#ifndef VALVE_TAIL_DIAG_H
#define VALVE_TAIL_DIAG_H
#ifndef _WIN32_WINNT
#define _WIN32_WINNT 0x0600
#endif
#include "local_probe.h"
#include <stdint.h>
#include <string.h>
#include <errno.h>
#include <xmmintrin.h>
#ifndef KD_TRACE
#define KD_TRACE 1
#endif
typedef struct {unsigned mxcsr;unsigned short cw,sw;int err;} KdEnv;
typedef struct {uint64_t key[5];double cm,vel;KdEnv env;int ready;} KdMemo;
typedef struct {uint64_t jac;int group,hit;uint64_t key[5];double cm,vel;KdEnv before,after;uint64_t tail,lookup,capture;int executed,pad;} KdRecord;
extern int kd_scope;
extern uint64_t kd_jac;
extern KdMemo kd_memo;
static inline uint64_t kd_clock(void){unsigned lo,hi;__asm__ __volatile__("lfence\n\trdtsc\n\tlfence":"=a"(lo),"=d"(hi)::"memory");return ((uint64_t)hi<<32)|lo;}
static inline KdEnv kd_env(void){KdEnv e;__asm__ __volatile__("stmxcsr %0;fnstcw %1;fnstsw %2":"=m"(e.mxcsr),"=m"(e.cw),"=m"(e.sw)::"memory");e.err=errno;return e;}
static inline int kd_env_equal(KdEnv a,KdEnv b){return a.mxcsr==b.mxcsr && a.cw==b.cw && a.sw==b.sw && a.err==b.err;}
static inline int kd_finite(const uint64_t *key){for(int i=0;i<5;i++)if((key[i]&UINT64_C(0x7ff0000000000000))==UINT64_C(0x7ff0000000000000))return 0;return 1;}
/* Diagnostic-only lookup: its answer NEVER bypasses the original tail.
 * Only a baseline entry is considered. Probe misses never populate the slot. */
static inline int kd_lookup(const uint64_t *key,KdEnv e,double *cm,double *vel){
    if(!kd_memo.ready || !kd_env_equal(e,kd_memo.env) || memcmp(key,kd_memo.key,40))return 0;
    *cm=kd_memo.cm;*vel=kd_memo.vel;return 1;
}
void kd_start(void);void kd_finish(void);void kd_new_jac(void);
void kd_before(KdRecord*,double,double,double,double,double);
void kd_after(KdRecord*,double,double,uint64_t,int);
void kd_matrix(double,const double*,const double*);
void kd_eval_exit(double,const double*,const double*,const double*,int,NativePropertyCache*,NativePipeCache*,ModelJacobianWorkspace*);
#endif