implemented truncf and modff in assembly, added log1pl and expm1l

src/libc/acoshl.c

@@ -1,9 +1,6 @@
 #include <math.h>
 #include "__float64_constants.h"
 
-/* enable when log1pl is implemented */
-#if 0
-
 /**
  * @remarks Minimum relative precision of:
  * 2^-42.37 at +1.000000358e+00 with ideal log1pl
@@ -15,14 +12,3 @@ long double acoshl(long double x) {
     }
     return logl(x) + F64_LN2;
 }
-
-#else
-
-long double acoshl(long double x) {
-    if (x < 0x1.0p+511L) {
-        return logl(x + sqrtl(x * x - 1.0L));
-    }
-    return logl(x) + F64_LN2;
-}
-
-#endif

src/libc/atanhl.c

@@ -1,23 +1,13 @@
 #include <math.h>
 
-/* enable when log1pl is implemented */
-#if 0
-
 /**
- * @remarks Relative precision of:
- * 2^-51 at +1.277396381e-01 with ideal log1pl
+ * @remarks Minimum ulp:
+ * ulp of -2    at +0x1.cf33c905a6324p-4 with ideal log1pl
+ * ulp of +4098 at +0x1.ffff1c95280b8p-15 with current log1pl and ideal logl
+ * ulp of +4135 at -0x1.f319bfec310c4p-14 with current log1pl and logl
  */
 long double atanhl(long double arg) {
     long double x = fabsl(arg);
     x = 0.5L * log1pl((2.0L * x) / (1.0L - x));
     return copysignl(x, arg);
 }
-
-#else
-
-long double atanhl(long double x)
-{
-    return 0.5L * logl((1.0L + x) / (1.0L - x));
-}
-
-#endif

src/libc/ceilf.src

@@ -30,7 +30,6 @@ _ceilf:
 	ld	e,a
 	push	bc
 	pop	hl
-	ld	sp,ix
 	pop	ix
 	ret
 

src/libc/expm1l.c

@@ -0,0 +1,14 @@
+#include <math.h>
+#include "__float64_constants.h"
+
+/**
+ * @remarks Approximate minimum ulp:
+ * ulp of -4098 at +0x1.3683c27b9c856p-13 with ideal expl
+ */
+long double expm1l(long double x) {
+    if (fabsl(x) < 0x1.0p-13L) {
+        long double x_sqr = x * x;
+        return (x_sqr * x) * F64_1_DIV_3 + 0.5L * (x_sqr) + x;
+    }
+    return expl(x) - 1.0L;
+}

src/libc/log1pl.c

@@ -0,0 +1,14 @@
+#include <math.h>
+#include "__float64_constants.h"
+
+/**
+ * @remarks Approximate minimum ulp:
+ * ulp of +4097 at +0x1.0003239899b0ap-13 with ideal expl
+ */
+long double log1pl(long double x) {
+    if (fabsl(x) < 0x1.2p-13L) {
+        long double x_sqr = x * x;
+        return (x_sqr * x) * F64_1_DIV_3 - 0.5L * (x_sqr) + x;
+    }
+    return logl(1.0L + x);
+}

src/libc/modff.src

@@ -11,10 +11,91 @@ _modf := _modff
 
 else
 
-_modff := __modff_c
-_modf := __modf_c
+; float modf(float x, float* iptr)
+_modff:
+_modf:
+	call	__frameset0
 
