boinc/lib/coproc.cpp

// This file is part of BOINC.
// http://boinc.berkeley.edu
// Copyright (C) 2007 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/>.

#if defined(_WIN32)
#include "boinc_win.h"
#else
#ifdef _USING_FCGI_
#include "boinc_fcgi.h"
#else
#include <cstdio>
#endif
#include <cstring>
#include <cstdlib>
#include <cmath>
#endif

#ifdef _WIN32
#include "win_util.h"
#else
#ifdef __APPLE__
// Suppress obsolete warning when building for OS 10.3.9
#define DLOPEN_NO_WARN
#include <mach-o/dyld.h>
#endif
#include "config.h"
#include <dlfcn.h>
#include <setjmp.h>
#include <signal.h>
#endif

#include "error_numbers.h"
#include "filesys.h"
#include "parse.h"
#include "str_replace.h"
#include "util.h"

#include "coproc.h"

#ifndef _USING_FCGI_
using std::perror;
#endif

int COPROC_REQ::parse(XML_PARSER& xp) {
    safe_strcpy(type, "");
    count = 0;
    while (!xp.get_tag()) {
        if (xp.match_tag("/coproc")) {
            if (!strlen(type)) return ERR_XML_PARSE;
            return 0;
        }
        if (xp.parse_str("type", type, sizeof(type))) continue;
        if (xp.parse_double("count", count)) continue;
    }
    return ERR_XML_PARSE;
}

int PCI_INFO::parse(XML_PARSER& xp) {
    present = false;
    bus_id = device_id = domain_id = 0;
    while (!xp.get_tag()) {
        if (xp.match_tag("/pci_info")) {
            return 0;
        }
        if (xp.parse_int("bus_id", bus_id)) continue;
        if (xp.parse_int("device_id", device_id)) continue;
        if (xp.parse_int("domain_id", domain_id)) continue;
    }
    return ERR_XML_PARSE;
}

#ifndef _USING_FCGI_

void PCI_INFO::write(MIOFILE& f) {
    f.printf(
        "<pci_info>\n"
        "   <bus_id>%d</bus_id>\n"
        "   <device_id>%d</device_id>\n"
        "   <domain_id>%d</domain_id>\n"
        "</pci_info>\n",
        bus_id,
        device_id,
        domain_id
    );
}

void COPROC::write_xml(MIOFILE& f, bool scheduler_rpc) {
    f.printf(
        "<coproc>\n"
        "   <type>%s</type>\n"
        "   <count>%d</count>\n"
        "   <peak_flops>%f</peak_flops>\n",
        type, count, peak_flops
    );
    
    if (scheduler_rpc) {
        write_request(f);
    }

    if (have_opencl) {
        opencl_prop.write_xml(f, "coproc_opencl");
    }
    
    f.printf("</coproc>\n");
}

void COPROC::write_request(MIOFILE& f) {
    f.printf(
        "   <req_secs>%f</req_secs>\n"
        "   <req_instances>%f</req_instances>\n"
        "   <estimated_delay>%f</estimated_delay>\n",
        req_secs,
        req_instances,
        estimated_delay
    );
}

#endif

int COPROC::parse(XML_PARSER& xp) {
    char buf[256];
    safe_strcpy(type, "");
    clear();
    for (int i=0; i<MAX_COPROC_INSTANCES; i++) {
        device_nums[i] = i;
    }
    while (!xp.get_tag()) {
        if (!xp.is_tag) continue;
        if (xp.match_tag("/coproc")) {
            if (!strlen(type)) return ERR_XML_PARSE;
            clear_usage();
            return 0;
        }
        if (xp.match_tag("coproc_opencl")) {
            opencl_prop.parse(xp, "/coproc_opencl");
            continue;
        }
        if (xp.parse_str("type", type, sizeof(type))) continue;
        if (xp.parse_int("count", count)) continue;
        if (xp.parse_double("req_secs", req_secs)) continue;
        if (xp.parse_double("req_instances", req_instances)) continue;
        if (xp.parse_double("peak_flops", peak_flops)) continue;
        if (xp.parse_str("device_nums", buf, sizeof(buf))) {
            int i=0;
            char* p = strtok(buf, " ");
            while (p && i<MAX_COPROC_INSTANCES) {
                device_nums[i++] = atoi(p);
                p = strtok(NULL, " ");
            }
            continue;
        }
        if (xp.parse_bool("non_gpu", non_gpu)) continue;
    }
    return ERR_XML_PARSE;
}

