• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*  Copyright (C) 2011 IBM
2 
3  Author: Maynard Johnson <maynardj@us.ibm.com>
4 
5  This program is free software; you can redistribute it and/or
6  modify it under the terms of the GNU General Public License as
7  published by the Free Software Foundation; either version 2 of the
8  License, or (at your option) any later version.
9 
10  This program is distributed in the hope that it will be useful, but
11  WITHOUT ANY WARRANTY; without even the implied warranty of
12  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  General Public License for more details.
14 
15  You should have received a copy of the GNU General Public License
16  along with this program; if not, write to the Free Software
17  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
18  02111-1307, USA.
19 
20  The GNU General Public License is contained in the file COPYING.
21  */
22 
23 #ifdef HAS_VSX
24 
25 #include <stdio.h>
26 #include <stdint.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <malloc.h>
30 #include <altivec.h>
31 #include <math.h>
32 
33 #ifndef __powerpc64__
34 typedef uint32_t HWord_t;
35 #else
36 typedef uint64_t HWord_t;
37 #endif /* __powerpc64__ */
38 
39 typedef unsigned char Bool;
40 #define True 1
41 #define False 0
42 
43 #ifdef VGP_ppc64le_linux
44 #define isLE 1
45 #else
46 #define isLE 0
47 #endif
48 
49 register HWord_t r14 __asm__ ("r14");
50 register HWord_t r15 __asm__ ("r15");
51 register HWord_t r16 __asm__ ("r16");
52 register HWord_t r17 __asm__ ("r17");
53 register double f14 __asm__ ("fr14");
54 register double f15 __asm__ ("fr15");
55 register double f16 __asm__ ("fr16");
56 register double f17 __asm__ ("fr17");
57 
58 static volatile unsigned int div_flags, div_xer;
59 
60 #define ALLCR "cr0","cr1","cr2","cr3","cr4","cr5","cr6","cr7"
61 
62 #define SET_CR(_arg) \
63       __asm__ __volatile__ ("mtcr  %0" : : "b"(_arg) : ALLCR );
64 
65 #define SET_XER(_arg) \
66       __asm__ __volatile__ ("mtxer %0" : : "b"(_arg) : "xer" );
67 
68 #define GET_CR(_lval) \
69       __asm__ __volatile__ ("mfcr %0"  : "=b"(_lval) )
70 
71 #define GET_XER(_lval) \
72       __asm__ __volatile__ ("mfxer %0" : "=b"(_lval) )
73 
74 #define GET_CR_XER(_lval_cr,_lval_xer) \
75    do { GET_CR(_lval_cr); GET_XER(_lval_xer); } while (0)
76 
77 #define SET_CR_ZERO \
78       SET_CR(0)
79 
80 #define SET_XER_ZERO \
81       SET_XER(0)
82 
83 #define SET_CR_XER_ZERO \
84    do { SET_CR_ZERO; SET_XER_ZERO; } while (0)
85 
86 #define SET_FPSCR_ZERO \
87    do { double _d = 0.0; \
88         __asm__ __volatile__ ("mtfsf 0xFF, %0" : : "f"(_d) ); \
89    } while (0)
90 
91 
92 typedef void (*test_func_t)(void);
93 typedef struct test_table test_table_t;
94 
95 
96 /* These functions below that construct a table of floating point
97  * values were lifted from none/tests/ppc32/jm-insns.c.
98  */
99 
100 #if defined (DEBUG_ARGS_BUILD)
101 #define AB_DPRINTF(fmt, args...) do { fprintf(stderr, fmt , ##args); } while (0)
102 #else
103 #define AB_DPRINTF(fmt, args...) do { } while (0)
104 #endif
105 
register_farg(void * farg,int s,uint16_t _exp,uint64_t mant)106 static inline void register_farg (void *farg,
107                                   int s, uint16_t _exp, uint64_t mant)
108 {
109    uint64_t tmp;
110 
111    tmp = ((uint64_t)s << 63) | ((uint64_t)_exp << 52) | mant;
112    *(uint64_t *)farg = tmp;
113    AB_DPRINTF("%d %03x %013llx => %016llx %0e\n",
114               s, _exp, mant, *(uint64_t *)farg, *(double *)farg);
115 }
116 
register_sp_farg(void * farg,int s,uint16_t _exp,uint32_t mant)117 static inline void register_sp_farg (void *farg,
118                                      int s, uint16_t _exp, uint32_t mant)
119 {
120    uint32_t tmp;
121    tmp = ((uint32_t)s << 31) | ((uint32_t)_exp << 23) | mant;
122    *(uint32_t *)farg = tmp;
123 }
124 
125 typedef struct fp_test_args {
126    int fra_idx;
127    int frb_idx;
128 } fp_test_args_t;
129 
130 
131 fp_test_args_t fp_cmp_tests[] = {
132                                    {8, 8},
133                                    {8, 14},
134                                    {8, 6},
135                                    {8, 5},
136                                    {8, 4},
137                                    {8, 7},
138                                    {8, 9},
139                                    {8, 11},
140                                    {14, 8},
141                                    {14, 14},
142                                    {14, 6},
143                                    {14, 5},
144                                    {14, 4},
145                                    {14, 7},
146                                    {14, 9},
147                                    {14, 11},
148                                    {6, 8},
149                                    {6, 14},
150                                    {6, 6},
151                                    {6, 5},
152                                    {6, 4},
153                                    {6, 7},
154                                    {6, 9},
155                                    {6, 11},
156                                    {5, 8},
157                                    {5, 14},
158                                    {5, 6},
159                                    {5, 5},
160                                    {5, 4},
161                                    {5, 7},
162                                    {5, 9},
163                                    {5, 11},
164                                    {4, 8},
165                                    {4, 14},
166                                    {4, 6},
167                                    {4, 5},
168                                    {4, 1},
169                                    {4, 7},
170                                    {4, 9},
171                                    {4, 11},
172                                    {7, 8},
173                                    {7, 14},
174                                    {7, 6},
175                                    {7, 5},
176                                    {7, 4},
177                                    {7, 7},
178                                    {7, 9},
179                                    {7, 11},
180                                    {10, 8},
181                                    {10, 14},
182                                    {10, 6},
183                                    {10, 5},
184                                    {10, 4},
185                                    {10, 7},
186                                    {10, 9},
187                                    {10, 10},
188                                    {12, 8},
189                                    {12, 14},
190                                    {12, 6},
191                                    {12, 5},
192                                    {1, 1},
193                                    {2, 2},
194                                    {3, 3},
195                                    {4, 4},
196 };
197 
198 
199 fp_test_args_t two_arg_fp_tests[] = {
200                                      {8, 8},
201                                      {8, 14},
202                                      {15, 16},
203                                      {8, 5},
204                                      {8, 4},
205                                      {8, 7},
206                                      {8, 9},
207                                      {8, 11},
208                                      {14, 8},
209                                      {14, 14},
210                                      {14, 6},
211                                      {14, 5},
212                                      {14, 4},
213                                      {14, 7},
214                                      {14, 9},
215                                      {14, 11},
216                                      {6, 8},
217                                      {6, 14},
218                                      {6, 6},
219                                      {6, 5},
220                                      {6, 4},
221                                      {6, 7},
222                                      {6, 9},
223                                      {6, 11},
224                                      {5, 8},
225                                      {5, 14},
226                                      {5, 6},
227                                      {5, 5},
228                                      {5, 4},
229                                      {5, 7},
230                                      {5, 9},
231                                      {5, 11},
232                                      {4, 8},
233                                      {4, 14},
234                                      {4, 6},
235                                      {4, 5},
236                                      {4, 1},
237                                      {4, 7},
238                                      {4, 9},
239                                      {4, 11},
240                                      {7, 8},
241                                      {7, 14},
242                                      {7, 6},
243                                      {7, 5},
244                                      {7, 4},
245                                      {7, 7},
246                                      {7, 9},
247                                      {7, 11},
248                                      {10, 8},
249                                      {10, 14},
250                                      {12, 6},
251                                      {12, 5},
252                                      {10, 4},
253                                      {10, 7},
254                                      {10, 9},
255                                      {10, 11},
256                                      {12, 8 },
257                                      {12, 14},
258                                      {12, 6},
259                                      {15, 16},
260                                      {15, 16},
261                                      {9, 11},
262                                      {11, 11},
263                                      {11, 12}
264 };
265 
266 
267 static int nb_special_fargs;
268 static double * spec_fargs;
269 static float * spec_sp_fargs;
270 
build_special_fargs_table(void)271 static void build_special_fargs_table(void)
272 {
273 /*
274   Entry  Sign Exp   fraction                  Special value
275    0      0   3fd   0x8000000000000ULL         Positive finite number
276    1      0   404   0xf000000000000ULL         ...
