1/*	$OpenBSD: fma_test.c,v 1.3 2021/12/13 18:04:28 deraadt Exp $	*/
2/*-
3 * Copyright (c) 2008 David Schultz <das@FreeBSD.org>
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include "macros.h"
29
30/*
31 * Tests for fma{,f,l}().
32 */
33
34#include <sys/types.h>
35#include <fenv.h>
36#include <float.h>
37#include <math.h>
38#include <stdio.h>
39#include <stdlib.h>
40
41#include "test-utils.h"
42
43#pragma STDC FENV_ACCESS ON
44
45/*
46 * Test that a function returns the correct value and sets the
47 * exception flags correctly. The exceptmask specifies which
48 * exceptions we should check. We need to be lenient for several
49 * reasons, but mainly because on some architectures it's impossible
50 * to raise FE_OVERFLOW without raising FE_INEXACT.
51 *
52 * These are macros instead of functions so that assert provides more
53 * meaningful error messages.
54 */
55#define	test(func, x, y, z, result, exceptmask, excepts) do {		\
56	volatile long double _vx = (x), _vy = (y), _vz = (z);		\
57	ATF_CHECK(feclearexcept(FE_ALL_EXCEPT) == 0);			\
58	CHECK_FPEQUAL((func)(_vx, _vy, _vz), (result));		\
59	CHECK_FP_EXCEPTIONS_MSG(excepts, exceptmask, "for %s(%s)",	\
60	    #func, #x);							\
61} while (0)
62
63#define	testall(x, y, z, result, exceptmask, excepts)	do {		\
64	test(fma, (double)(x), (double)(y), (double)(z),		\
65		(double)(result), (exceptmask), (excepts));		\
66	test(fmaf, (float)(x), (float)(y), (float)(z),			\
67		(float)(result), (exceptmask), (excepts));		\
68	test(fmal, (x), (y), (z), (result), (exceptmask), (excepts));	\
69} while (0)
70
71/* Test in all rounding modes. */
72#define	testrnd(func, x, y, z, rn, ru, rd, rz, exceptmask, excepts)	do { \
73	fesetround(FE_TONEAREST);					\
74	test((func), (x), (y), (z), (rn), (exceptmask), (excepts));	\
75	fesetround(FE_UPWARD);						\
76	test((func), (x), (y), (z), (ru), (exceptmask), (excepts));	\
77	fesetround(FE_DOWNWARD);					\
78	test((func), (x), (y), (z), (rd), (exceptmask), (excepts));	\
79	fesetround(FE_TOWARDZERO);					\
80	test((func), (x), (y), (z), (rz), (exceptmask), (excepts));	\
81} while (0)
82
83/*
84 * This is needed because clang constant-folds fma in ways that are incorrect
85 * in rounding modes other than FE_TONEAREST.
86 */
87static volatile double one = 1.0;
88
89static void
90test_zeroes(void)
91{
92	const int rd = (fegetround() == FE_DOWNWARD);
93
94	testall(0.0, 0.0, 0.0, 0.0, ALL_STD_EXCEPT, 0);
95	testall(1.0, 0.0, 0.0, 0.0, ALL_STD_EXCEPT, 0);
96	testall(0.0, 1.0, 0.0, 0.0, ALL_STD_EXCEPT, 0);
97	testall(0.0, 0.0, 1.0, 1.0, ALL_STD_EXCEPT, 0);
98
99	testall(-0.0, 0.0, 0.0, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
100	testall(0.0, -0.0, 0.0, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
101	testall(-0.0, -0.0, 0.0, 0.0, ALL_STD_EXCEPT, 0);
102	testall(0.0, 0.0, -0.0, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
103	testall(-0.0, -0.0, -0.0, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
104
105	testall(-0.0, 0.0, -0.0, -0.0, ALL_STD_EXCEPT, 0);
106	testall(0.0, -0.0, -0.0, -0.0, ALL_STD_EXCEPT, 0);
107