// return a string, to be stored in host.serialnum,
// describing the host's coprocessors
//
void COPROCS::summary_string(char* buf, int len) {
    char buf2[1024];

    strlcpy(buf, "", len);
    if (nvidia.count) {
        int mem = (int)(nvidia.prop.totalGlobalMem/MEGA);
        snprintf(buf2, sizeof(buf2),
            "[CUDA|%s|%d|%dMB|%d|%d]",
            nvidia.prop.name, nvidia.count,
            mem, nvidia.display_driver_version,
            nvidia.opencl_prop.opencl_device_version_int
        );
        strlcat(buf, buf2, len);
    }
    if (ati.count) {
        snprintf(buf2, sizeof(buf2),
            "[CAL|%s|%d|%uMB|%s|%d]",
            ati.name, ati.count,
            ati.attribs.localRAM, ati.version,
            ati.opencl_prop.opencl_device_version_int
        );
        strlcat(buf, buf2, len);
    }
    if (intel_gpu.count) {
        snprintf(buf2, sizeof(buf2),
            "[INTEL|%s|%d|%dMB|%s|%d]",
            intel_gpu.name, intel_gpu.count,
            (int)(intel_gpu.opencl_prop.global_mem_size/MEGA),
            intel_gpu.version,
            intel_gpu.opencl_prop.opencl_device_version_int
        );
        strlcat(buf, buf2, len);
    }

    // add OpenCL devices other than nvidia/amd/intel
    //
    for (int i=1; i<n_rsc; i++) {
        COPROC& cp = coprocs[i];
        int type = coproc_type_name_to_num(cp.type);
        if (type == PROC_TYPE_NVIDIA_GPU) continue;
        if (type == PROC_TYPE_AMD_GPU) continue;
        if (type == PROC_TYPE_INTEL_GPU) continue;
        if (!strlen(cp.opencl_prop.name)) continue;
        snprintf(buf2, sizeof(buf2),
            "[opencl_gpu|%s|%d|%dMB|%d]",
            cp.type,
            cp.count,
            (int)(cp.opencl_prop.global_mem_size/MEGA),
            cp.opencl_prop.opencl_device_version_int
        );
        strlcat(buf, buf2, len);
    }
}

int COPROCS::parse(XML_PARSER& xp) {
    int retval;

    clear();
    n_rsc = 1;
    safe_strcpy(coprocs[0].type, "CPU");
    while (!xp.get_tag()) {
        if (xp.match_tag("/coprocs")) {
            return 0;
        }
        if (xp.match_tag("coproc_cuda")) {
            retval = nvidia.parse(xp);
            if (retval) {
                nvidia.clear();
            } else {
                coprocs[n_rsc++] = nvidia;
            }
            continue;
        }
        if (xp.match_tag("coproc_ati")) {
            retval = ati.parse(xp);
            if (retval) {
                ati.clear();
            } else {
                coprocs[n_rsc++] = ati;
            }
            continue;
        }
        if (xp.match_tag("coproc_intel_gpu")) {
            retval = intel_gpu.parse(xp);
            if (retval) {
                intel_gpu.clear();
            } else {
                coprocs[n_rsc++] = intel_gpu;
            }
            continue;
        }
        if (xp.match_tag("coproc")) {
            COPROC cp;
            retval = cp.parse(xp);
            if (!retval) {
                coprocs[n_rsc++] = cp;
            } else {
                fprintf(stderr, "failed to parse <coproc>: %d\n", retval);
            }
        }
    }
    return ERR_XML_PARSE;
}

#ifdef _USING_FCGI_
void COPROCS::write_xml(MIOFILE&, bool) {
}
#else
void COPROCS::write_xml(MIOFILE& mf, bool scheduler_rpc) {
    mf.printf("    <coprocs>\n");
    
    for (int i=1; i<n_rsc; i++) {
        switch (coproc_type_name_to_num(coprocs[i].type)) {
        case PROC_TYPE_NVIDIA_GPU:
            nvidia.write_xml(mf, scheduler_rpc);
            break;
        case PROC_TYPE_AMD_GPU:
            ati.write_xml(mf, scheduler_rpc);
            break;
        case PROC_TYPE_INTEL_GPU:
            intel_gpu.write_xml(mf, scheduler_rpc);
            break;
        default:
            coprocs[i].write_xml(mf, scheduler_rpc);
        }
    }
    
