Checkpoint

runtime/build.sh

@@ -37,9 +37,14 @@ asm "$SRC/libgcc.s"
 cc  "$SRC/libc.c"
 cc  "$SRC/strtol.c"
 cc  "$SRC/snprintf.c"
+cc  "$SRC/sscanf.c"
 cc  "$SRC/qsort.c"
 cc  "$SRC/extras.c"
 cc  "$SRC/strtok.c"
+cc  "$SRC/timeExt.c" -O1
+# timeExt.c at -O1: -O2 generates code where strftime's directive
+# switch overflows the W65816's 8-bit signed stack-relative offset
+# range. -O1 keeps the per-function frame small enough.
 cc  "$SRC/math.c"
 cc  "$SRC/softFloat.c"
 cc  "$SRC/libcxxabi.c"

runtime/include/math.h

@@ -1,30 +1,93 @@
 #ifndef _MATH_H
 #define _MATH_H
 
+// ---- Special values --------------------------------------------------
+// Use compiler builtins so const-folding works in user code.
+#define HUGE_VAL  (__builtin_huge_val())
+#define HUGE_VALF (__builtin_huge_valf())
+#define INFINITY  (__builtin_inff())
+#define NAN       (__builtin_nanf(""))
+
+// ---- Classification (functions, exposed via macros below) -----------
+int __isnan_d   (double x);
+int __isinf_d   (double x);
+int __isfinite_d(double x);
+int __signbit_d (double x);
+
+#define isnan(x)    __isnan_d((double)(x))
+#define isinf(x)    __isinf_d((double)(x))
+#define isfinite(x) __isfinite_d((double)(x))
+#define signbit(x)  __signbit_d((double)(x))
+
+// ---- Absolute / sign ------------------------------------------------
 double fabs    (double x);
 float  fabsf   (float  x);
-double floor   (double x);
-float  floorf  (float  x);
-double ceil    (double x);
-float  ceilf   (float  x);
-double fmod    (double x, double y);
-float  fmodf   (float  x, float  y);
 double copysign (double x, double y);
 float  copysignf(float  x, float  y);
+
+// ---- Rounding -------------------------------------------------------
+double floor  (double x);
+float  floorf (float  x);
+double ceil   (double x);
+float  ceilf  (float  x);
+double trunc  (double x);
+float  truncf (float  x);
+double round  (double x);
+float  roundf (float  x);
+
+// ---- Min / max / positive difference --------------------------------
+double fmax (double x, double y);
+double fmin (double x, double y);
+double fdim (double x, double y);
+float  fmaxf(float  x, float  y);
+float  fminf(float  x, float  y);
+float  fdimf(float  x, float  y);
+
+// ---- Mod / remainder ------------------------------------------------
+double fmod (double x, double y);
+float  fmodf(float  x, float  y);
+
+// ---- FP decomposition -----------------------------------------------
+double ldexp(double x, int n);
+float  ldexpf(float x, int n);
+double frexp(double x, int *e);
+float  frexpf(float x, int *e);
+double modf (double x, double *iptr);
+float  modff(float  x, float  *iptr);
+
+// ---- Power / root ---------------------------------------------------
 double sqrt    (double x);
 float  sqrtf   (float  x);
+double cbrt    (double x);
+float  cbrtf   (float  x);
 double pow     (double x, double y);
 float  powf    (float  x, float  y);
+double hypot   (double x, double y);
+float  hypotf  (float  x, float  y);
+
+// ---- Exponential / log ----------------------------------------------
+double exp    (double x);
+float  expf   (float  x);
+double exp2   (double x);
+float  exp2f  (float  x);
+double expm1  (double x);
+float  expm1f (float  x);
+double log    (double x);
+float  logf   (float  x);
+double log10  (double x);
+float  log10f (float  x);
+double log2   (double x);
+float  log2f  (float  x);
+double log1p  (double x);
+float  log1pf (float  x);
+
+// ---- Trigonometric --------------------------------------------------
 double sin     (double x);
 float  sinf    (float  x);
 double cos     (double x);
 float  cosf    (float  x);
 double tan     (double x);
 float  tanf    (float  x);
-double exp     (double x);
-float  expf    (float  x);
-double log     (double x);
-float  logf    (float  x);
 double atan    (double x);
 float  atanf   (float  x);
 double atan2   (double y, double x);
@@ -33,6 +96,8 @@ double asin    (double x);
 float  asinf   (float  x);
 double acos    (double x);
 float  acosf   (float  x);
+
+// ---- Hyperbolic -----------------------------------------------------
 double sinh    (double x);
 float  sinhf   (float  x);
 double cosh    (double x);
@@ -40,4 +105,20 @@ float  coshf   (float  x);
 double tanh    (double x);
 float  tanhf   (float  x);
 
+// ---- Common constants -----------------------------------------------
+// (Not in C99 strict, but defined by glibc/BSD math.h and widely used.)
+#define M_E         2.7182818284590452354
+#define M_LOG2E     1.4426950408889634074
+#define M_LOG10E    0.43429448190325182765
+#define M_LN2       0.69314718055994530942
+#define M_LN10      2.30258509299404568402
+#define M_PI        3.14159265358979323846
+#define M_PI_2      1.57079632679489661923
+#define M_PI_4      0.78539816339744830962
+#define M_1_PI      0.31830988618379067154
+#define M_2_PI      0.63661977236758134308
+#define M_2_SQRTPI  1.12837916709551257390
+#define M_SQRT2     1.41421356237309504880
+#define M_SQRT1_2   0.70710678118654752440
+
 #endif

runtime/include/stdio.h

@@ -47,6 +47,14 @@ char *fgets(char *buf, int n, FILE *stream);
 int   ungetc(int c, FILE *stream);
 #define getc(s) fgetc(s)
 
+// scanf family — only sscanf and vsscanf are implemented (parsing
+// from a string buffer). scanf/fscanf would need a reliable byte-at-
+// a-time stdin which we don't have. Supports %d %i %u %x %X %o %s
+// %c %% with optional `l` long modifier.
+int sscanf (const char *str, const char *fmt, ...);
+int vsscanf(const char *str, const char *fmt, va_list ap);
+void  rewind(FILE *stream);  // = fseek(s, 0, SEEK_SET) + clearerr
+
 // Memory-backed FS: register a memory region as a named file so
 // fopen can open it.  `cap` should be >= size; use cap > size for
 // files that may grow on write.  `writable` controls whether

runtime/include/stdlib.h

@@ -10,12 +10,25 @@ void   free(void *p);
 
 int    abs(int n);
 long   labs(long n);
+long long llabs(long long n);
 int    atoi(const char *s);
 long   atol(const char *s);
-long   long llabs(long long n);
+long long atoll(const char *s);
+double atof(const char *s);
 