277    2      0   001   0x8000000b77501ULL         ...
278    3      0   7fe   0x800000000051bULL         ...
279    4      0   012   0x3214569900000ULL         ...
280    5      0   000   0x0000000000000ULL         +0.0 (+zero)
281    6      1   000   0x0000000000000ULL         -0.0 (-zero)
282    7      0   7ff   0x0000000000000ULL         +infinity
283    8      1   7ff   0x0000000000000ULL         -infinity
284    9      0   7ff   0x7FFFFFFFFFFFFULL         +SNaN
285    10     1   7ff   0x7FFFFFFFFFFFFULL         -SNaN
286    11     0   7ff   0x8000000000000ULL         +QNaN
287    12     1   7ff   0x8000000000000ULL         -QNaN
288    13     1   000   0x8340000078000ULL         Denormalized val (zero exp and non-zero fraction)
289    14     1   40d   0x0650f5a07b353ULL         Negative finite number
290    15     0   412   0x32585a9900000ULL         A couple more positive finite numbers
291    16     0   413   0x82511a2000000ULL         ...
292 */
293 
294    uint64_t mant;
295    uint32_t mant_sp;
296    uint16_t _exp;
297    int s;
298    int j, i = 0;
299 
300    if (spec_fargs)
301       return;
302 
303    spec_fargs = malloc( 17 * sizeof(double) );
304    spec_sp_fargs = malloc( 17 * sizeof(float) );
305 
306    // #0
307    s = 0;
308    _exp = 0x3fd;
309    mant = 0x8000000000000ULL;
310    register_farg(&spec_fargs[i++], s, _exp, mant);
311 
312    // #1
313    s = 0;
314    _exp = 0x404;
315    mant = 0xf000000000000ULL;
316    register_farg(&spec_fargs[i++], s, _exp, mant);
317 
318    /* None of the ftdiv tests succeed.
319     * FRA = value #0; FRB = value #1
320     * ea_ = -2; e_b = 5
321     * fl_flag || fg_flag || fe_flag = 100
322     */
323 
324    /*************************************************
325     *     fe_flag tests
326     *
327     *************************************************/
328 
329    /* fe_flag <- 1 if FRA is a NaN
330     * FRA = value #9; FRB = value #1
331     * e_a = 1024; e_b = 5
332     * fl_flag || fg_flag || fe_flag = 101
333     */
334 
335    /* fe_flag <- 1 if FRB is a NaN
336     * FRA = value #1; FRB = value #12
337     * e_a = 5; e_b = 1024
338     * fl_flag || fg_flag || fe_flag = 101
339     */
340 
341    /* fe_flag <- 1 if e_b <= -1022
342     * FRA = value #0; FRB = value #2
343     * e_a = -2; e_b = -1022
344     * fl_flag || fg_flag || fe_flag = 101
345     *
346     */
347    // #2
348    s = 0;
349    _exp = 0x001;
350    mant = 0x8000000b77501ULL;
351    register_farg(&spec_fargs[i++], s, _exp, mant);
352 
353    /* fe_flag <- 1 if e_b >= 1021
354     * FRA = value #1; FRB = value #3
355     * e_a = 5; e_b = 1023
356     * fl_flag || fg_flag || fe_flag = 101
357     */
358    // #3
359    s = 0;
360    _exp = 0x7fe;
361    mant = 0x800000000051bULL;
362    register_farg(&spec_fargs[i++], s, _exp, mant);
363 
364    /* fe_flag <- 1 if FRA != 0 && e_a - e_b >= 1023
365     * Let FRA = value #3 and FRB be value #0.
366     * e_a = 1023; e_b = -2
367     * fl_flag || fg_flag || fe_flag = 101
368     */
369 
370    /* fe_flag <- 1 if FRA != 0 && e_a - e_b <= -1023
371     * Let FRA = value #0 above and FRB be value #3 above
372     * e_a = -2; e_b = 1023
373     * fl_flag || fg_flag || fe_flag = 101
374     */
375 
376    /* fe_flag <- 1 if FRA != 0 && e_a <= -970
377     * Let FRA = value #4 and FRB be value #0
378     * e_a = -1005; e_b = -2
379     * fl_flag || fg_flag || fe_flag = 101
380    */
381    // #4
382    s = 0;
383    _exp = 0x012;
384    mant = 0x3214569900000ULL;
385    register_farg(&spec_fargs[i++], s, _exp, mant);
386 
387    /*************************************************
388     *     fg_flag tests
389     *
390     *************************************************/
391    /* fg_flag <- 1 if FRA is an Infinity
392     * NOTE: FRA = Inf also sets fe_flag
393     * Do two tests, using values #7 and #8 (+/- Inf) for FRA.
394     * Test 1:
395     *   Let FRA be value #7 and FRB be value #1
396     *   e_a = 1024; e_b = 5
397     *   fl_flag || fg_flag || fe_flag = 111
398     *
399     * Test 2:
400     *   Let FRA be value #8 and FRB be value #1
401     *   e_a = 1024; e_b = 5
402     *   fl_flag || fg_flag || fe_flag = 111
403     *
404     */
405 
406    /* fg_flag <- 1 if FRB is an Infinity
407     * NOTE: FRB = Inf also sets fe_flag
408     * Let FRA be value #1 and FRB be value #7
409     * e_a = 5; e_b = 1024
410     * fl_flag || fg_flag || fe_flag = 111
411     */
412 
413    /* fg_flag <- 1 if FRB is denormalized
414     * NOTE: e_b < -1022 ==> fe_flag <- 1
415     * Let FRA be value #0 and FRB be value #13
416     * e_a = -2; e_b = -1023
417     * fl_flag || fg_flag || fe_flag = 111
418     */
419 
420    /* fg_flag <- 1 if FRB is +zero
421     * NOTE: FRA = Inf also sets fe_flag
422     * Let FRA = val #5; FRB = val #5
423     * ea_ = -1023; e_b = -1023
424     * fl_flag || fg_flag || fe_flag = 111
425     */
426 
427    /* fg_flag <- 1 if FRB is -zero
428     * NOTE: FRA = Inf also sets fe_flag
429     * Let FRA = val #5; FRB = val #6
430     * ea_ = -1023; e_b = -1023
431     * fl_flag || fg_flag || fe_flag = 111
432     */
433 
434    /* Special values */
435    /* +0.0      : 0 0x000 0x0000000000000 */
436    // #5
437    s = 0;
438    _exp = 0x000;
439    mant = 0x0000000000000ULL;
440    register_farg(&spec_fargs[i++], s, _exp, mant);
441 
442    /* -0.0      : 1 0x000 0x0000000000000 */
443    // #6
444    s = 1;
445    _exp = 0x000;
446    mant = 0x0000000000000ULL;
447    register_farg(&spec_fargs[i++], s, _exp, mant);
448 
449    /* +infinity : 0 0x7FF 0x0000000000000  */
450    // #7
451    s = 0;
452    _exp = 0x7FF;
453    mant = 0x0000000000000ULL;
454    register_farg(&spec_fargs[i++], s, _exp, mant);
455 
456    /* -infinity : 1 0x7FF 0x0000000000000 */
457    // #8
458    s = 1;
459    _exp = 0x7FF;
460    mant = 0x0000000000000ULL;
461    register_farg(&spec_fargs[i++], s, _exp, mant);
462 
463    /*
464     * This comment applies to values #9 and #10 below:
465     * When src is a SNaN, it's converted to a QNaN first before rounding to single-precision,
466     * so we can't just copy the double-precision value to the corresponding slot in the
467     * single-precision array (i.e., in the loop at the end of this function).  Instead, we
468     * have to manually set the bits using register_sp_farg().
