpenCL.h
+++ b/include/ext_OpenCL.h
@@ -38,6 +38,7 @@ typedef union
#define CL_PLATFORMS_MAX 16
typedef cl_int (CL_API_CALL *OCL_CLBUILDPROGRAM) (cl_program, cl_uint, const cl_device_id *, const char *, void (CL_CALLBACK *)(cl_program, void *), void *);
+typedef cl_int (CL_API_CALL *OCL_CLCOMPILEPROGRAM) (cl_program, cl_uint, const cl_device_id *, const char *, cl_uint, const cl_program *, const char **, void (CL_CALLBACK *)(cl_program, void *), void *);
typedef cl_mem (CL_API_CALL *OCL_CLCREATEBUFFER) (cl_context, cl_mem_flags, size_t, void *, cl_int *);
typedef cl_command_queue (CL_API_CALL *OCL_CLCREATECOMMANDQUEUE) (cl_context, cl_device_id, cl_command_queue_properties, cl_int *);
typedef cl_context (CL_API_CALL *OCL_CLCREATECONTEXT) (const cl_context_properties *, cl_uint, const cl_device_id *, void (CL_CALLBACK *)(const char *, const void *, size_t, void *), void *, cl_int *);
@@ -61,6 +62,7 @@ typedef cl_int (CL_API_CALL *OCL_CLGETPLATFORMIDS) (cl_uint,
typedef cl_int (CL_API_CALL *OCL_CLGETPLATFORMINFO) (cl_platform_id, cl_platform_info, size_t, void *, size_t *);
typedef cl_int (CL_API_CALL *OCL_CLGETPROGRAMBUILDINFO) (cl_program, cl_device_id, cl_program_build_info, size_t, void *, size_t *);
typedef cl_int (CL_API_CALL *OCL_CLGETPROGRAMINFO) (cl_program, cl_program_info, size_t, void *, size_t *);
+typedef cl_program (CL_API_CALL *OCL_CLLINKPROGRAM) (cl_context, cl_uint, const cl_device_id *, const char *, cl_uint, const cl_program *, void (CL_CALLBACK *) (cl_program, void *), void *, cl_int *);
typedef cl_int (CL_API_CALL *OCL_CLRELEASECOMMANDQUEUE) (cl_command_queue);
typedef cl_int (CL_API_CALL *OCL_CLRELEASECONTEXT) (cl_context);
typedef cl_int (CL_API_CALL *OCL_CLRELEASEEVENT) (cl_event);
@@ -75,6 +77,7 @@ typedef struct hc_opencl_lib
hc_dynlib_t lib;
OCL_CLBUILDPROGRAM clBuildProgram;
+ OCL_CLCOMPILEPROGRAM clCompileProgram;
OCL_CLCREATEBUFFER clCreateBuffer;
OCL_CLCREATECOMMANDQUEUE clCreateCommandQueue;
OCL_CLCREATECONTEXT clCreateContext;
@@ -98,6 +101,7 @@ typedef struct hc_opencl_lib
OCL_CLGETPLATFORMINFO clGetPlatformInfo;
OCL_CLGETPROGRAMBUILDINFO clGetProgramBuildInfo;
OCL_CLGETPROGRAMINFO clGetProgramInfo;
+ OCL_CLLINKPROGRAM clLinkProgram;
OCL_CLRELEASECOMMANDQUEUE clReleaseCommandQueue;
OCL_CLRELEASECONTEXT clReleaseContext;
OCL_CLRELEASEEVENT clReleaseEvent;
diff --git a/src/backend.c b/src/backend.c
index 804787d57..bc5ce0fe1 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -2183,6 +2183,7 @@ int ocl_init (hashcat_ctx_t *hashcat_ctx)
if (ocl->lib == NULL) return -1;
HC_LOAD_FUNC (ocl, clBuildProgram, OCL_CLBUILDPROGRAM, OpenCL, 1);
+ HC_LOAD_FUNC (ocl, clCompileProgram, OCL_CLCOMPILEPROGRAM, OpenCL, 1);
HC_LOAD_FUNC (ocl, clCreateBuffer, OCL_CLCREATEBUFFER, OpenCL, 1);
HC_LOAD_FUNC (ocl, clCreateCommandQueue, OCL_CLCREATECOMMANDQUEUE, OpenCL, 1);
HC_LOAD_FUNC (ocl, clCreateContext, OCL_CLCREATECONTEXT, OpenCL, 1);
@@ -2205,6 +2206,7 @@ int ocl_init (hashcat_ctx_t *hashcat_ctx)
HC_LOAD_FUNC (ocl, clGetPlatformInfo, OCL_CLGETPLATFORMINFO, OpenCL, 1);
HC_LOAD_FUNC (ocl, clGetProgramBuildInfo, OCL_CLGETPROGRAMBUILDINFO, OpenCL, 1);
HC_LOAD_FUNC (ocl, clGetProgramInfo, OCL_CLGETPROGRAMINFO, OpenCL, 1);
+ HC_LOAD_FUNC (ocl, clLinkProgram, OCL_CLLINKPROGRAM, OpenCL, 1);
HC_LOAD_FUNC (ocl, clReleaseCommandQueue, OCL_CLRELEASECOMMANDQUEUE, OpenCL, 1);
HC_LOAD_FUNC (ocl, clReleaseContext, OCL_CLRELEASECONTEXT, OpenCL, 1);
HC_LOAD_FUNC (ocl, clReleaseKernel, OCL_CLRELEASEKERNEL, OpenCL, 1);
@@ -2571,6 +2573,44 @@ int hc_clBuildProgram (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint n
return 0;
}
+int hc_clCompileProgram (hashcat_ctx_t *hashcat_ctx, cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, cl_uint num_input_headers, const cl_program *input_headers, const char **header_include_names, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data)