108	testall(-one, one, one, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
109	testall(one, -one, one, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
110	testall(-one, -one, -one, rd ? -0.0 : 0.0, ALL_STD_EXCEPT, 0);
111
112	switch (fegetround()) {
113	case FE_TONEAREST:
114	case FE_TOWARDZERO:
115		test(fmaf, -FLT_MIN, FLT_MIN, 0.0, -0.0,
116		     ALL_STD_EXCEPT, FE_INEXACT | FE_UNDERFLOW);
117		test(fma, -DBL_MIN, DBL_MIN, 0.0, -0.0,
118		     ALL_STD_EXCEPT, FE_INEXACT | FE_UNDERFLOW);
119		test(fmal, -LDBL_MIN, LDBL_MIN, 0.0, -0.0,
120		     ALL_STD_EXCEPT, FE_INEXACT | FE_UNDERFLOW);
121	}
122}
123
124static void
125test_infinities(void)
126{
127	testall(INFINITY, 1.0, -1.0, INFINITY, ALL_STD_EXCEPT, 0);
128	testall(-1.0, INFINITY, 0.0, -INFINITY, ALL_STD_EXCEPT, 0);
129	testall(0.0, 0.0, INFINITY, INFINITY, ALL_STD_EXCEPT, 0);
130	testall(1.0, 1.0, INFINITY, INFINITY, ALL_STD_EXCEPT, 0);
131	testall(1.0, 1.0, -INFINITY, -INFINITY, ALL_STD_EXCEPT, 0);
132
133	testall(INFINITY, -INFINITY, 1.0, -INFINITY, ALL_STD_EXCEPT, 0);
134	testall(INFINITY, INFINITY, 1.0, INFINITY, ALL_STD_EXCEPT, 0);
135	testall(-INFINITY, -INFINITY, INFINITY, INFINITY, ALL_STD_EXCEPT, 0);
136
137	testall(0.0, INFINITY, 1.0, NAN, ALL_STD_EXCEPT, FE_INVALID);
138	testall(INFINITY, 0.0, -0.0, NAN, ALL_STD_EXCEPT, FE_INVALID);
139
140	/* The invalid exception is optional in this case. */
141	testall(INFINITY, 0.0, NAN, NAN, ALL_STD_EXCEPT & ~FE_INVALID, 0);
142
143	testall(INFINITY, INFINITY, -INFINITY, NAN,
144		ALL_STD_EXCEPT, FE_INVALID);
145	testall(-INFINITY, INFINITY, INFINITY, NAN,
146		ALL_STD_EXCEPT, FE_INVALID);
147	testall(INFINITY, -1.0, INFINITY, NAN,
148		ALL_STD_EXCEPT, FE_INVALID);
149
150	test(fmaf, FLT_MAX, FLT_MAX, -INFINITY, -INFINITY, ALL_STD_EXCEPT, 0);
151	test(fma, DBL_MAX, DBL_MAX, -INFINITY, -INFINITY, ALL_STD_EXCEPT, 0);
152	test(fmal, LDBL_MAX, LDBL_MAX, -INFINITY, -INFINITY,
153	     ALL_STD_EXCEPT, 0);
154	test(fmaf, FLT_MAX, -FLT_MAX, INFINITY, INFINITY, ALL_STD_EXCEPT, 0);
155	test(fma, DBL_MAX, -DBL_MAX, INFINITY, INFINITY, ALL_STD_EXCEPT, 0);
156	test(fmal, LDBL_MAX, -LDBL_MAX, INFINITY, INFINITY,
157	     ALL_STD_EXCEPT, 0);
158}
159
160static void
161test_nans(void)
162{
163	testall(NAN, 0.0, 0.0, NAN, ALL_STD_EXCEPT, 0);
164	testall(1.0, NAN, 1.0, NAN, ALL_STD_EXCEPT, 0);
165	testall(1.0, -1.0, NAN, NAN, ALL_STD_EXCEPT, 0);
166	testall(0.0, 0.0, NAN, NAN, ALL_STD_EXCEPT, 0);
167	testall(NAN, NAN, NAN, NAN, ALL_STD_EXCEPT, 0);
168
169	/* x*y should not raise an inexact/overflow/underflow if z is NaN. */
170	testall(M_PI, M_PI, NAN, NAN, ALL_STD_EXCEPT, 0);
171	test(fmaf, FLT_MIN, FLT_MIN, NAN, NAN, ALL_STD_EXCEPT, 0);
172	test(fma, DBL_MIN, DBL_MIN, NAN, NAN, ALL_STD_EXCEPT, 0);
173	test(fmal, LDBL_MIN, LDBL_MIN, NAN, NAN, ALL_STD_EXCEPT, 0);
174	test(fmaf, FLT_MAX, FLT_MAX, NAN, NAN, ALL_STD_EXCEPT, 0);
175	test(fma, DBL_MAX, DBL_MAX, NAN, NAN, ALL_STD_EXCEPT, 0);
176	test(fmal, LDBL_MAX, LDBL_MAX, NAN, NAN, ALL_STD_EXCEPT, 0);
177}
178
179/*
180 * Tests for cases where z is very small compared to x*y.