469     */
470 
471    /* +SNaN     : 0 0x7FF 0x7FFFFFFFFFFFF */
472    // #9
473    s = 0;
474    _exp = 0x7FF;
475    mant = 0x7FFFFFFFFFFFFULL;
476    register_farg(&spec_fargs[i++], s, _exp, mant);
477    _exp = 0xff;
478    mant_sp = 0x3FFFFF;
479    register_sp_farg(&spec_sp_fargs[i-1], s, _exp, mant_sp);
480 
481    /* -SNaN     : 1 0x7FF 0x7FFFFFFFFFFFF */
482    // #10
483    s = 1;
484    _exp = 0x7FF;
485    mant = 0x7FFFFFFFFFFFFULL;
486    register_farg(&spec_fargs[i++], s, _exp, mant);
487    _exp = 0xff;
488    mant_sp = 0x3FFFFF;
489    register_sp_farg(&spec_sp_fargs[i-1], s, _exp, mant_sp);
490 
491    /* +QNaN     : 0 0x7FF 0x8000000000000 */
492    // #11
493    s = 0;
494    _exp = 0x7FF;
495    mant = 0x8000000000000ULL;
496    register_farg(&spec_fargs[i++], s, _exp, mant);
497 
498    /* -QNaN     : 1 0x7FF 0x8000000000000 */
499    // #12
500    s = 1;
501    _exp = 0x7FF;
502    mant = 0x8000000000000ULL;
503    register_farg(&spec_fargs[i++], s, _exp, mant);
504 
505    /* denormalized value */
506    // #13
507    s = 1;
508    _exp = 0x000;
509    mant = 0x8340000078000ULL;
510    register_farg(&spec_fargs[i++], s, _exp, mant);
511 
512    /* Negative finite number */
513    // #14
514    s = 1;
515    _exp = 0x40d;
516    mant = 0x0650f5a07b353ULL;
517    register_farg(&spec_fargs[i++], s, _exp, mant);
518 
519    /* A couple positive finite numbers ... */
520    // #15
521    s = 0;
522    _exp = 0x412;
523    mant = 0x32585a9900000ULL;
524    register_farg(&spec_fargs[i++], s, _exp, mant);
525 
526    // #16
527    s = 0;
528    _exp = 0x413;
529    mant = 0x82511a2000000ULL;
530    register_farg(&spec_fargs[i++], s, _exp, mant);
531 
532    nb_special_fargs = i;
533    for (j = 0; j < i; j++) {
534       if (!(j == 9 || j == 10))
535          spec_sp_fargs[j] = spec_fargs[j];
536    }
537 }
538 
539 
540 struct test_table
541 {
542    test_func_t test_category;
543    char * name;
544 };
545 
546 typedef enum {
547    SINGLE_TEST,
548    DOUBLE_TEST
549 } precision_type_t;
550 
551 typedef enum {
552    VX_SCALAR_FP_NMSUB = 0,
553    // ALL VECTOR-TYPE OPS SHOULD BE ADDED AFTER THIS LINE
554    VX_VECTOR_FP_MULT_AND_OP2 = 10,
555    // and before this line
556    VX_BASIC_CMP = 30,
557    VX_CONV_WORD,
558    VX_DEFAULT
559 } vx_fp_test_type;
560 
561 typedef struct vx_fp_test
562 {
563    test_func_t test_func;
564    const char * name;
565    fp_test_args_t * targs;
566    int num_tests;
567    precision_type_t precision;
568    vx_fp_test_type type;
569    const char * op;
570 } vx_fp_test_t;
571 
572 static vector unsigned int vec_out, vec_inA, vec_inB, vec_inC;
573 
574 static Bool do_dot;
test_xvcmpeqdp(void)575 static void test_xvcmpeqdp(void)
576 {
577    if (do_dot)
578       __asm__ __volatile__ ("xvcmpeqdp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
579    else
580       __asm__ __volatile__ ("xvcmpeqdp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
581 }
582 
test_xvcmpgedp(void)583 static void test_xvcmpgedp(void)
584 {
585    if (do_dot)
586       __asm__ __volatile__ ("xvcmpgedp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
587    else
588       __asm__ __volatile__ ("xvcmpgedp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
589 }
590 
test_xvcmpgtdp(void)591 static void test_xvcmpgtdp(void)
592 {
593    if (do_dot)
594       __asm__ __volatile__ ("xvcmpgtdp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
595    else
596       __asm__ __volatile__ ("xvcmpgtdp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
597 }
598 
test_xvcmpeqsp(void)599 static void test_xvcmpeqsp(void)
600 {
601    if (do_dot)
602       __asm__ __volatile__ ("xvcmpeqsp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
603    else
604       __asm__ __volatile__ ("xvcmpeqsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
605 }
606 
test_xvcmpgesp(void)607 static void test_xvcmpgesp(void)
608 {
609    if (do_dot)
610       __asm__ __volatile__ ("xvcmpgesp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
611    else
612       __asm__ __volatile__ ("xvcmpgesp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
613 }
614 
test_xvcmpgtsp(void)615 static void test_xvcmpgtsp(void)
616 {
617    if (do_dot)
618       __asm__ __volatile__ ("xvcmpgtsp.          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
619    else
620       __asm__ __volatile__ ("xvcmpgtsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
621 }
622 
623 static Bool do_aXp;
624 static Bool do_dp;
test_xsnmsub(void)625 static void test_xsnmsub(void)
626 {
627    if (do_aXp)
628       __asm__ __volatile__ ("xsnmsubadp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
629    else
630       __asm__ __volatile__ ("xsnmsubmdp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
631 }
632 
test_xvmadd(void)633 static void test_xvmadd(void)
634 {
635    if (do_aXp)
636       if (do_dp)
637          __asm__ __volatile__ ("xvmaddadp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
638       else
639          __asm__ __volatile__ ("xvmaddasp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
640    else
641       if (do_dp)
642          __asm__ __volatile__ ("xvmaddmdp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
643       else
644          __asm__ __volatile__ ("xvmaddmsp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
645 }
646 
test_xvnmadd(void)647 static void test_xvnmadd(void)
648 {
649    if (do_aXp)
650       if (do_dp)
651          __asm__ __volatile__ ("xvnmaddadp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
652       else
653          __asm__ __volatile__ ("xvnmaddasp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
654    else
655       if (do_dp)
656          __asm__ __volatile__ ("xvnmaddmdp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
657       else
658          __asm__ __volatile__ ("xvnmaddmsp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
659 }
660 
test_xvnmsub(void)661 static void test_xvnmsub(void)
662 {
663    if (do_aXp)
664       if (do_dp)
665          __asm__ __volatile__ ("xvnmsubadp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
666       else
667          __asm__ __volatile__ ("xvnmsubasp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
668    else
669       if (do_dp)
670          __asm__ __volatile__ ("xvnmsubmdp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
671       else
672          __asm__ __volatile__ ("xvnmsubmsp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
673 }
674 
test_xvmsub(void)675 static void test_xvmsub(void)
676 {
677    if (do_aXp)
678       if (do_dp)
679          __asm__ __volatile__ ("xvmsubadp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
680       else
681          __asm__ __volatile__ ("xvmsubasp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
682    else
683       if (do_dp)
684          __asm__ __volatile__ ("xvmsubmdp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
685       else
686          __asm__ __volatile__ ("xvmsubmsp          %x0, %x1, %x2" : "+wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
687 }
688 
test_xssqrtdp(void)689 static void test_xssqrtdp(void)
690 {
691    __asm__ __volatile__ ("xssqrtdp   %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
692 }
693 
test_xsrdpim(void)694 static void test_xsrdpim(void)
695 {
696    __asm__ __volatile__ ("xsrdpim   %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
697 }
698 
test_xsrdpip(void)699 static void test_xsrdpip(void)
700 {
