Optimize long double routines

src/crt/dcmp.c

@@ -12,23 +12,31 @@ typedef union F64_pun {
 #define F64_CMP_GREATER    1 /* doesn't trigger flags */
 #define F64_CMP_UNORDERED  1 /* doesn't trigger flags */
 
+typedef struct f64_cmp_arg {
+    long double x;
+    bool x_sign; /* <-- unimplemented */
+    unsigned int const return_address;
+    long double y;
+    bool y_sign; /* <-- unimplemented */
+} f64_cmp_arg;
+
 // assumes no NaN
-int _dcmp_c(const long double *__restrict x, const long double *__restrict y) {
-    F64_pun arg_x, arg_y;
-    arg_x.flt = *x;
-    arg_y.flt = *y;
+int _dcmp_c(f64_cmp_arg *__restrict const arg) {
+    F64_pun x, y;
+    x.flt = arg->x;
+    y.flt = arg->y;
 
-    bool x_sign = signbit(arg_x.flt);
-    bool y_sign = signbit(arg_y.flt);
+    bool x_sign = signbit(x.flt);
+    bool y_sign = signbit(y.flt);
     if (x_sign != y_sign) {
-        if (iszero(arg_x.flt) && iszero(arg_y.flt)) {
+        if (iszero(x.flt) && iszero(y.flt)) {
             return F64_CMP_EQUAL;
         }
         return (x_sign ? F64_CMP_LESS : F64_CMP_GREATER);
     }
 
-    if (arg_x.bin == arg_y.bin) {
+    if (x.bin == y.bin) {
         return F64_CMP_EQUAL;
     }
-    return ((arg_x.bin < arg_y.bin) != x_sign) ? F64_CMP_LESS : F64_CMP_GREATER;
+    return ((x.bin < y.bin) != x_sign) ? F64_CMP_LESS : F64_CMP_GREATER;
 }

src/crt/dcmp.src

@@ -4,16 +4,18 @@
 
 	public	__dcmp
 
-; int _dcmp_c(const long double *__restrict x, const long double *__restrict y)
+; int _dcmp_c(f64_cmp_arg const *__restrict const arg)
 __dcmp:
-	; f64*_f64*_ret_i24
-	push	bc, de, hl, af, iy
-	ld	iy, 6
-	add	iy, sp
-	pea	iy + 12
-	push	iy
+	; f64_cmp_arg*_ret_i24
+	push	bc, de, hl
+	or	a, a
+	sbc	hl, hl
+	add	hl, sp
+	push	iy, af
+	push	hl	; f64_cmp_arg*
 	call	__dcmp_c
-	pop	af, af, iy, af
+	pop	af
+	pop	af, iy
 	; Set the comparison flags
 	add	hl, de
 	or	a, a

src/crt/ddiv.src

@@ -4,14 +4,23 @@
 
 	public	__ddiv
 
-; float64_t f64_div(float64_t, const float64_t*)
+; float64_t f64_div(bool, float64_t, const float64_t*)
 __ddiv:
 	push	af, iy
 	ld	iy, 9
-	add	iy, sp
+	add	iy, sp	
+
 	push	iy, bc, de, hl
+
+	ld	a, b	; signbit(x)
+	xor	a, (iy + 7)	; signbit(y)
+	rla
+	push	af	; Carry = (signbit(x) != signbit(y))
+
 	call	_f64_div
-	pop	af, af, af, af, iy, af
+
+	pop	af, af, af, af, af
+	pop	iy, af	; restore
 	ret
 
 	extern	_f64_div

src/crt/dmul.src

@@ -4,14 +4,23 @@
 
 	public	__dmul
 
-; float64_t f64_mul(float64_t, const float64_t*)
+; float64_t f64_mul(bool, float64_t, const float64_t*)
 __dmul:
 	push	af, iy
 	ld	iy, 9
-	add	iy, sp
+	add	iy, sp	
+
 	push	iy, bc, de, hl
+
+	ld	a, b	; signbit(x)
+	xor	a, (iy + 7)	; signbit(y)
+	rla
+	push	af	; Carry = (signbit(x) != signbit(y))
+
 	call	_f64_mul
-	pop	af, af, af, af, iy, af
+
+	pop	af, af, af, af, af
+	pop	iy, af	; restore
 	ret
 
