redgpu/ispc: Intel® Implicit SPMD Program Compiler

check_isa.cpp
216 строк9 KB
Arina Neshlyaeva
VNNI support (avx2vnni and avx512vnni targets)
20 апр 2024, 02:46
20 апр 2024, 02:46dda0f45
/*
  Copyright (c) 2013-2024, Intel Corporation

  SPDX-License-Identifier: BSD-3-Clause
*/

///////////////////////////////////////////////////////////////////////////////
//                                                                           //
// This file is a standalone program, which detects the best supported ISA.  //
//                                                                           //
///////////////////////////////////////////////////////////////////////////////

#include <stdio.h>

#if defined(_WIN32) || defined(_WIN64)
#define HOST_IS_WINDOWS
#include <intrin.h>
#elif defined(__APPLE__)
#define HOST_IS_APPLE
#endif

#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
#if !defined(HOST_IS_WINDOWS)
static void __cpuid(int info[4], int infoType) {
    __asm__ __volatile__("cpuid" : "=a"(info[0]), "=b"(info[1]), "=c"(info[2]), "=d"(info[3]) : "0"(infoType));
}

static void __cpuidex(int info[4], int level, int count) {
    __asm__ __volatile__("cpuid" : "=a"(info[0]), "=b"(info[1]), "=c"(info[2]), "=d"(info[3]) : "0"(level), "2"(count));
}
#endif // !HOST_IS_WINDOWS

static bool __os_has_avx_support() {
#if defined(HOST_IS_WINDOWS)
    // Check if the OS will save the YMM registers
    unsigned long long xcrFeatureMask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
    return (xcrFeatureMask & 6) == 6;
#else  // !defined(HOST_IS_WINDOWS)
    // Check xgetbv; this uses a .byte sequence instead of the instruction
    // directly because older assemblers do not include support for xgetbv and
    // there is no easy way to conditionally compile based on the assembler used.
    int rEAX, rEDX;
    __asm__ __volatile__(".byte 0x0f, 0x01, 0xd0" : "=a"(rEAX), "=d"(rEDX) : "c"(0));
    return (rEAX & 6) == 6;
#endif // !defined(HOST_IS_WINDOWS)
}

static bool __os_has_avx512_support() {
#if defined(HOST_IS_WINDOWS)
    // Check if the OS saves the XMM, YMM and ZMM registers, i.e. it supports AVX2 and AVX512.
    // See section 2.1 of software.intel.com/sites/default/files/managed/0d/53/319433-022.pdf
    unsigned long long xcrFeatureMask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
    return (xcrFeatureMask & 0xE6) == 0xE6;
#elif defined(HOST_IS_APPLE)
    // macOS has different way of dealing with AVX512 than Windows and Linux:
    // - by default AVX512 is off in the newly created thread, which means CPUID flags will
    //   indicate AVX512 availability, but OS support check (XCR0) will not succeed.
    // - AVX512 can be enabled either by calling thread_set_state() or by executing any
    //   AVX512 instruction, which would cause #UD exception handled by the OS.
    // The purpose of this check is to identify if AVX512 is potentially available, so we
    // need to bypass OS check and look at CPUID flags only.
    // See ispc issue #1854 for more details.
    return true;
#else  // !defined(HOST_IS_WINDOWS)
    // Check xgetbv; this uses a .byte sequence instead of the instruction
    // directly because older assemblers do not include support for xgetbv and
    // there is no easy way to conditionally compile based on the assembler used.
    int rEAX, rEDX;
    __asm__ __volatile__(".byte 0x0f, 0x01, 0xd0" : "=a"(rEAX), "=d"(rEDX) : "c"(0));
    return (rEAX & 0xE6) == 0xE6;
#endif // !defined(HOST_IS_WINDOWS)
}

static bool __os_enabled_amx_support() {
#if defined(HOST_IS_WINDOWS)
    // Check if the OS will save the YMM registers
    unsigned long long xcrFeatureMask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
    return (xcrFeatureMask & 0x60000) == 0x60000;
#else  // !defined(HOST_IS_WINDOWS)
    // Check xgetbv; this uses a .byte sequence instead of the instruction
    // directly because older assemblers do not include support for xgetbv and
    // there is no easy way to conditionally compile based on the assembler used.
    int rEAX, rEDX;
    __asm__ __volatile__(".byte 0x0f, 0x01, 0xd0" : "=a"(rEAX), "=d"(rEDX) : "c"(0));
    return (rEAX & 0x60000) == 0x60000;
#endif // !defined(HOST_IS_WINDOWS)
}
#endif // !__x86_64__

static const char *lGetSystemISA() {
#if defined(__arm__) || defined(__aarch64__) || defined(_M_ARM64)
    return "ARM NEON";
#elif defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
    int info[4];
    __cpuid(info, 1);

    int info2[4];
    // Call cpuid with eax=7, ecx=0
    __cpuidex(info2, 7, 0);

    int info3[4] = {0, 0, 0, 0};
    int max_subleaf = info2[0];
    // Call cpuid with eax=7, ecx=1
    if (max_subleaf >= 1)
        __cpuidex(info3, 7, 1);