-long   strtol (const char *nptr, char **endptr, int base);
+typedef struct { int       quot, rem; } div_t;
-unsigned long strtoul(const char *nptr, char **endptr, int base);
+typedef struct { long      quot, rem; } ldiv_t;
+typedef struct { long long quot, rem; } lldiv_t;
+div_t   div  (int       n, int       d);
+ldiv_t  ldiv (long      n, long      d);
+lldiv_t lldiv(long long n, long long d);
+
+long           strtol  (const char *nptr, char **endptr, int base);
+unsigned long  strtoul (const char *nptr, char **endptr, int base);
+long long      strtoll (const char *nptr, char **endptr, int base);
+unsigned long long strtoull(const char *nptr, char **endptr, int base);
+double         strtod  (const char *nptr, char **endptr);
+float          strtof  (const char *nptr, char **endptr);
 
 typedef int (*__cmp_fn)(const void *, const void *);
 void   qsort  (void *base, size_t nmemb, size_t size, __cmp_fn cmp);

runtime/include/string.h

@@ -22,6 +22,11 @@ char  *strrchr(const char *s, int c);
 char  *strstr(const char *haystack, const char *needle);
 char  *strcat(char *dst, const char *src);
 char  *strncat(char *dst, const char *src, size_t n);
+char  *strdup (const char *s);
+char  *strndup(const char *s, size_t maxlen);
+void  *memccpy(void *dst, const void *src, int c, size_t n);
+char  *stpcpy (char *dst, const char *src);
+char  *stpncpy(char *dst, const char *src, size_t n);
 char  *strpbrk(const char *s, const char *accept);
 size_t strspn (const char *s, const char *accept);
 size_t strcspn(const char *s, const char *reject);

runtime/include/time.h

@@ -3,11 +3,48 @@
 
 typedef long          time_t;
 typedef unsigned long clock_t;
+typedef unsigned int  size_t;
 
 #define CLOCKS_PER_SEC 60   // IIgs vsync tick (placeholder)
 
+struct tm {
+    int tm_sec;        // 0..60 (60 = leap second)
+    int tm_min;        // 0..59
+    int tm_hour;       // 0..23
+    int tm_mday;       // 1..31
+    int tm_mon;        // 0..11 (Jan=0)
+    int tm_year;       // years since 1900
+    int tm_wday;       // 0..6 (Sun=0)
+    int tm_yday;       // 0..365 (Jan 1 = 0)
+    int tm_isdst;      // > 0 = DST in effect, 0 = not, < 0 = unknown
+};
+
 time_t  time(time_t *t);
 clock_t clock(void);
+double  difftime(time_t end, time_t start);
+
+// Calendar conversions. gmtime and localtime are identical here —
+// no timezone support; "local" is treated as UTC.
+//
+// ⚠ gmtime/localtime are CURRENTLY KNOWN-BROKEN: the year-decomposition
+// loop hits a W65816 backend codegen issue that mis-iterates and
+// returns year=1970 regardless of input. Workaround: build the
+// struct tm by hand and call mktime/asctime/strftime, all of which
+// work correctly on a user-supplied struct tm.
+struct tm *gmtime   (const time_t *t);
+struct tm *localtime(const time_t *t);
+time_t     mktime   (struct tm *tm);
+
+// Text formatting. asctime returns "Sun Jan  1 00:00:00 1970\n" form.
+// Both functions return a pointer to a static buffer (overwritten on
+// each call — not thread-safe).
+char *asctime(const struct tm *tm);
+char *ctime  (const time_t *t);
+
+// strftime: format `tm` per `fmt` into `buf` (max `n` bytes incl.
+// terminator). Supports a useful subset: %Y %m %d %H %M %S %j %w %a
+// %A %b %B %p %% — sufficient for log timestamps and date display.
+size_t strftime(char *buf, size_t n, const char *fmt, const struct tm *tm);
 
 // Initialise the IIgs Tool Locator so time() can call ReadTimeHex.
 // Call once before any time() use.  Idempotent — repeated calls

runtime/src/crt0Gsos.s

@@ -26,6 +26,14 @@
 
 	.globl __start
 __start:
+	; Set DBR := PBR so absolute (DBR-relative) loads/stores reach
+	; symbols within our segment. The Loader's documented contract is
+	; "DBR set to your bank" but verified empirically NOT always true
+	; for KIND=0x1000 segments — `lda absConst` was reading from the
+	; wrong bank without this. PHK + PLB copies PBR into DBR.
+	phk
+	plb
+
 	; Set DP=0. The C compiler assumes DP=0 for all `sta dp` and
 	; `[dp],y`-style accesses; GS/OS hands us a Memory-Manager-
 	; allocated DP page that we discard.

runtime/src/libc.c

@@ -131,6 +131,19 @@ int tolower(int c) { return isupper(c) ? c + 32 : c; }
 int abs(int n)        { return n < 0 ? -n : n; }
 long labs(long n)     { return n < 0 ? -n : n; }
 
+// div/ldiv/lldiv: return both quotient and remainder in one struct.
+// Useful for code that wants a single libcall instead of paired / and %.
+// Per C99: quot is integer division truncated toward zero; rem has the
+// same sign as the numerator.
+
+typedef struct { int       quot, rem; } div_t;
+typedef struct { long      quot, rem; } ldiv_t;
+typedef struct { long long quot, rem; } lldiv_t;
+
+div_t   div  (int       n, int       d) { div_t   r; r.quot = n/d; r.rem = n%d; return r; }
+ldiv_t  ldiv (long      n, long      d) { ldiv_t  r; r.quot = n/d; r.rem = n%d; return r; }
+lldiv_t lldiv(long long n, long long d) { lldiv_t r; r.quot = n/d; r.rem = n%d; return r; }
+
 int atoi(const char *s) {
     int sign = 1;
     while (isspace(*s)) s++;
@@ -364,6 +377,59 @@ char *strstr(const char *haystack, const char *needle) {
     return 0;
 }
 
+// Forward declarations for strdup/strndup; the actual definitions
+// live further down in the file (malloc and strnlen are below).
+extern void *malloc(size_t);
+extern size_t strnlen(const char *, size_t);
+
+// strdup/strndup — POSIX, super common in real-world code.
+// Both allocate via malloc; caller frees.
+char *strdup(const char *s) {
+    size_t n = strlen(s);
+    char *r = (char *)malloc(n + 1);
+    if (!r) return 0;
+    memcpy(r, s, n + 1);
+    return r;
+}
+
+char *strndup(const char *s, size_t maxlen) {
+    size_t n = strnlen(s, maxlen);
+    char *r = (char *)malloc(n + 1);
+    if (!r) return 0;
+    memcpy(r, s, n);
+    r[n] = 0;
+    return r;
+}
+
+// memccpy — copy until either `n` bytes done OR byte `c` was copied.
+// Returns ptr to byte after the copied `c`, or NULL if `c` not found.
+void *memccpy(void *dst, const void *src, int c, size_t n) {
+    unsigned char *d = (unsigned char *)dst;
+    const unsigned char *s = (const unsigned char *)src;
+    while (n--) {
+        unsigned char v = *s++;
+        *d++ = v;
+        if (v == (unsigned char)c) return d;
+    }
+    return 0;
+}
+
+// stpcpy — like strcpy but returns ptr to terminating NUL of dst.
+char *stpcpy(char *dst, const char *src) {
+    while ((*dst = *src) != 0) { dst++; src++; }
+    return dst;
+}
+
+// stpncpy — like strncpy but returns ptr to terminator (or to dst+n
+// if no terminator was copied).
+char *stpncpy(char *dst, const char *src, size_t n) {
+    char *p = dst;
+    while (n && (*p = *src) != 0) { p++; src++; n--; }
+    char *end = p;
+    while (n--) *p++ = 0;
+    return end;
+}
+
 // ---- malloc/free — first-fit allocator with coalescing-on-free ----
 //
 // Heap lives between the static-data top (linker-supplied __heap_start)
@@ -1129,6 +1195,14 @@ void clearerr(FILE *stream) {
     if (stream) { stream->eof = 0; stream->err = 0; }
 }
 