181 */
182static void
183test_small_z(void)
184{
185	/* x*y positive, z positive */
186	if (fegetround() == FE_UPWARD) {
187		test(fmaf, one, one, 0x1.0p-100, 1.0 + FLT_EPSILON,
188		     ALL_STD_EXCEPT, FE_INEXACT);
189		test(fma, one, one, 0x1.0p-200, 1.0 + DBL_EPSILON,
190		     ALL_STD_EXCEPT, FE_INEXACT);
191		test(fmal, one, one, 0x1.0p-200, 1.0 + LDBL_EPSILON,
192		     ALL_STD_EXCEPT, FE_INEXACT);
193	} else {
194		testall(0x1.0p100, one, 0x1.0p-100, 0x1.0p100,
195			ALL_STD_EXCEPT, FE_INEXACT);
196	}
197
198	/* x*y negative, z negative */
199	if (fegetround() == FE_DOWNWARD) {
200		test(fmaf, -one, one, -0x1.0p-100, -(1.0 + FLT_EPSILON),
201		     ALL_STD_EXCEPT, FE_INEXACT);
202		test(fma, -one, one, -0x1.0p-200, -(1.0 + DBL_EPSILON),
203		     ALL_STD_EXCEPT, FE_INEXACT);
204		test(fmal, -one, one, -0x1.0p-200, -(1.0 + LDBL_EPSILON),
205		     ALL_STD_EXCEPT, FE_INEXACT);
206	} else {
207		testall(0x1.0p100, -one, -0x1.0p-100, -0x1.0p100,
208			ALL_STD_EXCEPT, FE_INEXACT);
209	}
210
211	/* x*y positive, z negative */
212	if (fegetround() == FE_DOWNWARD || fegetround() == FE_TOWARDZERO) {
213		test(fmaf, one, one, -0x1.0p-100, 1.0 - FLT_EPSILON / 2,
214		     ALL_STD_EXCEPT, FE_INEXACT);
215		test(fma, one, one, -0x1.0p-200, 1.0 - DBL_EPSILON / 2,
216		     ALL_STD_EXCEPT, FE_INEXACT);
217		test(fmal, one, one, -0x1.0p-200, 1.0 - LDBL_EPSILON / 2,
218		     ALL_STD_EXCEPT, FE_INEXACT);
219	} else {
220		testall(0x1.0p100, one, -0x1.0p-100, 0x1.0p100,
221			ALL_STD_EXCEPT, FE_INEXACT);
222	}
223
224	/* x*y negative, z positive */
225	if (fegetround() == FE_UPWARD || fegetround() == FE_TOWARDZERO) {
226		test(fmaf, -one, one, 0x1.0p-100, -1.0 + FLT_EPSILON / 2,
227		     ALL_STD_EXCEPT, FE_INEXACT);
228		test(fma, -one, one, 0x1.0p-200, -1.0 + DBL_EPSILON / 2,
229		     ALL_STD_EXCEPT, FE_INEXACT);
230		test(fmal, -one, one, 0x1.0p-200, -1.0 + LDBL_EPSILON / 2,
231		     ALL_STD_EXCEPT, FE_INEXACT);
232	} else {
233		testall(-0x1.0p100, one, 0x1.0p-100, -0x1.0p100,
234			ALL_STD_EXCEPT, FE_INEXACT);
235	}
236}
237
238/*
239 * Tests for cases where z is very large compared to x*y.