    mf.printf("    </coprocs>\n");
}
#endif

void COPROC_NVIDIA::description(char* buf, int buflen) {
    char vers[256], cuda_vers[256];
    if (display_driver_version) {
#ifdef __APPLE__
     	int maj = display_driver_version >> 16;
    	int min = (display_driver_version >> 8) & 0xff;
    	int rev = display_driver_version & 0xff;
        snprintf(vers, sizeof(vers), "%d.%d.%d", maj, min, rev);
#else
        int maj = display_driver_version/100;
        int min = display_driver_version%100;
        snprintf(vers, sizeof(vers), "%d.%02d", maj, min);
#endif
    } else {
        safe_strcpy(vers, "unknown");
    }
    if (cuda_version) {
        int maj = cuda_version/1000;
        int min = (cuda_version%1000)/10;
        snprintf(cuda_vers, sizeof(cuda_vers), "%d.%d", maj, min);
    } else {
        safe_strcpy(cuda_vers, "unknown");
    }
    snprintf(buf, buflen,
        "%s (driver version %s, CUDA version %s, compute capability %d.%d, %.0fMB, %.0fMB available, %.0f GFLOPS peak)",
        prop.name, vers, cuda_vers, prop.major, prop.minor,
        prop.totalGlobalMem/MEGA, available_ram/MEGA, peak_flops/1e9
    );
}

#ifndef _USING_FCGI_
void COPROC_NVIDIA::write_xml(MIOFILE& f, bool scheduler_rpc) {
    f.printf(
        "<coproc_cuda>\n"
        "   <count>%d</count>\n"
        "   <name>%s</name>\n"
        "   <available_ram>%f</available_ram>\n"
        "   <have_cuda>%d</have_cuda>\n"
        "   <have_opencl>%d</have_opencl>\n",
        count,
        prop.name,
        available_ram,
        have_cuda ? 1 : 0,
        have_opencl ? 1 : 0
    );
    if (scheduler_rpc) {
        write_request(f);
    }
    f.printf(
        "   <peak_flops>%f</peak_flops>\n"
        "   <cudaVersion>%d</cudaVersion>\n"
        "   <drvVersion>%d</drvVersion>\n"
        "   <totalGlobalMem>%f</totalGlobalMem>\n"
        "   <sharedMemPerBlock>%f</sharedMemPerBlock>\n"
        "   <regsPerBlock>%d</regsPerBlock>\n"
        "   <warpSize>%d</warpSize>\n"
        "   <memPitch>%f</memPitch>\n"
        "   <maxThreadsPerBlock>%d</maxThreadsPerBlock>\n"
        "   <maxThreadsDim>%d %d %d</maxThreadsDim>\n"
        "   <maxGridSize>%d %d %d</maxGridSize>\n"
        "   <clockRate>%d</clockRate>\n"
        "   <totalConstMem>%f</totalConstMem>\n"
        "   <major>%d</major>\n"
        "   <minor>%d</minor>\n"
        "   <textureAlignment>%f</textureAlignment>\n"
        "   <deviceOverlap>%d</deviceOverlap>\n"
        "   <multiProcessorCount>%d</multiProcessorCount>\n",
        peak_flops,
        cuda_version,
        display_driver_version,
        prop.totalGlobalMem,
        prop.sharedMemPerBlock,
        prop.regsPerBlock,
        prop.warpSize,
        prop.memPitch,
        prop.maxThreadsPerBlock,
        prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2],
        prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2],
        prop.clockRate,
        prop.totalConstMem,
        prop.major,
        prop.minor,
        prop.textureAlignment,
        prop.deviceOverlap,
        prop.multiProcessorCount
    );

    if (have_opencl) {
        opencl_prop.write_xml(f, "coproc_opencl");
    }

    if (!scheduler_rpc) {
        for (int i=0; i<count; i++) {
            pci_infos[i].write(f);
        }
    }
    
    f.printf("</coproc_cuda>\n");
}
#endif

void COPROC_NVIDIA::clear() {
    static const COPROC_NVIDIA x(0);
    *this = x;
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_NVIDIA_GPU));
    estimated_delay = -1;   // mark as absent
    cuda_version = 0;
    display_driver_version = 0;
    safe_strcpy(prop.name, "");
    prop.totalGlobalMem = 0.0;
    prop.sharedMemPerBlock = 0;
    prop.regsPerBlock = 0;
    prop.warpSize = 0;
    prop.memPitch = 0;
    prop.maxThreadsPerBlock = 0;
    prop.maxThreadsDim[0] = 0;
    prop.maxThreadsDim[1] = 0;
    prop.maxThreadsDim[2] = 0;
    prop.maxGridSize[0] = 0;
    prop.maxGridSize[1] = 0;
    prop.maxGridSize[2] = 0;
    prop.clockRate = 0;
    prop.totalConstMem = 0;
    prop.major = 0;
    prop.minor = 0;
    prop.textureAlignment = 0;
    prop.deviceOverlap = 0;
    prop.multiProcessorCount = 0;
    is_used = COPROC_USED;
}

