boinc/sched/plan_class_spec.cpp

733 lines
24 KiB
C++

// This file is part of BOINC.
// http://boinc.berkeley.edu
// Copyright (C) 2012 University of California
//
// BOINC is free software; you can redistribute it and/or modify it
// under the terms of the GNU Lesser General Public License
// as published by the Free Software Foundation,
// either version 3 of the License, or (at your option) any later version.
//
// BOINC is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
// See the GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with BOINC. If not, see <http://www.gnu.org/licenses/>.
#include <cmath>
#include "util.h"
#include "coproc.h"
#include "sched_config.h"
#include "sched_customize.h"
#include "plan_class_spec.h"
using std::string;
int PLAN_CLASS_SPECS::parse_file(const char* path) {
#ifndef _USING_FCGI_
FILE* f = fopen(path, "r");
#else
FCGI_FILE *f = FCGI::fopen(path, "r");
#endif
if (!f) return ERR_FOPEN;
int retval = parse_specs(f);
fclose(f);
return retval;
}
bool PLAN_CLASS_SPEC::check(SCHEDULER_REQUEST& sreq, HOST_USAGE& hu) {
COPROC* cpp = NULL;
bool can_use_multicore = true;
hu.sequential_app(sreq.host.p_fpops);
// CPU features
//
// older clients report CPU features in p_model,
// within square brackets
//
// the requested features are surrounded by spaces,
// so we can look for them with strstr()
//
if (!cpu_features.empty()) {
char buf[8192], buf2[512];
sprintf(buf, " %s ", sreq.host.p_features);
char* p = strrchr(sreq.host.p_model, '[');
if (p) {
sprintf(buf2, " %s", p+1);
p = strchr(buf2, ']');
if (p) {
*p = 0;
}
strcat(buf2, " ");
strcat(buf, buf2);
}
downcase_string(buf);
for (unsigned int i=0; i<cpu_features.size(); i++) {
if (!strstr(buf, cpu_features[i].c_str())) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CPU lacks feature '%s' (got '%s')\n",
cpu_features[i].c_str(), sreq.host.p_features
);
}
return false;
}
}
}
// min NCPUS
//
if (min_ncpus && g_wreq->effective_ncpus < min_ncpus) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: not enough CPUs: %d < %f\n",
g_wreq->effective_ncpus, min_ncpus
);
}
return false;
}
// OS version
//
if (have_os_regex && regexec(&(os_regex), sreq.host.os_version, 0, NULL, 0)) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: OS version '%s' didn't match regexp\n",
sreq.host.os_version
);
}
return false;
}
if (virtualbox) {
// host must run 7.0+ client
//
if (sreq.core_client_major_version < 7) {
add_no_work_message("BOINC client 7.0+ required for Virtualbox jobs");
return false;
}
// host must have VirtualBox 3.2 or later
//
if (strlen(sreq.host.virtualbox_version) == 0) {
add_no_work_message("VirtualBox is not installed");
return false;
}
int n, maj, min, rel;
n = sscanf(sreq.host.virtualbox_version, "%d.%d.%d", &maj, &min, &rel);
if (n != 3) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: can't parse vbox version\n"
);
}
return false;
}
int v = maj*10000 + min*100 + rel;
if (min_vbox_version && v < min_vbox_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: vbox version too low: %d < %d\n",
v, min_vbox_version
);
}
return false;
}
if (max_vbox_version && v > max_vbox_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: vbox version too high: %d > %d\n",
v, max_vbox_version
);
}
return false;
}
// host must have VM acceleration in order to run multi-core jobs
//
if (max_threads > 1) {
if ((!strstr(sreq.host.p_features, "vmx") && !strstr(sreq.host.p_features, "svm"))
|| sreq.host.p_vm_extensions_disabled
) {
can_use_multicore = false;
}
}
// only send the version for host's primary platform.