240 */
241static void
242test_big_z(void)
243{
244	/* z positive, x*y positive */
245	if (fegetround() == FE_UPWARD) {
246		test(fmaf, 0x1.0p-50, 0x1.0p-50, 1.0, 1.0 + FLT_EPSILON,
247		     ALL_STD_EXCEPT, FE_INEXACT);
248		test(fma, 0x1.0p-100, 0x1.0p-100, 1.0, 1.0 + DBL_EPSILON,
249		     ALL_STD_EXCEPT, FE_INEXACT);
250		test(fmal, 0x1.0p-100, 0x1.0p-100, 1.0, 1.0 + LDBL_EPSILON,
251		     ALL_STD_EXCEPT, FE_INEXACT);
252	} else {
253		testall(-0x1.0p-50, -0x1.0p-50, 0x1.0p100, 0x1.0p100,
254			ALL_STD_EXCEPT, FE_INEXACT);
255	}
256
257	/* z negative, x*y negative */
258	if (fegetround() == FE_DOWNWARD) {
259		test(fmaf, -0x1.0p-50, 0x1.0p-50, -1.0, -(1.0 + FLT_EPSILON),
260		     ALL_STD_EXCEPT, FE_INEXACT);
261		test(fma, -0x1.0p-100, 0x1.0p-100, -1.0, -(1.0 + DBL_EPSILON),
262		     ALL_STD_EXCEPT, FE_INEXACT);
263		test(fmal, -0x1.0p-100, 0x1.0p-100, -1.0, -(1.0 + LDBL_EPSILON),
264		     ALL_STD_EXCEPT, FE_INEXACT);
265	} else {
266		testall(0x1.0p-50, -0x1.0p-50, -0x1.0p100, -0x1.0p100,
267			ALL_STD_EXCEPT, FE_INEXACT);
268	}
269
270	/* z negative, x*y positive */
271	if (fegetround() == FE_UPWARD || fegetround() == FE_TOWARDZERO) {
272		test(fmaf, -0x1.0p-50, -0x1.0p-50, -1.0,
273		     -1.0 + FLT_EPSILON / 2, ALL_STD_EXCEPT, FE_INEXACT);
274		test(fma, -0x1.0p-100, -0x1.0p-100, -1.0,
275		     -1.0 + DBL_EPSILON / 2, ALL_STD_EXCEPT, FE_INEXACT);
276		test(fmal, -0x1.0p-100, -0x1.0p-100, -1.0,
277		     -1.0 + LDBL_EPSILON / 2, ALL_STD_EXCEPT, FE_INEXACT);
278	} else {
279		testall(0x1.0p-50, 0x1.0p-50, -0x1.0p100, -0x1.0p100,
280			ALL_STD_EXCEPT, FE_INEXACT);
281	}
282
283	/* z positive, x*y negative */
284	if (fegetround() == FE_DOWNWARD || fegetround() == FE_TOWARDZERO) {
285		test(fmaf, 0x1.0p-50, -0x1.0p-50, 1.0, 1.0 - FLT_EPSILON / 2,
286		     ALL_STD_EXCEPT, FE_INEXACT);
287		test(fma, 0x1.0p-100, -0x1.0p-100, 1.0, 1.0 - DBL_EPSILON / 2,
288		     ALL_STD_EXCEPT, FE_INEXACT);
289		test(fmal, 0x1.0p-100, -0x1.0p-100, 1.0, 1.0 - LDBL_EPSILON / 2,
290		     ALL_STD_EXCEPT, FE_INEXACT);
291	} else {