+{
+ backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+ OCL_PTR *ocl = (OCL_PTR *) backend_ctx->ocl;
+
+ const cl_int CL_err = ocl->clCompileProgram (program, num_devices, device_list, options, num_input_headers, input_headers, header_include_names, pfn_notify, user_data);
+
+ if (CL_err != CL_SUCCESS)
+ {
+ event_log_error (hashcat_ctx, "clCompileProgram(): %s", val2cstr_cl (CL_err));
+
+ return -1;
+ }
+
+ return 0;
+}
+
+int hc_clLinkProgram (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint num_devices, const cl_device_id *device_list, const char *options, cl_uint num_input_programs, const cl_program *input_programs, void (CL_CALLBACK *pfn_notify) (cl_program program, void *user_data), void *user_data, cl_program *program)
+{
+ backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
+
+ OCL_PTR *ocl = (OCL_PTR *) backend_ctx->ocl;
+
+ cl_int CL_err;
+
+ *program = ocl->clLinkProgram (context, num_devices, device_list, options, num_input_programs, input_programs, pfn_notify, user_data, &CL_err);
+
+ if (CL_err != CL_SUCCESS)
+ {
+ event_log_error (hashcat_ctx, "clLinkProgram(): %s", val2cstr_cl (CL_err));
+
+ return -1;
+ }
+
+ return 0;
+}
+
int hc_clCreateKernel (hashcat_ctx_t *hashcat_ctx, cl_program program, const char *kernel_name, cl_kernel *kernel)
{
backend_ctx_t *backend_ctx = hashcat_ctx->backend_ctx;
@@ -7707,17 +7747,17 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
if (device_param->is_opencl == true)
{
- if (hc_clCreateProgramWithSource (hashcat_ctx, device_param->opencl_context, 1, (const char **) kernel_sources, NULL, opencl_program) == -1) return false;
-
- const int CL_rc = hc_clBuildProgram (hashcat_ctx, *opencl_program, 1, &device_param->opencl_device, build_options_buf, NULL, NULL);
-
- //if (CL_rc == -1) return -1;
-
size_t build_log_size = 0;
- hc_clGetProgramBuildInfo (hashcat_ctx, *opencl_program, device_param->opencl_device, CL_PROGRAM_BUILD_LOG, 0, NULL, &build_log_size);
+ int CL_rc;
- //if (CL_rc == -1) return -1;
+ cl_program p1 = NULL;
+
+ if (hc_clCreateProgramWithSource (hashcat_ctx, device_param->opencl_context, 1, (const char **) kernel_sources, NULL, &p1) == -1) return false;
+
+ CL_rc = hc_clCompileProgram (hashcat_ctx, p1, 1, &device_param->opencl_device, build_options_buf, 0, NULL, NULL, NULL, NULL);
+
+ hc_clGetProgramBuildInfo (hashcat_ctx, p1, device_param->opencl_device, CL_PROGRAM_BUILD_LOG, 0, NULL, &build_log_size);
#if defined (DEBUG)
if ((build_log_size > 1) || (CL_rc == -1))
@@ -7727,7 +7767,7 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
{
char *build_log = (char *) hcmalloc (build_log_size + 1);
- const int rc_clGetProgramBuildInfo = hc_clGetProgramBuildInfo (hashcat_ctx, *opencl_program, device_param->opencl_device, CL_PROGRAM_BUILD_LOG, build_log_size, build_log, NULL);
+ const int rc_clGetProgramBuildInfo = hc_clGetProgramBuildInfo (hashcat_ctx, p1, device_param->opencl_device, CL_PROGRAM_BUILD_LOG, build_log_size, build_log, NULL);
if (rc_clGetProgramBuildInfo == -1) return false;
@@ -7738,6 +7778,21 @@ static bool load_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_p
if (CL_rc == -1) return false;
+ cl_program t2[1];
+
+ t2[0] = p1;
+
+ cl_program fin;
+
+ if (hc_clLinkProgram (hashcat_ctx, device_param->opencl_context, 1, &device_param->opencl_device, NULL, 1, t2, NULL, NULL, &fin) == -1) return false;
+
+ // it seems errors caused by clLinkProgram() do not go into CL_PROGRAM_BUILD
+ // I couldn't find any information on the web explaining how else to retrieve the error messages from the linker
+
+ *opencl_program = fin;
+
+ hc_clReleaseProgram (hashcat_ctx, p1);
+
if (cache_disable == false)
{
size_t binary_size;