Implement async run_cuda_kernel_memset() and run_cuda_kernel_memset32()

This commit is contained in:
Jukka Ojanen
2021-07-27 18:56:59 +03:00
parent e8be7028cd
commit cdf27a1cb3
5 changed files with 210 additions and 106 deletions
+11 -3
View File
@@ -72,6 +72,8 @@ int hc_cuMemcpyDtoDAsync (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDev
int hc_cuMemcpyDtoHAsync (hashcat_ctx_t *hashcat_ctx, void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
int hc_cuMemcpyHtoDAsync (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream);
int hc_cuMemFree (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dptr);
int hc_cuMemsetD32Async (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream);
int hc_cuMemsetD8Async (hashcat_ctx_t *hashcat_ctx, CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream);
int hc_cuModuleGetFunction (hashcat_ctx_t *hashcat_ctx, CUfunction *hfunc, CUmodule hmod, const char *name);
int hc_cuModuleLoadDataEx (hashcat_ctx_t *hashcat_ctx, CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues);
int hc_cuModuleUnload (hashcat_ctx_t *hashcat_ctx, CUmodule hmod);
@@ -120,6 +122,8 @@ int hc_hipMemcpyDtoDAsync (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDe
int hc_hipMemcpyDtoHAsync (hashcat_ctx_t *hashcat_ctx, void *dstHost, HIPdeviceptr srcDevice, size_t ByteCount, HIPstream hStream);
int hc_hipMemcpyHtoDAsync (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, const void *srcHost, size_t ByteCount, HIPstream hStream);
int hc_hipMemFree (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dptr);
int hc_hipMemsetD32Async (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, unsigned int ui, size_t N, HIPstream hStream);
int hc_hipMemsetD8Async (hashcat_ctx_t *hashcat_ctx, HIPdeviceptr dstDevice, unsigned char uc, size_t N, HIPstream hStream);
int hc_hipModuleGetFunction (hashcat_ctx_t *hashcat_ctx, HIPfunction *hfunc, HIPmodule hmod, const char *name);
int hc_hipModuleLoadDataEx (hashcat_ctx_t *hashcat_ctx, HIPmodule *module, const void *image, unsigned int numOptions, HIPjit_option *options, void **optionValues);
int hc_hipModuleUnload (hashcat_ctx_t *hashcat_ctx, HIPmodule hmod);
@@ -142,6 +146,7 @@ int hc_clCreateKernel (hashcat_ctx_t *hashcat_ctx, cl_program program
int hc_clCreateProgramWithBinary (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint num_devices, const cl_device_id *device_list, const size_t *lengths, const unsigned char **binaries, cl_int *binary_status, cl_program *program);
int hc_clCreateProgramWithSource (hashcat_ctx_t *hashcat_ctx, cl_context context, cl_uint count, const char **strings, const size_t *lengths, cl_program *program);
int hc_clEnqueueCopyBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem src_buffer, cl_mem dst_buffer, size_t src_offset, size_t dst_offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
int hc_clEnqueueFillBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, const void *pattern, size_t pattern_size, size_t offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
int hc_clEnqueueMapBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, cl_bool blocking_map, cl_map_flags map_flags, size_t offset, size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event, void **buf);
int hc_clEnqueueNDRangeKernel (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_kernel kernel, cl_uint work_dim, const size_t *global_work_offset, const size_t *global_work_size, const size_t *local_work_size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
int hc_clEnqueueReadBuffer (hashcat_ctx_t *hashcat_ctx, cl_command_queue command_queue, cl_mem buffer, cl_bool blocking_read, size_t offset, size_t size, void *ptr, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event);
@@ -177,17 +182,20 @@ void rebuild_pws_compressed_append (hc_device_param_t *device_param, const u64 p
int run_cuda_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 num);
int run_cuda_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 num);
int run_cuda_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u32 value, const u64 size);
int run_cuda_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u8 value, const u64 size);
int run_cuda_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 offset, const u32 value, const u64 size);
int run_cuda_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, CUdeviceptr buf, const u64 size);
int run_hip_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num);
int run_hip_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 num);
int run_hip_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u32 value, const u64 size);
int run_hip_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 offset, const u8 value, const u64 size);
int run_hip_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 offset, const u32 value, const u64 size);
int run_hip_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, HIPdeviceptr buf, const u64 size);
int run_opencl_kernel_atinit (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 num);