+// rewind — convenience wrapper: seek to start + clear error/EOF.
+void rewind(FILE *stream) {
+    if (!stream) return;
+    fseek(stream, 0L, 0 /* SEEK_SET */);
+    stream->eof = 0;
+    stream->err = 0;
+}
+
 // ---- locale.h stubs ----
 //
 // No real locale support — IIgs is single-locale.  setlocale always
@@ -1225,3 +1299,5 @@ int raise(int sig) {
     h(sig);
     return 0;
 }
+
+

runtime/src/math.c

@@ -503,3 +503,251 @@ double tanh(double x) {
 float tanhf(float x) {
     return (float)tanh((double)x);
 }
+
+
+// ---- Classification ------------------------------------------------
+//
+// Implemented as functions rather than the C99 macros so they can be
+// referenced from outside math.h. The math.h header also exposes them
+// as macros that expand to these calls.
+
+int __isnan_d(double x) {
+    uint64_t b = dToBits(x);
+    return ((b >> 52) & 0x7FF) == 0x7FF && (b & 0xFFFFFFFFFFFFFULL) != 0;
+}
+
+
+int __isinf_d(double x) {
+    uint64_t b = dToBits(x);
+    return ((b >> 52) & 0x7FF) == 0x7FF && (b & 0xFFFFFFFFFFFFFULL) == 0;
+}
+
+
+int __isfinite_d(double x) {
+    return ((dToBits(x) >> 52) & 0x7FF) != 0x7FF;
+}
+
+
+int __signbit_d(double x) {
+    return (int)(dToBits(x) >> 63);
+}
+
+
+// ---- Rounding ------------------------------------------------------
+
+// trunc: round toward zero. = floor for positive, ceil for negative.
+double trunc(double x) {
+    return (dToBits(x) >> 63) ? ceil(x) : floor(x);
+}
+
+
+float truncf(float x) {
+    return (float)trunc((double)x);
+}
+
+
+// round: round half away from zero.
+double round(double x) {
+    return (dToBits(x) >> 63) ? -floor(-x + 0.5) : floor(x + 0.5);
+}
+
+
+float roundf(float x) {
+    return (float)round((double)x);
+}
+
+
+// ---- Min / max / positive difference -------------------------------
+
+double fmax(double x, double y) {
+    if (__isnan_d(x)) return y;
+    if (__isnan_d(y)) return x;
+    return x > y ? x : y;
+}
+
+
+double fmin(double x, double y) {
+    if (__isnan_d(x)) return y;
+    if (__isnan_d(y)) return x;
+    return x < y ? x : y;
+}
+
+
+double fdim(double x, double y) {
+    return x > y ? x - y : 0.0;
+}
+
+
+float fmaxf(float x, float y) { return (float)fmax((double)x, (double)y); }
+float fminf(float x, float y) { return (float)fmin((double)x, (double)y); }
+float fdimf(float x, float y) { return (float)fdim((double)x, (double)y); }
+
+
+// ---- FP decomposition ----------------------------------------------
+
+// ldexp(x, n) = x * 2^n. Implemented by adjusting the exponent field
+// in place. Handles subnormals and overflow only crudely (overflow →
+// infinity; underflow → zero).
+double ldexp(double x, int n) {
+    uint64_t b = dToBits(x);
+    int e = (int)((b >> 52) & 0x7FF);
+    if (e == 0 || e == 0x7FF) {
+        // 0 or denorm or inf/nan — return as-is for inf/nan, else
+        // fall back to multiplication (handles subnormals correctly
+        // enough for typical use).
+        if (e == 0x7FF) return x;
+        // For zero / subnormal, multiply by 2^n via repeated *2.
+        if (n > 0) {
+            while (n--) x *= 2.0;
+            return x;
+        }
+        if (n < 0) {
+            while (n++) x *= 0.5;
+            return x;
+        }
+        return x;
+    }
+    int newE = e + n;
+    if (newE >= 0x7FF) {
+        // Overflow → ±infinity.
+        return (b >> 63) ? -1e308 * 1e308 : 1e308 * 1e308;
+    }
+    if (newE <= 0) {
+        // Underflow → ±0 (skip subnormal handling).
+        return (b >> 63) ? -0.0 : 0.0;
+    }
+    b = (b & 0x800FFFFFFFFFFFFFULL) | ((uint64_t)newE << 52);
+    return dFromBits(b);
+}
+
+
+float ldexpf(float x, int n) {
+    return (float)ldexp((double)x, n);
+}
+
+
+// frexp(x, *e): split x into normalized fraction in [0.5, 1) (or 0)
+// and integer exponent. x = frac * 2^e.
+double frexp(double x, int *eOut) {
+    uint64_t b = dToBits(x);
+    int e = (int)((b >> 52) & 0x7FF);
+    if (e == 0) { *eOut = 0; return x; }       // zero or subnormal
+    if (e == 0x7FF) { *eOut = 0; return x; }   // inf or nan
+    *eOut = e - 1022;                          // bias adjustment for [0.5,1)
+    // Force exponent to 1022 (so result is in [0.5, 1)).
+    b = (b & 0x800FFFFFFFFFFFFFULL) | ((uint64_t)1022 << 52);
+    return dFromBits(b);
+}
+
+
+float frexpf(float x, int *eOut) {
+    double r = frexp((double)x, eOut);
+    return (float)r;
+}
+
+
+// modf(x, *iptr): split into integer and fractional parts, both with
+// the same sign as x.
+double modf(double x, double *iptr) {
+    double ip = trunc(x);
+    *iptr = ip;
+    return x - ip;
+}
+
+
+float modff(float x, float *iptr) {
+    double ipd;
+    double frac = modf((double)x, &ipd);
+    *iptr = (float)ipd;
+    return (float)frac;
+}
+
+
+// ---- log10 / log2 / exp2 -------------------------------------------
+//
+// All routed through log() / exp() with the conversion constant.
+
+double log10(double x) {
+    // 1 / ln(10) = 0.43429448190325182765112891891660508229439700580366
+    return log(x) * 0.43429448190325182765;
+}
+
+
+double log2(double x) {
+    // 1 / ln(2) = 1.4426950408889634073599246810018921374266459541530
+    return log(x) * 1.4426950408889634074;
+}
+
+
+double exp2(double x) {
+    // ln(2) = 0.69314718055994530941723212145817656807550013436026
+    return exp(x * 0.69314718055994530942);
+}
+
+
+float log10f(float x) { return (float)log10((double)x); }
+float log2f(float x)  { return (float)log2((double)x); }
+float exp2f(float x)  { return (float)exp2((double)x); }
+
+
+// log1p(x) = log(1 + x).  For |x| small the naive form loses precision;
+// we accept that for now since this is a standalone runtime, not a
+// numerics library.
+double log1p(double x) { return log(1.0 + x); }
+float  log1pf(float x) { return (float)log1p((double)x); }
+
+
+// expm1(x) = exp(x) - 1.  Same loss-of-precision caveat near 0.
+double expm1(double x) { return exp(x) - 1.0; }
+float  expm1f(float x) { return (float)expm1((double)x); }
+
+
+// ---- hypot ---------------------------------------------------------
+//
+// hypot(x, y) = sqrt(x*x + y*y).  This implementation does NOT scale
+// to avoid overflow — for |x|, |y| < ~1e150 the naive form is fine,
+// past that you'd want the standard scale-by-max trick.
+
+// hypot — naive sqrt(x*x + y*y). NO `volatile` on the temps —
+// clang's codegen for volatile-double locals on this target generates
+// stack-relative loads/stores that crash under the GS/OS Loader (the
+// chain executes correctly under runInMame but not via Finder). The
+// volatile-free version works in both contexts.
+__attribute__((noinline))
+double hypot(double x, double y) {
+    double xx = x * x;
+    double yy = y * y;
+    double s  = xx + yy;
+    return sqrt(s);
+}
+
+
+float hypotf(float x, float y) {
+    return (float)hypot((double)x, (double)y);
+}
+
+
+// ---- cbrt ----------------------------------------------------------
+//
+// Newton-Raphson for cube root: r_{n+1} = (2*r_n + a/r_n²) / 3.
+// Converges quadratically; 30 iters more than enough for double.
+// Implemented WITHOUT calling pow because clang treats pow as a
+// known builtin and either inlines it (with bad fold of pow(x,1/3))
+// or DCEs the call entirely (cbrt body collapses to "return 0").
+// This implementation has no pow dependency and is immune.
+__attribute__((noinline))
+double cbrt(double x) {
+    if (x == 0.0) return x;
+    int neg = (int)(dToBits(x) >> 63) & 1;
+    double a = neg ? -x : x;
+    double r = a;                        // crude initial guess
+    for (int i = 0; i < 30; i++) {
+        r = (2.0 * r + a / (r * r)) * (1.0 / 3.0);
+    }
+    return neg ? -r : r;
+}
+
+
+float cbrtf(float x) {
+    return (float)cbrt((double)x);
+}

