static GLuint make_shader(GLenum type, const char *filename) { GLint length; GLchar *source = file_contents(filename, &length); GLuint shader; GLint shader_ok; if (!source) return 0; shader = glCreateShader(type); glShaderSource(shader, 1, (const GLchar**)&source, &length); free(source); glCompileShader(shader); glGetShaderiv(shader, GL_COMPILE_STATUS, &shader_ok); if (!shader_ok) { glGetShaderInfoLog(shader, sizeof(logbuf), NULL, logbuf); fprintf(stderr, "Failed to compile %s:\n", filename); fputs(logbuf, stderr); glDeleteShader(shader); return 0; } return shader; }
void CL::loadProgram(const char* relative_path) { // Program Setup int pl; size_t program_length; printf("load the program\n"); //CL_SOURCE_DIR is set in the CMakeLists.txt std::string path(CL_SOURCE_DIR); path += "/" + std::string(relative_path); printf("path: %s\n", path.c_str()); //file_contents is defined in util.cpp //it loads the contents of the file at the given path char* cSourceCL = file_contents(path.c_str(), &pl); //printf("file: %s\n", cSourceCL); program_length = (size_t)pl; // create the program program = clCreateProgramWithSource(context, 1, (const char **) &cSourceCL, &program_length, &err); printf("clCreateProgramWithSource: %s\n", oclErrorString(err)); buildExecutable(); }
JNIEXPORT jboolean JNICALL Java_com_example_LiveFeatureActivity_compileKernels(JNIEnv *env, jclass clazz) { // Find OCL devices and compile kernels cl_int err = CL_SUCCESS; try { std::vector<cl::Platform> platforms; cl::Platform::get(&platforms); if (platforms.size() == 0) { return false; } cl_context_properties properties[] = { CL_CONTEXT_PLATFORM, (cl_context_properties)(platforms[0])(), 0}; gContext = cl::Context(CL_DEVICE_TYPE_GPU, properties); std::vector<cl::Device> devices = gContext.getInfo<CL_CONTEXT_DEVICES>(); gQueue = cl::CommandQueue(gContext, devices[0], 0, &err); int src_length = 0; const char* src = file_contents("/data/data/com.example/app_execdir/kernels.cl",&src_length); cl::Program::Sources sources(1,std::make_pair(src, src_length) ); cl::Program program(gContext, sources); program.build(devices,NULL,cb); while(program.getBuildInfo<CL_PROGRAM_BUILD_STATUS>(devices[0]) != CL_BUILD_SUCCESS); gNV21Kernel = cl::Kernel(program, "nv21torgba", &err); gLaplacianK = cl::Kernel(program, "laplacian", &err); gNegative = cl::Kernel(program, "negative", &err); return true; } catch (cl::Error e) { if( !throwJavaException(env,"decode",e.what(),e.err()) ) LOGI("@decode: %s \n",e.what()); return false; } }
static char * desktop_from_dir(const char *directory, char **error_message) { char *environ_contents = file_contents(directory, "environ", error_message); if (!environ_contents) return NULL; char *desktop = strstr(environ_contents, "DESKTOP_SESSION="); if (!desktop) { free(environ_contents); return NULL; } /* avoid prefixes - DESKTOP_SESSION= must either be the very first variable or preceeded by a newline */ if (desktop != environ_contents && *(desktop - 1) != '\n') { free(environ_contents); return NULL; } desktop += strlen("DESKTOP_SESSION="); char *newline = strchrnul(desktop, '\n'); *newline = '\0'; char *result = sr_strdup(desktop); free(environ_contents); return result; }
int main(int argc, char **argv){ bt_zero("~~~Userspace test program start!~~~\n"); print_string("TEST Command Prompt!\n"); while(true){ char input[128]={0}; print_string("[TEST]>"); get_string(input, 128); if(input[0]=='d') dir_listing(); else if(input[0]=='b') ata_test(); else if(input[0]=='l') dir_listing2(input); else if(input[0]=='f') file_contents(); else if(input[0]=='c') file_contents2(input); else if(input[0]=='m') mount_test(); else if(input[0]=='v') version(); else if(input[0]=='r') run_program(input); else if(input[0]=='p') path(input); else if(input[0]=='t') thread_test(); else if(input[0]=='x') crash_test(); else if(input[0]=='q') break; else { if(strlen(input) && input[0]!='\n') print_string("Unrecognised command.\n"); } } bt_zero("~~~Userspace test program done!~~~\n"); bt_exit(0); return 0; }
static GLuint make_shader(GLenum type, const std::string& filename)/*{{{*/ { /* Utility function to create a shader object from a file*/ GLint length; GLchar *source = (GLchar*)file_contents(filename.c_str(), &length); GLuint shader; GLint shader_ok; if (!source) return 0; // shader = glCreateShader(type); glShaderSource(shader, 1, (const GLchar**)&source, &length); free(source); glCompileShader(shader); // glGetShaderiv(shader, GL_COMPILE_STATUS, &shader_ok); if (!shader_ok) { std::cerr<<"Failed to compile "<<filename<<":"<<std::endl; show_info_log(shader, glGetShaderiv, glGetShaderInfoLog); glDeleteShader(shader); return 0; } else { // std::cout << filename << " compiled" << std::endl; } return shader; }/*}}}*/
struct sr_operating_system * sr_abrt_operating_system_from_dir(const char *directory, char **error_message) { bool success = false; struct sr_operating_system *os = sr_operating_system_new(); char *osinfo_contents = file_contents(directory, "os_info", error_message); if (osinfo_contents) { success = sr_operating_system_parse_etc_os_release(osinfo_contents, os); free(osinfo_contents); } /* fall back to os_release if parsing os_info fails */ if (!success) { char *release_contents = file_contents(directory, "os_release", error_message); if (release_contents) { success = sr_operating_system_parse_etc_system_release(release_contents, &os->name, &os->version); free(release_contents); } } if (!success) { sr_operating_system_free(os); *error_message = sr_strdup("Failed to parse operating system release string"); return NULL; } os->architecture = file_contents(directory, "architecture", error_message); if (!os->architecture) { sr_operating_system_free(os); return NULL; } /* optional - failure is not fatal */ os->desktop = desktop_from_dir(directory, error_message); return os; }
struct sr_rpm_package * sr_abrt_rpm_packages_from_dir(const char *directory, char **error_message) { char *epoch_str = file_contents(directory, "pkg_epoch", error_message); if (!epoch_str) { return NULL; } unsigned long epoch = strtoul(epoch_str, NULL, 10); if (epoch == ULONG_MAX) { *error_message = sr_asprintf("Epoch '%s' is not a number", epoch_str); return NULL; } free(epoch_str); struct sr_rpm_package *packages = sr_rpm_package_new(); packages->epoch = (uint32_t)epoch; packages->name = file_contents(directory, "pkg_name", error_message); packages->version = file_contents(directory, "pkg_version", error_message); packages->release = file_contents(directory, "pkg_release", error_message); packages->architecture = file_contents(directory, "pkg_arch", error_message); packages->role = SR_ROLE_AFFECTED; if (!(packages->name && packages->version && packages->release && packages->architecture)) { sr_rpm_package_free(packages, false); return NULL; } /* Workaround abrt-action-save-kernel-data appending trailing \n for * koopses. */ strip_newline(packages->name); strip_newline(packages->version); strip_newline(packages->release); strip_newline(packages->architecture); char *dso_list_contents = file_contents(directory, "dso_list", error_message); if (dso_list_contents) { struct sr_rpm_package *dso_packages = sr_abrt_parse_dso_list(dso_list_contents); if (dso_packages) { packages = sr_rpm_package_append(packages, dso_packages); packages = sr_rpm_package_sort(packages); packages = sr_rpm_package_uniq(packages); } free(dso_list_contents); } return packages; }
static bool create_core_stacktrace(const char *directory, const char *gdb_output, bool hash_fingerprints, char **error_message) { char *executable_contents = file_contents(directory, "executable", error_message); if (!executable_contents) return NULL; char *coredump_filename = sr_build_path(directory, "coredump", NULL); struct sr_core_stacktrace *core_stacktrace; if (gdb_output) core_stacktrace = sr_core_stacktrace_from_gdb(gdb_output, coredump_filename, executable_contents, error_message); else core_stacktrace = sr_parse_coredump(coredump_filename, executable_contents, error_message); free(executable_contents); free(coredump_filename); if (!core_stacktrace) return false; fulfill_missing_values(core_stacktrace); #if 0 sr_core_fingerprint_generate(core_stacktrace, error_message); if (hash_fingerprints) sr_core_fingerprint_hash(core_stacktrace); #endif char *json = sr_core_stacktrace_to_json(core_stacktrace); // Add newline to the end of core stacktrace file to make text // editors happy. json = sr_realloc(json, strlen(json) + 2); strcat(json, "\n"); char *core_backtrace_filename = sr_build_path(directory, "core_backtrace", NULL); bool success = sr_string_to_file(core_backtrace_filename, json, error_message); free(core_backtrace_filename); free(json); sr_core_stacktrace_free(core_stacktrace); return success; }
void Shader::set_source_from_file(const char * filename) { std::ifstream ifs; ifs.open(filename); if (!ifs.is_open()) { throw Error(std::string("Error opening ") + filename); } ifs.seekg(0, ifs.end); int length = ifs.tellg(); ifs.seekg(0, ifs.beg); std::vector<char> file_contents(length); ifs.read(&file_contents[0], length); set_source(&file_contents[0], length); }
int main(){ std::string file_contents("adsf"); x3_ast::declaration result; boost::spirit::x3::ascii::space_type space; auto it = file_contents.begin(); auto end = file_contents.end(); bool r = x3::phrase_parse(it, end, x3_grammar::parser, space, result); if(r && it == end){ return true; } else { return false; } }
