Fixed __ftoull

Author: ZERICO2005

src/crt/ftoll.src

@@ -1,9 +1,9 @@
 	assume	adl=1
 
 	section	.text
-	public	__ftoll, __ftoull
+
+	public	__ftoll
 __ftoll:
-__ftoull:
 	ld	d, a
 	push	iy, de, hl
 	call	__ftoll_c

src/crt/ftoull.c

@@ -0,0 +1,43 @@
+#include <stdbool.h>
+#include <stdint.h>
+
+typedef union F32_pun {
+    float flt;
+    uint32_t bin;
+    struct {
+        uint24_t UHL;
+        uint8_t E;
+    } reg;
+} F32_pun;
+
+#define Float32_pos_one      UINT32_C(0x3F800000)
+#define Float32_bias         127
+#define Float32_mant_bits    23
+#define Float32_u64_max_exp  64
+
+// this should be easy to rewrite in assembly
+uint64_t _ftoull_c(float x) {
+    F32_pun val;
+    val.flt = x;
+    // if trunc(x) is zero or x is negative
+    if ((int32_t)val.bin < (int32_t)Float32_pos_one) {
+        /* undefined return value for negative inputs */
+        return 0;
+    }
+    // signbit is zero here
+    uint8_t expon = (uint8_t)(val.bin >> 23);
+    // doesn't underflow since 1.0 - Float32_bias gives us an exponent of zero
+    expon -= Float32_bias;
+    if (expon >= Float32_u64_max_exp) {
+        /* undefined return value for overflow */
+        return 0;
+    }
+    // sets the LSB of the exponent since x is normalized
+    uint24_t mant = val.reg.UHL | 0x800000;
+    if (expon < Float32_mant_bits) {
+        mant >>= Float32_mant_bits - expon;
+        return mant;
+    }
+    /* expon >= 23 or [23, 63] */
+    return (uint64_t)mant << (expon - Float32_mant_bits);
+}

src/crt/ftoull.src

@@ -0,0 +1,13 @@
+	assume	adl=1
+
+	section	.text
+
+	public	__ftoull
+__ftoull:
+	ld	d, a
+	push	iy, de, hl
+	call	__ftoull_c
+	pop	af, af, iy
+	ret
+
+	extern	__ftoull_c