292		testall(-0x1.0p-50, 0x1.0p-50, 0x1.0p100, 0x1.0p100,
293			ALL_STD_EXCEPT, FE_INEXACT);
294	}
295}
296
297static void
298test_accuracy(void)
299{
300
301	/* ilogb(x*y) - ilogb(z) = 20 */
302	testrnd(fmaf, -0x1.c139d8p-51, -0x1.600e7ap32, 0x1.26558cp-38,
303		0x1.34e48ap-18, 0x1.34e48cp-18, 0x1.34e48ap-18, 0x1.34e48ap-18,
304		ALL_STD_EXCEPT, FE_INEXACT);
305	testrnd(fma, -0x1.c139d7b84f1a3p-51, -0x1.600e7a2a16484p32,
306		0x1.26558cac31580p-38, 0x1.34e48a78aae97p-18,
307		0x1.34e48a78aae97p-18, 0x1.34e48a78aae96p-18,
308		0x1.34e48a78aae96p-18, ALL_STD_EXCEPT, FE_INEXACT);
309#if LDBL_MANT_DIG == 113
310	testrnd(fmal, -0x1.c139d7b84f1a3079263afcc5bae3p-51L,
311		-0x1.600e7a2a164840edbe2e7d301a72p32L,
312		0x1.26558cac315807eb07e448042101p-38L,
313		0x1.34e48a78aae96c76ed36077dd387p-18L,
314		0x1.34e48a78aae96c76ed36077dd388p-18L,
315		0x1.34e48a78aae96c76ed36077dd387p-18L,
316		0x1.34e48a78aae96c76ed36077dd387p-18L,
317		ALL_STD_EXCEPT, FE_INEXACT);
318#elif LDBL_MANT_DIG == 64
319	testrnd(fmal, -0x1.c139d7b84f1a307ap-51L, -0x1.600e7a2a164840eep32L,
320		0x1.26558cac315807ecp-38L, 0x1.34e48a78aae96c78p-18L,
321		0x1.34e48a78aae96c78p-18L, 0x1.34e48a78aae96c76p-18L,
322		0x1.34e48a78aae96c76p-18L, ALL_STD_EXCEPT, FE_INEXACT);
323#elif LDBL_MANT_DIG == 53
324	testrnd(fmal, -0x1.c139d7b84f1a3p-51L, -0x1.600e7a2a16484p32L,
325		0x1.26558cac31580p-38L, 0x1.34e48a78aae97p-18L,
326		0x1.34e48a78aae97p-18L, 0x1.34e48a78aae96p-18L,
327		0x1.34e48a78aae96p-18L, ALL_STD_EXCEPT, FE_INEXACT);
328#endif
329
330	/* ilogb(x*y) - ilogb(z) = -40 */
331	testrnd(fmaf, 0x1.98210ap53, 0x1.9556acp-24, 0x1.d87da4p70,
332		0x1.d87da4p70, 0x1.d87da6p70, 0x1.d87da4p70, 0x1.d87da4p70,
333		ALL_STD_EXCEPT, FE_INEXACT);
334	testrnd(fma, 0x1.98210ac83fe2bp53, 0x1.9556ac1475f0fp-24,
335		0x1.d87da3aafc60ep70, 0x1.d87da3aafda40p70,
336		0x1.d87da3aafda40p70, 0x1.d87da3aafda3fp70,
337		0x1.d87da3aafda3fp70, ALL_STD_EXCEPT, FE_INEXACT);
338#if LDBL_MANT_DIG == 113
339	testrnd(fmal, 0x1.98210ac83fe2a8f65b6278b74cebp53L,
340		0x1.9556ac1475f0f28968b61d0de65ap-24L,