GLuint CreateShader(GLenum type, const char *filename) { GLint length; GLchar *source = (GLchar*) file_contents(filename, &length); GLuint shader; GLint shader_ok; if (!source){ return 0; } shader = glCreateShader(type); glShaderSource(shader, 1, (const GLchar**)&source, &length); delete source; glCompileShader(shader); GLint status; glGetShaderiv(shader, GL_COMPILE_STATUS, &status); if (status == GL_FALSE) { GLint infoLogLength; glGetShaderiv(shader, GL_INFO_LOG_LENGTH, &infoLogLength); GLchar *strInfoLog = new GLchar[infoLogLength + 1]; glGetShaderInfoLog(shader, infoLogLength, NULL, strInfoLog); const char *strShaderType = NULL; switch(type) { case GL_VERTEX_SHADER: strShaderType = "vertex"; break; case GL_GEOMETRY_SHADER: strShaderType = "geometry"; break; case GL_FRAGMENT_SHADER: strShaderType = "fragment"; break; } printf("Compile failure in %s shader:\n%s\n", strShaderType, strInfoLog); delete[] strInfoLog; } return shader; }
cl_program build_opencl_kernel(build_kernel_data *data, const char *filename) { int pl; char *source = file_contents(data->source_filename, &pl); size_t sourceSize[] = {(size_t)pl}; cl_int status; cl_program program = NULL; cl_program ret = NULL; if (!source) goto out; program = clCreateProgramWithSource(data->context, 1, (const char **)&source, sourceSize, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithSource)", status); goto out; } applog(LOG_DEBUG, "CompilerOptions: %s", data->compiler_options); status = clBuildProgram(program, 1, data->device, data->compiler_options, NULL, NULL); if (status != CL_SUCCESS) { size_t log_size; applog(LOG_ERR, "Error %d: Building Program (clBuildProgram)", status); status = clGetProgramBuildInfo(program, *data->device, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size); char *sz_log = (char *)malloc(log_size + 1); status = clGetProgramBuildInfo(program, *data->device, CL_PROGRAM_BUILD_LOG, log_size, sz_log, NULL); sz_log[log_size] = '\0'; applogsiz(LOG_ERR, log_size, "%s", sz_log); free(sz_log); goto out; } ret = program; out: if (source) free(source); return ret; }
static GLuint createShader(GLenum eShaderType , const char *filename) { GLint length; GLuint shader; GLint status; GLchar *source = file_contents(filename, &length); if (!source) return 0; shader = glCreateShader(eShaderType); glShaderSource(shader, 1, (const GLchar**)&source, &length); glCompileShader(shader); glGetShaderiv(shader, GL_COMPILE_STATUS, &status); if (status == GL_FALSE) { GLint infoLogLength; glGetShaderiv(shader, GL_INFO_LOG_LENGTH, &infoLogLength); GLchar *strInfoLog = malloc(infoLogLength); glGetShaderInfoLog(shader, infoLogLength, NULL, strInfoLog); const char *strShaderType = NULL; switch(eShaderType) { case GL_VERTEX_SHADER: strShaderType = "vertex"; break; #ifdef GL_GEOMETRY_SHADER case GL_GEOMETRY_SHADER: strShaderType = "geometry"; break; #endif case GL_FRAGMENT_SHADER: strShaderType = "fragment"; break; } fprintf(stderr, "Compile failure in %s shader:\n%s\n", strShaderType, strInfoLog); free(strInfoLog); } return shader; }
GLuint make_shader(GLenum type, const char *filename) { GLint length; GLchar *source = (GLchar *)file_contents(filename, &length); GLuint shader; GLint shader_ok; if (!source) return 0; shader = glCreateShader(type); glShaderSource(shader, 1, (const GLchar**)&source, &length); free(source); glCompileShader(shader); glGetShaderiv(shader, GL_COMPILE_STATUS, &shader_ok); if (!shader_ok) { fprintf(stderr, "Failed to compile %s:\n", filename); show_info_log(shader, glGetShaderiv, glGetShaderInfoLog); glDeleteShader(shader); return 0; } return shader; }
void http_cache::get (const std::string& url, int timeout_msec, std::function<void(std::string)> callback) const { auto file (cache_file(url)); if (fs::exists(file)) { callback(file_contents(url)); return; } boost::thread bg ([=]{ http::client::request rq (url); rq << header("Connection", "close"); http::client temp_client; std::string data (http::body(temp_client.get(rq))); std::ofstream out (file.string(), std::ios::binary); out.write(&data[0], data.size()); out.close(); callback(std::move(data)); }); bg.detach(); }
struct sr_report * sr_abrt_report_from_dir(const char *directory, char **error_message) { struct sr_report *report = sr_report_new(); /* Report type. */ char *type_contents = file_contents(directory, "type", error_message); if (!type_contents) { sr_report_free(report); return NULL; } report->report_type = sr_abrt_type_from_type(type_contents); free(type_contents); /* Operating system. */ report->operating_system = sr_abrt_operating_system_from_dir( directory, error_message); if (!report->operating_system) { sr_report_free(report); return NULL; } /* Component name. */ report->component_name = file_contents(directory, "component", error_message); /* RPM packages. */ report->rpm_packages = sr_abrt_rpm_packages_from_dir( directory, error_message); if (!report->rpm_packages) { sr_report_free(report); return NULL; } /* Core stacktrace. */ if (report->report_type == SR_REPORT_CORE) { char *core_backtrace_contents = file_contents(directory, "core_backtrace", error_message); if (!core_backtrace_contents) { sr_report_free(report); return NULL; } report->stacktrace = (struct sr_stacktrace *)sr_core_stacktrace_from_json_text( core_backtrace_contents, error_message); free(core_backtrace_contents); if (!report->stacktrace) { sr_report_free(report); return NULL; } } /* Python stacktrace. */ if (report->report_type == SR_REPORT_PYTHON) { char *backtrace_contents = file_contents(directory, "backtrace", error_message); if (!backtrace_contents) { sr_report_free(report); return NULL; } /* Parse the Python stacktrace. */ struct sr_location location; sr_location_init(&location); const char *contents_pointer = backtrace_contents; report->stacktrace = (struct sr_stacktrace *)sr_python_stacktrace_parse( &contents_pointer, &location); free(backtrace_contents); if (!report->stacktrace) { *error_message = sr_location_to_string(&location); sr_report_free(report); return NULL; } } /* Kerneloops stacktrace. */ if (report->report_type == SR_REPORT_KERNELOOPS) { /* Determine kernel version */ char *kernel_contents = file_contents(directory, "kernel", error_message); if (!kernel_contents) { sr_report_free(report); return NULL; } /* Load the Kerneloops stacktrace */ char *backtrace_contents = file_contents(directory, "backtrace", error_message); if (!backtrace_contents) { sr_report_free(report); return NULL; } /* Parse the Kerneloops stacktrace. */ struct sr_location location; sr_location_init(&location); const char *contents_pointer = backtrace_contents; struct sr_koops_stacktrace *stacktrace = sr_koops_stacktrace_parse( &contents_pointer, &location); stacktrace->version = kernel_contents; report->stacktrace = (struct sr_stacktrace *)stacktrace; free(backtrace_contents); if (!report->stacktrace) { *error_message = sr_location_to_string(&location); sr_report_free(report); return NULL; } } /* Java stacktrace. */ if (report->report_type == SR_REPORT_JAVA) { char *backtrace_contents = file_contents(directory, "backtrace", error_message); if (!backtrace_contents) { sr_report_free(report); return NULL; } /* Parse the Java stacktrace. */ struct sr_location location; sr_location_init(&location); const char *contents_pointer = backtrace_contents; report->stacktrace = (struct sr_stacktrace *)sr_java_stacktrace_parse( &contents_pointer, &location); free(backtrace_contents); if (!report->stacktrace) { *error_message = sr_location_to_string(&location); sr_report_free(report); return NULL; } } /* Ruby stacktrace. */ if (report->report_type == SR_REPORT_RUBY) { char *backtrace_contents = file_contents(directory, "backtrace", error_message); if (!backtrace_contents) { sr_report_free(report); return NULL; } /* Parse the Ruby stacktrace. */ struct sr_location location; sr_location_init(&location); const char *contents_pointer = backtrace_contents; report->stacktrace = (struct sr_stacktrace *)sr_ruby_stacktrace_parse( &contents_pointer, &location); free(backtrace_contents); if (!report->stacktrace) { *error_message = sr_location_to_string(&location); sr_report_free(report); return NULL; } } return report; }
//---------------------------------------------------------------------- GLuint Render::compileShaders(const char* vertex_file, const char* fragment_file, const char* geometry_file, GLenum* geom_param, GLint* geom_value, int geom_param_len) { //this may not be the cleanest implementation //#include "shaders.cpp" printf("vertex_file: %s\n", vertex_file); printf("fragment_file: %s\n", fragment_file); //printf("vertex shader:\n%s\n", vertex_shader_source); //printf("fragment shader:\n%s\n", fragment_shader_source); char *vertex_shader_source = NULL,*fragment_shader_source= NULL,*geometry_shader_source=NULL; int vert_size,frag_size,geom_size; if (vertex_file) { vertex_shader_source = file_contents(vertex_file,&vert_size); if (!vertex_shader_source) { printf("Vertex shader file not found or is empty! Cannot compile shader"); return -1; } } else { printf("No vertex file specified! Cannot compile shader!"); return -1; } if (fragment_file) { fragment_shader_source = file_contents(fragment_file,&frag_size); if (!fragment_shader_source) { printf("Fragment shader file not found or is empty! Cannot compile shader"); free(vertex_shader_source); return -1; } } else { printf("No fragment file specified! Cannot compile shader!"); free(vertex_shader_source); return -1; } if (geometry_file) { geometry_shader_source = file_contents(fragment_file,&frag_size); if (!geometry_shader_source) { printf("Geometry shader file not found or is empty! Cannot compile shader"); free(vertex_shader_source); free(fragment_shader_source); return -1; } } GLint len; GLuint program = glCreateProgram(); GLuint vertex_shader = glCreateShader(GL_VERTEX_SHADER); glShaderSource(vertex_shader, 1, (const GLchar**)&vertex_shader_source, 0); glCompileShader(vertex_shader); glGetShaderiv(vertex_shader, GL_INFO_LOG_LENGTH, &len); if (len > 0) { char log[1024]; glGetShaderInfoLog(vertex_shader, 1024, 0, log); printf("Vertex Shader log:\n %s\n", log); } glAttachShader(program, vertex_shader); GLuint fragment_shader = glCreateShader(GL_FRAGMENT_SHADER); glShaderSource(fragment_shader, 1, (const GLchar**)&fragment_shader_source, 0); glCompileShader(fragment_shader); glGetShaderiv(fragment_shader, GL_INFO_LOG_LENGTH, &len); if (len > 0) { char log[1024]; glGetShaderInfoLog(fragment_shader, 1024, 0, log); printf("Fragment Shader log:\n %s\n", log); } glAttachShader(program, fragment_shader); GLuint geometry_shader=0; if (geometry_shader_source) { geometry_shader = glCreateShader(GL_GEOMETRY_SHADER_EXT); glShaderSource(geometry_shader, 1, (const GLchar**)&geometry_shader_source, 0); glCompileShader(geometry_shader); glGetShaderiv(geometry_shader, GL_INFO_LOG_LENGTH, &len); printf("geometry len %d\n", len); if (len > 0) { char log[1024]; glGetShaderInfoLog(geometry_shader, 1024, 0, log); printf("Geometry Shader log:\n %s\n", log); } glAttachShader(program, geometry_shader); for (int i = 0;i < geom_param_len; i++) { glProgramParameteriEXT(program,geom_param[i],geom_value[i]); } } glLinkProgram(program); // check if program linked GLint success = 0; glGetProgramiv(program, GL_LINK_STATUS, &success); if (!success) { char temp[256]; glGetProgramInfoLog(program, 256, 0, temp); printf("Failed to link program:\n%s\n", temp); glDeleteProgram(program); program = 0; } //cleanup glDeleteShader(vertex_shader); glDeleteShader(fragment_shader); if (geometry_shader) { glDeleteShader(geometry_shader); } free(vertex_shader_source); free(fragment_shader_source); free(geometry_shader_source); return program; }