// A Win64 host can't run a 32-bit VM app:
// it will look in the 32-bit half of the registry and fail
//
PLATFORM* p = g_request->platforms.list[0];
if (is_64b_platform(p->name)) {
if (!is64bit) return false;
} else {
if (is64bit) return false;
}
}
// project-specific preference
//
if (have_project_prefs_regex && strlen(project_prefs_tag)) {
char tag[256], value[256];
char buf[65536];
extract_venue(g_reply->user.project_prefs, g_reply->host.venue, buf);
sprintf(tag,"<%s>",project_prefs_tag);
bool p = parse_str(buf, tag, value, sizeof(value));
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: parsed project prefs setting '%s' : %s\n",
project_prefs_tag, p?"true":"false"
);
}
if (regexec(&(project_prefs_regex), value, 0, NULL, 0)) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: project prefs setting '%s' prevents using plan class.\n",
project_prefs_tag
);
}
return false;
}
}
double gpu_ram = 0;
int driver_version = 0;
double gpu_utilization = 1.0;
// user defined gpu_utilization
//
if (strlen(gpu_utilization_tag)) {
char tag[256];
char buf[65536];
double v = 0;
extract_venue(g_reply->user.project_prefs, g_reply->host.venue, buf);
sprintf(tag,"<%s>",gpu_utilization_tag);
bool p = parse_double(buf, tag, v);
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: parsed project prefs setting '%s' : %s : %f\n",
gpu_utilization_tag, p?"true":"false", v
);
}
if (v) {
gpu_utilization = v;
}
}
// AMD
//
if (!strcmp(gpu_type, "amd") || !strcmp(gpu_type, "ati")) {
COPROC_ATI& cp = sreq.coprocs.ati;
cpp = &cp;
if (!cp.count) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: No AMD GPUs found\n"
);
}
return false;
}
if (min_gpu_ram_mb) {
gpu_requirements[PROC_TYPE_AMD_GPU].update(0, min_gpu_ram_mb * MEGA);
}
if (min_driver_version) {
gpu_requirements[PROC_TYPE_AMD_GPU].update(abs(min_driver_version), 0);
}
if (need_ati_libs) {
if (!cp.atirt_detected) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: ATI libraries not found\n"
);
}
return false;
}
} else {
if (!cp.amdrt_detected) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: AMD libraries not found\n"
);
}
return false;
}
}
if (min_cal_target && cp.attribs.target < min_cal_target) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CAL target less than minimum (%d < %d)\n",
cp.attribs.target, min_cal_target
);
}
return false;
}
if (max_cal_target && cp.attribs.target > max_cal_target) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CAL target greater than maximum (%d > %d)\n",
cp.attribs.target, max_cal_target
);
}
return false;
}
cp.set_peak_flops();
gpu_ram = cp.opencl_prop.global_mem_size;
driver_version = 0;
if (cp.have_cal) {
int major, minor, release, scanned;
scanned = sscanf(cp.version, "%d.%d.%d", &major, &minor, &release);
if (scanned != 3) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: driver version '%s' couldn't be parsed\n",
cp.version
);
}
return false;
} else {
driver_version = ati_version_int(major,minor,release);
}
} else {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: no CAL, driver version couldn't be determined\n"
);
}
}
// NVIDIA
//
} else if (!strcmp(gpu_type, "nvidia")) {
COPROC_NVIDIA& cp = sreq.coprocs.nvidia;
cpp = &cp;
if (!cp.count) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: No NVIDIA GPUs found\n"
);
}
return false;
}
if (min_gpu_ram_mb) {
