//
// Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// defs.hfa -- Commen macros, functions and typedefs
// Most files depend on them and they are always useful to have.
//
//  *** Must not contain code specific to libcfathread ***
//
// Author           : Thierry Delisle
// Created On       : Thu Nov  9 13:24:10 2017
// Last Modified By : Peter A. Buhr
// Last Modified On : Sat Oct 24 10:53:15 2020
// Update Count     : 21
//

#pragma once

#include <stdint.h>
#include <assert.h>


#define likely(x)   __builtin_expect(!!(x), 1)
#define unlikely(x) __builtin_expect(!!(x), 0)
#define thread_local _Thread_local

typedef void (*fptr_t)();
typedef int_fast16_t __lock_size_t;

#ifdef __cforall
#define __cfa_anonymous_object(x) inline struct x
#define __cfa_dlink(x) inline dlink(x)
#else
#define __cfa_anonymous_object(x) struct x __cfa_anonymous_object
#define __cfa_dlink(x) struct { struct x * next; struct x * back; } __dlink_substitute
#endif

#define libcfa_public __attribute__((visibility("default")))

#ifdef __cforall
void abort( const char fmt[], ... ) __attribute__ (( format(printf, 1, 2), __nothrow__, __leaf__, __noreturn__ ));
void abort( bool signalAbort, const char fmt[], ... ) __attribute__ (( format(printf, 2, 3), __nothrow__, __leaf__, __noreturn__ ));
extern "C" {
	#include <pthread.h>
	void __cabi_abort( const char fmt[], ... ) __attribute__ (( format(printf, 1, 2), __nothrow__, __leaf__, __noreturn__ ));
	int real_pthread_create(pthread_t *_thread, const pthread_attr_t *attr, void *(*start_routine) (void *), void *arg);
	int real_pthread_join(pthread_t _thread, void **retval);
	pthread_t real_pthread_self(void);
	int real_pthread_attr_init(pthread_attr_t *attr);
	int real_pthread_attr_destroy(pthread_attr_t *attr);
	int real_pthread_attr_setstack( pthread_attr_t *attr, void *stackaddr, size_t stacksize );
	int real_pthread_attr_getstacksize( const pthread_attr_t *attr, size_t *stacksize );
	//int real_pthread_sigqueue(pthread_t _thread, int sig, const union sigval value);
	//int real_pthread_sigmask( int /* how */, const sigset_t * /* set */, sigset_t * /* oset */ );
}
#endif

#if defined(__cforall_thread__)
#define OPTIONAL_THREAD
#else
#define OPTIONAL_THREAD __attribute__((weak))
#endif

static inline long long int rdtscl(void) {
	#if defined( __i386 ) || defined( __x86_64 )
	unsigned int lo, hi;
	__asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
	return ( (unsigned long long)lo)|( ((unsigned long long)hi)<<32 );
	#elif defined( __aarch64__ ) || defined( __arm__ )
	// https://github.com/google/benchmark/blob/v1.1.0/src/cycleclock.h#L116
	long long int virtual_timer_value;
	asm volatile("mrs %0, cntvct_el0" : "=r"(virtual_timer_value));
	return virtual_timer_value;
	#else
		#error unsupported hardware architecture
	#endif
}

// pause to prevent excess processor bus usage
#if defined( __i386 ) || defined( __x86_64 )
	#define Pause() __asm__ __volatile__ ( "pause" : : : )
#elif defined( __ARM_ARCH )
	#define Pause() __asm__ __volatile__ ( "YIELD" : : : )
#else
	#error unsupported architecture
#endif

#define CFA_IO_LAZY (1_l64u << 32_l64u)