void setup_opencl(const char* cl_source_filename, const char* cl_source_main, cl_device_id* device_id, cl_kernel* kernel, cl_context* context, cl_command_queue* queue) { cl_int err; // error code returned from api calls cl_platform_id platform_id; // compute device id cl_program program; // compute program cl_device_id devices[MAX_RESOURCES]; cl_platform_id platforms[MAX_RESOURCES]; unsigned int best_platform = 0; unsigned int best_device = 0; print_devices(0); if(!get_best_device(&best_platform, &best_device)) { printf("No suitable device was found! Try using an OpenCL1.1 compatible device.\n"); exit(1); } printf("Initiating platform-%d device-%d.\n", best_platform, best_device); // Platform err = clGetPlatformIDs(MAX_RESOURCES, platforms, NULL); ocl_error("Getting platform id", err); platform_id = platforms[best_platform]; // Device err = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_ALL, sizeof(devices), devices, NULL); //NULL, ignore number returned devices. ocl_error("Getting device ids", err); *device_id = devices[best_device]; // Context *context = clCreateContext(0, 1, device_id, NULL, NULL, &err); ocl_error("Creating context", err); // Command-queue *queue = clCreateCommandQueue(*context, *device_id, 0, &err); ocl_error("Creating command queue", err); // Read .cl source into memory int cl_source_len = 0; char* cl_source = file_contents(cl_source_filename, &cl_source_len); // Create thes compute program from the source buffer program = clCreateProgramWithSource(*context, 1, (const char **) &cl_source, NULL, &err); ocl_error("Failed to create compute program", err); // Build the program executable err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL); if (err != CL_SUCCESS) { char* build_log; size_t log_size; // First call to know the proper size clGetProgramBuildInfo(program, *device_id, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size); build_log = malloc(sizeof(char)*(log_size+1)); if(log_size > 0 && build_log != NULL) { // Second call to get the log clGetProgramBuildInfo(program, *device_id, CL_PROGRAM_BUILD_LOG, log_size, build_log, NULL); build_log[log_size] = '\0'; printf("%s\n", build_log); free(build_log); } exit(err); } // Create the compute kernel in the program we wish to run *kernel = clCreateKernel(program, cl_source_main, &err); ocl_error("Failed to create compute kernel", err); }
_clState *initCl(unsigned int gpu, char *name, size_t nameSize) { _clState *clState = calloc(1, sizeof(_clState)); bool patchbfi = false, prog_built = false; struct cgpu_info *cgpu = &gpus[gpu]; cl_platform_id platform = NULL; char pbuff[256], vbuff[255]; cl_platform_id* platforms; cl_uint preferred_vwidth; cl_device_id *devices; cl_uint numPlatforms; cl_uint numDevices; cl_int status; status = clGetPlatformIDs(0, NULL, &numPlatforms); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platforms. (clGetPlatformsIDs)", status); return NULL; } platforms = (cl_platform_id *)alloca(numPlatforms*sizeof(cl_platform_id)); status = clGetPlatformIDs(numPlatforms, platforms, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platform Ids. (clGetPlatformsIDs)", status); return NULL; } if (opt_platform_id >= (int)numPlatforms) { applog(LOG_ERR, "Specified platform that does not exist"); return NULL; } status = clGetPlatformInfo(platforms[opt_platform_id], CL_PLATFORM_VENDOR, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platform Info. (clGetPlatformInfo)", status); return NULL; } platform = platforms[opt_platform_id]; if (platform == NULL) { perror("NULL platform found!\n"); return NULL; } applog(LOG_INFO, "CL Platform vendor: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_NAME, sizeof(pbuff), pbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform name: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_VERSION, sizeof(vbuff), vbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform version: %s", vbuff); status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device IDs (num)", status); return NULL; } if (numDevices > 0 ) { devices = (cl_device_id *)malloc(numDevices*sizeof(cl_device_id)); /* Now, get the device list data */ status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, numDevices, devices, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device IDs (list)", status); return NULL; } applog(LOG_INFO, "List of devices:"); unsigned int i; for (i = 0; i < numDevices; i++) { status = clGetDeviceInfo(devices[i], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device Info", status); return NULL; } applog(LOG_INFO, "\t%i\t%s", i, pbuff); } if (gpu < numDevices) { status = clGetDeviceInfo(devices[gpu], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device Info", status); return NULL; } applog(LOG_INFO, "Selected %i: %s", gpu, pbuff); strncpy(name, pbuff, nameSize); } else { applog(LOG_ERR, "Invalid GPU %i", gpu); return NULL; } } else return NULL; cl_context_properties cps[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platform, 0 }; clState->context = clCreateContextFromType(cps, CL_DEVICE_TYPE_GPU, NULL, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Context. (clCreateContextFromType)", status); return NULL; } ///////////////////////////////////////////////////////////////// // Create an OpenCL command queue ///////////////////////////////////////////////////////////////// clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &status); if (status != CL_SUCCESS) /* Try again without OOE enable */ clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], 0 , &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Command Queue. (clCreateCommandQueue)", status); return NULL; } /* Check for BFI INT support. Hopefully people don't mix devices with * and without it! */ char * extensions = malloc(1024); const char * camo = "cl_amd_media_ops"; char *find; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_EXTENSIONS, 1024, (void *)extensions, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_EXTENSIONS", status); return NULL; } find = strstr(extensions, camo); if (find) clState->hasBitAlign = true; /* Check for OpenCL >= 1.0 support, needed for global offset parameter usage. */ char * devoclver = malloc(1024); const char * ocl10 = "OpenCL 1.0"; const char * ocl11 = "OpenCL 1.1"; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_VERSION, 1024, (void *)devoclver, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_VERSION", status); return NULL; } find = strstr(devoclver, ocl10); if (!find) { clState->hasOpenCL11plus = true; find = strstr(devoclver, ocl11); if (!find) clState->hasOpenCL12plus = true; } status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&preferred_vwidth, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT", status); return NULL; } applog(LOG_DEBUG, "Preferred vector width reported %d", preferred_vwidth); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void *)&clState->max_work_size, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_WORK_GROUP_SIZE", status); return NULL; } applog(LOG_DEBUG, "Max work group size reported %d", (int)(clState->max_work_size)); size_t compute_units = 0; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(size_t), (void *)&compute_units, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_COMPUTE_UNITS", status); return NULL; } // AMD architechture got 64 compute shaders per compute unit. // Source: http://www.amd.com/us/Documents/GCN_Architecture_whitepaper.pdf clState->compute_shaders = compute_units * 64; applog(LOG_DEBUG, "Max shaders calculated %d", (int)(clState->compute_shaders)); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_MEM_ALLOC_SIZE , sizeof(cl_ulong), (void *)&cgpu->max_alloc, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_MEM_ALLOC_SIZE", status); return NULL; } applog(LOG_DEBUG, "Max mem alloc size is %lu", (long unsigned int)(cgpu->max_alloc)); /* Create binary filename based on parameters passed to opencl * compiler to ensure we only load a binary that matches what would * have otherwise created. The filename is: * name + kernelname +/- g(offset) + v + vectors + w + work_size + l + sizeof(long) + .bin * For scrypt the filename is: * name + kernelname + g + lg + lookup_gap + tc + thread_concurrency + w + work_size + l + sizeof(long) + .bin */ char binaryfilename[255]; char filename[255]; char numbuf[16]; if (cgpu->kernel == KL_NONE) { applog(LOG_INFO, "Selecting kernel ckolivas"); clState->chosen_kernel = KL_CKOLIVAS; cgpu->kernel = clState->chosen_kernel; } else { clState->chosen_kernel = cgpu->kernel; } /* For some reason 2 vectors is still better even if the card says * otherwise, and many cards lie about their max so use 256 as max * unless explicitly set on the command line. Tahiti prefers 1 */ if (strstr(name, "Tahiti")) preferred_vwidth = 1; else if (preferred_vwidth > 2) preferred_vwidth = 2; /* All available kernels only support vector 1 */ cgpu->vwidth = 1; switch (clState->chosen_kernel) { case KL_ALEXKARNEW: applog(LOG_WARNING, "Kernel alexkarnew is experimental."); strcpy(filename, ALEXKARNEW_KERNNAME".cl"); strcpy(binaryfilename, ALEXKARNEW_KERNNAME); break; case KL_ALEXKAROLD: applog(LOG_WARNING, "Kernel alexkarold is experimental."); strcpy(filename, ALEXKAROLD_KERNNAME".cl"); strcpy(binaryfilename, ALEXKAROLD_KERNNAME); break; case KL_CKOLIVAS: strcpy(filename, CKOLIVAS_KERNNAME".cl"); strcpy(binaryfilename, CKOLIVAS_KERNNAME); break; case KL_ZUIKKIS: applog(LOG_WARNING, "Kernel zuikkis is experimental."); strcpy(filename, ZUIKKIS_KERNNAME".cl"); strcpy(binaryfilename, ZUIKKIS_KERNNAME); /* Kernel only supports lookup-gap 2 */ cgpu->lookup_gap = 2; /* Kernel only supports worksize 256 */ cgpu->work_size = 256; break; case KL_NONE: /* Shouldn't happen */ break; } if (cgpu->vwidth) clState->vwidth = cgpu->vwidth; else { clState->vwidth = preferred_vwidth; cgpu->vwidth = preferred_vwidth; } clState->goffset = true; if (cgpu->work_size && cgpu->work_size <= clState->max_work_size) clState->wsize = cgpu->work_size; else clState->wsize = 256; if (!cgpu->opt_lg) { applog(LOG_DEBUG, "GPU %d: selecting lookup gap of 2", gpu); cgpu->lookup_gap = 2; } else cgpu->lookup_gap = cgpu->opt_lg; if (!cgpu->opt_tc) { unsigned int sixtyfours; sixtyfours = cgpu->max_alloc / 131072 / 64 - 1; cgpu->thread_concurrency = sixtyfours * 64; if (cgpu->shaders && cgpu->thread_concurrency > cgpu->shaders) { cgpu->thread_concurrency -= cgpu->thread_concurrency % cgpu->shaders; if (cgpu->thread_concurrency > cgpu->shaders * 5) cgpu->thread_concurrency = cgpu->shaders * 5; } applog(LOG_DEBUG, "GPU %d: selecting thread concurrency of %d", gpu, (int)(cgpu->thread_concurrency)); } else cgpu->thread_concurrency = cgpu->opt_tc; FILE *binaryfile; size_t *binary_sizes; char **binaries; int pl; char *source = file_contents(filename, &pl); size_t sourceSize[] = {(size_t)pl}; cl_uint slot, cpnd; slot = cpnd = 0; if (!source) return NULL; binary_sizes = calloc(sizeof(size_t) * MAX_GPUDEVICES * 4, 1); if (unlikely(!binary_sizes)) { applog(LOG_ERR, "Unable to calloc binary_sizes"); return NULL; } binaries = calloc(sizeof(char *) * MAX_GPUDEVICES * 4, 1); if (unlikely(!binaries)) { applog(LOG_ERR, "Unable to calloc binaries"); return NULL; } strcat(binaryfilename, name); if (clState->goffset) strcat(binaryfilename, "g"); sprintf(numbuf, "lg%utc%u", cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency); strcat(binaryfilename, numbuf); sprintf(numbuf, "w%d", (int)clState->wsize); strcat(binaryfilename, numbuf); sprintf(numbuf, "l%d", (int)sizeof(long)); strcat(binaryfilename, numbuf); strcat(binaryfilename, ".bin"); binaryfile = fopen(binaryfilename, "rb"); if (!binaryfile) { applog(LOG_DEBUG, "No binary found, generating from source"); } else { struct stat binary_stat; if (unlikely(stat(binaryfilename, &binary_stat))) { applog(LOG_DEBUG, "Unable to stat binary, generating from source"); fclose(binaryfile); goto build; } if (!binary_stat.st_size) goto build; binary_sizes[slot] = binary_stat.st_size; binaries[slot] = (char *)calloc(binary_sizes[slot], 1); if (unlikely(!binaries[slot])) { applog(LOG_ERR, "Unable to calloc binaries"); fclose(binaryfile); return NULL; } if (fread(binaries[slot], 1, binary_sizes[slot], binaryfile) != binary_sizes[slot]) { applog(LOG_ERR, "Unable to fread binaries"); fclose(binaryfile); free(binaries[slot]); goto build; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[slot], (const unsigned char **)binaries, &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithBinary)", status); fclose(binaryfile); free(binaries[slot]); goto build; } fclose(binaryfile); applog(LOG_DEBUG, "Loaded binary image %s", binaryfilename); goto built; } ///////////////////////////////////////////////////////////////// // Load CL file, build CL program object, create CL kernel object ///////////////////////////////////////////////////////////////// build: clState->program = clCreateProgramWithSource(clState->context, 1, (const char **)&source, sourceSize, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithSource)", status); return NULL; } /* create a cl program executable for all the devices specified */ char *CompilerOptions = calloc(1, 256); sprintf(CompilerOptions, "-D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize); applog(LOG_DEBUG, "Setting worksize to %d", (int)(clState->wsize)); if (clState->vwidth > 1) applog(LOG_DEBUG, "Patched source to suit %d vectors", clState->vwidth); if (clState->hasBitAlign) { strcat(CompilerOptions, " -D BITALIGN"); applog(LOG_DEBUG, "cl_amd_media_ops found, setting BITALIGN"); if (!clState->hasOpenCL12plus && (strstr(name, "Cedar") || strstr(name, "Redwood") || strstr(name, "Juniper") || strstr(name, "Cypress" ) || strstr(name, "Hemlock" ) || strstr(name, "Caicos" ) || strstr(name, "Turks" ) || strstr(name, "Barts" ) || strstr(name, "Cayman" ) || strstr(name, "Antilles" ) || strstr(name, "Wrestler" ) || strstr(name, "Zacate" ) || strstr(name, "WinterPark" ))) patchbfi = true; } else applog(LOG_DEBUG, "cl_amd_media_ops not found, will not set BITALIGN"); if (patchbfi) { strcat(CompilerOptions, " -D BFI_INT"); applog(LOG_DEBUG, "BFI_INT patch requiring device found, patched source with BFI_INT"); } else applog(LOG_DEBUG, "BFI_INT patch requiring device not found, will not BFI_INT patch"); if (clState->goffset) strcat(CompilerOptions, " -D GOFFSET"); if (!clState->hasOpenCL11plus) strcat(CompilerOptions, " -D OCL1"); applog(LOG_DEBUG, "CompilerOptions: %s", CompilerOptions); status = clBuildProgram(clState->program, 1, &devices[gpu], CompilerOptions , NULL, NULL); free(CompilerOptions); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Building Program (clBuildProgram)", status); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_ERR, "%s", log); return NULL; } prog_built = true; #ifdef __APPLE__ /* OSX OpenCL breaks reading off binaries with >1 GPU so always build * from source. */ goto built; #endif status = clGetProgramInfo(clState->program, CL_PROGRAM_NUM_DEVICES, sizeof(cl_uint), &cpnd, NULL); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info CL_PROGRAM_NUM_DEVICES. (clGetProgramInfo)", status); return NULL; } status = clGetProgramInfo(clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*cpnd, binary_sizes, NULL); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info CL_PROGRAM_BINARY_SIZES. (clGetProgramInfo)", status); return NULL; } /* The actual compiled binary ends up in a RANDOM slot! Grr, so we have * to iterate over all the binary slots and find where the real program * is. What the heck is this!? */ for (slot = 0; slot < cpnd; slot++) if (binary_sizes[slot]) break; /* copy over all of the generated binaries. */ applog(LOG_DEBUG, "Binary size for gpu %d found in binary slot %d: %d", gpu, slot, (int)(binary_sizes[slot])); if (!binary_sizes[slot]) { applog(LOG_ERR, "OpenCL compiler generated a zero sized binary, FAIL!"); return NULL; } binaries[slot] = calloc(sizeof(char) * binary_sizes[slot], 1); status = clGetProgramInfo(clState->program, CL_PROGRAM_BINARIES, sizeof(char *) * cpnd, binaries, NULL ); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info. CL_PROGRAM_BINARIES (clGetProgramInfo)", status); return NULL; } /* Patch the kernel if the hardware supports BFI_INT but it needs to * be hacked in */ if (patchbfi) { unsigned remaining = binary_sizes[slot]; char *w = binaries[slot]; unsigned int start, length; /* Find 2nd incidence of .text, and copy the program's * position and length at a fixed offset from that. Then go * back and find the 2nd incidence of \x7ELF (rewind by one * from ELF) and then patch the opcocdes */ if (!advance(&w, &remaining, ".text")) goto build; w++; remaining--; if (!advance(&w, &remaining, ".text")) { /* 32 bit builds only one ELF */ w--; remaining++; } memcpy(&start, w + 285, 4); memcpy(&length, w + 289, 4); w = binaries[slot]; remaining = binary_sizes[slot]; if (!advance(&w, &remaining, "ELF")) goto build; w++; remaining--; if (!advance(&w, &remaining, "ELF")) { /* 32 bit builds only one ELF */ w--; remaining++; } w--; remaining++; w += start; remaining -= start; applog(LOG_DEBUG, "At %p (%u rem. bytes), to begin patching", w, remaining); patch_opcodes(w, length); status = clReleaseProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Releasing program. (clReleaseProgram)", status); return NULL; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[slot], (const unsigned char **)&binaries[slot], &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithBinary)", status); return NULL; } /* Program needs to be rebuilt */ prog_built = false; } free(source); /* Save the binary to be loaded next time */ binaryfile = fopen(binaryfilename, "wb"); if (!binaryfile) { /* Not a fatal problem, just means we build it again next time */ applog(LOG_DEBUG, "Unable to create file %s", binaryfilename); } else { if (unlikely(fwrite(binaries[slot], 1, binary_sizes[slot], binaryfile) != binary_sizes[slot])) { applog(LOG_ERR, "Unable to fwrite to binaryfile"); return NULL; } fclose(binaryfile); } built: if (binaries[slot]) free(binaries[slot]); free(binaries); free(binary_sizes); applog(LOG_INFO, "Initialising kernel %s with%s bitalign, %d vectors and worksize %d", filename, clState->hasBitAlign ? "" : "out", clState->vwidth, (int)(clState->wsize)); if (!prog_built) { /* create a cl program executable for all the devices specified */ status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Building Program (clBuildProgram)", status); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_ERR, "%s", log); return NULL; } } /* get a kernel object handle for a kernel with the given name */ clState->kernel = clCreateKernel(clState->program, "search", &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Kernel from program. (clCreateKernel)", status); return NULL; } size_t ipt = (1024 / cgpu->lookup_gap + (1024 % cgpu->lookup_gap > 0)); size_t bufsize = 128 * ipt * cgpu->thread_concurrency; /* Use the max alloc value which has been rounded to a power of * 2 greater >= required amount earlier */ if (bufsize > cgpu->max_alloc) { applog(LOG_WARNING, "Maximum buffer memory device %d supports says %lu", gpu, (long unsigned int)(cgpu->max_alloc)); applog(LOG_WARNING, "Your scrypt settings come to %d", (int)bufsize); } applog(LOG_DEBUG, "Creating scrypt buffer sized %d", (int)bufsize); clState->padbufsize = bufsize; /* This buffer is weird and might work to some degree even if * the create buffer call has apparently failed, so check if we * get anything back before we call it a failure. */ clState->padbuffer8 = NULL; clState->padbuffer8 = clCreateBuffer(clState->context, CL_MEM_READ_WRITE, bufsize, NULL, &status); if (status != CL_SUCCESS && !clState->padbuffer8) { applog(LOG_ERR, "Error %d: clCreateBuffer (padbuffer8), decrease TC or increase LG", status); return NULL; } clState->CLbuffer0 = clCreateBuffer(clState->context, CL_MEM_READ_ONLY, 128, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: clCreateBuffer (CLbuffer0)", status); return NULL; } clState->outputBuffer = clCreateBuffer(clState->context, CL_MEM_WRITE_ONLY, BUFFERSIZE, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: clCreateBuffer (outputBuffer)", status); return NULL; } return clState; }
_clState *initCl(unsigned int gpu, char *name, size_t nameSize) { _clState *clState = calloc(1, sizeof(_clState)); bool patchbfi = false, prog_built = false; cl_platform_id platform = NULL; cl_platform_id* platforms; cl_device_id *devices; cl_uint numPlatforms; cl_uint numDevices; char pbuff[256]; cl_int status; status = clGetPlatformIDs(0, NULL, &numPlatforms); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Platforms. (clGetPlatformsIDs)"); return NULL; } platforms = (cl_platform_id *)alloca(numPlatforms*sizeof(cl_platform_id)); status = clGetPlatformIDs(numPlatforms, platforms, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Platform Ids. (clGetPlatformsIDs)"); return NULL; } if (opt_platform_id >= numPlatforms) { applog(LOG_ERR, "Specified platform that does not exist"); return NULL; } status = clGetPlatformInfo(platforms[opt_platform_id], CL_PLATFORM_VENDOR, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Platform Info. (clGetPlatformInfo)"); return NULL; } platform = platforms[opt_platform_id]; if (platform == NULL) { perror("NULL platform found!\n"); return NULL; } applog(LOG_INFO, "CL Platform vendor: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_NAME, sizeof(pbuff), pbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform name: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_VERSION, sizeof(pbuff), pbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform version: %s", pbuff); status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Device IDs (num)"); return NULL; } if (numDevices > 0 ) { devices = (cl_device_id *)malloc(numDevices*sizeof(cl_device_id)); /* Now, get the device list data */ status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, numDevices, devices, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Device IDs (list)"); return NULL; } applog(LOG_INFO, "List of devices:"); unsigned int i; for (i = 0; i < numDevices; i++) { status = clGetDeviceInfo(devices[i], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Device Info"); return NULL; } applog(LOG_INFO, "\t%i\t%s", i, pbuff); } if (gpu < numDevices) { status = clGetDeviceInfo(devices[gpu], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Getting Device Info"); return NULL; } applog(LOG_INFO, "Selected %i: %s", gpu, pbuff); strncpy(name, pbuff, nameSize); } else { applog(LOG_ERR, "Invalid GPU %i", gpu); return NULL; } } else return NULL; cl_context_properties cps[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platform, 0 }; clState->context = clCreateContextFromType(cps, CL_DEVICE_TYPE_GPU, NULL, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Creating Context. (clCreateContextFromType)"); return NULL; } /* Check for BFI INT support. Hopefully people don't mix devices with * and without it! */ char * extensions = malloc(1024); const char * camo = "cl_amd_media_ops"; char *find; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_EXTENSIONS, 1024, (void *)extensions, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Failed to clGetDeviceInfo when trying to get CL_DEVICE_EXTENSIONS"); return NULL; } find = strstr(extensions, camo); if (find) clState->hasBitAlign = true; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&clState->preferred_vwidth, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Failed to clGetDeviceInfo when trying to get CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT"); return NULL; } if (opt_debug) applog(LOG_DEBUG, "Preferred vector width reported %d", clState->preferred_vwidth); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void *)&clState->max_work_size, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_WORK_GROUP_SIZE"); return NULL; } if (opt_debug) applog(LOG_DEBUG, "Max work group size reported %d", clState->max_work_size); /* For some reason 2 vectors is still better even if the card says * otherwise, and many cards lie about their max so use 256 as max * unless explicitly set on the command line. 79x0 cards perform * better without vectors */ if (clState->preferred_vwidth > 1) { if (strstr(name, "Tahiti")) clState->preferred_vwidth = 1; else clState->preferred_vwidth = 2; } if (opt_vectors) clState->preferred_vwidth = opt_vectors; if (opt_worksize && opt_worksize <= clState->max_work_size) clState->work_size = opt_worksize; else clState->work_size = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->preferred_vwidth; /* Create binary filename based on parameters passed to opencl * compiler to ensure we only load a binary that matches what would * have otherwise created. The filename is: * name + kernelname +/i bitalign + v + vectors + w + work_size + sizeof(long) + .bin */ char binaryfilename[255]; char numbuf[10]; char filename[16]; if (chosen_kernel == KL_NONE) { if (!clState->hasBitAlign || strstr(name, "Tahiti")) chosen_kernel = KL_POCLBM; else chosen_kernel = KL_PHATK; } switch (chosen_kernel) { case KL_POCLBM: strcpy(filename, "poclbm120203.cl"); strcpy(binaryfilename, "poclbm120203"); break; case KL_NONE: /* Shouldn't happen */ case KL_PHATK: strcpy(filename, "phatk120203.cl"); strcpy(binaryfilename, "phatk120203"); break; } FILE *binaryfile; size_t *binary_sizes; char **binaries; int pl; char *source = file_contents(filename, &pl); size_t sourceSize[] = {(size_t)pl}; if (!source) return NULL; binary_sizes = (size_t *)malloc(sizeof(size_t)*numDevices); if (unlikely(!binary_sizes)) { applog(LOG_ERR, "Unable to malloc binary_sizes"); return NULL; } binaries = (char **)malloc(sizeof(char *)*numDevices); if (unlikely(!binaries)) { applog(LOG_ERR, "Unable to malloc binaries"); return NULL; } strcat(binaryfilename, name); if (clState->hasBitAlign) strcat(binaryfilename, "bitalign"); strcat(binaryfilename, "v"); sprintf(numbuf, "%d", clState->preferred_vwidth); strcat(binaryfilename, numbuf); strcat(binaryfilename, "w"); sprintf(numbuf, "%d", (int)clState->work_size); strcat(binaryfilename, numbuf); strcat(binaryfilename, "long"); sprintf(numbuf, "%d", (int)sizeof(long)); strcat(binaryfilename, numbuf); strcat(binaryfilename, ".bin"); binaryfile = fopen(binaryfilename, "rb"); if (!binaryfile) { if (opt_debug) applog(LOG_DEBUG, "No binary found, generating from source"); } else { struct stat binary_stat; if (unlikely(stat(binaryfilename, &binary_stat))) { if (opt_debug) applog(LOG_DEBUG, "Unable to stat binary, generating from source"); fclose(binaryfile); goto build; } if (!binary_stat.st_size) goto build; binary_sizes[gpu] = binary_stat.st_size; binaries[gpu] = (char *)malloc(binary_sizes[gpu]); if (unlikely(!binaries[gpu])) { applog(LOG_ERR, "Unable to malloc binaries"); fclose(binaryfile); return NULL; } if (fread(binaries[gpu], 1, binary_sizes[gpu], binaryfile) != binary_sizes[gpu]) { applog(LOG_ERR, "Unable to fread binaries[gpu]"); fclose(binaryfile); free(binaries[gpu]); goto build; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[gpu], (const unsigned char **)&binaries[gpu], &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithBinary)"); fclose(binaryfile); free(binaries[gpu]); goto build; } fclose(binaryfile); if (opt_debug) applog(LOG_DEBUG, "Loaded binary image %s", binaryfilename); /* We don't need to patch this already loaded image, but need to * set the flag for status later */ if (clState->hasBitAlign) patchbfi = true; free(binaries[gpu]); goto built; } ///////////////////////////////////////////////////////////////// // Load CL file, build CL program object, create CL kernel object ///////////////////////////////////////////////////////////////// build: clState->program = clCreateProgramWithSource(clState->context, 1, (const char **)&source, sourceSize, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithSource)"); return NULL; } clRetainProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Retaining Program (clRetainProgram)"); return