gpu_requirements[PROC_TYPE_NVIDIA_GPU].update(0, min_gpu_ram_mb * MEGA);
}
if (min_driver_version) {
gpu_requirements[PROC_TYPE_NVIDIA_GPU].update(abs(min_driver_version), 0);
}
// compute capability
int v = (cp.prop.major)*100 + cp.prop.minor;
if (min_nvidia_compcap && min_nvidia_compcap > v) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: NVIDIA compute capability required min: %d, supplied: %d\n",
min_nvidia_compcap, v
);
}
return false;
}
if (max_nvidia_compcap && max_nvidia_compcap < v) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CUDA compute capability required max: %d, supplied: %d\n",
max_nvidia_compcap, v
);
}
return false;
}
if (cuda) {
// CUDA version
if (min_cuda_version && min_cuda_version > cp.cuda_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CUDA version required min: %d, supplied: %d\n",
min_cuda_version, cp.cuda_version
);
}
return false;
}
if (max_cuda_version && max_cuda_version < cp.cuda_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: CUDA version required max: %d, supplied: %d\n",
max_cuda_version, cp.cuda_version
);
}
return false;
}
}
gpu_ram = cp.prop.totalGlobalMem;
cp.set_peak_flops();
// Intel GPU
//
} else if (!strcmp(gpu_type, "intel")) {
COPROC& cp = sreq.coprocs.intel_gpu;
cpp = &cp;
if (!cp.count) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] [version] No Intel GPUs found\n"
);
}
return false;
}
if (min_gpu_ram_mb) {
gpu_requirements[PROC_TYPE_INTEL_GPU].update(0, min_gpu_ram_mb * MEGA);
}
}
if (opencl) {
// check for OpenCL at all
if (!cpp->have_opencl) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] GPU doesn't support OpenCL\n"
);
}
return false;
}
// OpenCL device version
//
if (min_opencl_version && min_opencl_version > cpp->opencl_prop.opencl_device_version_int) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] OpenCL device version required min: %d, supplied: %d\n",
min_opencl_version, cpp->opencl_prop.opencl_device_version_int
);
}
return false;
}
gpu_ram = cpp->opencl_prop.global_mem_size;
}
// general GPU
//
if (strlen(gpu_type)) {
// GPU RAM
if (min_gpu_ram_mb && min_gpu_ram_mb * MEGA > gpu_ram) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: GPU RAM required min: %f, supplied: %f\n",
min_gpu_ram_mb * MEGA, gpu_ram
);
}
return false;
}
// (display) driver version
if (min_driver_version && driver_version) {
if (min_driver_version > driver_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: driver version required min: %d, supplied: %d\n",
abs(min_driver_version), driver_version
);
}
return false;
}
}
if (max_driver_version && driver_version) {
if (max_driver_version < driver_version) {
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: driver version required max: %d, supplied: %d\n",
abs(max_driver_version), driver_version
);
}
return false;
}
}
hu.gpu_ram = (gpu_ram_used_mb?gpu_ram_used_mb:min_gpu_ram_mb) * MEGA;
double gpu_usage;
// if ngpus < 0, set gpu_usage by the fraction of the total
// video RAM a tasks would take
// i.e. fill the device memory with tasks
//
if (ngpus < 0) {
gpu_usage = (floor(gpu_ram/ hu.gpu_ram) * hu.gpu_ram) / gpu_ram ;
} else if (ngpus > 0) {
gpu_usage = ngpus * gpu_utilization;
} else {
gpu_usage = gpu_utilization;
}
coproc_perf(
capped_host_fpops(),
gpu_peak_flops_scale * gpu_usage * cpp->peak_flops,
cpu_frac,
hu.projected_flops,
hu.avg_ncpus
);
if (avg_ncpus) {
hu.avg_ncpus = avg_ncpus;
}