701    __asm__ __volatile__ ("xsrdpip   %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
702 }
703 
test_xstdivdp(void)704 static void test_xstdivdp(void)
705 {
706    __asm__ __volatile__ ("xstdivdp   6, %x0, %x1" : : "wa" (vec_inA), "wa" (vec_inB));
707 }
708 
test_xsmaxdp(void)709 static void test_xsmaxdp(void)
710 {
711    __asm__ __volatile__ ("xsmaxdp   %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
712 }
713 
test_xsmindp(void)714 static void test_xsmindp(void)
715 {
716    __asm__ __volatile__ ("xsmindp   %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
717 }
718 
test_xvadddp(void)719 static void test_xvadddp(void)
720 {
721    __asm__ __volatile__ ("xvadddp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
722 }
723 
test_xvaddsp(void)724 static void test_xvaddsp(void)
725 {
726    __asm__ __volatile__ ("xvaddsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
727 }
728 
test_xvdivdp(void)729 static void test_xvdivdp(void)
730 {
731    __asm__ __volatile__ ("xvdivdp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
732 }
733 
test_xvdivsp(void)734 static void test_xvdivsp(void)
735 {
736    __asm__ __volatile__ ("xvdivsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
737 }
738 
test_xvmuldp(void)739 static void test_xvmuldp(void)
740 {
741    __asm__ __volatile__ ("xvmuldp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
742 }
743 
test_xvmulsp(void)744 static void test_xvmulsp(void)
745 {
746    __asm__ __volatile__ ("xvmulsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
747 }
748 
test_xvsubdp(void)749 static void test_xvsubdp(void)
750 {
751    __asm__ __volatile__ ("xvsubdp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
752 }
753 
test_xvmaxdp(void)754 static void test_xvmaxdp(void)
755 {
756    __asm__ __volatile__ ("xvmaxdp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
757 }
758 
test_xvmindp(void)759 static void test_xvmindp(void)
760 {
761    __asm__ __volatile__ ("xvmindp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
762 }
763 
test_xvmaxsp(void)764 static void test_xvmaxsp(void)
765 {
766    __asm__ __volatile__ ("xvmaxsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
767 }
768 
test_xvminsp(void)769 static void test_xvminsp(void)
770 {
771    __asm__ __volatile__ ("xvminsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
772 }
773 
test_xvsubsp(void)774 static void test_xvsubsp(void)
775 {
776    __asm__ __volatile__ ("xvsubsp          %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
777 }
778 
test_xvresp(void)779 static void test_xvresp(void)
780 {
781    __asm__ __volatile__ ("xvresp   %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
782 }
783 
test_xxsel(void)784 static void test_xxsel(void)
785 {
786    unsigned long long * dst;
787    unsigned long long xa[] =  { 0xa12bc37de56f9708ULL, 0x3894c1fddeadbeefULL};
788    unsigned long long xb[] =  { 0xfedc432124681235ULL, 0xf1e2d3c4e0057708ULL};
789    unsigned long long xc[] =  { 0xffffffff01020304ULL, 0x128934bd00000000ULL};
790 
791    memcpy(&vec_inA, xa, 16);
792    memcpy(&vec_inB, xb, 16);
793    memcpy(&vec_inC, xc, 16);
794 
795 
796    __asm__ __volatile__ ("xxsel   %x0, %x1, %x2, %x3" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB),"wa" (vec_inC));
797    dst = (unsigned long long *) &vec_out;
798    printf("xxsel %016llx,%016llx,%016llx => %016llx\n", xa[0], xb[0], xc[0], *dst);
799    dst++;
800    printf("xxsel %016llx,%016llx,%016llx => %016llx\n", xa[1], xb[1], xc[1], *dst);
801    printf("\n");
802 }
803 
test_xxspltw(void)804 static void test_xxspltw(void)
805 {
806    int uim;
807    unsigned long long * dst = NULL;
808    unsigned int xb[] =  { 0xfedc4321, 0x24681235, 0xf1e2d3c4, 0xe0057708};
809    int i;
810    void * vecB_ptr = &vec_inB;
811    if (isLE) {
812       for (i = 3; i >=0; i--) {
813          memcpy(vecB_ptr, &xb[i], 4);
814          vecB_ptr+=4;
815       }
816    } else {
817       for (i = 0; i < 4; i++) {
818          memcpy(vecB_ptr, &xb[i], 4);
819          vecB_ptr+=4;
820       }
821    }
822 
823    for (uim = 0; uim < 4; uim++) {
824       switch (uim) {
825          case 0:
826             __asm__ __volatile__ ("xxspltw   %x0, %x1, 0" : "=wa" (vec_out): "wa" (vec_inB));
827             break;
828          case 1:
829             __asm__ __volatile__ ("xxspltw   %x0, %x1, 1" : "=wa" (vec_out): "wa" (vec_inB));
830             break;
831          case 2:
832             __asm__ __volatile__ ("xxspltw   %x0, %x1, 2" : "=wa" (vec_out): "wa" (vec_inB));
833             break;
834          case 3:
835             __asm__ __volatile__ ("xxspltw   %x0, %x1, 3" : "=wa" (vec_out): "wa" (vec_inB));
836             break;
837       }
838       dst = (unsigned long long *) &vec_out;
839       printf("xxspltw 0x%08x%08x%08x%08x %d=> 0x%016llx",  xb[0], xb[1],
840              xb[2], xb[3], uim, *dst);
841       dst++;
842       printf("%016llx\n", *dst);
843    }
844    printf("\n");
845 }
846 
test_xscvdpsxws(void)847 static void test_xscvdpsxws(void)
848 {
849    __asm__ __volatile__ ("xscvdpsxws  %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
850 }
851 
test_xscvdpuxds(void)852 static void test_xscvdpuxds(void)
853 {
854    __asm__ __volatile__ ("xscvdpuxds  %x0, %x1" : "=wa" (vec_out): "wa" (vec_inB));
855 }
856 
test_xvcpsgndp(void)857 static void test_xvcpsgndp(void)
858 {
859    __asm__ __volatile__  ("xvcpsgndp  %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
860 }
861 
test_xvcpsgnsp(void)862 static void test_xvcpsgnsp(void)
863 {
864    __asm__ __volatile__  ("xvcpsgnsp  %x0, %x1, %x2" : "=wa" (vec_out): "wa" (vec_inA),"wa" (vec_inB));
865 }
866 
test_xvcvdpsxws(void)867 static void test_xvcvdpsxws(void)
868 {
869    __asm__ __volatile__ ("xvcvdpsxws  %x0, %x1 " : "=wa" (vec_out): "wa" (vec_inB));
870 }
871 
test_xvcvspsxws(void)872 static void test_xvcvspsxws(void)
873 {
874    __asm__ __volatile__ ("xvcvspsxws  %x0, %x1 " : "=wa" (vec_out): "wa" (vec_inB));
875 }
876 
877 static vx_fp_test_t
878 vx_vector_one_fp_arg_tests[] = {
879                                 { &test_xvresp, "xvresp", NULL, 16, SINGLE_TEST, VX_BASIC_CMP, "1/x"},
880                                 { &test_xvcvdpsxws, "xvcvdpsxws", NULL, 16, DOUBLE_TEST, VX_CONV_WORD, "conv"},
881                                 { &test_xvcvspsxws, "xvcvspsxws", NULL, 16, SINGLE_TEST, VX_CONV_WORD, "conv"},
882                                 { NULL, NULL, NULL, 0 , 0, 0, NULL}
883 };
884 
885 static vx_fp_test_t
886 vx_vector_fp_tests[] = {
887                         { &test_xvcmpeqdp, "xvcmpeqdp", fp_cmp_tests, 64, DOUBLE_TEST, VX_BASIC_CMP, "eq"},
888                         { &test_xvcmpgedp, "xvcmpgedp", fp_cmp_tests, 64, DOUBLE_TEST, VX_BASIC_CMP, "ge"},
889                         { &test_xvcmpgtdp, "xvcmpgtdp", fp_cmp_tests, 64, DOUBLE_TEST, VX_BASIC_CMP, "gt"},
890                         { &test_xvcmpeqsp, "xvcmpeqsp", fp_cmp_tests, 64, SINGLE_TEST, VX_BASIC_CMP, "eq"},
891                         { &test_xvcmpgesp, "xvcmpgesp", fp_cmp_tests, 64, SINGLE_TEST, VX_BASIC_CMP, "ge"},
892                         { &test_xvcmpgtsp, "xvcmpgtsp", fp_cmp_tests, 64, SINGLE_TEST, VX_BASIC_CMP, "gt"},
893                         { &test_xvadddp, "xvadddp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "+" },
894                         { &test_xvaddsp, "xvaddsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "+" },
895                         { &test_xvdivdp, "xvdivdp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "/" },