NULL; } /* create a cl program executable for all the devices specified */ char *CompilerOptions = calloc(1, 256); sprintf(CompilerOptions, "-D WORKSIZE=%d -D VECTORS%d", (int)clState->work_size, clState->preferred_vwidth); if (opt_debug) applog(LOG_DEBUG, "Setting worksize to %d", clState->work_size); if (clState->preferred_vwidth > 1 && opt_debug) applog(LOG_DEBUG, "Patched source to suit %d vectors", clState->preferred_vwidth); if (clState->hasBitAlign) { strcat(CompilerOptions, " -D BITALIGN"); if (opt_debug) applog(LOG_DEBUG, "cl_amd_media_ops found, setting BITALIGN"); if (strstr(name, "Cedar") || strstr(name, "Redwood") || strstr(name, "Juniper") || strstr(name, "Cypress" ) || strstr(name, "Hemlock" ) || strstr(name, "Caicos" ) || strstr(name, "Turks" ) || strstr(name, "Barts" ) || strstr(name, "Cayman" ) || strstr(name, "Antilles" ) || strstr(name, "Wrestler" ) || strstr(name, "Zacate" ) || strstr(name, "WinterPark" ) || strstr(name, "BeaverCreek" )) patchbfi = true; } else if (opt_debug) applog(LOG_DEBUG, "cl_amd_media_ops not found, will not set BITALIGN"); if (patchbfi) { strcat(CompilerOptions, " -D BFI_INT"); if (opt_debug) applog(LOG_DEBUG, "BFI_INT patch requiring device found, patched source with BFI_INT"); } else if (opt_debug) applog(LOG_DEBUG, "BFI_INT patch requiring device not found, will not BFI_INT patch"); if (opt_debug) applog(LOG_DEBUG, "CompilerOptions: %s", CompilerOptions); status = clBuildProgram(clState->program, 1, &devices[gpu], CompilerOptions , NULL, NULL); free(CompilerOptions); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Building Program (clBuildProgram)"); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_INFO, "%s", log); return NULL; } prog_built = true; status = clGetProgramInfo( clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*numDevices, binary_sizes, NULL ); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error: Getting program info CL_PROGRAM_BINARY_SIZES. (clGetPlatformInfo)"); return NULL; } /* copy over all of the generated binaries. */ if (opt_debug) applog(LOG_DEBUG, "binary size %d : %d", gpu, binary_sizes[gpu]); if (!binary_sizes[gpu]) { applog(LOG_ERR, "OpenCL compiler generated a zero sized binary, may need to reboot!"); return NULL; } binaries[gpu] = (char *)malloc( sizeof(char)*binary_sizes[gpu]); status = clGetProgramInfo( clState->program, CL_PROGRAM_BINARIES, sizeof(char *)*numDevices, binaries, NULL ); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error: Getting program info. (clGetPlatformInfo)"); return NULL; } /* Patch the kernel if the hardware supports BFI_INT but it needs to * be hacked in */ if (patchbfi) { unsigned remaining = binary_sizes[gpu]; char *w = binaries[gpu]; unsigned int start, length; /* Find 2nd incidence of .text, and copy the program's * position and length at a fixed offset from that. Then go * back and find the 2nd incidence of \x7ELF (rewind by one * from ELF) and then patch the opcocdes */ if (!advance(&w, &remaining, ".text")) {patchbfi = 0; goto build;} w++; remaining--; if (!advance(&w, &remaining, ".text")) { /* 32 bit builds only one ELF */ w--; remaining++; } memcpy(&start, w + 285, 4); memcpy(&length, w + 289, 4); w = binaries[gpu]; remaining = binary_sizes[gpu]; if (!advance(&w, &remaining, "ELF")) {patchbfi = 0; goto build;} w++; remaining--; if (!advance(&w, &remaining, "ELF")) { /* 32 bit builds only one ELF */ w--; remaining++; } w--; remaining++; w += start; remaining -= start; if (opt_debug) applog(LOG_DEBUG, "At %p (%u rem. bytes), to begin patching", w, remaining); patch_opcodes(w, length); status = clReleaseProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Releasing program. (clReleaseProgram)"); return NULL; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[gpu], (const unsigned char **)&binaries[gpu], &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithBinary)"); return NULL; } clRetainProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Retaining Program (clRetainProgram)"); return NULL; } /* Program needs to be rebuilt */ prog_built = false; } free(source); /* Save the binary to be loaded next time */ binaryfile = fopen(binaryfilename, "wb"); if (!binaryfile) { /* Not a fatal problem, just means we build it again next time */ if (opt_debug) applog(LOG_DEBUG, "Unable to create file %s", binaryfilename); } else { if (unlikely(fwrite(binaries[gpu], 1, binary_sizes[gpu], binaryfile) != binary_sizes[gpu])) { applog(LOG_ERR, "Unable to fwrite to binaryfile"); return NULL; } fclose(binaryfile); } if (binaries[gpu]) free(binaries[gpu]); built: free(binaries); free(binary_sizes); applog(LOG_INFO, "Initialising kernel %s with%s BFI_INT, %d vectors and worksize %d", filename, patchbfi ? "" : "out", clState->preferred_vwidth, clState->work_size); if (!prog_built) { /* create a cl program executable for all the devices specified */ status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Building Program (clBuildProgram)"); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_INFO, "%s", log); return NULL; } clRetainProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Retaining Program (clRetainProgram)"); return NULL; } } /* get a kernel object handle for a kernel with the given name */ clState->kernel = clCreateKernel(clState->program, "search", &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: Creating Kernel from program. (clCreateKernel)"); return NULL; } ///////////////////////////////////////////////////////////////// // Create an OpenCL command queue ///////////////////////////////////////////////////////////////// clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &status); if (status != CL_SUCCESS) /* Try again without OOE enable */ clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], 0 , &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Creating Command Queue. (clCreateCommandQueue)"); return NULL; } clState->outputBuffer = clCreateBuffer(clState->context, CL_MEM_READ_WRITE, BUFFERSIZE, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error: clCreateBuffer (outputBuffer)"); return NULL; } return clState; }
_clState *initCl(unsigned int gpu, char *name, size_t nameSize) { _clState *clState = calloc(1, sizeof(_clState)); bool patchbfi = false, prog_built = false; struct cgpu_info *cgpu = &gpus[gpu]; cl_platform_id platform = NULL; char pbuff[256], vbuff[255]; cl_platform_id* platforms; cl_uint preferred_vwidth; cl_device_id *devices; cl_uint numPlatforms; cl_uint numDevices; cl_int status; status = clGetPlatformIDs(0, NULL, &numPlatforms); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platforms. (clGetPlatformsIDs)", status); return NULL; } platforms = (cl_platform_id *)alloca(numPlatforms*sizeof(cl_platform_id)); status = clGetPlatformIDs(numPlatforms, platforms, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platform Ids. (clGetPlatformsIDs)", status); return NULL; } if (opt_platform_id >= (int)numPlatforms) { applog(LOG_ERR, "Specified platform that does not exist"); return NULL; } status = clGetPlatformInfo(platforms[opt_platform_id], CL_PLATFORM_VENDOR, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Platform Info. (clGetPlatformInfo)", status); return NULL; } platform = platforms[opt_platform_id]; if (platform == NULL) { perror("NULL platform found!\n"); return NULL; } applog(LOG_INFO, "CL Platform vendor: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_NAME, sizeof(pbuff), pbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform name: %s", pbuff); status = clGetPlatformInfo(platform, CL_PLATFORM_VERSION, sizeof(vbuff), vbuff, NULL); if (status == CL_SUCCESS) applog(LOG_INFO, "CL Platform version: %s", vbuff); status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device IDs (num)", status); return NULL; } if (numDevices > 0 ) { devices = (cl_device_id *)malloc(numDevices*sizeof(cl_device_id)); /* Now, get the device list data */ status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, numDevices, devices, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device IDs (list)", status); return NULL; } applog(LOG_INFO, "List of devices:"); unsigned int i; for (i = 0; i < numDevices; i++) { status = clGetDeviceInfo(devices[i], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device Info", status); return NULL; } applog(LOG_INFO, "\t%i\t%s", i, pbuff); } if (gpu < numDevices) { status = clGetDeviceInfo(devices[gpu], CL_DEVICE_NAME, sizeof(pbuff), pbuff, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Getting Device Info", status); return NULL; } applog(LOG_INFO, "Selected %i: %s", gpu, pbuff); strncpy(name, pbuff, nameSize); } else { applog(LOG_ERR, "Invalid GPU %i", gpu); return NULL; } } else return NULL; cl_context_properties cps[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platform, 0 }; clState->context = clCreateContextFromType(cps, CL_DEVICE_TYPE_GPU, NULL, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Context. (clCreateContextFromType)", status); return NULL; } ///////////////////////////////////////////////////////////////// // Create an OpenCL command queue ///////////////////////////////////////////////////////////////// clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &status); if (status != CL_SUCCESS) /* Try again without OOE enable */ clState->commandQueue = clCreateCommandQueue(clState->context, devices[gpu], 0 , &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Command Queue. (clCreateCommandQueue)", status); return NULL; } /* Check for BFI INT support. Hopefully people don't mix devices with * and without it! */ char * extensions = malloc(1024); const char * camo = "cl_amd_media_ops"; char *find; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_EXTENSIONS, 1024, (void *)extensions, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_EXTENSIONS", status); return