-	extern	__modff_c
-	extern	__modf_c
+	; copy the mantissa into HL and BC
+	sbc	hl, hl
+	ld	bc, (ix + 6) ; mantissa
+	add	hl, bc
+	adc	hl, hl
+	ld	a, (ix + 9) ; exponent
+	ld	e, a
+	rla
+	ld	d, a
+	jr	z, .maybe_inf
+	inc	a
+	ld	a, e
+	jr	z, .ret_nan
+.not_inf:
+	ld	a, d
+	sub	a, 127	; |x| < 1.0f
+	jr	c, .less_than_one
+	; fintie and normal
+	ld	hl, (ix + 6) ; mantissa
+	; E has exponent
+	push	de
+	push	hl
+	call	_truncf
+	pop	af, af
+
+	ld	a, e
+	ex	hl, de
+	ld	hl, (ix + 12)	; float *iptr
+	ld	(hl), de
+	inc	hl
+	inc	hl
+	inc	hl
+	ld	(hl), a
+	ex	hl, de
+	ld	e, a
+	ld	bc, (ix + 6)
+	ld	a, (ix + 9)
+
+	call	__fsub
+	push	bc
+	pop	hl
+	ld	e, a
+	pop	ix
+	ret
+
+.maybe_inf:
+	inc	a
+	jr	nz, .not_inf
+	ld	a, e
+	ld	hl, (ix + 12)	; float *iptr
+	ld	(hl), bc
+	inc	hl
+	inc	hl
+	inc	hl
+	ld	(hl), a	; inf exp
+	xor	a, a
+	ld	e, a
+	sbc	hl, hl
+	pop	ix
+	ret
+
+.less_than_one:
+	ld	bc, 0
+	ld	a, e
+	and	a, $80
+	; A:UBC = copysignf(0.0f, x)
+.ret_nan:
+	ld	hl, (ix + 12)	; float *iptr
+	ld	(hl), bc
+	inc	hl
+	inc	hl
+	inc	hl
+	ld	(hl), a
+	ld	hl, (ix + 6)	; mantissa
+	; E has exponent
+	pop	ix
+	ret
+
+	extern	__frameset0
+	extern	_truncf
+	extern	__fsub
 
 end if

src/libc/truncf.src

@@ -0,0 +1,46 @@
+	assume	adl=1
+
+	section	.text
+
+	public	_truncf, _trunc
+
+_trunc:
+_truncf:
+	ld	iy, 0
+	add	iy, sp
+	ld	e, (iy + 6)
+	ld	bc, (iy + 3)
+	sbc	hl, hl
+	add	hl, bc	; copy BC to HL
+	add	hl, hl
+	ld	a, e
+	rla
+
+	sub	a, 127	; exponent bias
+	jr	c, .ret_zero
+	sub	a, 23	; bits in the mantissa
+	jr	nc, .ret_self
+	sbc	hl, hl	; HL = 0xFFFFFF
+	neg
+	ld	d, c	; store C
+	ld	c, a
+	call	__ishl
+	ld	c, d	; restore C
+	jp	__iand
+
+.ret_self:
+	; already (large) integer, inf, or NaN
+	sbc	hl, hl
+	add	hl, bc
+	ret
+
+.ret_zero:
+	; return signed zero
+	ld	a, e
+	and	a, $80
+	ld	e, a
+	sbc	hl, hl
+	ret
+
+	extern	__ishl
+	extern	__iand

test/floating_point/float32_modf/autotest.json

@@ -0,0 +1,43 @@
+{
+	"transfer_files": [
+	  "bin/DEMO.8xp"
+	],
+	"target": {
+	  "name": "DEMO",
+	  "isASM": true
+	},
+	"sequence": [
+	  "action|launch",
+	  "delay|1000",
+	  "hashWait|1",
+	  "key|enter",
+	  "delay|300",
+	  "hashWait|2"
+	],
+	"hashes": {
+	  "1": {
+		"description": "All tests passed or GDB1 error",
+		"timeout": 5000,
+		"start": "vram_start",
+		"size": "vram_16_size",
+		"expected_CRCs": [
+		  "38E2AD5A",
+		  "2C812DC2"
+		]
+	  },
+	  "2": {
+		"description": "Exit or GDB1 error",
+		"start": "vram_start",
+		"size": "vram_16_size",
+		"expected_CRCs": [
+		  "FFAF89BA",
+		  "101734A5",
+		  "9DA19F44",
+		  "A32840C8",
+		  "349F4775",
+		  "271A9FBF",
+		  "82FD0B1E"
+		]
+	  }
+	}
+}

test/floating_point/float32_modf/makefile

@@ -0,0 +1,17 @@
+# ----------------------------
+# Makefile Options
+# ----------------------------
+
+NAME = DEMO
+ICON = icon.png
+DESCRIPTION = "CE C Toolchain Demo"
+COMPRESSED = NO
+
+CFLAGS = -Wall -Wextra -Wshadow -Wfloat-conversion -Wimplicit-float-conversion -Wimplicit-int-float-conversion -Oz -std=c11
+CXXFLAGS = -Wall -Wextra -Wshadow -Wfloat-conversion -Wimplicit-float-conversion -Wimplicit-int-float-conversion -Oz -std=c++11
+
+PREFER_OS_LIBC = NO
+
+# ----------------------------
+
+include $(shell cedev-config --makefile)