    // clang-format off
    bool sse2 =                (info[3] & (1 << 26))  != 0;
    bool sse41 =               (info[2] & (1 << 19))  != 0;
    bool sse42 =               (info[2] & (1 << 20))  != 0;
    bool avx_f16c =            (info[2] & (1 << 29))  != 0;
    bool avx_rdrand =          (info[2] & (1 << 30))  != 0;
    bool osxsave =             (info[2] & (1 << 27))  != 0;
    bool avx =                 (info[2] & (1 << 28))  != 0;
    bool avx2 =                (info2[1] & (1 << 5))  != 0;
    bool avx_vnni =            (info3[0] & (1 << 4))  != 0;
    bool avx512_f =            (info2[1] & (1 << 16)) != 0;
    // clang-format on

    if (osxsave && avx2 && avx512_f && __os_has_avx512_support()) {
        // We need to verify that AVX2 is also available,
        // as well as AVX512, because our targets are supposed
        // to use both.
        // clang-format off
        bool avx512_dq =           (info2[1] & (1 << 17)) != 0;
        bool avx512_pf =           (info2[1] & (1 << 26)) != 0;
        bool avx512_er =           (info2[1] & (1 << 27)) != 0;
        bool avx512_cd =           (info2[1] & (1 << 28)) != 0;
        bool avx512_bw =           (info2[1] & (1 << 30)) != 0;
        bool avx512_vl =           (info2[1] & (1 << 31)) != 0;
        bool avx512_vbmi2 =        (info2[2] & (1 << 6))  != 0;
        bool avx512_gfni =         (info2[2] & (1 << 8))  != 0;
        bool avx512_vaes =         (info2[2] & (1 << 9))  != 0;
        bool avx512_vpclmulqdq =   (info2[2] & (1 << 10)) != 0;
        bool avx512_vnni =         (info2[2] & (1 << 11)) != 0;
        bool avx512_bitalg =       (info2[2] & (1 << 12)) != 0;
        bool avx512_vpopcntdq =    (info2[2] & (1 << 14)) != 0;
        bool avx512_bf16 =         (info3[0] & (1 << 5))  != 0;
        bool avx512_vp2intersect = (info2[3] & (1 << 8))  != 0;
        bool avx512_amx_bf16 =     (info2[3] & (1 << 22)) != 0;
        bool avx512_amx_tile =     (info2[3] & (1 << 24)) != 0;
        bool avx512_amx_int8 =     (info2[3] & (1 << 25)) != 0;
        bool avx512_fp16 =         (info2[3] & (1 << 23)) != 0;
        // clang-format on

        // Knights Landing:          KNL = F + PF + ER + CD
        // Skylake server:           SKX = F + DQ + CD + BW + VL
        // Cascade Lake server:      CLX = SKX + VNNI
        // Cooper Lake server:       CPX = CLX + BF16
        // Ice Lake client & server: ICL = CLX + VBMI2 + GFNI + VAES + VPCLMULQDQ + BITALG + VPOPCNTDQ
        // Tiger Lake:               TGL = ICL + VP2INTERSECT
        // Sapphire Rapids:          SPR = ICL + BF16 + AMX_BF16 + AMX_TILE + AMX_INT8 + AVX_VNNI + FP16
        bool knl = avx512_pf && avx512_er && avx512_cd;
        bool skx = avx512_dq && avx512_cd && avx512_bw && avx512_vl;
        bool clx = skx && avx512_vnni;
        [[maybe_unused]] bool cpx = clx && avx512_bf16;
        bool icl =
            clx && avx512_vbmi2 && avx512_gfni && avx512_vaes && avx512_vpclmulqdq && avx512_bitalg && avx512_vpopcntdq;
        [[maybe_unused]] bool tgl = icl && avx512_vp2intersect;
        bool spr =
            icl && avx512_bf16 && avx512_amx_bf16 && avx512_amx_tile && avx512_amx_int8 && avx_vnni && avx512_fp16;
        if (spr) {
            if (__os_enabled_amx_support()) {
                return "SPR (AMX on)";
            } else {
                return "SPR (AMX off)";
            }
        } else if (icl) {
            return "ICL";
        } else if (skx) {
            return "SKX";
        } else if (knl) {
            return "KNL";
        }
        // If it's unknown AVX512 target, fall through and use AVX2
        // or whatever is available in the machine.
    }

    if (osxsave && avx && __os_has_avx_support()) {
        if (avx_vnni) {
            return "AVX2VNNI (codename Alder Lake)";
        }
        // AVX1 for sure....
        // Ivy Bridge?
        if (avx_f16c && avx_rdrand) {
            // So far, so good.  AVX2?
            if (avx2) {
                return "AVX2 (codename Haswell)";
            } else {
                // Ivy Bridge specific target was deprecated in ISPC, but
                // no harm detecting it in standalone tool.
                return "AVX1.1 (codename Ivy Bridge)";
            }
        }
        // Regular AVX
        return "AVX (codename Sandy Bridge)";
    } else if (sse42) {
        return "SSE4.2";
    } else if (sse41) {
        return "SSE4.1";
    } else if (sse2) {
        return "SSE2";
    } else {
        return "Error";
    }
#else
#error "Unsupported host CPU architecture."
#endif
}

int main() {
    const char *isa = lGetSystemISA();
    printf("ISA: %s\n", isa);

    return 0;
}