NULL; } find = strstr(extensions, camo); if (find) clState->hasBitAlign = true; /* Check for OpenCL >= 1.0 support, needed for global offset parameter usage. */ char * devoclver = malloc(1024); const char * ocl10 = "OpenCL 1.0"; const char * ocl11 = "OpenCL 1.1"; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_VERSION, 1024, (void *)devoclver, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_VERSION", status); return NULL; } find = strstr(devoclver, ocl10); if (!find) { clState->hasOpenCL11plus = true; find = strstr(devoclver, ocl11); if (!find) clState->hasOpenCL12plus = true; } status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&preferred_vwidth, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT", status); return NULL; } applog(LOG_DEBUG, "Preferred vector width reported %d", preferred_vwidth); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void *)&clState->max_work_size, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_WORK_GROUP_SIZE", status); return NULL; } applog(LOG_DEBUG, "Max work group size reported %d", (int)(clState->max_work_size)); size_t compute_units = 0; status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(size_t), (void *)&compute_units, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_COMPUTE_UNITS", status); return NULL; } // AMD architechture got 64 compute shaders per compute unit. // Source: http://www.amd.com/us/Documents/GCN_Architecture_whitepaper.pdf clState->compute_shaders = compute_units * 64; applog(LOG_DEBUG, "Max shaders calculated %d", (int)(clState->compute_shaders)); status = clGetDeviceInfo(devices[gpu], CL_DEVICE_MAX_MEM_ALLOC_SIZE , sizeof(cl_ulong), (void *)&cgpu->max_alloc, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Failed to clGetDeviceInfo when trying to get CL_DEVICE_MAX_MEM_ALLOC_SIZE", status); return NULL; } applog(LOG_DEBUG, "Max mem alloc size is %lu", (long unsigned int)(cgpu->max_alloc)); /* Create binary filename based on parameters passed to opencl * compiler to ensure we only load a binary that matches what would * have otherwise created. The filename is: * name + kernelname +/- g(offset) + v + vectors + w + work_size + l + sizeof(long) + .bin * For scrypt the filename is: * name + kernelname + g + lg + lookup_gap + tc + thread_concurrency + w + work_size + l + sizeof(long) + .bin */ char binaryfilename[255]; char filename[255]; char numbuf[16]; if (cgpu->kernel == KL_NONE) { if (opt_scrypt) { if (opt_scrypt_chacha) { applog(LOG_INFO, "Selecting scrypt-chacha kernel"); clState->chosen_kernel = KL_SCRYPT_CHACHA; } else if (opt_n_scrypt) { applog(LOG_INFO, "Selecting N-scrypt kernel"); clState->chosen_kernel = KL_N_SCRYPT; } else { applog(LOG_INFO, "Selecting standard scrypt kernel"); clState->chosen_kernel = KL_SCRYPT; } } else if (!strstr(name, "Tahiti") && /* Detect all 2.6 SDKs not with Tahiti and use diablo kernel */ (strstr(vbuff, "844.4") || // Linux 64 bit ATI 2.6 SDK strstr(vbuff, "851.4") || // Windows 64 bit "" strstr(vbuff, "831.4") || strstr(vbuff, "898.1") || // 12.2 driver SDK strstr(vbuff, "923.1") || // 12.4 strstr(vbuff, "938.2") || // SDK 2.7 strstr(vbuff, "1113.2"))) {// SDK 2.8 applog(LOG_INFO, "Selecting diablo kernel"); clState->chosen_kernel = KL_DIABLO; /* Detect all 7970s, older ATI and NVIDIA and use poclbm */ } else if (strstr(name, "Tahiti") || !clState->hasBitAlign) { applog(LOG_INFO, "Selecting poclbm kernel"); clState->chosen_kernel = KL_POCLBM; /* Use phatk for the rest R5xxx R6xxx */ } else { applog(LOG_INFO, "Selecting phatk kernel"); clState->chosen_kernel = KL_PHATK; } cgpu->kernel = clState->chosen_kernel; } else { clState->chosen_kernel = cgpu->kernel; if (clState->chosen_kernel == KL_PHATK && (strstr(vbuff, "844.4") || strstr(vbuff, "851.4") || strstr(vbuff, "831.4") || strstr(vbuff, "898.1") || strstr(vbuff, "923.1") || strstr(vbuff, "938.2") || strstr(vbuff, "1113.2"))) { applog(LOG_WARNING, "WARNING: You have selected the phatk kernel."); applog(LOG_WARNING, "You are running SDK 2.6+ which performs poorly with this kernel."); applog(LOG_WARNING, "Downgrade your SDK and delete any .bin files before starting again."); applog(LOG_WARNING, "Or allow cgminer to automatically choose a more suitable kernel."); } } /* For some reason 2 vectors is still better even if the card says * otherwise, and many cards lie about their max so use 256 as max * unless explicitly set on the command line. Tahiti prefers 1 */ if (strstr(name, "Tahiti")) preferred_vwidth = 1; else if (preferred_vwidth > 2) preferred_vwidth = 2; switch (clState->chosen_kernel) { case KL_POCLBM: strcpy(filename, POCLBM_KERNNAME".cl"); strcpy(binaryfilename, POCLBM_KERNNAME); break; case KL_PHATK: strcpy(filename, PHATK_KERNNAME".cl"); strcpy(binaryfilename, PHATK_KERNNAME); break; case KL_DIAKGCN: strcpy(filename, DIAKGCN_KERNNAME".cl"); strcpy(binaryfilename, DIAKGCN_KERNNAME); break; case KL_SCRYPT: strcpy(filename, SCRYPT_KERNNAME".cl"); strcpy(binaryfilename, SCRYPT_KERNNAME); /* Scrypt only supports vector 1 */ cgpu->vwidth = 1; break; case KL_N_SCRYPT: strcpy(filename, N_SCRYPT_KERNNAME".cl"); strcpy(binaryfilename, N_SCRYPT_KERNNAME); /* Scrypt only supports vector 1 */ cgpu->vwidth = 1; break; case KL_SCRYPT_CHACHA: strcpy(filename, SCRYPT_CHACHA_KERNNAME".cl"); strcpy(binaryfilename, SCRYPT_CHACHA_KERNNAME); /* Scrypt only supports vector 1 */ cgpu->vwidth = 1; break; case KL_NONE: /* Shouldn't happen */ case KL_DIABLO: strcpy(filename, DIABLO_KERNNAME".cl"); strcpy(binaryfilename, DIABLO_KERNNAME); break; } if (cgpu->vwidth) clState->vwidth = cgpu->vwidth; else { clState->vwidth = preferred_vwidth; cgpu->vwidth = preferred_vwidth; } if (((clState->chosen_kernel == KL_POCLBM || clState->chosen_kernel == KL_DIABLO || clState->chosen_kernel == KL_DIAKGCN) && clState->vwidth == 1 && clState->hasOpenCL11plus) || opt_scrypt) clState->goffset = true; if (cgpu->work_size && cgpu->work_size <= clState->max_work_size) clState->wsize = cgpu->work_size; else if (opt_scrypt) clState->wsize = 256; else if (strstr(name, "Tahiti")) clState->wsize = 64; else clState->wsize = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->vwidth; cgpu->work_size = clState->wsize; #ifdef USE_SCRYPT if (opt_scrypt) { if (!cgpu->opt_lg) { applog(LOG_NOTICE, "GPU %d: selecting lookup gap of 4", gpu); cgpu->lookup_gap = 4; } else cgpu->lookup_gap = cgpu->opt_lg; unsigned int bsize = opt_n_scrypt ? 2048 : 1024; size_t ipt = (bsize / cgpu->lookup_gap + (bsize % cgpu->lookup_gap > 0)); // if we do not have TC and we do not have BS, then calculate some conservative numbers if ((!cgpu->opt_tc) && (!cgpu->buffer_size)) { unsigned int base_alloc; // default to 88% of the available memory and find the closest MB value divisible by 8 base_alloc = (int)(cgpu->max_alloc * 88 / 100 / 1024 / 1024 / 8) * 8 * 1024 * 1024 / cgpu->threads; // base_alloc is now the number of bytes to allocate. // 2 threads of 336 MB did not fit into dedicated VRAM while 1 thread of 772MB did. 334 MB each did // to be safe, reduce by 2MB per thread beyond the first base_alloc -= (cgpu->threads - 1) * 2 * 1024 * 1024; cgpu->thread_concurrency = base_alloc / 128 / ipt; cgpu->buffer_size = base_alloc / 1024 / 1024; applog(LOG_DEBUG,"88%% Max Allocation: %u",base_alloc); applog(LOG_NOTICE, "GPU %d: selecting buffer_size of %zu", gpu, cgpu->buffer_size); } else cgpu->thread_concurrency = cgpu->opt_tc; if (cgpu->buffer_size) { // use the buffer-size to overwrite the thread-concurrency cgpu->thread_concurrency = (int)((cgpu->buffer_size * 1024 * 1024) / ipt / 128); applog(LOG_DEBUG, "GPU %d: setting thread_concurrency to %d based on buffer size %d and lookup gap %d", gpu, (int)(cgpu->thread_concurrency),(int)(cgpu->buffer_size),(int)(cgpu->lookup_gap)); } } #endif FILE *binaryfile; size_t *binary_sizes; char **binaries; int pl; char *source = file_contents(filename, &pl); size_t sourceSize[] = {(size_t)pl}; cl_uint slot, cpnd; slot = cpnd = 0; if (!source) return NULL; binary_sizes = calloc(sizeof(size_t) * MAX_GPUDEVICES * 4, 1); if (unlikely(!binary_sizes)) { applog(LOG_ERR, "Unable to calloc binary_sizes"); return NULL; } binaries = calloc(sizeof(char *) * MAX_GPUDEVICES * 4, 1); if (unlikely(!binaries)) { applog(LOG_ERR, "Unable to calloc binaries"); return NULL; } strcat(binaryfilename, name); if (clState->goffset) strcat(binaryfilename, "g"); if (opt_scrypt) { #ifdef USE_SCRYPT sprintf(numbuf, "lg%utc%u", cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency); strcat(binaryfilename, numbuf); #endif } else { sprintf(numbuf, "v%d", clState->vwidth); strcat(binaryfilename, numbuf); } sprintf(numbuf, "w%d", (int)clState->wsize); strcat(binaryfilename, numbuf); sprintf(numbuf, "l%d", (int)sizeof(long)); strcat(binaryfilename, numbuf); strcat(binaryfilename, ".bin"); binaryfile = fopen(binaryfilename, "rb"); if (!binaryfile) { applog(LOG_DEBUG, "No binary found, generating from source"); } else { struct stat binary_stat; if (unlikely(stat(binaryfilename, &binary_stat))) { applog(LOG_DEBUG, "Unable to stat binary, generating from source"); fclose(binaryfile); goto build; } if (!binary_stat.st_size) goto build; binary_sizes[slot] = binary_stat.st_size; binaries[slot] = (char *)calloc(binary_sizes[slot], 1); if (unlikely(!binaries[slot])) { applog(LOG_ERR, "Unable to calloc binaries"); fclose(binaryfile); return NULL; } if (fread(binaries[slot], 1, binary_sizes[slot], binaryfile) != binary_sizes[slot]) { applog(LOG_ERR, "Unable to fread binaries"); fclose(binaryfile); free(binaries[slot]); goto build; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[slot], (const unsigned char **)binaries, &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithBinary)", status); fclose(binaryfile); free(binaries[slot]); goto build; } fclose(binaryfile); applog(LOG_DEBUG, "Loaded