896                         { &test_xvdivsp, "xvdivsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "/" },
897                         { &test_xvmuldp, "xvmuldp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "*" },
898                         { &test_xvmulsp, "xvmulsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "*" },
899                         { &test_xvsubdp, "xvsubdp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "-" },
900                         { &test_xvsubsp, "xvsubsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "-" },
901                         { &test_xvmaxdp, "xvmaxdp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "@max@" },
902                         { &test_xvmindp, "xvmindp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "@min@" },
903                         { &test_xvmaxsp, "xvmaxsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "@max@" },
904                         { &test_xvminsp, "xvminsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "@min@" },
905                         { &test_xvcpsgndp, "xvcpsgndp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, "+-cp"},
906                         { &test_xvcpsgnsp, "xvcpsgnsp", two_arg_fp_tests, 64, SINGLE_TEST, VX_DEFAULT, "+-cp"},
907                         { NULL, NULL, NULL, 0 , 0, 0, NULL}
908 };
909 
910 
911 static vx_fp_test_t
912 vx_aORm_fp_tests[] = {
913                        { &test_xsnmsub, "xsnmsub", two_arg_fp_tests, 64, DOUBLE_TEST, VX_SCALAR_FP_NMSUB, "!*-"},
914                        { &test_xvmadd, "xvmadd", two_arg_fp_tests, 64, DOUBLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "*+"},
915                        { &test_xvmadd, "xvmadd", two_arg_fp_tests, 64, SINGLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "*+"},
916                        { &test_xvnmadd, "xvnmadd", two_arg_fp_tests, 64, DOUBLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "!*+"},
917                        { &test_xvnmadd, "xvnmadd", two_arg_fp_tests, 64, SINGLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "!*+"},
918                        { &test_xvmsub, "xvmsub", two_arg_fp_tests, 64, DOUBLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "*-"},
919                        { &test_xvmsub, "xvmsub", two_arg_fp_tests, 64, SINGLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "*-"},
920                        { &test_xvnmsub, "xvnmsub", two_arg_fp_tests, 64, DOUBLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "!*-"},
921                        { &test_xvnmsub, "xvnmsub", two_arg_fp_tests, 64, SINGLE_TEST, VX_VECTOR_FP_MULT_AND_OP2, "!*-"},
922                        { NULL, NULL, NULL, 0, 0, 0,  NULL }
923 };
924 
925 static vx_fp_test_t
926 vx_simple_scalar_fp_tests[] = {
927                                { &test_xssqrtdp, "xssqrtdp", NULL, 17, DOUBLE_TEST, VX_DEFAULT, NULL},
928                                { &test_xsrdpim, "xsrdpim", NULL, 17, DOUBLE_TEST, VX_DEFAULT, NULL},
929                                { &test_xsrdpip, "xsrdpip", NULL, 17, DOUBLE_TEST, VX_DEFAULT, NULL},
930                                { &test_xstdivdp, "xstdivdp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, NULL},
931                                { &test_xsmaxdp, "xsmaxdp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, NULL},
932                                { &test_xsmindp, "xsmindp", two_arg_fp_tests, 64, DOUBLE_TEST, VX_DEFAULT, NULL},
933                                { &test_xscvdpsxws, "xscvdpsxws", NULL, 17, DOUBLE_TEST, VX_CONV_WORD, NULL},
934                                { &test_xscvdpuxds, "xscvdpuxds", NULL, 17, DOUBLE_TEST, VX_DEFAULT, NULL},
935                                { NULL, NULL, NULL, 0, 0, 0, NULL }
936 };
937 
938 
939 #ifdef __powerpc64__
test_bpermd(void)940 static void test_bpermd(void)
941 {
942    /* NOTE: Bit number is '0 . . . 63'
943     *
944     * Permuted bits are generated bit 0 -7 as follows:
945     *    index = (r14)8*i:8*i+7
946     *    perm[i] = (r15)index
947     *
948     * So, for i = 0, index is (r14)8*0:8*0+7, or (r14)0:7, which is the MSB
949     * byte of r14, 0x1b(27/base 10).  This identifies bit 27 of r15, which is '1'.
950     * For i = 1, index is 0x2c, identifying bit 44 of r15, which is '1'.
951     * So the result of the first two iterations of i are:
952     *   perm = 0b01xxxxxx
953     *
954     */
955    r15 = 0xa12bc37de56f9708ULL;
956    r14 = 0x1b2c31f030000001ULL;
957    __asm__ __volatile__ ("bpermd %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
958    printf("bpermd: 0x%016llx : 0x%016llx => 0x%llx\n", (unsigned long long)r14,
959           (unsigned long long)r15, (unsigned long long)r17);
960    printf("\n");
961 }
962 #endif
963 
964 static Bool do_OE;
965 typedef enum {
966    DIV_BASE = 1,
967    DIV_OE = 2,
968    DIV_DOT = 4,
969 } div_type_t;
970 /* Possible divde type combinations are:
971  *   - base
972  *   - base+dot
973  *   - base+OE
974  *   - base+OE+dot
975  */
976 #ifdef __powerpc64__
test_divde(void)977 static void test_divde(void)
978 {
979    int divde_type = DIV_BASE;
980    if (do_OE)
981       divde_type |= DIV_OE;
982    if (do_dot)
983       divde_type |= DIV_DOT;
984 
985    switch (divde_type) {
986       case 1:
987         SET_CR_XER_ZERO;
988          __asm__ __volatile__ ("divde %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
989          GET_CR_XER(div_flags, div_xer);
990          break;
991       case 3:
992         SET_CR_XER_ZERO;
993          __asm__ __volatile__ ("divdeo %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
994          GET_CR_XER(div_flags, div_xer);
995          break;
996       case 5:
997         SET_CR_XER_ZERO;
998          __asm__ __volatile__ ("divde. %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
999          GET_CR_XER(div_flags, div_xer);
1000          break;
1001       case 7:
1002         SET_CR_XER_ZERO;
1003          __asm__ __volatile__ ("divdeo. %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
1004          GET_CR_XER(div_flags, div_xer);
1005          break;
1006       default:
1007          fprintf(stderr, "Invalid divde type. Exiting\n");
1008          exit(1);
1009    }
1010 }
1011 #endif
1012 
test_divweu(void)1013 static void test_divweu(void)
1014 {
1015    int divweu_type = DIV_BASE;
1016    if (do_OE)
1017       divweu_type |= DIV_OE;
1018    if (do_dot)
1019       divweu_type |= DIV_DOT;
1020 
1021    switch (divweu_type) {
1022       case 1:
1023         SET_CR_XER_ZERO;
1024          __asm__ __volatile__ ("divweu %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
1025          GET_CR_XER(div_flags, div_xer);
1026          break;
1027       case 3:
1028         SET_CR_XER_ZERO;
1029          __asm__ __volatile__ ("divweuo %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
1030          GET_CR_XER(div_flags, div_xer);
1031          break;
1032       case 5:
1033         SET_CR_XER_ZERO;
1034          __asm__ __volatile__ ("divweu. %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
1035          GET_CR_XER(div_flags, div_xer);
1036          break;
1037       case 7:
1038         SET_CR_XER_ZERO;
1039          __asm__ __volatile__ ("divweuo. %0, %1, %2" : "=r" (r17) : "r" (r14),"r" (r15));
1040          GET_CR_XER(div_flags, div_xer);
1041          break;
1042       default:
1043          fprintf(stderr, "Invalid divweu type. Exiting\n");
1044          exit(1);
1045    }
1046 }
1047 
test_fctiduz(void)1048 static void test_fctiduz(void)
1049 {
1050    if (do_dot)
1051       __asm__ __volatile__ ("fctiduz. %0, %1" : "=d" (f17) : "d" (f14));