int COPROC_NVIDIA::parse(XML_PARSER& xp) {
    char buf2[256];
    int retval;
    int ipci = 0;

    clear();
    while (!xp.get_tag()) {
        if (xp.match_tag("/coproc_cuda")) {
            if (!peak_flops) {
                set_peak_flops();
            }
            if (!available_ram) {
                available_ram = prop.totalGlobalMem;
            }
            return 0;
        }
        if (xp.parse_int("count", count)) continue;
        if (xp.parse_double("peak_flops", peak_flops)) continue;
        if (xp.parse_bool("have_cuda", have_cuda)) continue;
        if (xp.parse_bool("have_opencl", have_opencl)) continue;
        if (xp.parse_double("available_ram", available_ram)) continue;
        if (xp.parse_double("req_secs", req_secs)) continue;
        if (xp.parse_double("req_instances", req_instances)) continue;
        if (xp.parse_double("estimated_delay", estimated_delay)) continue;
        if (xp.parse_int("cudaVersion", cuda_version)) continue;
        if (xp.parse_int("drvVersion", display_driver_version)) continue;
        if (xp.parse_str("name", prop.name, sizeof(prop.name))) continue;
        if (xp.parse_double("totalGlobalMem", prop.totalGlobalMem)) continue;
        if (xp.parse_double("sharedMemPerBlock", prop.sharedMemPerBlock)) continue;
        if (xp.parse_int("regsPerBlock", prop.regsPerBlock)) continue;
        if (xp.parse_int("warpSize", prop.warpSize)) continue;
        if (xp.parse_double("memPitch", prop.memPitch)) continue;
        if (xp.parse_int("maxThreadsPerBlock", prop.maxThreadsPerBlock)) continue;
        if (xp.parse_str("maxThreadsDim", buf2, sizeof(buf2))) {
            // can't use sscanf here (FCGI)
            //
            prop.maxThreadsDim[0] = atoi(buf2);
            char* p = strchr(buf2, ' ');
            if (p) {
                p++;
                prop.maxThreadsDim[1] = atoi(p);
                p = strchr(p, ' ');
                if (p) {
                    p++;
                    prop.maxThreadsDim[2] = atoi(p);
                }
            }
            continue;
        }
        if (xp.parse_str("maxGridSize", buf2, sizeof(buf2))) {
            prop.maxGridSize[0] = atoi(buf2);
            char* p = strchr(buf2, ' ');
            if (p) {
                p++;
                prop.maxGridSize[1] = atoi(p);
                p = strchr(p, ' ');
                if (p) {
                    p++;
                    prop.maxGridSize[2] = atoi(p);
                }
            }
            continue;
        }
        if (xp.parse_int("clockRate", prop.clockRate)) continue;
        if (xp.parse_double("totalConstMem", prop.totalConstMem)) continue;
        if (xp.parse_int("major", prop.major)) continue;
        if (xp.parse_int("minor", prop.minor)) continue;
        if (xp.parse_double("textureAlignment", prop.textureAlignment)) continue;
        if (xp.parse_int("deviceOverlap", prop.deviceOverlap)) continue;
        if (xp.parse_int("multiProcessorCount", prop.multiProcessorCount)) continue;
        if (xp.match_tag("pci_info")) {
            PCI_INFO p;
            p.parse(xp);
            if (ipci < MAX_COPROC_INSTANCES) {
                pci_infos[ipci++] = p;
            }
        }
        if (xp.match_tag("coproc_opencl")) {
            retval = opencl_prop.parse(xp, "/coproc_opencl");
            if (retval) return retval;
            continue;
        }
    }
    return ERR_XML_PARSE;
}

void COPROC_NVIDIA::set_peak_flops() {
    double x=0;
    int flops_per_clock=0, cores_per_proc=0;

    if (prop.major || opencl_prop.nv_compute_capability_major) {
        int major = prop.major;
        int minor = prop.minor;

