cupva_device_debug.h

Fully qualified name: src/device/vpu_runtime/include/cupva_device_debug.h

File members: src/device/vpu_runtime/include/cupva_device_debug.h

/*
 * Copyright (c) 2022 NVIDIA Corporation.  All rights reserved.
 *
 * NVIDIA Corporation and its licensors retain all intellectual property
 * and proprietary rights in and to this software, related documentation
 * and any modifications thereto.  Any use, reproduction, disclosure or
 * distribution of this software and related documentation without an express
 * license agreement from NVIDIA Corporation is strictly prohibited.
 */
#ifndef CUPVA_DEVICE_DEBUG_H
#define CUPVA_DEVICE_DEBUG_H

#if CUPVA_ALLOW_NONSAFETY != 1
#    error \
        "cupva_device_debug.h contains non safety certified functions, you must set CUPVA_ALLOW_NONSAFETY preprocessor token to 1 to use this header."
#endif

#include <stdint.h>

#if CUPVA_BUILD_MODE == CUPVA_NATIVE
#    include <signal.h>
//include stdio.h for printf so that we get same functionality from this header in both native and non-native modes
#    include <stdio.h>
inline void swbrk(void)
{
    raise(SIGTRAP);
}
#else

#    ifdef __cplusplus
inline void swbrk(void)
{
    asm("SWBRK"
        "\n NOP");
}
#    else
inline assembly void swbrk(void)
{
    asm_begin SWBRK;
    NOP;
    asm_end
}
#    endif

#    define printf printf_

#endif

int printf_(const char *format, ...);

enum PerfmonCounters
{
    PERFMON_NO_INSTRUCTION = 0,
    PERFMON_VALID_INSTRUCTION,
    PERFMON_STALL_INSTRUCTION_DECODE,
    PERFMON_STALL_SUPERBANK_LOAD_CONFLICT,
    PERFMON_STALL_DATA_HAZARD_LOAD,
    PERFMON_STALL_DMA_LOAD_CONFLICT,
    PERFMON_STALL_SUPERBANK_STORE_CONFLICT,
    PERFMON_STALL_DMA_STORE_CONFLICT,
    PERFMON_VPU_WAITING,
    PERFMON_ICACHE_MISSES,
    PERFMON_ICACHE_MISS_DURATION,
    PERFMON_DLUT_BUSY,
    PERFMON_DLUT_WAIT_FOR_VPU,
    PERFMON_COUNT
};

struct PerfmonSample
{
    int32_t counters[PERFMON_COUNT];
};

#define MASK_PERFMON_GPO (1U << 3U)
void cupvaPerfmonTakeSampleImpl(PerfmonSample &sample);
inline void cupvaPerfmonTakeSample(PerfmonSample &sample)
{
    chess_separator_scheduler();
    gpo_clr(MASK_PERFMON_GPO);
    cupvaPerfmonTakeSampleImpl(sample);
    gpo_set(MASK_PERFMON_GPO);
    chess_separator_scheduler();
}

inline int32_t cupvaPerfmonReportRaw(PerfmonSample const &before, PerfmonSample const &after, PerfmonCounters counter)
{
    return after.counters[counter] - before.counters[counter];
}

#endif //CUPVA_DEVICE_DEBUG_H