runtime/src/softDouble.c

@@ -377,3 +377,24 @@ s32 __fixdfsi(u64 x) {
     if (shift >= 0) m >>= shift; else m <<= -shift;
     return sign ? -(s32)m : (s32)m;
 }
+
+
+// __fixunsdfsi: unsigned double → uint32. Saturates to 0 for negative
+// inputs, to 0xFFFFFFFF for inputs >= 2^32. Used by clang when casting
+// double values to unsigned integer types.
+u32 __fixunsdfsi(u64 x) {
+    if (x & DSIGN_BIT) return 0;                 // negative → 0
+    u16 e = (u16)((x >> DEXP_SHIFT) & 0x7FF);
+    if (e == 0) return 0;
+    if (e == 0x7FF) return 0xFFFFFFFF;
+    s16 unbiased = (s16)e - DEXP_BIAS;
+    if (unbiased < 0) return 0;
+    if (unbiased > 31) return 0xFFFFFFFF;
+    u64 m = (x & DMANT_MASK) | DMANT_LEAD;
+    if (unbiased >= 52) {
+        m <<= (unbiased - 52);
+    } else {
+        m >>= (52 - unbiased);
+    }
+    return (u32)m;
+}

runtime/src/softFloat.c

@@ -220,6 +220,15 @@ s16 __gtsf2(u32 a, u32 b) { return __cmpsf2(a, b);     }
 s16 __lesf2(u32 a, u32 b) { return __cmpsf2(a, b);     }
 s16 __gesf2(u32 a, u32 b) { return __cmpsf2(a, b);     }
 
+// __unordsf2: 1 if either arg is NaN, else 0. Used for IEEE 754
+// unordered comparisons (a < b is false if either is NaN, etc.).
+s16 __unordsf2(u32 a, u32 b) {
+    u32 ax = a & 0x7FFFFFFFUL;
+    u32 bx = b & 0x7FFFFFFFUL;
+    // NaN: exp=0xFF, mantissa != 0  →  ax > 0x7F800000
+    return (ax > 0x7F800000UL) || (bx > 0x7F800000UL) ? 1 : 0;
+}
+
 u32 __floatsisf(s32 i) {
     if (i == 0) return 0;
     u32 sign = 0;

runtime/src/sscanf.c

@@ -0,0 +1,146 @@
+// sscanf — minimal subset for the W65816 runtime.
+// Supports format directives:
+//   %d / %i  signed int (decimal)
+//   %u       unsigned int (decimal)
+//   %x %X    unsigned int (hex; "0x" prefix optional)
+//   %o       unsigned int (octal)
+//   %ld %lu %lx  long-int variants (32-bit)
+//   %s       whitespace-terminated string into char*
+//   %c       single char into char*
+//   %%       literal %
+// Whitespace in the format matches zero or more whitespace chars
+// in the input. Returns the number of successful conversions or
+// EOF (-1) if input ends before any match.
+
+typedef __builtin_va_list va_list;
+#define va_start(ap, last) __builtin_va_start(ap, last)
+#define va_arg(ap, ty)     __builtin_va_arg(ap, ty)
+#define va_end(ap)         __builtin_va_end(ap)
+
+extern int isspace(int);
+
+// Skip leading whitespace, return the first non-space char ptr.
+static const char *skipWs(const char *s) {
+    while (*s && isspace(*s)) s++;
+    return s;
+}
+
+// Parse an unsigned integer in the given base. Updates *pp to the
+// first unconsumed char. Returns 1 if any digit was consumed, else 0.
+static int parseUL(const char **pp, int base, unsigned long *out) {
+    const char *p = *pp;
+    unsigned long v = 0;
+    int saw = 0;
+    while (*p) {
+        int c = *p, d;
+        if (c >= '0' && c <= '9')      d = c - '0';
+        else if (c >= 'a' && c <= 'z') d = 10 + c - 'a';
+        else if (c >= 'A' && c <= 'Z') d = 10 + c - 'A';
+        else break;
+        if (d < 0 || d >= base) break;
+        v = v * (unsigned long)base + (unsigned long)d;
+        saw = 1;
+        p++;
+    }
+    *pp = p;
+    *out = v;
+    return saw;
+}
+
+int vsscanf(const char *str, const char *fmt, va_list ap) {
+    int matched = 0;
+    const char *s = str;
+    while (*fmt) {
+        if (isspace(*fmt)) {
+            // Whitespace in format: skip 0+ whitespace in input.
+            while (*fmt && isspace(*fmt)) fmt++;
+            s = skipWs(s);
+            continue;
+        }
+        if (*fmt != '%') {
+            if (*s != *fmt) break;
+            fmt++; s++;
+            continue;
+        }
+        fmt++;
+        if (*fmt == 0) break;
+        // Long modifier?
+        int isLong = 0;
+        if (*fmt == 'l') { isLong = 1; fmt++; if (*fmt == 0) break; }
+        char spec = *fmt;
+
+        if (spec == '%') {
+            if (*s != '%') break;
+            s++; fmt++; continue;
+        }
+        if (spec == 'c') {
+            char *out = va_arg(ap, char *);
+            if (!*s) break;
+            *out = *s++;
+            matched++;
+            fmt++;
+            continue;
+        }
+        if (spec == 's') {
+            char *out = va_arg(ap, char *);
+            s = skipWs(s);
+            if (!*s) break;
+            int n = 0;
+            while (*s && !isspace(*s)) { *out++ = *s++; n++; }
+            *out = 0;
+            if (n) matched++;
+            fmt++;
+            continue;
+        }
+        // Numeric conversions: skip whitespace first.
+        s = skipWs(s);
+        int neg = 0;
+        if ((spec == 'd' || spec == 'i') && (*s == '+' || *s == '-')) {
+            neg = (*s == '-');
+            s++;
+        }
+        int base = 10;
+        if (spec == 'x' || spec == 'X') base = 16;
+        else if (spec == 'o') base = 8;
+        else if (spec == 'i') {
+            if (s[0] == '0' && (s[1] == 'x' || s[1] == 'X')) {
+                base = 16; s += 2;
+            } else if (s[0] == '0') {
+                base = 8;
+            }
+        }
+        if ((spec == 'x' || spec == 'X') && s[0] == '0' &&
+            (s[1] == 'x' || s[1] == 'X')) s += 2;
+        unsigned long v;
+        if (!parseUL(&s, base, &v)) break;
+        if (isLong) {
+            if (spec == 'd' || spec == 'i') {
+                long *out = va_arg(ap, long *);
+                *out = neg ? -(long)v : (long)v;
+            } else {
+                unsigned long *out = va_arg(ap, unsigned long *);
+                *out = v;
+            }
+        } else {
+            if (spec == 'd' || spec == 'i') {
+                int *out = va_arg(ap, int *);
+                *out = neg ? -(int)v : (int)v;
+            } else {
+                unsigned int *out = va_arg(ap, unsigned int *);
+                *out = (unsigned int)v;
+            }
+        }
+        matched++;
+        fmt++;
+    }
+    if (matched == 0 && !*s) return -1;   // EOF: no chars consumed
+    return matched;
+}
+
+int sscanf(const char *str, const char *fmt, ...) {
+    va_list ap;
+    va_start(ap, fmt);
+    int r = vsscanf(str, fmt, ap);
+    va_end(ap);
+    return r;
+}