// I believe the first term here is just hu.projected_flops,
// but I'm leaving it spelled out to match GPU scheduling
// code in sched_customize.cpp
//
hu.peak_flops = gpu_peak_flops_scale*gpu_usage*cpp->peak_flops + hu.avg_ncpus*capped_host_fpops();
if (!strcmp(gpu_type, "amd") || !strcmp(gpu_type, "ati")) {
hu.proc_type = PROC_TYPE_AMD_GPU;
hu.gpu_usage = gpu_usage;
} else if (!strcmp(gpu_type, "nvidia")) {
hu.proc_type = PROC_TYPE_NVIDIA_GPU;
hu.gpu_usage = gpu_usage;
} else if (!strcmp(gpu_type, "intel")) {
hu.proc_type = PROC_TYPE_INTEL_GPU;
hu.gpu_usage = gpu_usage;
}
// CPU only
//
} else {
if (avg_ncpus) {
hu.avg_ncpus = avg_ncpus;
} else {
if (can_use_multicore) {
if (max_threads > g_wreq->effective_ncpus) {
hu.avg_ncpus = g_wreq->effective_ncpus;
} else {
hu.avg_ncpus = max_threads;
}
} else {
hu.avg_ncpus = 1;
}
}
hu.peak_flops = capped_host_fpops() * hu.avg_ncpus;
hu.projected_flops = capped_host_fpops() * hu.avg_ncpus * projected_flops_scale;
}
hu.max_ncpus = hu.avg_ncpus;
if (config.debug_version_select) {
log_messages.printf(MSG_NORMAL,
"[version] plan_class_spec: host_flops: %e, \tscale: %.2f, \tprojected_flops: %e, \tpeak_flops: %e\n",
sreq.host.p_fpops, projected_flops_scale, hu.projected_flops,
hu.peak_flops
);
}
return true;
}
bool PLAN_CLASS_SPECS::check(
SCHEDULER_REQUEST& sreq, char* plan_class, HOST_USAGE& hu
) {
for (unsigned int i=0; i<classes.size(); i++) {
if (!strcmp(classes[i].name, plan_class)) {
return classes[i].check(sreq, hu);
}
}
log_messages.printf(MSG_CRITICAL, "Unknown plan class: %s\n", plan_class);
return false;
}
int PLAN_CLASS_SPEC::parse(XML_PARSER& xp) {
char buf[256];
while (!xp.get_tag()) {
if (xp.match_tag("/plan_class")) {
return 0;
}
if (xp.parse_str("name", name, sizeof(name))) continue;
if (xp.parse_str("gpu_type", gpu_type, sizeof(gpu_type))) continue;
if (xp.parse_bool("cuda", cuda)) continue;
if (xp.parse_bool("cal", cal)) continue;
if (xp.parse_bool("opencl", opencl)) continue;
if (xp.parse_bool("virtualbox", virtualbox)) continue;
if (xp.parse_bool("is64bit", is64bit)) continue;
if (xp.parse_str("cpu_feature", buf, sizeof(buf))) {
cpu_features.push_back(" " + (string)buf + " ");
continue;
}
if (xp.parse_double("min_ncpus", min_ncpus)) continue;
if (xp.parse_int("max_threads", max_threads)) continue;
if (xp.parse_double("projected_flops_scale", projected_flops_scale)) continue;
if (xp.parse_str("os_regex", buf, sizeof(buf))) {
if (regcomp(&(os_regex), buf, REG_EXTENDED|REG_NOSUB) ) {
log_messages.printf(MSG_CRITICAL, "BAD REGEXP: %s\n", buf);
return ERR_XML_PARSE;
}
have_os_regex = true;
continue;
}
if (xp.parse_str("project_prefs_tag", project_prefs_tag, sizeof(project_prefs_tag))) continue;
if (xp.parse_str("project_prefs_regex", buf, sizeof(buf))) {
if (regcomp(&(project_prefs_regex), buf, REG_EXTENDED|REG_NOSUB) ) {
log_messages.printf(MSG_CRITICAL, "BAD REGEXP: %s\n", buf);
return ERR_XML_PARSE;
}
have_project_prefs_regex = true;
continue;
}
if (xp.parse_double("avg_ncpus", avg_ncpus)) continue;
if (xp.parse_double("cpu_frac", cpu_frac)) continue;
if (xp.parse_double("min_gpu_ram_mb", min_gpu_ram_mb)) continue;
if (xp.parse_double("gpu_ram_used_mb", gpu_ram_used_mb)) continue;
if (xp.parse_double("gpu_peak_flops_scale", gpu_peak_flops_scale)) continue;
if (xp.parse_double("ngpus", ngpus)) continue;
if (xp.parse_int("min_driver_version", min_driver_version)) continue;