1052    else
1053       __asm__ __volatile__ ("fctiduz %0, %1" : "=d" (f17) : "d" (f14));
1054 }
1055 
test_fctidu(void)1056 static void test_fctidu(void)
1057 {
1058    if (do_dot)
1059       __asm__ __volatile__ ("fctidu. %0, %1" : "=d" (f17) : "d" (f14));
1060    else
1061       __asm__ __volatile__ ("fctidu %0, %1" : "=d" (f17) : "d" (f14));
1062 }
1063 
test_fctiwuz(void)1064 static void test_fctiwuz(void)
1065 {
1066    if (do_dot)
1067       __asm__ __volatile__ ("fctiwuz. %0, %1" : "=d" (f17) : "d" (f14));
1068    else
1069       __asm__ __volatile__ ("fctiwuz %0, %1" : "=d" (f17) : "d" (f14));
1070 }
1071 
test_fctiwu(void)1072 static void test_fctiwu(void)
1073 {
1074    if (do_dot)
1075       __asm__ __volatile__ ("fctiwu. %0, %1" : "=d" (f17) : "d" (f14));
1076    else
1077       __asm__ __volatile__ ("fctiwu %0, %1" : "=d" (f17) : "d" (f14));
1078 }
1079 
1080 typedef struct simple_test {
1081    test_func_t test_func;
1082    char * name;
1083    precision_type_t precision;
1084 } simple_test_t;
1085 
1086 static simple_test_t fct_tests[] = {
1087                                     { &test_fctiduz, "fctiduz", DOUBLE_TEST },
1088                                     { &test_fctidu, "fctidu", DOUBLE_TEST },
1089                                     { &test_fctiwuz, "fctiwuz", SINGLE_TEST },
1090                                     { &test_fctiwu, "fctiwu", SINGLE_TEST },
1091                                    { NULL, NULL }
1092 };
1093 
setup_sp_fp_args(fp_test_args_t * targs,Bool swap_inputs)1094 static void setup_sp_fp_args(fp_test_args_t * targs, Bool swap_inputs)
1095 {
1096    int a_idx, b_idx, i;
1097    void * inA, * inB;
1098    void * vec_src = swap_inputs ? &vec_out : &vec_inB;
1099 
1100    for (i = 0; i < 4; i++) {
1101       a_idx = targs->fra_idx;
1102       b_idx = targs->frb_idx;
1103       inA = (void *)&spec_sp_fargs[a_idx];
1104       inB = (void *)&spec_sp_fargs[b_idx];
1105       // copy single precision FP  into vector element i
1106       memcpy(((void *)&vec_inA) + (i * 4), inA, 4);
1107       memcpy(vec_src + (i * 4), inB, 4);
1108       targs++;
1109    }
1110 }
1111 
setup_dp_fp_args(fp_test_args_t * targs,Bool swap_inputs)1112 static void setup_dp_fp_args(fp_test_args_t * targs, Bool swap_inputs)
1113 {
1114    int a_idx, b_idx, i;
1115    void * inA, * inB;
1116    void * vec_src = swap_inputs ? (void *)&vec_out : (void *)&vec_inB;
1117 
1118    for (i = 0; i < 2; i++) {
1119       a_idx = targs->fra_idx;
1120       b_idx = targs->frb_idx;
1121       inA = (void *)&spec_fargs[a_idx];
1122       inB = (void *)&spec_fargs[b_idx];
1123       // copy double precision FP  into vector element i
1124       memcpy(((void *)&vec_inA) + (i * 8), inA, 8);
1125       memcpy(vec_src + (i * 8), inB, 8);
1126       targs++;
1127    }
1128 }
1129 
1130 #define VX_NOT_CMP_OP 0xffffffff
print_vector_fp_result(unsigned int cc,vx_fp_test_t * test_group,int i)1131 static void print_vector_fp_result(unsigned int cc, vx_fp_test_t * test_group, int i)
1132 {
1133    int a_idx, b_idx, k;
1134    char * name = malloc(20);
1135    int dp = test_group->precision == DOUBLE_TEST ? 1 : 0;
1136    int loops = dp ? 2 : 4;
1137    fp_test_args_t * targs = &test_group->targs[i];
1138    unsigned long long * frA_dp, * frB_dp, * dst_dp;
1139    unsigned int * frA_sp, *frB_sp, * dst_sp;
1140    strcpy(name, test_group->name);
1141    printf("#%d: %s%s ", dp? i/2 : i/4, name, (do_dot ? "." : ""));
1142    for (k = 0; k < loops; k++) {
1143       a_idx = targs->fra_idx;
1144       b_idx = targs->frb_idx;
1145       if (k)
1146          printf(" AND ");
1147       if (dp) {
1148          frA_dp = (unsigned long long *)&spec_fargs[a_idx];
1149          frB_dp = (unsigned long long *)&spec_fargs[b_idx];
1150          printf("%016llx %s %016llx", *frA_dp, test_group->op, *frB_dp);
1151       } else {
1152          frA_sp = (unsigned int *)&spec_sp_fargs[a_idx];
1153          frB_sp = (unsigned int *)&spec_sp_fargs[b_idx];
1154          printf("%08x %s %08x", *frA_sp, test_group->op, *frB_sp);
1155       }
1156       targs++;
1157    }
1158    if (cc != VX_NOT_CMP_OP)
1159       printf(" ? cc=%x", cc);
1160 
1161    if (dp) {
1162       dst_dp = (unsigned long long *) &vec_out;
1163       printf(" => %016llx %016llx\n", dst_dp[0], dst_dp[1]);
1164    } else {
1165       dst_sp = (unsigned int *) &vec_out;
1166       printf(" => %08x %08x %08x %08x\n", dst_sp[0], dst_sp[1], dst_sp[2], dst_sp[3]);
1167    }
1168    free(name);
1169 }
1170 
1171 
print_vx_aORm_fp_result(unsigned long long * XT_arg,unsigned long long * XB_arg,vx_fp_test_t * test_group,int i)1172 static void print_vx_aORm_fp_result(unsigned long long * XT_arg, unsigned long long * XB_arg,
1173                                     vx_fp_test_t * test_group, int i)
1174 {
1175    int a_idx, k;
1176    char * name = malloc(20);
1177    int dp = test_group->precision == DOUBLE_TEST ? 1 : 0;
1178    int loops = dp ? 2 : 4;
1179    fp_test_args_t * targs = &test_group->targs[i];
1180    unsigned long long frA_dp, * dst_dp;
1181    unsigned int frA_sp, * dst_sp;
1182 
1183    strcpy(name, test_group->name);
1184    if (do_aXp)
1185       if (dp)
1186          strcat(name, "adp");
1187       else
1188          strcat(name, "asp");
1189    else
1190       if (dp)
1191          strcat(name, "mdp");
1192       else
1193          strcat(name, "msp");
1194 
1195    printf("#%d: %s ", dp? i/2 : i/4, name);
1196    for (k = 0; k < loops; k++) {
1197       a_idx = targs->fra_idx;
1198       if (k)
1199          printf(" AND ");
1200       if (dp) {
1201          frA_dp = *((unsigned long long *)&spec_fargs[a_idx]);
1202          printf("%s(%016llx,%016llx,%016llx)", test_group->op, XT_arg[k], frA_dp, XB_arg[k]);
1203       } else {
1204          unsigned int * xt_sp = (unsigned int *)XT_arg;
1205          unsigned int * xb_sp = (unsigned int *)XB_arg;
1206          frA_sp = *((unsigned int *)&spec_sp_fargs[a_idx]);
1207          printf("%s(%08x,%08x,%08x)", test_group->op, xt_sp[k], frA_sp, xb_sp[k]);
1208       }
1209       targs++;
1210    }
1211 
1212    if (dp) {
1213       dst_dp = (unsigned long long *) &vec_out;
1214       printf(" => %016llx %016llx\n", dst_dp[0], dst_dp[1]);
1215    } else {
1216       dst_sp = (unsigned int *) &vec_out;
1217       printf(" => %08x %08x %08x %08x\n", dst_sp[0], dst_sp[1], dst_sp[2], dst_sp[3]);
1218    }
1219    free(name);
1220 }
1221 
1222 /* This function currently only supports double precision input arguments. */
test_vx_simple_scalar_fp_ops(void)1223 static void test_vx_simple_scalar_fp_ops(void)
1224 {
1225    test_func_t func;
1226    int k = 0;
1227 
1228    build_special_fargs_table();
1229    while ((func = vx_simple_scalar_fp_tests[k].test_func)) {
1230       unsigned long long * frap, * frbp, * dst;
1231       unsigned int * pv;
1232       int idx;
1233       vx_fp_test_t test_group = vx_simple_scalar_fp_tests[k];
1234       Bool convToWord = (test_group.type == VX_CONV_WORD);
1235       if (test_group.precision != DOUBLE_TEST) {
1236          fprintf(stderr, "Unsupported single precision for scalar op in test_vx_aORm_fp_ops\n");
1237          exit(1);
1238       }
1239       pv = (unsigned int *)&vec_out;
1240       // clear vec_out
1241       for (idx = 0; idx < 4; idx++, pv++)
1242          *pv = 0;
1243 
1244       /* If num_tests is exactly equal to nb_special_fargs, this implies the
1245        * instruction being tested only requires one floating point argument
1246        * (e.g. xssqrtdp).