runtime/src/strtol.c

@@ -60,3 +60,127 @@ long strtol(const char *nptr, char **endptr, int base) {
     // ("-2147483648" — the magnitude doesn't fit in long).
     return neg ? (long)(0ul - n) : (long)n;
 }
+
+
+// ---- Long-long (64-bit) variants ------------------------------------
+
+unsigned long long strtoull(const char *nptr, char **endptr, int base) {
+    const char *s = nptr;
+    while (isspace(*s)) s++;
+    int neg = 0;
+    if (*s == '-') { neg = 1; s++; }
+    else if (*s == '+') s++;
+    if ((base == 0 || base == 16) && s[0] == '0' &&
+        (s[1] == 'x' || s[1] == 'X') && charDigit(s[2], 16) >= 0) {
+        base = 16;
+        s += 2;
+    } else if (base == 0 && *s == '0') {
+        base = 8;
+        s++;
+    } else if (base == 0) {
+        base = 10;
+    }
+    unsigned long long n = 0;
+    int saw_digit = 0;
+    for (;;) {
+        int d = charDigit(*s, base);
+        if (d < 0) break;
+        n = n * (unsigned long long)base + (unsigned long long)d;
+        saw_digit = 1;
+        s++;
+    }
+    if (endptr) *endptr = (char *)(saw_digit ? s : nptr);
+    return neg ? (0ull - n) : n;
+}
+
+
+long long strtoll(const char *nptr, char **endptr, int base) {
+    const char *s = nptr;
+    while (isspace(*s)) s++;
+    int neg = (*s == '-');
+    if (*s == '+' || *s == '-') s++;
+    char *ep = 0;
+    unsigned long long n = strtoull(s, &ep, base);
+    if (ep == s) {
+        if (endptr) *endptr = (char *)nptr;
+        return 0;
+    }
+    if (endptr) *endptr = ep;
+    return neg ? (long long)(0ull - n) : (long long)n;
+}
+
+
+long long atoll(const char *s) {
+    return strtoll(s, 0, 10);
+}
+
+
+// ---- Float parsing (atof / strtod / strtof) -------------------------
+
+extern double pow(double, double);   // for 10^N scaling
+
+
+// strtod parses [whitespace][+|-]<digits>[.digits][e|E[+|-]<digits>].
+// Pure-double accumulation; precision degrades for very long inputs but
+// sufficient for typical config / data parsing.
+double strtod(const char *nptr, char **endptr) {
+    const char *s = nptr;
+    while (isspace(*s)) s++;
+    int neg = 0;
+    if (*s == '-') { neg = 1; s++; }
+    else if (*s == '+') s++;
+
+    double whole = 0.0;
+    int saw_digit = 0;
+    while (*s >= '0' && *s <= '9') {
+        whole = whole * 10.0 + (double)(*s - '0');
+        s++;
+        saw_digit = 1;
+    }
+    double frac = 0.0;
+    double scale = 1.0;
+    if (*s == '.') {
+        s++;
+        while (*s >= '0' && *s <= '9') {
+            frac = frac * 10.0 + (double)(*s - '0');
+            scale *= 10.0;
+            s++;
+            saw_digit = 1;
+        }
+    }
+    if (!saw_digit) {
+        if (endptr) *endptr = (char *)nptr;
+        return 0.0;
+    }
+    double v = whole + frac / scale;
+    if (*s == 'e' || *s == 'E') {
+        const char *eMark = s;
+        s++;
+        int eNeg = 0;
+        if (*s == '-') { eNeg = 1; s++; }
+        else if (*s == '+') s++;
+        if (*s < '0' || *s > '9') {
+            // No exponent digits — leave 'e' unconsumed.
+            s = eMark;
+        } else {
+            int e = 0;
+            while (*s >= '0' && *s <= '9') {
+                e = e * 10 + (*s - '0');
+                s++;
+            }
+            v *= pow(10.0, eNeg ? -e : e);
+        }
+    }
+    if (endptr) *endptr = (char *)s;
+    return neg ? -v : v;
+}
+
+
+float strtof(const char *nptr, char **endptr) {
+    return (float)strtod(nptr, endptr);
+}
+
+
+double atof(const char *s) {
+    return strtod(s, 0);
+}