 	extern	_f64_mul

src/crt/drem.src

@@ -4,14 +4,20 @@
 
 	public	__drem
 
-; float64_t f64_rem(float64_t, const float64_t*)
+; float64_t f64_rem(bool, float64_t, const float64_t*)
 __drem:
 	push	af, iy
 	ld	iy, 9
 	add	iy, sp
 	push	iy, bc, de, hl
+
+	rl	b
+	push	af	; Carry = signbit(x)
+
 	call	_f64_rem
-	pop	af, af, af, af, iy, af
+
+	pop	af, af, af, af, af
+	pop	iy, af	; restore
 	ret
 
 	extern	_f64_rem

src/crt/dtol.src

@@ -6,9 +6,17 @@
 
 __dtol:
 	; f64_ret_i32
-	push	af, iy, bc, de, hl
+	push	af, iy
+	ld	a, b
+	push	bc, de, hl
+	ld	hl, 7
+	add	hl, sp
+	res	7, (hl)	; fabsl(x)
+	inc	hl
+	rlca
+	ld	(hl), a	; store the sign of x in the padding byte
 	call	__dtol_c
-	pop	af, af, af, iy, af
+	pop	af, af, bc, iy, af
 	ret
 
 	extern	__dtol_c

src/crt/dtoll.src

@@ -6,7 +6,14 @@
 
 __dtoll:
 	; f64_ret_i64
-	push	af, iy, bc, de, hl
+	push	af, iy
+	ld	a, b
+	res	7, b	; fabsl(x)
+	push	bc, de, hl
+	ld	hl, 8
+	add	hl, sp
+	rlca
+	ld	(hl), a	; store the sign of x in the padding byte
 	call	__dtoll_c
 	pop	af, af, af, iy, af
 	ret

src/crt/dtoul.src

@@ -8,7 +8,7 @@ __dtoul:
 	; f64_ret_u32
 	push	af, iy, bc, de, hl
 	call	__dtoul_c
-	pop	af, af, af, iy, af
+	pop	af, af, bc, iy, af
 	ret
 
 	extern	__dtoul_c

src/crt/float64_to_int.c

@@ -74,6 +74,11 @@ uint64_t _dtoull_c(long double x) {
  */
 #define HANDLE_INT_MIN 1
 
+typedef struct f64_sign {
+    long double flt;
+    bool sign;
+} f64_sign;
+
 /**
  * @note val must have the signbit cleared
  */
@@ -123,10 +128,10 @@ uint32_t _dtoul_c(long double x) {
     return (uint32_t)f64_to_unsigned(val);
 }
 
-int64_t _dtoll_c(long double x) {
+int64_t _dtoll_c(f64_sign arg) {
     F64_pun val;
-    bool x_sign = signbit(x);
-    val.flt = fabsl(x);
+    bool x_sign = arg.sign;
+    val.flt = arg.flt;
 
     /* overflow || isinf(x) || isnan(x) */
     if (val.reg.BC >= ((Float64_bias + Float64_i64_max_exp) << Float64_exp_BC_shift)) {
@@ -145,10 +150,10 @@ int64_t _dtoll_c(long double x) {
     return ret;
 }
 
-int32_t _dtol_c(long double x) {
+int32_t _dtol_c(f64_sign arg) {
     F64_pun val;
-    bool x_sign = signbit(x);
-    val.flt = fabsl(x);
+    bool x_sign = arg.sign;
+    val.flt = arg.flt;
 
     /* overflow || isinf(x) || isnan(x) */
     if (val.reg.BC >= ((Float64_bias + Float64_i32_max_exp) << Float64_exp_BC_shift)) {

src/libc/ceill.c

@@ -1,9 +1,5 @@
 #include <math.h>
 
-#ifdef ceill
-#undef ceill
-#endif
-
 long double ceill(long double x) {
     if (signbit(x)) {
         return truncl(x);
@@ -17,5 +13,3 @@ long double ceill(long double x) {
     }
     return x;
 }
-
-long double _debug_ceill(long double) __attribute__((alias("ceill")));