        if (opencl_prop.nv_compute_capability_major) major = opencl_prop.nv_compute_capability_major;
        if (opencl_prop.nv_compute_capability_minor) minor = opencl_prop.nv_compute_capability_minor;

        switch (major) {
        case 1:
            flops_per_clock = 3;
            cores_per_proc = 8;
            break;
        case 2:
            flops_per_clock = 2;
            switch (minor) {
            case 0:
                cores_per_proc = 32;
                break;
            default:
                cores_per_proc = 48;
                break;
            }
            break;
        case 3:
            flops_per_clock = 2;
            cores_per_proc = 192;
            break;
        case 5:
            flops_per_clock = 2;
            cores_per_proc = 128;
            break;
        case 6:
            flops_per_clock = 2;
            switch (minor) {
            case 0:    // special for Tesla P100 (GP100)
                cores_per_proc = 64;
                break;
            default:
                cores_per_proc = 128;
                break;
            }
            break;
        case 7:    // for both cc7.0 (Titan V, Tesla V100) and cc7.5 (RTX, Tesla T4)
        default:
            flops_per_clock = 2;
            cores_per_proc = 64;
            break;
        }

    }

    if (prop.clockRate) {
        // clock rate is scaled down by 1000
        //
        x = (1000.*prop.clockRate) * prop.multiProcessorCount * cores_per_proc * flops_per_clock;
    } else if (opencl_prop.nv_compute_capability_major) {

        // OpenCL w/ cl_nv_device_attribute_query extension
        // Per: https://www.khronos.org/registry/cl/extensions/nv/cl_nv_device_attribute_query.txt
        //
        // The theoretical single-precision processing power of a Maxwell GPU in GFLOPS is computed as 2 (operations per FMA instruction per CUDA core per cycle) × number of CUDA cores × core clock speed (in GHz).
        // Per: https://en.wikipedia.org/wiki/Maxwell_(microarchitecture)#Performance
        // Per: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
        //
        // clock is in MHz
        //
        x = opencl_prop.max_compute_units * cores_per_proc * flops_per_clock * (opencl_prop.max_clock_frequency * 1e6);

    } else if (opencl_prop.max_compute_units) {
        // OpenCL doesn't give us compute capability.
        // assume CC 2: cores_per_proc is 48 and flops_per_clock is 2
        //
        x = opencl_prop.max_compute_units * 48 * 2 * opencl_prop.max_clock_frequency * 1e6;
    }
    peak_flops = x;
}

// fake a NVIDIA GPU (for debugging)
//
void COPROC_NVIDIA::fake(
    int driver_version, double ram, double avail_ram, int n
) {
    static const COPROC_NVIDIA x;
    *this = x;
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_NVIDIA_GPU));
    count = n;
    for (int i=0; i<count; i++) {
        device_nums[i] = i;
    }
    available_ram = avail_ram;
    display_driver_version = driver_version;
    cuda_version = 5000;
    have_cuda = true;
    safe_strcpy(prop.name, "Fake NVIDIA GPU");
    prop.totalGlobalMem = ram;
    prop.sharedMemPerBlock = 100;
    prop.regsPerBlock = 8;
    prop.warpSize = 10;
    prop.memPitch = 10;
    prop.maxThreadsPerBlock = 20;
    prop.maxThreadsDim[0] = 2;
    prop.maxThreadsDim[1] = 2;
    prop.maxThreadsDim[2] = 2;
    prop.maxGridSize[0] = 10;
    prop.maxGridSize[1] = 10;
    prop.maxGridSize[2] = 10;
    prop.totalConstMem = 10;
    prop.major = 1;
    prop.minor = 2;
    prop.clockRate = 1250000;
    prop.textureAlignment = 1000;
    prop.multiProcessorCount = 14;
    have_opencl = true;
    safe_strcpy(opencl_prop.opencl_device_version, "OpenCL 3.17");
    opencl_prop.opencl_device_version_int = 317;
    set_peak_flops();
}