runtime/src/timeExt.c

@@ -0,0 +1,207 @@
+// time.h additions (struct tm, gmtime, localtime, mktime, asctime,
+// ctime, difftime, strftime). Split out from libc.c to keep that
+// translation unit's frame size below the W65816 stack-relative range.
+
+typedef long          time_t;
+typedef unsigned long clock_t;
+typedef unsigned int  size_t;
+
+extern size_t strlen(const char *);
+
+static const unsigned short __monthDays[12] = {
+    0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334
+};
+
+static int __isLeap(int y) {
+    return (y % 4 == 0 && y % 100 != 0) || (y % 400 == 0);
+}
+
+// ---- time.h: struct tm + calendar conversions -----------------------
+
+struct tm {
+    int tm_sec, tm_min, tm_hour;
+    int tm_mday, tm_mon, tm_year;
+    int tm_wday, tm_yday, tm_isdst;
+};
+
+// Internal mutable buffer for gmtime/localtime/asctime/ctime.
+// Not thread-safe but matches the C standard's spec.
+static struct tm __gmtimeBuf;
+static char      __asctimeBuf[32];
+
+double difftime(time_t end, time_t start) {
+    return (double)(end - start);
+}
+
+// gmtime / localtime: convert seconds-since-1970 to broken-down time.
+// "local" is identical to "gm" — no timezone support.
+// Convert days-since-1970 to (year, days-into-year). Uses 4-year
+// cycles where possible to keep the loop short and to avoid clang
+// generating code that misbehaves on this target.
+// gmtime: KNOWN BROKEN at all -O levels. The year-decomposition loop
+// (subtract years from `days` until what's left fits in one year)
+// triggers a W65816 backend codegen issue — the loop doesn't iterate
+// correctly under either -O2 (frame overflow) or -O1/-O0 (wrong
+// values returned). For now, gmtime fills in fields with zeros and
+// just stashes the input epoch in tm_sec/tm_min as low/mid 16-bit.
+// asctime/strftime/mktime work correctly on a user-supplied struct
+// tm. Workaround for callers that need decomposition: build the
+// struct tm manually.
+struct tm *gmtime(const time_t *t) {
+    long secs = *t;
+    __gmtimeBuf.tm_sec  = (int)(secs & 0xFFFF);
+    __gmtimeBuf.tm_min  = (int)((secs >> 16) & 0xFFFF);
+    __gmtimeBuf.tm_hour = 0;
+    __gmtimeBuf.tm_mday = 1;
+    __gmtimeBuf.tm_mon  = 0;
+    __gmtimeBuf.tm_year = 70;          // 1970, sentinel "not decomposed"
+    __gmtimeBuf.tm_wday = 4;           // Jan 1 1970 was Thursday
+    __gmtimeBuf.tm_yday = 0;
+    __gmtimeBuf.tm_isdst = -1;
+    return &__gmtimeBuf;
+}
+
+struct tm *localtime(const time_t *t) {
+    return gmtime(t);
+}
+
+// mktime: convert broken-down time → seconds-since-1970. Also fills
+// in tm_wday and tm_yday if the caller didn't bother.
+time_t mktime(struct tm *tm) {
+    int year = tm->tm_year + 1900;
+    int mon  = tm->tm_mon;
+    long days = 0;
+    for (int y = 1970; y < year; y++) {
+        days += __isLeap(y) ? 366 : 365;
+    }
+    for (int m = 0; m < mon; m++) {
+        days += __monthDays[m + 1] - __monthDays[m];
+        // (the table starts at 0 for Jan, so each month is the diff)
+    }
+    if (mon > 1 && __isLeap(year)) days++;
+    days += tm->tm_mday - 1;
+    long secs = days * 86400L
+              + (long)tm->tm_hour * 3600L
+              + (long)tm->tm_min  * 60L
+              + tm->tm_sec;
+    tm->tm_wday = (int)((days + 4) % 7);
+    if (tm->tm_wday < 0) tm->tm_wday += 7;
+    tm->tm_yday = (int)(__monthDays[mon]
+                        + ((mon > 1 && __isLeap(year)) ? 1 : 0)
+                        + tm->tm_mday - 1);
+    return secs;
+}
+
+static const char *const __wdayShort[7] = {
+    "Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat"
+};
+static const char *const __wdayLong[7] = {
+    "Sunday", "Monday", "Tuesday", "Wednesday",
+    "Thursday", "Friday", "Saturday"
+};
+static const char *const __monShort[12] = {
+    "Jan", "Feb", "Mar", "Apr", "May", "Jun",
+    "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
+};
+static const char *const __monLong[12] = {
+    "January", "February", "March", "April",
+    "May", "June", "July", "August",
+    "September", "October", "November", "December"
+};
+
+// Format N-digit zero-padded. Forces u32 (long) arithmetic — clang's
+// strength reducer otherwise lowers /10 and %10 on small types into
+// i8 mulhu by 0xCD (magic constant for div-by-10), which the W65816
+// backend has no select pattern for.
+static char *fmtN(char *p, unsigned long v, int n) {
+    p += n;
+    char *end = p;
+    while (n--) {
+        unsigned long q = v / 10ul;
+        unsigned long r = v - q * 10ul;
+        p--;
+        *p = (char)('0' + (int)r);
+        v = q;
+    }
+    return end;
+}
+static char *fmt02(char *p, int v) { return fmtN(p, (unsigned long)v, 2); }
+static char *fmt04(char *p, int v) { return fmtN(p, (unsigned long)v, 4); }
+
+char *asctime(const struct tm *tm) {
+    // "Sun Jan  1 00:00:00 1970\n\0" = 26 bytes
+    char *p = __asctimeBuf;
+    int wday = tm->tm_wday & 7; if (wday > 6) wday = 0;
+    int mon  = tm->tm_mon  & 15; if (mon  > 11) mon = 0;
+    *p++ = __wdayShort[wday][0];
+    *p++ = __wdayShort[wday][1];
+    *p++ = __wdayShort[wday][2];
+    *p++ = ' ';
+    *p++ = __monShort[mon][0];
+    *p++ = __monShort[mon][1];
+    *p++ = __monShort[mon][2];
+    *p++ = ' ';
+    int mday = tm->tm_mday;
+    *p++ = (mday >= 10) ? ('0' + mday/10) : ' ';
+    *p++ = '0' + mday % 10;
+    *p++ = ' ';
+    p = fmt02(p, tm->tm_hour); *p++ = ':';
+    p = fmt02(p, tm->tm_min);  *p++ = ':';
+    p = fmt02(p, tm->tm_sec);  *p++ = ' ';
+    p = fmt04(p, tm->tm_year + 1900);
+    *p++ = '\n';
+    *p   = 0;
+    return __asctimeBuf;
+}
+
+char *ctime(const time_t *t) {
+    return asctime(gmtime(t));
+}
+
+// strftime — directive expansion is split into a helper so the main
+// loop's frame stays small (W65816 stack-relative offsets are 8-bit
+// signed).
+__attribute__((noinline))
+static int strftimeOne(char dst[8], char spec, const struct tm *tm,
+                       const char **strOut) {
+    *strOut = 0;
+    switch (spec) {
+    case 'Y': fmt04(dst, tm->tm_year + 1900); return 4;
+    case 'm': fmt02(dst, tm->tm_mon + 1);     return 2;
+    case 'd': fmt02(dst, tm->tm_mday);        return 2;
+    case 'H': fmt02(dst, tm->tm_hour);        return 2;
+    case 'M': fmt02(dst, tm->tm_min);         return 2;
+    case 'S': fmt02(dst, tm->tm_sec);         return 2;
+    case 'j': fmtN(dst, (unsigned long)(tm->tm_yday + 1), 3); return 3;
+    case 'w': dst[0] = (char)('0' + (tm->tm_wday & 7));       return 1;
+    case 'a': *strOut = __wdayShort[tm->tm_wday & 7];         return 3;
+    case 'A': *strOut = __wdayLong [tm->tm_wday & 7];
+              return (int)strlen(*strOut);
+    case 'b':
+    case 'h': *strOut = __monShort[tm->tm_mon & 15];          return 3;
+    case 'B': *strOut = __monLong [tm->tm_mon & 15];
+              return (int)strlen(*strOut);
+    case 'p': *strOut = (tm->tm_hour < 12) ? "AM" : "PM";     return 2;
+    case '%': dst[0] = '%';                                   return 1;
+    default:  dst[0] = '%'; dst[1] = spec;                    return 2;
+    }
+}
+
+size_t strftime(char *buf, size_t n, const char *fmt, const struct tm *tm) {
+    char *p   = buf;
+    char *end = buf + n;
+    char  tmp[8];
+    while (*fmt && p + 1 < end) {
+        if (*fmt != '%') { *p++ = *fmt++; continue; }
+        fmt++;
+        if (!*fmt) break;
+        const char *ins;
+        int len = strftimeOne(tmp, *fmt, tm, &ins);
+        const char *src = ins ? ins : tmp;
+        for (int i = 0; i < len && p + 1 < end; i++) *p++ = src[i];
+        fmt++;
+    }
+    if (p < end) *p = 0;
+    else if (n > 0) buf[n - 1] = 0;
+    return (size_t)(p - buf);
+}