src/libc/copysignl.src

@@ -2,9 +2,7 @@
 
 	section	.text
 
-	public	_copysignl, __debug_copysignl
-
-__debug_copysignl:
+	public	_copysignl
 _copysignl:
 	ld	hl, 19	; upper 8 bits of y
 	add	hl, sp

src/libc/fabsl.src

@@ -2,9 +2,8 @@
 
 	section	.text
 
-	public	_fabsl, __debug_fabsl
+	public	_fabsl
 
-__debug_fabsl:
 _fabsl:
 	pop	iy, hl ,de, bc
 	push	bc, de, hl

src/libc/float64_rounding.c

@@ -47,10 +47,6 @@ long double roundevenl(long double x) {
     return ret.flt;
 }
 
-#ifdef roundl
-#undef roundl
-#endif
-
 long double roundl(long double x) {
     F64_pun arg_x, ret;
     arg_x.flt = x;
@@ -93,21 +89,13 @@ long long llroundl(long double x) {
     }
 #endif
 
-#ifdef nearbyintl
-#undef nearbyintl
-#endif
-
 long double nearbyintl(long double x) {
     F64_pun arg_x, ret;
     arg_x.flt = x;
     ret.soft = f64_roundToInt(arg_x.soft, GET_FENV_SOFTFLOAT_ROUNDING(), false);
     return ret.flt;
 }
 
-#ifdef rintl
-#undef rintl
-#endif
-
 /* flags handled by softfloat */
 long double rintl(long double x) {
     F64_pun arg_x, ret;
@@ -129,8 +117,3 @@ long long llrintl(long double x) {
     arg_x.flt = x;
     return f64_to_i64(arg_x.soft, GET_FENV_SOFTFLOAT_ROUNDING(), true);
 }
-
-long double _debug_roundl(long double) __attribute__((alias("roundl")));
-long double _debug_rintl(long double) __attribute__((alias("rintl")));
-long double _debug_nearbyintl(long double) __attribute__((alias("nearbyintl")));
-

src/libc/floorl.c

@@ -1,9 +1,5 @@
 #include <math.h>
 
-#ifdef floorl
-#undef floorl
-#endif
-
 long double floorl(long double x) {
     if (!signbit(x)) {
         return truncl(x);
@@ -17,5 +13,3 @@ long double floorl(long double x) {
     }
     return x;
 }
-
-long double _debug_floorl(long double) __attribute__((alias("floorl")));

src/libc/fmal.src

@@ -2,10 +2,19 @@
 
 	section	.text
 
-	public	_fmal, __debug_fmal
+	public	_fmal
 
-_fmal := _softfloat_mulAddF64
-__debug_fmal := _softfloat_mulAddF64
+_fmal:
 	; flags handled by softfloat
+	ld	iy, 0
+	add	iy, sp
+	ld	a, (iy + 28)
+	rlca
+	ld	(iy + 11), a	; signC
+	ld	a, (iy + 10)
+	xor	a, (iy + 19)
+	rlca
+	ld	(iy + 20), a	; signZ
+	jq	_softfloat_mulAddF64
 
 	extern	_softfloat_mulAddF64

src/libc/fmax.c

@@ -13,10 +13,6 @@ float fmaxf(float x, float y) {
 
 double fmax(double, double) __attribute__((alias("fmaxf")));
 
-#ifdef fmaxl
-#undef fmaxl
-#endif
-
 long double fmaxl(long double x, long double y) {
     return
         isless(x, y) ? y :
@@ -27,5 +23,3 @@ long double fmaxl(long double x, long double y) {
         /* arguments are equal or signed zero */
         signbit(x) ? y : x;
 }
-
-long double _debug_fmaxl(long double, long double) __attribute__((alias("fmaxl")));

src/libc/fmin.c

@@ -13,10 +13,6 @@ float fminf(float x, float y) {
 
 double fmin(double, double) __attribute__((alias("fminf")));
 
-#ifdef fminl
-#undef fminl
-#endif
-
 long double fminl(long double x, long double y) {
     return
         isless(x, y) ? x :
@@ -27,5 +23,3 @@ long double fminl(long double x, long double y) {
         /* arguments are equal or signed zero */
         signbit(x) ? x : y;
 }
-
-long double _debug_fminl(long double, long double) __attribute__((alias("fminl")));