////////////////// ATI STARTS HERE /////////////////

#ifndef _USING_FCGI_
void COPROC_ATI::write_xml(MIOFILE& f, bool scheduler_rpc) {
    f.printf(
        "<coproc_ati>\n"
        "   <count>%d</count>\n"
        "   <name>%s</name>\n"
        "   <available_ram>%f</available_ram>\n"
        "   <have_cal>%d</have_cal>\n"
        "   <have_opencl>%d</have_opencl>\n",
        count,
        name,
        available_ram,
        have_cal ? 1 : 0,
        have_opencl ? 1 : 0
    );
    if (scheduler_rpc) {
        write_request(f);
    }
    f.printf(
        "   <peak_flops>%f</peak_flops>\n"
        "   <CALVersion>%s</CALVersion>\n"
        "   <target>%d</target>\n"
        "   <localRAM>%d</localRAM>\n"
        "   <uncachedRemoteRAM>%d</uncachedRemoteRAM>\n"
        "   <cachedRemoteRAM>%d</cachedRemoteRAM>\n"
        "   <engineClock>%u</engineClock>\n"
        "   <memoryClock>%d</memoryClock>\n"
        "   <wavefrontSize>%d</wavefrontSize>\n"
        "   <numberOfSIMD>%d</numberOfSIMD>\n"
        "   <doublePrecision>%d</doublePrecision>\n"
        "   <pitch_alignment>%d</pitch_alignment>\n"
        "   <surface_alignment>%d</surface_alignment>\n"
        "   <maxResource1DWidth>%d</maxResource1DWidth>\n"
        "   <maxResource2DWidth>%d</maxResource2DWidth>\n"
        "   <maxResource2DHeight>%d</maxResource2DHeight>\n",
        peak_flops,
        version,
        attribs.target,
        attribs.localRAM,
        attribs.uncachedRemoteRAM,
        attribs.cachedRemoteRAM,
        attribs.engineClock,
        attribs.memoryClock,
        attribs.wavefrontSize,
        attribs.numberOfSIMD,
        attribs.doublePrecision,
        attribs.pitch_alignment,
        attribs.surface_alignment,
        info.maxResource1DWidth,
        info.maxResource2DWidth,
        info.maxResource2DHeight
    );

    if (atirt_detected) {
        f.printf("    <atirt_detected/>\n");
    }

    if (amdrt_detected) {
        f.printf("    <amdrt_detected/>\n");
    }

    if (have_opencl) {
        opencl_prop.write_xml(f, "coproc_opencl");
    }
        
    f.printf("</coproc_ati>\n");
}
#endif

void COPROC_ATI::clear() {
    static const COPROC_ATI x(0);
    *this = x;
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_AMD_GPU));
    estimated_delay = -1;
    safe_strcpy(name, "");
    safe_strcpy(version, "");
    atirt_detected = false;
    amdrt_detected = false;
    version_num = 0;
    is_used = COPROC_USED;
}

int COPROC_ATI::parse(XML_PARSER& xp) {
    int n, retval;

    clear();

    while (!xp.get_tag()) {
        if (xp.match_tag("/coproc_ati")) {
            if (strlen(version)) {
                int major, minor, release;
                n = sscanf(version, "%d.%d.%d", &major, &minor, &release);
                if (n ==3) {
                    version_num = ati_version_int(major, minor, release);
                }
            }

            if (!peak_flops) {
				set_peak_flops();
            }
            if (!available_ram) {
                available_ram = attribs.localRAM*MEGA;
            }
            return 0;
        }
        if (xp.parse_int("count", count)) continue;
        if (xp.parse_double("peak_flops", peak_flops)) continue;
        if (xp.parse_bool("have_cal", have_cal)) continue;
        if (xp.parse_bool("have_opencl", have_opencl)) continue;
        if (xp.parse_double("available_ram", available_ram)) continue;
        if (xp.parse_double("req_secs", req_secs)) continue;
        if (xp.parse_double("req_instances", req_instances)) continue;
        if (xp.parse_double("estimated_delay", estimated_delay)) continue;
        if (xp.parse_str("name", name, sizeof(name))) continue;
        if (xp.parse_str("CALVersion", version, sizeof(version))) continue;
        if (xp.parse_bool("amdrt_detected", amdrt_detected)) continue;
        if (xp.parse_bool("atirt_detected", atirt_detected)) continue;