scripts/runViaFinder.sh

@@ -36,8 +36,12 @@ WORK=$(mktemp -d -t finderlaunch.XXXXXX)
 trap 'rm -rf "$WORK"' EXIT
 
 cp "$SYSDISK" "$WORK/disk.po"
-cp "$OMF"    "$WORK/HELLO#B30000"
+# Create a separate 800K data disk and put HELLO on it. Keeps the
-"$CADIUS" ADDFILE "$WORK/disk.po" /SYSTEM.DISK "$WORK/HELLO#B30000" >/dev/null
+# boot disk untouched (and avoids the "20K free" limit on sys602.po
+# that fails for OMFs > ~15K).
+"$CADIUS" CREATEVOLUME "$WORK/data.po" DATA 800KB >/dev/null
+cp "$OMF" "$WORK/HELLO#B30000"
+"$CADIUS" ADDFILE "$WORK/data.po" /DATA "$WORK/HELLO#B30000" >/dev/null
 
 LUA_CHECKS=""
 EXPECTS=()
@@ -65,9 +69,9 @@ local key_cmd = get_field(":macadb:KEY3", "Command / Open Apple")
 local function press(f)   if f then f:set_value(1) end end
 local function release(f) if f then f:set_value(0) end end
 
--- Keystroke timeline: open System.Disk, then launch HELLO.
+-- Keystroke timeline: open DATA volume (the second disk), then launch HELLO.
 local steps = {
-    {3300, function() nat:post("S") end},     -- select System.Disk
+    {3300, function() nat:post("D") end},     -- select Data
     {3540, function() press(key_cmd) end},
     {3546, function() nat:post("o") end},     -- Cmd-O opens volume
     {3600, function() release(key_cmd) end},
@@ -75,7 +79,7 @@ local steps = {
     {4500, function() press(key_cmd) end},
     {4506, function() nat:post("o") end},     -- Cmd-O launches
     {4560, function() release(key_cmd) end},
-    {5400, function()
+    {6000, function()
 $LUA_CHECKS
         manager.machine:exit()
     end},
@@ -89,9 +93,9 @@ emu.register_frame_done(function()
 end)
 LUA
 
-OUT=$(timeout 130 mame apple2gs -rompath "$PROJECT_ROOT/tools/mame/roms" \
+OUT=$(timeout 150 mame apple2gs -rompath "$PROJECT_ROOT/tools/mame/roms" \
     -window -nothrottle -sound none \
-    -seconds_to_run 110 -flop3 "$WORK/disk.po" \
+    -seconds_to_run 130 -flop3 "$WORK/disk.po" -flop4 "$WORK/data.po" \
     -autoboot_script "$WORK/finder.lua" </dev/null 2>&1)
 
 # Verify each expected value.

scripts/smokeTest.sh

@@ -4610,8 +4610,10 @@ EOF
     oGsCrt0="$(mktemp --suffix=.o)"
     "$PROJECT_ROOT/tools/llvm-mos-build/bin/llvm-mc" -arch=w65816 -filetype=obj \
         "$PROJECT_ROOT/runtime/src/crt0.s" -o "$oGsCrt0"
+    # extras.o supplies strnlen (used by libc.c's strdup wrapper).
     if ! "$PROJECT_ROOT/tools/link816" -o "$binGs" --text-base 0x1000 \
             "$oGsCrt0" "$oGsLibc" "$oGsSnp" "$oGsSf" "$oGsSd" \
+            "$PROJECT_ROOT/runtime/extras.o" \
             "$oGsFile" "$oGsAsm" "$oLibgccFile" \
             --no-gc-sections 2>&1; then
         die "iigs/gsos.h + iigsGsos.s failed to link"
@@ -5148,7 +5150,8 @@ while post < bytecnt:
     op = b[post]
     if op == 0xF5:
         # 1 + 1 (ByteCnt) + 1 (BitShift) + 2 (OffsetPatch) + 2 (OffsetReference) = 7 bytes
-        if b[post+1] != 3: print(f'FAIL: cRELOC ByteCnt {b[post+1]} != 3'); sys.exit(1)
+        # ByteCnt 2 = IMM16 (data refs), 3 = IMM24 (JSL/JML/STAlong).
+        if b[post+1] not in (2, 3): print(f'FAIL: cRELOC ByteCnt {b[post+1]} not in {{2,3}}'); sys.exit(1)
         nCreloc += 1
         post += 7
     elif op == 0x00:

src/link816/link816.cpp

@@ -313,14 +313,15 @@ struct Layout {
     std::vector<TextSeg> segments;
 };
 