1247        */
1248       if (test_group.num_tests == nb_special_fargs && !test_group.targs) {
1249          void * inB, * vec_void_ptr = (void *)&vec_inB;
1250          int i;
1251          if (isLE)
1252             vec_void_ptr += 8;
1253          for (i = 0; i < nb_special_fargs; i++) {
1254             inB = (void *)&spec_fargs[i];
1255             frbp = (unsigned long long *)&spec_fargs[i];
1256             memcpy(vec_void_ptr, inB, 8);
1257             (*func)();
1258             dst = (unsigned long long *) &vec_out;
1259             if (isLE)
1260                dst++;
1261             printf("#%d: %s %016llx => %016llx\n", i, test_group.name, *frbp,
1262                    convToWord ? (*dst & 0x00000000ffffffffULL) : *dst);
1263          }
1264       } else {
1265          void * inA, * inB, * vecA_void_ptr, * vecB_void_ptr;
1266          unsigned int condreg, flags;
1267          int isTdiv = (strstr(test_group.name, "xstdivdp") != NULL) ? 1 : 0;
1268          int i;
1269          if (isLE) {
1270             vecA_void_ptr = (void *)&vec_inA + 8;
1271             vecB_void_ptr = (void *)&vec_inB + 8;
1272          } else {
1273             vecA_void_ptr = (void *)&vec_inA;
1274             vecB_void_ptr = (void *)&vec_inB;
1275          }
1276          for (i = 0; i < test_group.num_tests; i++) {
1277             fp_test_args_t aTest = test_group.targs[i];
1278             inA = (void *)&spec_fargs[aTest.fra_idx];
1279             inB = (void *)&spec_fargs[aTest.frb_idx];
1280             frap = (unsigned long long *)&spec_fargs[aTest.fra_idx];
1281             frbp = (unsigned long long *)&spec_fargs[aTest.frb_idx];
1282             // Only need to copy one doubleword into each vector's element 0
1283             memcpy(vecA_void_ptr, inA, 8);
1284             memcpy(vecB_void_ptr, inB, 8);
1285             SET_FPSCR_ZERO;
1286             SET_CR_XER_ZERO;
1287             (*func)();
1288             GET_CR(flags);
1289             if (isTdiv) {
1290                condreg = (flags & 0x000000f0) >> 4;
1291                printf("#%d: %s %016llx,%016llx => cr %x\n", i, test_group.name, *frap, *frbp, condreg);
1292             } else {
1293                dst = (unsigned long long *) &vec_out;
1294                if (isLE)
1295                   dst++;
1296                printf("#%d: %s %016llx,%016llx => %016llx\n", i, test_group.name,
1297                       *frap, *frbp, *dst);
1298             }
1299          }
1300       }
1301       printf( "\n" );
1302       k++;
1303    }
1304 }
1305 
test_vx_aORm_fp_ops(void)1306 static void test_vx_aORm_fp_ops(void)
1307 {
1308    /* These ops need a third src argument, which is stored in element 0 of
1309     * VSX[XT] -- i.e., vec_out.  For the xs<ZZZ>m{d|s}p cases, VSX[XT] holds
1310     * src3 and VSX[XB] holds src2; for the xs<ZZZ>a{d|s}p cases, VSX[XT] holds
1311     * src2 and VSX[XB] holds src3.  The fp_test_args_t that holds the test
1312     * data (input args, result) contain only two inputs, so I arbitrarily
1313     * choose some spec_fargs elements for the third source argument.
1314     * Note that that by using the same input data for a given pair of
1315     * a{d|s}p/m{d|s}p-type instructions (by swapping the src2 and src3
1316     * arguments), the expected result should be the same.
1317     */
1318 
1319    test_func_t func;
1320    int k;
1321    char * test_name = (char *)malloc(20);
1322    k = 0;
1323    do_dot = False;
1324 
1325    build_special_fargs_table();
1326    while ((func = vx_aORm_fp_tests[k].test_func)) {
1327       int i, stride;
1328       Bool repeat = False;
1329       Bool scalar = False;
1330       unsigned long long * frap, * frbp, * dst;
1331       vx_fp_test_t test_group = vx_aORm_fp_tests[k];
1332       vx_fp_test_type test_type = test_group.type;
1333       do_dp = test_group.precision == DOUBLE_TEST ? True : False;
1334       frap = frbp = NULL;
1335 
1336       if (test_type < VX_VECTOR_FP_MULT_AND_OP2) {
1337             scalar = True;
1338             strcpy(test_name, test_group.name);
1339             if (!repeat) {
1340                repeat = 1;
1341                stride = 1;
1342                // Only support double precision scalar ops in this function
1343                if (do_dp) {
1344                   strcat(test_name, "adp");
1345                } else {
1346                   fprintf(stderr, "Unsupported single precision for scalar op in test_vx_aORm_fp_ops\n");
1347                   exit(1);
1348                }
1349                do_aXp = True;
1350             }
1351       } else if (test_type < VX_BASIC_CMP) {
1352          // Then it must be a VX_VECTOR_xxx type
1353             stride = do_dp ? 2 : 4;
1354             if (!repeat) {
1355                // No need to work up the testcase name here, since that will be done in
1356                // the print_vx_aORm_fp_result() function we'll call for vector-type ops.
1357                repeat = 1;
1358                do_aXp = True;
1359             }
1360       } else {
1361             printf("ERROR:  Invalid VX FP test type %d\n", test_type);
1362             exit(1);
1363       }
1364 
1365 again:
1366       for (i = 0; i < test_group.num_tests; i+=stride) {
1367          void  * inA, * inB;
1368          int m, fp_idx[4];
1369          unsigned long long vsr_XT[2];
1370          unsigned long long vsr_XB[2];
1371          fp_test_args_t aTest = test_group.targs[i];
1372          for (m = 0; m < stride; m++)
1373             fp_idx[m] = i % (nb_special_fargs - stride) + m;
1374 
1375          /* When repeat == True, we're on the first time through of one of the VX_FP_SMx
1376           * test types, meaning we're testing a xs<ZZZ>adp case, thus we have to swap
1377           * inputs as described above:
1378           *    src2 <= VSX[XT]
1379           *    src3 <= VSX[XB]
1380           */
1381          if (scalar) {
1382 #ifdef VGP_ppc64le_linux
1383 #define VECTOR_ADDR(_v) ((void *)&_v) + 8
1384 #else
1385 #define VECTOR_ADDR(_v) ((void *)&_v)
1386 #endif
1387             // For scalar op, only need to copy one doubleword into each vector's element 0
1388             inA = (void *)&spec_fargs[aTest.fra_idx];
1389             inB = (void *)&spec_fargs[aTest.frb_idx];
1390             frap = (unsigned long long *)&spec_fargs[aTest.fra_idx];
1391             memcpy(VECTOR_ADDR(vec_inA), inA, 8);
1392             if (repeat) {
1393                memcpy(VECTOR_ADDR(vec_out), inB, 8);  // src2
1394                memcpy(VECTOR_ADDR(vec_inB), &spec_fargs[fp_idx[0]], 8);  //src3
1395                frbp = (unsigned long long *)&spec_fargs[fp_idx[0]];
1396             } else {
1397                frbp = (unsigned long long *)&spec_fargs[aTest.frb_idx];
1398                memcpy(VECTOR_ADDR(vec_inB), inB, 8);  // src2
1399                memcpy(VECTOR_ADDR(vec_out), &spec_fargs[fp_idx[0]], 8);  //src3
1400             }
1401             memcpy(vsr_XT, VECTOR_ADDR(vec_out), 8);
1402          } else {
1403             int j, loops = do_dp ? 2 : 4;
1404             size_t len = do_dp ? 8 : 4;
1405             void * vec_src = repeat ? (void *)&vec_inB : (void *)&vec_out;
1406             for (j = 0; j < loops; j++) {
1407                if (do_dp)
1408                   memcpy(vec_src + (j * len), &spec_fargs[fp_idx[j]], len);
1409                else
1410                   memcpy(vec_src + (j * len), &spec_sp_fargs[fp_idx[j]], len);
1411             }
1412             if (do_dp)
1413                setup_dp_fp_args(&test_group.targs[i], repeat);
1414             else
1415                setup_sp_fp_args(&test_group.targs[i], repeat);
1416 
1417             memcpy(vsr_XT, &vec_out, 16);
1418             memcpy(vsr_XB, &vec_inB, 16);
1419          }
1420 
1421          (*func)();
1422          dst = (unsigned long long *) &vec_out;
1423          if (isLE)
1424             dst++;
1425          if (test_type < VX_VECTOR_FP_MULT_AND_OP2)
1426             printf( "#%d: %s %s(%016llx,%016llx,%016llx) = %016llx\n", i,
1427                     test_name, test_group.op, vsr_XT[0], *frap, *frbp, *dst );
1428          else
1429             print_vx_aORm_fp_result(vsr_XT, vsr_XB, &test_group, i);
1430       }
1431       printf( "\n" );
1432 
1433       if (repeat) {
1434          repeat = 0;
1435          if (test_type < VX_VECTOR_FP_MULT_AND_OP2) {
1436                strcpy(test_name, test_group.name);
1437                strcat(test_name, "mdp");
1438          }
1439          do_aXp = False;
1440          goto again;
1441       }
1442       k++;
1443    }
1444    printf( "\n" );
1445    free(test_name);
1446 }
1447 