if (xp.parse_int("max_driver_version", max_driver_version)) continue;
if (xp.parse_str("gpu_utilization_tag", gpu_utilization_tag, sizeof(gpu_utilization_tag))) continue;
if (xp.parse_bool("need_ati_libs", need_ati_libs)) continue;
if (xp.parse_int("min_cal_target", min_cal_target)) continue;
if (xp.parse_int("max_cal_target", max_cal_target)) continue;
if (xp.parse_int("min_nvidia_compcap", min_nvidia_compcap)) continue;
if (xp.parse_int("max_nvidia_compcap", max_nvidia_compcap)) continue;
if (xp.parse_int("min_cuda_version", min_cuda_version)) continue;
if (xp.parse_int("max_cuda_version", max_cuda_version)) continue;
if (xp.parse_int("min_opencl_version", min_opencl_version)) continue;
if (xp.parse_int("max_opencl_version", max_opencl_version)) continue;
if (xp.parse_int("min_vbox_version", min_vbox_version)) continue;
if (xp.parse_int("max_vbox_version", max_vbox_version)) continue;
}
return ERR_XML_PARSE;
}
int PLAN_CLASS_SPECS::parse_specs(FILE* f) {
MIOFILE mf;
XML_PARSER xp(&mf);
mf.init_file(f);
if (!xp.parse_start("plan_classes")) return ERR_XML_PARSE;
while (!xp.get_tag()) {
if (!xp.is_tag) {
fprintf(stderr, "PLAN_CLASS_SPECS::parse(): unexpected text %s\n", xp.parsed_tag);
continue;
}
if (xp.match_tag("/plan_classes")) {
return 0;
}
if (xp.match_tag("plan_class")) {
PLAN_CLASS_SPEC pc;
int retval = pc.parse(xp);
if (retval) return retval;
classes.push_back(pc);
}
}
return ERR_XML_PARSE;
}
PLAN_CLASS_SPEC::PLAN_CLASS_SPEC() {
strcpy(name, "");
strcpy(gpu_type, "");
cuda = false;
cal = false;
opencl = false;
virtualbox = false;
is64bit = false;
min_ncpus = 0;
max_threads = 1;
projected_flops_scale = 1;
have_os_regex = false;
strcpy(project_prefs_tag, "");
avg_ncpus = 0;
cpu_frac = .1;
min_gpu_ram_mb = 0;
gpu_ram_used_mb = 0;
gpu_peak_flops_scale = 1;
ngpus = 1;
min_driver_version = 0;
max_driver_version = 0;
strcpy(gpu_utilization_tag, "");
need_ati_libs = false;
min_nvidia_compcap = 0;
max_nvidia_compcap = 0;
min_cuda_version = 0;
max_cuda_version = 0;
min_opencl_version = 0;
max_opencl_version = 0;
min_vbox_version = 0;
max_vbox_version = 0;
}
#ifdef PLAN_CLASS_TEST
int main() {
SCHEDULER_REQUEST sreq;
SCHEDULER_REPLY sreply;
g_request = &sreq;
g_reply = &sreply;
g_wreq = &sreply.wreq;
PLAN_CLASS_SPECS pcs;
int retval = pcs.parse_file("plan_class_spec.xml.sample");
if (retval) {
printf("parse_file: %d\n");
exit(1);
}
config.debug_version_select = true;
HOST_USAGE hu;
strcpy(sreq.host.p_features, "sse3");
sreq.host.p_fpops =5e9;
g_wreq->effective_ncpus = 4;
if (1) {
sreq.coprocs.nvidia.fake(18000, 512*MEGA, 490*MEGA, 1);
sreq.coprocs.nvidia.opencl_prop.opencl_device_version_int = 0;
} else {
sreq.coprocs.nvidia.count = 0;
}
if (1) {
sreq.coprocs.ati.fake(512*MEGA, 256*MEGA, 1);
sreq.coprocs.ati.have_cal = true;
sreq.coprocs.ati.opencl_prop.opencl_device_version_int = 0;
} else {
sreq.coprocs.ati.count = 0;
}
for (unsigned int i=0; i<pcs.classes.size(); i++) {
bool b = pcs.check(sreq, pcs.classes[i].name, hu);
if (b) {
printf("%s: check succeeded\n", pcs.classes[i].name);
printf("\tncudas: %f\n\tnatis: %f\n\tgpu_ram: %fMB\n\tavg_ncpus: %f\n\tprojected_flops: %fG\n\tpeak_flops: %fG\n",
hu.ncudas,
hu.natis,
hu.gpu_ram/1e6,
hu.avg_ncpus,
hu.projected_flops/1e9,
hu.peak_flops/1e9
);
} else {
printf("%s: check failed\n", pcs.classes[i].name);
}
}
}
#endif