-// One IMM24 (3-byte absolute) relocation site, recorded for OMF
+// One IMM-N relocation site (N=2 or 3), recorded for OMF cRELOC
-// cRELOC emission. The Loader will rewrite the 3 bytes at `patchOff`
+// emission. The Loader rewrites the N bytes at `patchOff` to be
-// to be (segPlacedBase + offsetRef) when the segment is placed at
+// (segPlacedBase + offsetRef) when the segment is placed at runtime
-// runtime — this is what makes our compiled C runnable from Finder
+// — this fixes both 24-bit JSL/JML/STAlong AND 16-bit absolute data
-// when the segment lands at e.g. bank $1F instead of bank 0.
+// references when the segment doesn't land at link-time-base.
 struct Imm24Site {
     uint32_t patchOff;   // offset within text image (== patchAddr - textBase)
     uint32_t offsetRef;  // offset within text image of target symbol
+    uint8_t  byteCnt;    // 2 = IMM16, 3 = IMM24
 };
 static std::vector<Imm24Site> gImm24Sites;
 static uint32_t gTextBaseForSites = 0;
@@ -346,6 +347,25 @@ static void applyReloc(std::vector<uint8_t> &buf, uint32_t off,
         // caller's DBR points at the target's bank.
         buf[off]     = static_cast<uint8_t>(target & 0xFF);
         buf[off + 1] = static_cast<uint8_t>((target >> 8) & 0xFF);
+        // Record for cRELOC emission so the Loader patches this
+        // 2-byte operand to (segPlacedBase + offsetRef) at load
+        // time. Without this, `lda absConst` reads from the wrong
+        // address when the segment doesn't land at link-time-base
+        // (e.g., link-time-base=0x1000 but Loader places at bank:0).
+        if (gRecordSites) {
+            uint32_t targetBank = target & 0xFF0000;
+            uint32_t baseBank   = gTextBaseForSites & 0xFF0000;
+            if (targetBank == baseBank) {
+                Imm24Site s;
+                s.patchOff  = patchAddr - gTextBaseForSites;
+                s.offsetRef = target    - gTextBaseForSites;
+                // Use type field width = 2 to distinguish from IMM24
+                // (3). Imm24Site struct is reused — emitOmf will
+                // emit cRELOC ByteCnt=2 for this.
+                s.byteCnt   = 2;
+                gImm24Sites.push_back(s);
+            }
+        }
         break;
     case R_W65816_IMM24:
         if (target > 0xFFFFFF)
@@ -368,6 +388,7 @@ static void applyReloc(std::vector<uint8_t> &buf, uint32_t off,
                 Imm24Site s;
                 s.patchOff  = patchAddr - gTextBaseForSites;
                 s.offsetRef = target    - gTextBaseForSites;
+                s.byteCnt   = 3;
                 gImm24Sites.push_back(s);
             }
         }
@@ -1361,20 +1382,23 @@ int main(int argc, char **argv) {
     if (!mapPath.empty()) linker.writeMap(mapPath);
     if (!debugOutPath.empty()) linker.writeDebugSidecar(debugOutPath);
     if (!relocOutPath.empty()) {
-        // Sidecar binary format:
+        // Sidecar binary format (v2):
         //   u32 count
-        //   { u32 patchOff; u32 offsetRef; } × count
+        //   { u32 patchOff; u32 offsetRef; u8 byteCnt; u8 pad[3]; } × count
         // Both offsets are within the segment image (== link-time addr
-        // minus textBase).  Consumed by omfEmit --relocs to emit cRELOC
+        // minus textBase).  byteCnt = 2 for IMM16, 3 for IMM24.
-        // opcodes after the LCONST data.
+        // Consumed by omfEmit --relocs to emit cRELOC opcodes.
         std::ofstream rf(relocOutPath, std::ios::binary);
         if (!rf) die("cannot open '" + relocOutPath + "' for writing");
         uint32_t count = (uint32_t)gImm24Sites.size();
         rf.write(reinterpret_cast<const char *>(&count), 4);
         for (const auto &s : gImm24Sites) {
             uint32_t po = s.patchOff, off = s.offsetRef;
+            uint8_t  bc = s.byteCnt, pad[3] = {0, 0, 0};
             rf.write(reinterpret_cast<const char *>(&po), 4);
             rf.write(reinterpret_cast<const char *>(&off), 4);
+            rf.write(reinterpret_cast<const char *>(&bc), 1);
+            rf.write(reinterpret_cast<const char *>(pad), 3);
         }
     }
     // Multi-segment: write per-segment images + manifest if there's

src/link816/omfEmit.cpp

@@ -38,11 +38,17 @@ namespace {
 }
 
 // Populated by --relocs from a link816 sidecar.  Each entry is
-// (OffsetPatch, OffsetReference) — the in-segment offset to patch
+// (OffsetPatch, OffsetReference, ByteCnt) — the in-segment offset
-// (3 bytes wide) and the in-segment offset of the target. Consumed
+// to patch, the in-segment offset of the target, and the byte width
-// by emitOneSeg to write cRELOC opcodes between LCONST and END.
+// of the patch (2 for IMM16, 3 for IMM24). Consumed by emitOneSeg
+// to write cRELOC opcodes between LCONST and END.
+struct RelocSite {
+    uint16_t patchOff;
+    uint16_t offsetRef;
+    uint8_t  byteCnt;
+};
 } // close namespace
-std::vector<std::pair<uint16_t, uint16_t>> gReloc24Sites;
+std::vector<RelocSite> gReloc24Sites;
 namespace {
 
 static std::vector<uint8_t> readFile(const std::string &path) {
@@ -128,10 +134,10 @@ static std::vector<uint8_t> emitOneSeg(const std::vector<uint8_t> &image,
     // the Loader places us at non-zero bank.
     for (const auto &s : ::gReloc24Sites) {
         body.push_back(0xF5);
-        body.push_back(3);                          // ByteCnt
+        body.push_back(s.byteCnt);                  // ByteCnt (2 or 3)
         body.push_back(0);                          // BitShift
-        put16(body, s.first);                       // OffsetPatch
+        put16(body, s.patchOff);                    // OffsetPatch
-        put16(body, s.second);                      // OffsetReference
+        put16(body, s.offsetRef);                   // OffsetReference
     }
     body.push_back(0x00);                           // END opcode
 
@@ -550,26 +556,36 @@ int main(int argc, char **argv) {
     }
     if (output.empty()) usage(argv[0]);
 
-    // Load IMM24 reloc list, if provided.
+    // Load reloc list, if provided.
+    // Sidecar v2 layout: u32 count + 12 bytes per entry
+    //   { u32 patchOff; u32 offsetRef; u8 byteCnt; u8 pad[3]; }
     if (!relocFile.empty()) {
         auto raw = readFile(relocFile);
         if (raw.size() < 4) die("--relocs file too small");
         uint32_t cnt = (uint32_t)raw[0] | ((uint32_t)raw[1] << 8)
                      | ((uint32_t)raw[2] << 16) | ((uint32_t)raw[3] << 24);
-        if (raw.size() != 4 + 8 * cnt)
+        const size_t entrySize = 12;
+        if (raw.size() != 4 + entrySize * cnt)
             die("--relocs file size mismatch: count=" + std::to_string(cnt)
-                + " expected " + std::to_string(4 + 8*cnt) + " bytes, got "
+                + " expected " + std::to_string(4 + entrySize*cnt) + " bytes, got "
                 + std::to_string(raw.size()));
         gReloc24Sites.reserve(cnt);
         for (uint32_t k = 0; k < cnt; k++) {
-            size_t off = 4 + k * 8;
+            size_t off = 4 + k * entrySize;
             uint32_t patchOff = (uint32_t)raw[off]   | ((uint32_t)raw[off+1] << 8)
                               | ((uint32_t)raw[off+2] << 16) | ((uint32_t)raw[off+3] << 24);
             uint32_t offRef   = (uint32_t)raw[off+4] | ((uint32_t)raw[off+5] << 8)
                               | ((uint32_t)raw[off+6] << 16) | ((uint32_t)raw[off+7] << 24);
+            uint8_t  byteCnt  = raw[off+8];
             if (patchOff > 0xFFFF || offRef > 0xFFFF)
                 die("reloc site out of 16-bit range — segment too large?");
-            gReloc24Sites.emplace_back((uint16_t)patchOff, (uint16_t)offRef);
+            if (byteCnt != 2 && byteCnt != 3)
+                die("reloc site byteCnt=" + std::to_string(byteCnt) + " (must be 2 or 3)");
+            RelocSite s;
+            s.patchOff  = (uint16_t)patchOff;
+            s.offsetRef = (uint16_t)offRef;
+            s.byteCnt   = byteCnt;
+            gReloc24Sites.push_back(s);
         }
     }