341		0x1.d87da3aafc60d830aa4c6d73b749p70L,
342		0x1.d87da3aafda3f36a69eb86488224p70L,
343		0x1.d87da3aafda3f36a69eb86488225p70L,
344		0x1.d87da3aafda3f36a69eb86488224p70L,
345		0x1.d87da3aafda3f36a69eb86488224p70L,
346		ALL_STD_EXCEPT, FE_INEXACT);
347#elif LDBL_MANT_DIG == 64
348	testrnd(fmal, 0x1.98210ac83fe2a8f6p53L, 0x1.9556ac1475f0f28ap-24L,
349		0x1.d87da3aafc60d83p70L, 0x1.d87da3aafda3f36ap70L,
350		0x1.d87da3aafda3f36ap70L, 0x1.d87da3aafda3f368p70L,
351		0x1.d87da3aafda3f368p70L, ALL_STD_EXCEPT, FE_INEXACT);
352#elif LDBL_MANT_DIG == 53
353	testrnd(fmal, 0x1.98210ac83fe2bp53L, 0x1.9556ac1475f0fp-24L,
354		0x1.d87da3aafc60ep70L, 0x1.d87da3aafda40p70L,
355		0x1.d87da3aafda40p70L, 0x1.d87da3aafda3fp70L,
356		0x1.d87da3aafda3fp70L, ALL_STD_EXCEPT, FE_INEXACT);
357#endif
358
359	/* ilogb(x*y) - ilogb(z) = 0 */
360	testrnd(fmaf, 0x1.31ad02p+100, 0x1.2fbf7ap-42, -0x1.c3e106p+58,
361		-0x1.64c27cp+56, -0x1.64c27ap+56, -0x1.64c27cp+56,
362		-0x1.64c27ap+56, ALL_STD_EXCEPT, FE_INEXACT);
363	testrnd(fma, 0x1.31ad012ede8aap+100, 0x1.2fbf79c839067p-42,
364		-0x1.c3e106929056ep+58, -0x1.64c282b970a5fp+56,
365		-0x1.64c282b970a5ep+56, -0x1.64c282b970a5fp+56,
366		-0x1.64c282b970a5ep+56, ALL_STD_EXCEPT, FE_INEXACT);
367#if LDBL_MANT_DIG == 113
368	testrnd(fmal, 0x1.31ad012ede8aa282fa1c19376d16p+100L,
369		 0x1.2fbf79c839066f0f5c68f6d2e814p-42L,
370		-0x1.c3e106929056ec19de72bfe64215p+58L,
371		-0x1.64c282b970a612598fc025ca8cddp+56L,
372		-0x1.64c282b970a612598fc025ca8cddp+56L,
373		-0x1.64c282b970a612598fc025ca8cdep+56L,
374		-0x1.64c282b970a612598fc025ca8cddp+56L,
375		ALL_STD_EXCEPT, FE_INEXACT);
376#elif LDBL_MANT_DIG == 64
377	testrnd(fmal, 0x1.31ad012ede8aa4eap+100L, 0x1.2fbf79c839066aeap-42L,
378		-0x1.c3e106929056e61p+58L, -0x1.64c282b970a60298p+56L,
379		-0x1.64c282b970a60298p+56L, -0x1.64c282b970a6029ap+56L,
380		-0x1.64c282b970a60298p+56L, ALL_STD_EXCEPT, FE_INEXACT);
381#elif LDBL_MANT_DIG == 53
382	testrnd(fmal, 0x1.31ad012ede8aap+100L, 0x1.2fbf79c839067p-42L,
383		-0x1.c3e106929056ep+58L, -0x1.64c282b970a5fp+56L,
