/*
 * This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/.
 *
 * Copyright 2018 Danny Robson <danny@nerdcruft.net>
 */

#include "x86.hpp"

#include "../std.hpp"
#include "../endian.hpp"
#include "../view.hpp"
#include "../bitwise.hpp"

#include <fmt/ostream.h>

#include <ostream>

using cruft::cpu::x86;


///////////////////////////////////////////////////////////////////////////////
namespace {
    struct cpuid_t {
        u32 a, b, c, d;
    };
}


//-----------------------------------------------------------------------------
static cpuid_t
cpuid (u32 query, u32 param = 0)
{
    ::cpuid_t res;

    asm (
        "cpuid"
        : "=a" (res.a), "=b" (res.b), "=c" (res.c), "=d" (res.d)
        :  "a" (query),                "c" (param)
    );

    return res;
}


///////////////////////////////////////////////////////////////////////////////
namespace {
    enum vendor_t {
        INTEL,
        AMD,

        UNKNOWN
    };
}


//-----------------------------------------------------------------------------
static constexpr
vendor_t
find_vendor (cpuid_t const &leaf0)
{
    constexpr struct {
        u32 b, c, d;
        vendor_t vendor;
    } VALUES[] = {
        { 0x68747541, 0x444d4163, 0x69746e65, AMD },
        { 0x756e6547, 0x6c65746e, 0x49656e69, INTEL },
    };

    for (auto const &v: VALUES) {
        if (v.b == leaf0.b && v.c == leaf0.c && v.d == leaf0.d)
            return v.vendor;
    }

    return UNKNOWN;
}


///////////////////////////////////////////////////////////////////////////////
x86::x86 ()
{
    auto const leaf0 = cpuid (0);
    memcpy (vendor_name.data () + 0, &leaf0.b, sizeof (leaf0.b));
    memcpy (vendor_name.data () + 4, &leaf0.d, sizeof (leaf0.d));
    memcpy (vendor_name.data () + 8, &leaf0.c, sizeof (leaf0.c));

    auto const vendor = find_vendor (leaf0);

    {
        auto features = cpuid (1);

        simd.sse3  = features.c & (1u <<  0);
        simd.ssse3 = features.c & (1u <<  9);
        simd.sse41 = features.c & (1u << 19);
        simd.sse42 = features.c & (1u << 20);
        simd.avx   = features.c & (1u << 28);
        simd.sse   = features.d & (1u << 25);
        simd.sse2  = features.d & (1u << 26);

        cores.hyper_threading = features.d & (1u << 28);
        if (cores.hyper_threading)
            cores.logical = from_bits (features.b, 23, 16);
        else
            cores.logical = 0;
    }

    {
        const auto product0 = cpuid (0x8000'0002u);
        const auto product1 = cpuid (0x8000'0003u);
        const auto product2 = cpuid (0x8000'0004u);

        memcpy (&product_name[0x00], &product0, sizeof (product0));
        memcpy (&product_name[0x10], &product1, sizeof (product1));
        memcpy (&product_name[0x20], &product2, sizeof (product2));
    }

    if (vendor == INTEL) {
        cores.logical = cores.physical = 0;

        for (int level = 0; true; ++level) {
            auto const param = cpuid (0x4, level);

            // The final level will set the 'cache type' bits to zero to
            // indicate there are no more caches.
            if ((param.a & 0b1111) == 0)
                break;

            cores.physical = max (
                cores.physical,
                cruft::cast::lossless<int> (from_bits (param.a, 31, 26)) + 1
            );
            cores.logical = max (
                cores.physical,
                cruft::cast::lossless<int> (from_bits (param.a, 25, 14)) + 1
            );
        }
    } else if (vendor == AMD) {
        auto const size_identifiers = cpuid (0x8000'0008);

        if (auto const apic_id_size = from_bits (size_identifiers.c, 15, 12); apic_id_size) {
            cores.physical = 1 << (apic_id_size - 1);
        } else {
            cores.physical = cruft::cast::lossless<int> (size_identifiers.c & 0xff) + 1;
        }
    } else {
        cores.physical = 0;
    }
}


///////////////////////////////////////////////////////////////////////////////
std::ostream&
cruft::cpu::operator<< (std::ostream &os, const x86 &val)
{
    auto const to_string = [] (auto const &obj) {
        return std::string_view {
            obj.begin (),
            std::find (obj.begin (), obj.end (), '\0')
        };
    };

    fmt::print (
        os,
        FMT_STRING(
            "{{ name: {{ vendor: '{}', product: '{}' }}"
            ", cores: {{ logical: {}, physical: {}, hyper_threading: {} }}"
            ", simd: {{ sse: {}, sse2: {}, sse3: {}, ssse3: {}, sse41: {}, sse42: {}, avx: {} }}"
            " }}"
        ),
        to_string (val.vendor_name),
        to_string (val.product_name),
        val.cores.logical,
        val.cores.physical,
        (val.cores.hyper_threading ? "true" : "false"),
        (val.simd.sse   ? "true" : "false"),
        (val.simd.sse2  ? "true" : "false"),
        (val.simd.sse3  ? "true" : "false"),
        (val.simd.ssse3 ? "true" : "false"),
        (val.simd.sse41 ? "true" : "false"),
        (val.simd.sse42 ? "true" : "false"),
        (val.simd.avx   ? "true" : "false")
    );

    return os;
}