        if (xp.parse_int("target", n)) {
            attribs.target = (CALtarget)n;
            continue;
        }
        if (xp.parse_int("localRAM", n)) {
            attribs.localRAM = n;
            continue;
        }
        if (xp.parse_int("uncachedRemoteRAM", n)) {
            attribs.uncachedRemoteRAM = n;
            continue;
        }
        if (xp.parse_int("cachedRemoteRAM", n)) {
            attribs.cachedRemoteRAM = n;
            continue;
        }
        if (xp.parse_int("engineClock", n)) {
            attribs.engineClock = n;
            continue;
        }
        if (xp.parse_int("memoryClock", n)) {
            attribs.memoryClock = n;
            continue;
        }
        if (xp.parse_int("wavefrontSize", n)) {
            attribs.wavefrontSize = n;
            continue;
        }
        if (xp.parse_int("numberOfSIMD"  , n)) {
            attribs.numberOfSIMD = n;
            continue;
        }
        if (xp.parse_int("doublePrecision", n)) {
            attribs.doublePrecision = n?CAL_TRUE:CAL_FALSE;
            continue;
        }
        if (xp.parse_int("pitch_alignment", n)) {
            attribs.pitch_alignment = n;
            continue;
        }
        if (xp.parse_int("surface_alignment", n)) {
            attribs.surface_alignment = n;
            continue;
        }
        if (xp.parse_int("maxResource1DWidth", n)) {
            info.maxResource1DWidth = n;
            continue;
        }
        if (xp.parse_int("maxResource2DWidth", n)) {
            info.maxResource2DWidth = n;
            continue;
        }
        if (xp.parse_int("maxResource2DHeight", n)) {
            info.maxResource2DHeight = n;
            continue;
        }
        if (xp.match_tag("coproc_opencl")) {
            retval = opencl_prop.parse(xp, "/coproc_opencl");
            if (retval) return retval;
            continue;
        }
    }
    return ERR_XML_PARSE;
}

void COPROC_ATI::description(char* buf, int buflen) {
    snprintf(buf, buflen,
        "%s (CAL version %s, %uMB, %.0fMB available, %.0f GFLOPS peak)",
        name, version, attribs.localRAM,
        available_ram/MEGA, peak_flops/1.e9
    );
}

void COPROC_ATI::set_peak_flops() {
    double x = 0;
    if (attribs.numberOfSIMD) {
        x = attribs.numberOfSIMD * attribs.wavefrontSize * 5 * attribs.engineClock * 1.e6;
        // clock is in MHz
    } else if (opencl_prop.amd_simd_per_compute_unit) {

        // OpenCL w/ cl_amd_device_attribute_query extension
        // Per: https://www.khronos.org/registry/cl/extensions/amd/cl_amd_device_attribute_query.txt
        //
        // Single precision performance is calculated as two times the number of shaders multiplied by the base core clock speed.
        // Per: https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units
        //
        // clock is in MHz
        x = opencl_prop.max_compute_units * 
            opencl_prop.amd_simd_per_compute_unit * 
            opencl_prop.amd_simd_width *
            opencl_prop.amd_simd_instruction_width *
            2 *
            (opencl_prop.max_clock_frequency * 1.e6);

    } else if (opencl_prop.max_compute_units) {
        // OpenCL gives us only:
        // - max_compute_units
        //   (which I'll assume is the same as attribs.numberOfSIMD)
        // - max_clock_frequency (which I'll assume is the same as engineClock)
        // It doesn't give wavefrontSize, which can be 16/32/64.
        // So let's be conservative and use 16
        //
        x = opencl_prop.max_compute_units * 16 * 5 * opencl_prop.max_clock_frequency * 1e6;
    }
    peak_flops = x;
}

void COPROC_ATI::fake(double ram, double avail_ram, int n) {
    clear();
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_AMD_GPU));
    safe_strcpy(version, "1.4.3");
    safe_strcpy(name, "foobar");
    count = n;
    available_ram = avail_ram;
    have_cal = true;
    attribs.localRAM = (int)(ram/MEGA);
    attribs.numberOfSIMD = 32;
    attribs.wavefrontSize = 32;
    attribs.engineClock = 50;
    for (int i=0; i<count; i++) {
        device_nums[i] = i;
    }
    set_peak_flops();
}

////////////////// INTEL GPU STARTS HERE /////////////////

#ifndef _USING_FCGI_
void COPROC_INTEL::write_xml(MIOFILE& f, bool scheduler_rpc) {
    f.printf(
        "<coproc_intel_gpu>\n"
        "   <count>%d</count>\n"
        "   <name>%s</name>\n"
        "   <available_ram>%f</available_ram>\n"
        "   <have_opencl>%d</have_opencl>\n",
        count,
        name,
        available_ram,
        have_opencl ? 1 : 0
    );
    if (scheduler_rpc) {
        write_request(f);
    }
    f.printf(
        "   <peak_flops>%f</peak_flops>\n"
        "   <version>%s</version>\n",
        peak_flops,
        version
    );