384		-0x1.64c282b970a5ep+56L, -0x1.64c282b970a5fp+56L,
385		-0x1.64c282b970a5ep+56L, ALL_STD_EXCEPT, FE_INEXACT);
386#endif
387
388	/* x*y (rounded) ~= -z */
389	/* XXX spurious inexact exceptions */
390	testrnd(fmaf, 0x1.bbffeep-30, -0x1.1d164cp-74, 0x1.ee7296p-104,
391		-0x1.c46ea8p-128, -0x1.c46ea8p-128, -0x1.c46ea8p-128,
392		-0x1.c46ea8p-128, ALL_STD_EXCEPT & ~FE_INEXACT, 0);
393	testrnd(fma, 0x1.bbffeea6fc7d6p-30, 0x1.1d164c6cbf078p-74,
394		-0x1.ee72993aff948p-104, -0x1.71f72ac7d9d8p-159,
395		-0x1.71f72ac7d9d8p-159, -0x1.71f72ac7d9d8p-159,
396		-0x1.71f72ac7d9d8p-159, ALL_STD_EXCEPT & ~FE_INEXACT, 0);
397#if LDBL_MANT_DIG == 113
398	testrnd(fmal, 0x1.bbffeea6fc7d65927d147f437675p-30L,
399		0x1.1d164c6cbf078b7a22607d1cd6a2p-74L,
400		-0x1.ee72993aff94973876031bec0944p-104L,
401		0x1.64e086175b3a2adc36e607058814p-217L,
402		0x1.64e086175b3a2adc36e607058814p-217L,
403		0x1.64e086175b3a2adc36e607058814p-217L,
404		0x1.64e086175b3a2adc36e607058814p-217L,
405		ALL_STD_EXCEPT & ~FE_INEXACT, 0);
406#elif LDBL_MANT_DIG == 64
407	testrnd(fmal, 0x1.bbffeea6fc7d6592p-30L, 0x1.1d164c6cbf078b7ap-74L,
408		-0x1.ee72993aff949736p-104L, 0x1.af190e7a1ee6ad94p-168L,
409		0x1.af190e7a1ee6ad94p-168L, 0x1.af190e7a1ee6ad94p-168L,
410		0x1.af190e7a1ee6ad94p-168L, ALL_STD_EXCEPT & ~FE_INEXACT, 0);
411#elif LDBL_MANT_DIG == 53
412	testrnd(fmal, 0x1.bbffeea6fc7d6p-30L, 0x1.1d164c6cbf078p-74L,
413		-0x1.ee72993aff948p-104L, -0x1.71f72ac7d9d8p-159L,
414		-0x1.71f72ac7d9d8p-159L, -0x1.71f72ac7d9d8p-159L,
415		-0x1.71f72ac7d9d8p-159L, ALL_STD_EXCEPT & ~FE_INEXACT, 0);
416#endif
417}
418
419static void
420test_double_rounding(void)
421{
422
423	/*
424	 *     a =  0x1.8000000000001p0
425	 *     b =  0x1.8000000000001p0
426	 *     c = -0x0.0000000000000000000000000080...1p+1
427	 * a * b =  0x1.2000000000001800000000000080p+1
428	 *
429	 * The correct behavior is to round DOWN to 0x1.2000000000001p+1 in
430	 * round-to-nearest mode.  An implementation that computes a*b+c in
431	 * double+double precision, however, will get 0x1.20000000000018p+1,
432	 * and then round UP.