binary image %s", binaryfilename); goto built; } ///////////////////////////////////////////////////////////////// // Load CL file, build CL program object, create CL kernel object ///////////////////////////////////////////////////////////////// build: clState->program = clCreateProgramWithSource(clState->context, 1, (const char **)&source, sourceSize, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithSource)", status); return NULL; } /* create a cl program executable for all the devices specified */ char *CompilerOptions = calloc(1, 256); #ifdef USE_SCRYPT if (opt_scrypt) { sprintf(CompilerOptions, "-D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize); } else #endif { sprintf(CompilerOptions, "-D WORKSIZE=%d -D VECTORS%d -D WORKVEC=%d", (int)clState->wsize, clState->vwidth, (int)clState->wsize * clState->vwidth); } applog(LOG_DEBUG, "Setting worksize to %d", (int)(clState->wsize)); if (clState->vwidth > 1) applog(LOG_DEBUG, "Patched source to suit %d vectors", clState->vwidth); if (clState->hasBitAlign) { strcat(CompilerOptions, " -D BITALIGN"); applog(LOG_DEBUG, "cl_amd_media_ops found, setting BITALIGN"); if (!clState->hasOpenCL12plus && (strstr(name, "Cedar") || strstr(name, "Redwood") || strstr(name, "Juniper") || strstr(name, "Cypress" ) || strstr(name, "Hemlock" ) || strstr(name, "Caicos" ) || strstr(name, "Turks" ) || strstr(name, "Barts" ) || strstr(name, "Cayman" ) || strstr(name, "Antilles" ) || strstr(name, "Wrestler" ) || strstr(name, "Zacate" ) || strstr(name, "WinterPark" ))) patchbfi = true; } else applog(LOG_DEBUG, "cl_amd_media_ops not found, will not set BITALIGN"); if (patchbfi) { strcat(CompilerOptions, " -D BFI_INT"); applog(LOG_DEBUG, "BFI_INT patch requiring device found, patched source with BFI_INT"); } else applog(LOG_DEBUG, "BFI_INT patch requiring device not found, will not BFI_INT patch"); if (clState->goffset) strcat(CompilerOptions, " -D GOFFSET"); if (!clState->hasOpenCL11plus) strcat(CompilerOptions, " -D OCL1"); applog(LOG_DEBUG, "CompilerOptions: %s", CompilerOptions); status = clBuildProgram(clState->program, 1, &devices[gpu], CompilerOptions , NULL, NULL); free(CompilerOptions); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Building Program (clBuildProgram)", status); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_ERR, "%s", log); return NULL; } else { applog(LOG_DEBUG, "Success: Building Program (clBuildProgram)"); } prog_built = true; #ifdef __APPLE__ /* OSX OpenCL breaks reading off binaries with >1 GPU so always build * from source. */ goto built; #endif status = clGetProgramInfo(clState->program, CL_PROGRAM_NUM_DEVICES, sizeof(cl_uint), &cpnd, NULL); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info CL_PROGRAM_NUM_DEVICES. (clGetProgramInfo)", status); return NULL; } status = clGetProgramInfo(clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*cpnd, binary_sizes, NULL); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info CL_PROGRAM_BINARY_SIZES. (clGetProgramInfo)", status); return NULL; } /* The actual compiled binary ends up in a RANDOM slot! Grr, so we have * to iterate over all the binary slots and find where the real program * is. What the heck is this!? */ for (slot = 0; slot < cpnd; slot++) if (binary_sizes[slot]) break; /* copy over all of the generated binaries. */ applog(LOG_DEBUG, "Binary size for gpu %d found in binary slot %d: %d", gpu, slot, (int)(binary_sizes[slot])); if (!binary_sizes[slot]) { applog(LOG_ERR, "OpenCL compiler generated a zero sized binary, FAIL!"); return NULL; } binaries[slot] = calloc(sizeof(char) * binary_sizes[slot], 1); status = clGetProgramInfo(clState->program, CL_PROGRAM_BINARIES, sizeof(char *) * cpnd, binaries, NULL ); if (unlikely(status != CL_SUCCESS)) { applog(LOG_ERR, "Error %d: Getting program info. CL_PROGRAM_BINARIES (clGetProgramInfo)", status); return NULL; } /* Patch the kernel if the hardware supports BFI_INT but it needs to * be hacked in */ if (patchbfi) { unsigned remaining = binary_sizes[slot]; char *w = binaries[slot]; unsigned int start, length; /* Find 2nd incidence of .text, and copy the program's * position and length at a fixed offset from that. Then go * back and find the 2nd incidence of \x7ELF (rewind by one * from ELF) and then patch the opcocdes */ if (!advance(&w, &remaining, ".text")) goto build; w++; remaining--; if (!advance(&w, &remaining, ".text")) { /* 32 bit builds only one ELF */ w--; remaining++; } memcpy(&start, w + 285, 4); memcpy(&length, w + 289, 4); w = binaries[slot]; remaining = binary_sizes[slot]; if (!advance(&w, &remaining, "ELF")) goto build; w++; remaining--; if (!advance(&w, &remaining, "ELF")) { /* 32 bit builds only one ELF */ w--; remaining++; } w--; remaining++; w += start; remaining -= start; applog(LOG_DEBUG, "At %p (%u rem. bytes), to begin patching", w, remaining); patch_opcodes(w, length); status = clReleaseProgram(clState->program); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Releasing program. (clReleaseProgram)", status); return NULL; } clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[slot], (const unsigned char **)&binaries[slot], &status, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Loading Binary into cl_program (clCreateProgramWithBinary)", status); return NULL; } /* Program needs to be rebuilt */ prog_built = false; } free(source); /* Save the binary to be loaded next time */ binaryfile = fopen(binaryfilename, "wb"); if (!binaryfile) { /* Not a fatal problem, just means we build it again next time */ applog(LOG_DEBUG, "Unable to create file %s", binaryfilename); } else { if (unlikely(fwrite(binaries[slot], 1, binary_sizes[slot], binaryfile) != binary_sizes[slot])) { applog(LOG_ERR, "Unable to fwrite to binaryfile"); return NULL; } fclose(binaryfile); } built: if (binaries[slot]) free(binaries[slot]); free(binaries); free(binary_sizes); applog(LOG_INFO, "Initialising kernel %s with%s bitalign, %d vectors and worksize %d", filename, clState->hasBitAlign ? "" : "out", clState->vwidth, (int)(clState->wsize)); if (!prog_built) { /* create a cl program executable for all the devices specified */ status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Building Program (clBuildProgram)", status); size_t logSize; status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize); char *log = malloc(logSize); status = clGetProgramBuildInfo(clState->program, devices[gpu], CL_PROGRAM_BUILD_LOG, logSize, log, NULL); applog(LOG_ERR, "%s", log); return NULL; } } /* get a kernel object handle for a kernel with the given name */ clState->kernel = clCreateKernel(clState->program, "search", &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: Creating Kernel from program. (clCreateKernel)", status); return NULL; } #ifdef USE_SCRYPT if (opt_scrypt) { unsigned int bsize = opt_n_scrypt ? 2048 : 1024; size_t ipt = (bsize / cgpu->lookup_gap + (bsize % cgpu->lookup_gap > 0)); size_t bufsize = 128 * ipt * cgpu->thread_concurrency; if (!cgpu->buffer_size) { applog(LOG_NOTICE, "GPU %d: bufsize for thread @ %dMB based on TC of %zu", gpu, (int)(bufsize/1048576),cgpu->thread_concurrency); } else { applog(LOG_NOTICE, "GPU %d: bufsize for thread @ %dMB based on buffer-size", gpu, (int)(cgpu->buffer_size)); bufsize = (size_t)(cgpu->buffer_size)*(1048576); } /* Use the max alloc value which has been rounded to a power of * 2 greater >= required amount earlier */ if (bufsize > cgpu->max_alloc) { applog(LOG_WARNING, "Maximum buffer memory device %d supports says %lu", gpu, (long unsigned int)(cgpu->max_alloc)); applog(LOG_WARNING, "Your scrypt settings come to %d", (int)bufsize); } applog(LOG_INFO, "Creating scrypt buffer sized %d", (int)bufsize); clState->padbufsize = bufsize; /* This buffer is weird and might work to some degree even if * the create buffer call has apparently failed, so check if we * get anything back before we call it a failure. */ clState->padbuffer8 = NULL; clState->padbuffer8 = clCreateBuffer(clState->context, CL_MEM_READ_WRITE, bufsize, NULL, &status); if (status != CL_SUCCESS && !clState->padbuffer8) { applog(LOG_ERR, "Error %d: clCreateBuffer (padbuffer8), decrease TC or increase LG", status); return NULL; } clState->CLbuffer0 = clCreateBuffer(clState->context, CL_MEM_READ_ONLY, 128, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: clCreateBuffer (CLbuffer0)", status); return NULL; } clState->outputBuffer = clCreateBuffer(clState->context, CL_MEM_WRITE_ONLY, SCRYPT_BUFFERSIZE, NULL, &status); } else #endif clState->outputBuffer = clCreateBuffer(clState->context, CL_MEM_WRITE_ONLY, BUFFERSIZE, NULL, &status); if (status != CL_SUCCESS) { applog(LOG_ERR, "Error %d: clCreateBuffer (outputBuffer)", status); return NULL; } return clState; }
shader_manager::resource shader_manager::load (const std::string& location) { return std::make_shared<std::string>(file_contents(resource_file(res_shader, location))); }
{ prepare_logdir(); std::string filename = "logs/simple_log.txt"; auto logger = spdlog::create<spdlog::sinks::simple_file_sink_mt>("logger", filename, true); logger->set_pattern("%v"); #if !defined(SPDLOG_FMT_PRINTF) logger->info("Test message {} {}", 1); logger->info("Test message {}", 2); #else logger->info("Test message %d %d", 1); logger->info("Test message %d", 2); #endif logger->flush(); REQUIRE(file_contents(filename) == std::string("Test message 2\n")); REQUIRE(count_lines(filename) == 1); } struct custom_ex { }; TEST_CASE("custom_error_handler", "[errors]]") { prepare_logdir(); std::string filename = "logs/simple_log.txt"; auto logger = spdlog::create<spdlog::sinks::simple_file_sink_mt>("logger", filename, true); logger->flush_on(spdlog::level::info); logger->set_error_handler([=](const std::string &msg) { throw custom_ex(); }); logger->info("Good message #1"); #if !defined(SPDLOG_FMT_PRINTF)