    if (have_opencl) {
        opencl_prop.write_xml(f, "coproc_opencl");
    }
        
    f.printf("</coproc_intel_gpu>\n");
}
#endif

void COPROC_INTEL::clear() {
    static const COPROC_INTEL x(0);
    *this = x;
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_INTEL_GPU));
    estimated_delay = -1;
    safe_strcpy(name, "");
    safe_strcpy(version, "");
    global_mem_size = 0;
    is_used = COPROC_USED;
}

int COPROC_INTEL::parse(XML_PARSER& xp) {
    int retval;

    clear();

    while (!xp.get_tag()) {
        if (xp.match_tag("/coproc_intel_gpu")) {
            if (!peak_flops) {
				set_peak_flops();
            }
            if (!available_ram) {
                available_ram = opencl_prop.global_mem_size;
            }
            return 0;
        }
        if (xp.parse_int("count", count)) continue;
        if (xp.parse_double("peak_flops", peak_flops)) continue;
        if (xp.parse_bool("have_opencl", have_opencl)) continue;
        if (xp.parse_double("available_ram", available_ram)) continue;
        if (xp.parse_double("req_secs", req_secs)) continue;
        if (xp.parse_double("req_instances", req_instances)) continue;
        if (xp.parse_double("estimated_delay", estimated_delay)) continue;
        if (xp.parse_str("name", name, sizeof(name))) continue;
        if (xp.parse_str("version", version, sizeof(version))) continue;

        if (xp.match_tag("coproc_opencl")) {
            retval = opencl_prop.parse(xp, "/coproc_opencl");
            if (retval) return retval;
            continue;
        }
    }
    return ERR_XML_PARSE;
}

// http://en.wikipedia.org/wiki/Comparison_of_Intel_graphics_processing_units says:
// The raw performance of integrated GPU, in single-precision FLOPS,
// can be calculated as follows:
// EU * 4 [dual-issue x 2 SP] * 2 [multiply + accumulate] * clock speed.
//
// However, there is some question of the accuracy of this due to Intel's
// Turbo Boost and Dynamic Frequency technologies.
// 
void COPROC_INTEL::set_peak_flops() {
    double x = 0;
    if (opencl_prop.max_compute_units) {
        x = opencl_prop.max_compute_units * 8 * opencl_prop.max_clock_frequency * 1e6;
    }
    peak_flops = x;
}

void COPROC_INTEL::fake(double ram, double avail_ram, int n) {
    safe_strcpy(type, proc_type_name_xml(PROC_TYPE_INTEL_GPU));
    safe_strcpy(version, "1.4.3");
    safe_strcpy(name, "foobar");
    count = n;
    available_ram = avail_ram;
    have_opencl = true;
    for (int i=0; i<count; i++) {
        device_nums[i] = i;
    }
    set_peak_flops();
    opencl_prop.global_mem_size = (cl_ulong)ram;
}

// used wherever a processor type is specified in XML, e.g.
// <coproc>
//    <type>xxx</type>
//
// Don't confuse this with the element names used for GPUS within <coprocs>,
// namely:
// coproc_cuda
// coproc_ati
// coproc_intel_gpu
//
const char* proc_type_name_xml(int pt) {
    switch(pt) {
    case PROC_TYPE_CPU: return "CPU";
    case PROC_TYPE_NVIDIA_GPU: return "NVIDIA";
    case PROC_TYPE_AMD_GPU: return "ATI";
    case PROC_TYPE_INTEL_GPU: return "intel_gpu";
    case PROC_TYPE_MINER_ASIC: return "miner_asic";
    }
    return "unknown";
}

const char* proc_type_name(int pt) {
    switch(pt) {
    case PROC_TYPE_CPU: return "CPU";
    case PROC_TYPE_NVIDIA_GPU: return "NVIDIA GPU";
    case PROC_TYPE_AMD_GPU: return "AMD/ATI GPU";
    case PROC_TYPE_INTEL_GPU: return "Intel GPU";
    case PROC_TYPE_MINER_ASIC: return "Miner ASIC";
    }
    return "unknown";
}

int coproc_type_name_to_num(const char* name) {
    if (!strcmp(name, "CUDA")) return PROC_TYPE_NVIDIA_GPU;
    if (!strcmp(name, "NVIDIA")) return PROC_TYPE_NVIDIA_GPU;
    if (!strcmp(name, "ATI")) return PROC_TYPE_AMD_GPU;
    if (!strcmp(name, "intel_gpu")) return PROC_TYPE_INTEL_GPU;
    if (!strcmp(name, "miner_asic")) return PROC_TYPE_MINER_ASIC;
    return -1;      // Some other type
}