int run_opencl_kernel_utf8toutf16le (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 num);
int run_opencl_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u32 value, const u64 size);
int run_opencl_kernel_memset (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u8 value, const u64 size);
int run_opencl_kernel_memset32 (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 offset, const u32 value, const u64 size);
int run_opencl_kernel_bzero (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, cl_mem buf, const u64 size);
int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, const u32 kern_run, const u64 pws_pos, const u64 num, const u32 event_update, const u32 iteration);
+2
View File
@@ -46,6 +46,7 @@ typedef cl_context (CL_API_CALL *OCL_CLCREATECONTEXT) (const cl_
typedef cl_kernel (CL_API_CALL *OCL_CLCREATEKERNEL) (cl_program, const char *, cl_int *);
typedef cl_program (CL_API_CALL *OCL_CLCREATEPROGRAMWITHBINARY) (cl_context, cl_uint, const cl_device_id *, const size_t *, const unsigned char **, cl_int *, cl_int *);
typedef cl_program (CL_API_CALL *OCL_CLCREATEPROGRAMWITHSOURCE) (cl_context, cl_uint, const char **, const size_t *, cl_int *);
typedef cl_int (CL_API_CALL *OCL_CLENQUEUEFILLBUFFER) (cl_command_queue, cl_mem, const void *, size_t, size_t, size_t, cl_uint, const cl_event *, cl_event *);
typedef cl_int (CL_API_CALL *OCL_CLENQUEUECOPYBUFFER) (cl_command_queue, cl_mem, cl_mem, size_t, size_t, size_t, cl_uint, const cl_event *, cl_event *);
typedef void * (CL_API_CALL *OCL_CLENQUEUEMAPBUFFER) (cl_command_queue, cl_mem, cl_bool, cl_map_flags, size_t, size_t, cl_uint, const cl_event *, cl_event *, cl_int *);
typedef cl_int (CL_API_CALL *OCL_CLENQUEUENDRANGEKERNEL) (cl_command_queue, cl_kernel, cl_uint, const size_t *, const size_t *, const size_t *, cl_uint, const cl_event *, cl_event *);
@@ -87,6 +88,7 @@ typedef struct hc_opencl_lib
OCL_CLCREATEPROGRAMWITHBINARY clCreateProgramWithBinary;
OCL_CLCREATEPROGRAMWITHSOURCE clCreateProgramWithSource;
OCL_CLENQUEUECOPYBUFFER clEnqueueCopyBuffer;
OCL_CLENQUEUEFILLBUFFER clEnqueueFillBuffer;
OCL_CLENQUEUEMAPBUFFER clEnqueueMapBuffer;
OCL_CLENQUEUENDRANGEKERNEL clEnqueueNDRangeKernel;
OCL_CLENQUEUEREADBUFFER clEnqueueReadBuffer;
+4 -4
View File
@@ -1034,8 +1034,8 @@ typedef CUresult (CUDA_API_CALL *CUDA_CUMEMCPYHTODASYNC) (CUdeviceptr, co
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMFREE) (CUdeviceptr);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMFREEHOST) (void *);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMGETINFO) (size_t *, size_t *);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD32) (CUdeviceptr, unsigned int, size_t);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD8) (CUdeviceptr, unsigned char, size_t);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD32ASYNC) (CUdeviceptr, unsigned int, size_t, CUstream);
typedef CUresult (CUDA_API_CALL *CUDA_CUMEMSETD8ASYNC) (CUdeviceptr, unsigned char, size_t, CUstream);
typedef CUresult (CUDA_API_CALL *CUDA_CUMODULEGETFUNCTION) (CUfunction *, CUmodule, const char *);
typedef CUresult (CUDA_API_CALL *CUDA_CUMODULEGETGLOBAL) (CUdeviceptr *, size_t *, CUmodule, const char *);
typedef CUresult (CUDA_API_CALL *CUDA_CUMODULELOAD) (CUmodule *, const char *);
@@ -1096,8 +1096,8 @@ typedef struct hc_cuda_lib
CUDA_CUMEMFREE cuMemFree;
CUDA_CUMEMFREEHOST cuMemFreeHost;
CUDA_CUMEMGETINFO cuMemGetInfo;
CUDA_CUMEMSETD32 cuMemsetD32;
CUDA_CUMEMSETD8 cuMemsetD8;
CUDA_CUMEMSETD32ASYNC cuMemsetD32Async;
CUDA_CUMEMSETD8ASYNC cuMemsetD8Async;
CUDA_CUMODULEGETFUNCTION cuModuleGetFunction;
CUDA_CUMODULEGETGLOBAL cuModuleGetGlobal;
CUDA_CUMODULELOAD cuModuleLoad;
+4 -4
View File
@@ -1043,8 +1043,8 @@ typedef HIPresult (HIP_API_CALL *HIP_HIPMEMCPYHTODASYNC) (HIPdeviceptr, c
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMFREE) (HIPdeviceptr);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMFREEHOST) (void *);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMGETINFO) (size_t *, size_t *);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD32) (HIPdeviceptr, unsigned int, size_t);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD8) (HIPdeviceptr, unsigned char, size_t);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD32ASYNC) (HIPdeviceptr, unsigned int, size_t, HIPstream);
typedef HIPresult (HIP_API_CALL *HIP_HIPMEMSETD8ASYNC) (HIPdeviceptr, unsigned char, size_t, HIPstream);
typedef HIPresult (HIP_API_CALL *HIP_HIPMODULEGETFUNCTION) (HIPfunction *, HIPmodule, const char *);
typedef HIPresult (HIP_API_CALL *HIP_HIPMODULEGETGLOBAL) (HIPdeviceptr *, size_t *, HIPmodule, const char *);
typedef HIPresult (HIP_API_CALL *HIP_HIPMODULELOAD) (HIPmodule *, const char *);
@@ -1105,8 +1105,8 @@ typedef struct hc_hip_lib
HIP_HIPMEMFREE hipMemFree;
HIP_HIPMEMFREEHOST hipMemFreeHost;
HIP_HIPMEMGETINFO hipMemGetInfo;
HIP_HIPMEMSETD32 hipMemsetD32;
HIP_HIPMEMSETD8 hipMemsetD8;
HIP_HIPMEMSETD32ASYNC hipMemsetD32Async;
HIP_HIPMEMSETD8ASYNC hipMemsetD8Async;
HIP_HIPMODULEGETFUNCTION hipModuleGetFunction;
HIP_HIPMODULEGETGLOBAL hipModuleGetGlobal;
HIP_HIPMODULELOAD hipModuleLoad;