433	 */
434	fesetround(FE_TONEAREST);
435	test(fma, 0x1.8000000000001p0, 0x1.8000000000001p0,
436	     -0x1.0000000000001p-104, 0x1.2000000000001p+1,
437	     ALL_STD_EXCEPT, FE_INEXACT);
438	fesetround(FE_DOWNWARD);
439	test(fma, 0x1.8000000000001p0, 0x1.8000000000001p0,
440	     -0x1.0000000000001p-104, 0x1.2000000000001p+1,
441	     ALL_STD_EXCEPT, FE_INEXACT);
442	fesetround(FE_UPWARD);
443	test(fma, 0x1.8000000000001p0, 0x1.8000000000001p0,
444	     -0x1.0000000000001p-104, 0x1.2000000000002p+1,
445	     ALL_STD_EXCEPT, FE_INEXACT);
446
447	fesetround(FE_TONEAREST);
448	test(fmaf, 0x1.800002p+0, 0x1.800002p+0, -0x1.000002p-46, 0x1.200002p+1,
449	     ALL_STD_EXCEPT, FE_INEXACT);
450	fesetround(FE_DOWNWARD);
451	test(fmaf, 0x1.800002p+0, 0x1.800002p+0, -0x1.000002p-46, 0x1.200002p+1,
452	     ALL_STD_EXCEPT, FE_INEXACT);
453	fesetround(FE_UPWARD);
454	test(fmaf, 0x1.800002p+0, 0x1.800002p+0, -0x1.000002p-46, 0x1.200004p+1,
455	     ALL_STD_EXCEPT, FE_INEXACT);
456
457	fesetround(FE_TONEAREST);
458#if LDBL_MANT_DIG == 64
459	test(fmal, 0x1.4p+0L, 0x1.0000000000000004p+0L, 0x1p-128L,
460	     0x1.4000000000000006p+0L, ALL_STD_EXCEPT, FE_INEXACT);
461#elif LDBL_MANT_DIG == 113
462	test(fmal, 0x1.8000000000000000000000000001p+0L,
463	     0x1.8000000000000000000000000001p+0L,
464	     -0x1.0000000000000000000000000001p-224L,
465	     0x1.2000000000000000000000000001p+1L, ALL_STD_EXCEPT, FE_INEXACT);
466#endif
467
468}
469
470static const int rmodes[] = {
471	FE_TONEAREST, FE_UPWARD, FE_DOWNWARD, FE_TOWARDZERO
472};
473
474ATF_TC_WITHOUT_HEAD(zeroes);
475ATF_TC_BODY(zeroes, tc)
476{
477	size_t i;
478	for (i = 0; i < nitems(rmodes); i++) {
479		printf("rmode = %d\n", rmodes[i]);
480		fesetround(rmodes[i]);
481		test_zeroes();
482	}
483}
484
485ATF_TC_WITHOUT_HEAD(infinities);
486ATF_TC_BODY(infinities, tc)
487{
488	size_t i;
489	for (i = 0; i < nitems(rmodes); i++) {
490		printf("rmode = %d\n", rmodes[i]);
491		fesetround(rmodes[i]);
492		test_infinities();
493	}
494}
495
496ATF_TC_WITHOUT_HEAD(nans);
497ATF_TC_BODY(nans, tc)
498{
499	fesetround(FE_TONEAREST);
500	test_nans();
501}
502
503
504ATF_TC_WITHOUT_HEAD(small_z);
505ATF_TC_BODY(small_z, tc)
506{
507	size_t i;
508	for (i = 0; i < nitems(rmodes); i++) {
509		printf("rmode = %d\n", rmodes[i]);
510		fesetround(rmodes[i]);
511		test_small_z();
512	}
513}
514
515
516ATF_TC_WITHOUT_HEAD(big_z);
517ATF_TC_BODY(big_z, tc)
518{
519	size_t i;
520	for (i = 0; i < nitems(rmodes); i++) {
521		printf("rmode = %d\n", rmodes[i]);
522		fesetround(rmodes[i]);
523		test_big_z();
524	}
525}
526
527ATF_TC_WITHOUT_HEAD(accuracy);
528ATF_TC_BODY(accuracy, tc)
529{
530	fesetround(FE_TONEAREST);
531	test_accuracy();
532}
533
534ATF_TC_WITHOUT_HEAD(double_rounding);
535ATF_TC_BODY(double_rounding, tc) {
536	test_double_rounding();
537}
538
539ATF_TP_ADD_TCS(tp)
540{
541	ATF_TP_ADD_TC(tp, zeroes);
542	ATF_TP_ADD_TC(tp, infinities);
543	ATF_TP_ADD_TC(tp, nans);
544	ATF_TP_ADD_TC(tp, small_z);
545	ATF_TP_ADD_TC(tp, big_z);
546	ATF_TP_ADD_TC(tp, accuracy);
547	ATF_TP_ADD_TC(tp, double_rounding);
548	/*
549	 * TODO:
550	 * - Tests for subnormals
551	 * - Cancellation tests (e.g., z = (double)x*y, but x*y is inexact)
552	 */
553	return (atf_no_error());
554}
555