test_vx_vector_one_fp_arg(void)1448 static void test_vx_vector_one_fp_arg(void)
1449 {
1450    test_func_t func;
1451    int k;
1452    k = 0;
1453    build_special_fargs_table();
1454 
1455    while ((func = vx_vector_one_fp_arg_tests[k].test_func)) {
1456       int idx, i;
1457       vx_fp_test_t test_group = vx_vector_one_fp_arg_tests[k];
1458       Bool convToWord = (test_group.type == VX_CONV_WORD);
1459       Bool dp = (test_group.precision == DOUBLE_TEST) ? True : False;
1460       Bool xvrespTest = (strstr(test_group.name , "xvresp") != NULL) ? True: False;
1461       int stride = dp ? 2 : 4;
1462 
1463       for (i = 0; i < test_group.num_tests; i+=stride) {
1464          unsigned int * pv;
1465          void * inB;
1466 
1467          pv = (unsigned int *)&vec_out;
1468          // clear vec_out
1469          for (idx = 0; idx < 4; idx++, pv++)
1470             *pv = 0;
1471 
1472          if (dp) {
1473             int j;
1474             unsigned long long * frB_dp, *dst_dp;
1475             for (j = 0; j < 2; j++) {
1476                inB = (void *)&spec_fargs[i + j];
1477                // copy double precision FP into vector element i
1478                memcpy(((void *)&vec_inB) + (j * 8), inB, 8);
1479             }
1480             // execute test insn
1481             (*func)();
1482             dst_dp = (unsigned long long *) &vec_out;
1483             printf("#%d: %s ", i/2, test_group.name);
1484             for (j = 0; j < 2; j++) {
1485                if (j)
1486                   printf("; ");
1487                frB_dp = (unsigned long long *)&spec_fargs[i + j];
1488                printf("%s(%016llx)", test_group.op, *frB_dp);
1489                printf(" = %016llx", convToWord ? (dst_dp[j] & 0x00000000ffffffffULL) : dst_dp[j]);
1490             }
1491             printf("\n");
1492          } else {
1493             int j;
1494             unsigned int * frB_sp, * dst_sp;
1495 
1496             for (j = 0; j < 4; j++) {
1497                inB = (void *)&spec_sp_fargs[i + j];
1498                // copy single precision FP into vector element i
1499                memcpy(((void *)&vec_inB) + (j * 4), inB, 4);
1500             }
1501             // execute test insn
1502             (*func)();
1503             dst_sp = (unsigned int *) &vec_out;
1504             // print result
1505             printf("#%d: %s ", i/4, test_group.name);
1506             for (j = 0; j < 4; j++) {
1507                if (j)
1508                   printf("; ");
1509                frB_sp = (unsigned int *)&spec_sp_fargs[i + j];
1510                printf("%s(%08x)", test_group.op, *frB_sp);
1511                if (xvrespTest) {
1512                   float calc_diff = fabs(spec_sp_fargs[i + j]/256);
1513                   float sp_res;
1514                   memcpy(&sp_res, &dst_sp[j], 4);
1515                   float div_result = 1/spec_sp_fargs[i + j];
1516                   float real_diff = fabs(sp_res - div_result);
1517                   printf( " ==> %s",
1518                           ( ( sp_res == div_result )
1519                                    || ( isnan(sp_res) && isnan(div_result) )
1520                                    || ( real_diff <= calc_diff ) ) ? "PASS"
1521                                                                      : "FAIL");
1522                } else {
1523                   printf(" = %08x", dst_sp[j]);
1524                }
1525             }
1526             printf("\n");
1527          }
1528       }
1529       k++;
1530       printf( "\n" );
1531    }
1532 
1533 }
1534 
1535 /* This function assumes the instruction being tested requires two args. */
test_vx_vector_fp_ops(void)1536 static void test_vx_vector_fp_ops(void)
1537 {
1538    test_func_t func;
1539    int k;
1540    k = 0;
1541    build_special_fargs_table();
1542 
1543    while ((func = vx_vector_fp_tests[k].test_func)) {
1544       int idx, i, repeat = 1;
1545       vx_fp_test_t test_group = vx_vector_fp_tests[k];
1546       int stride = test_group.precision == DOUBLE_TEST ? 2 : 4;
1547       do_dot = False;
1548 
1549 again:
1550       for (i = 0; i < test_group.num_tests; i+=stride) {
1551          unsigned int * pv, condreg;
1552          unsigned int flags;
1553 
1554          pv = (unsigned int *)&vec_out;
1555          if (test_group.precision == DOUBLE_TEST)
1556             setup_dp_fp_args(&test_group.targs[i], False);
1557          else
1558             setup_sp_fp_args(&test_group.targs[i], False);
1559 
1560          // clear vec_out
1561          for (idx = 0; idx < 4; idx++, pv++)
1562             *pv = 0;
1563 
1564          // execute test insn
1565          SET_FPSCR_ZERO;
1566          SET_CR_XER_ZERO;
1567          (*func)();
1568          GET_CR(flags);
1569          if (test_group.type == VX_BASIC_CMP) {
1570             condreg = (flags & 0x000000f0) >> 4;
1571          } else {
1572             condreg = VX_NOT_CMP_OP;
1573          }
1574          print_vector_fp_result(condreg, &test_group, i);
1575       }
1576       printf("\n");
1577       if (repeat && test_group.type == VX_BASIC_CMP) {
1578          repeat = 0;
1579          do_dot = True;
1580          goto again;
1581       }
1582       k++;
1583       printf( "\n" );
1584    }
1585 }
1586 
1587 
1588 // The div doubleword test data
1589 signed long long div_dw_tdata[13][2] = {
1590                                        { 4, -4 },
1591                                        { 4, -3 },
1592                                        { 4, 4 },
1593                                        { 4, -5 },
1594                                        { 3, 8 },
1595                                        { 0x8000000000000000ULL, 0xa },
1596                                        { 0x50c, -1 },
1597                                        { 0x50c, -4096 },
1598                                        { 0x1234fedc, 0x8000a873 },
1599                                        { 0xabcd87651234fedcULL, 0xa123b893 },
1600                                        { 0x123456789abdcULL, 0 },
1601                                        { 0, 2 },
1602                                        { 0x77, 0xa3499 }
1603 };
1604 #define dw_tdata_len (sizeof(div_dw_tdata)/sizeof(signed long long)/2)
1605 
1606 // The div word test data
1607 unsigned int div_w_tdata[6][2] = {
1608                               { 0, 2 },
1609                               { 2, 0 },
1610                               { 0x7abc1234, 0xf0000000 },
1611                               { 0xfabc1234, 5 },
1612                               { 77, 66 },
1613                               { 5, 0xfabc1234 },
1614 };
1615 #define w_tdata_len (sizeof(div_w_tdata)/sizeof(unsigned int)/2)
1616 
1617 typedef struct div_ext_test
1618 {
1619    test_func_t test_func;
1620    const char *name;
1621    int num_tests;
1622    div_type_t div_type;
1623    precision_type_t precision;
1624 } div_ext_test_t;
1625 
1626 static div_ext_test_t div_tests[] = {
1627 #ifdef __powerpc64__
1628                                    { &test_divde, "divde", dw_tdata_len, DIV_BASE, DOUBLE_TEST },
1629                                    { &test_divde, "divdeo", dw_tdata_len, DIV_OE, DOUBLE_TEST },
1630 #endif
1631                                    { &test_divweu, "divweu", w_tdata_len, DIV_BASE, SINGLE_TEST },
1632                                    { &test_divweu, "divweuo", w_tdata_len, DIV_OE, SINGLE_TEST },
1633                                    { NULL, NULL, 0, 0, 0 }
1634 };
1635 
test_div_extensions(void)1636 static void test_div_extensions(void)
1637 {
1638    test_func_t func;
1639    int k;
1640    k = 0;
1641 
1642    while ((func = div_tests[k].test_func)) {
1643       int i, repeat = 1;
1644       div_ext_test_t test_group = div_tests[k];
1645       do_dot = False;
1646 
1647 again:
1648       for (i = 0; i < test_group.num_tests; i++) {
1649          unsigned int condreg;
1650 
1651          if (test_group.div_type == DIV_OE)
1652             do_OE = True;
1653          else
1654             do_OE = False;
1655 
1656          if (test_group.precision == DOUBLE_TEST) {
1657             r14 = div_dw_tdata[i][0];
1658             r15 = div_dw_tdata[i][1];
1659          } else {
1660             r14 = div_w_tdata[i][0];
1661             r15 = div_w_tdata[i][1];
1662          }
1663          // execute test insn
1664          (*func)();
1665          condreg = (div_flags & 0xf0000000) >> 28;
1666          printf("#%d: %s%s: ", i, test_group.name, do_dot ? "." : "");
1667          if (test_group.precision == DOUBLE_TEST) {
1668             printf("0x%016llx / 0x%016llx = 0x%016llx;",
1669                    div_dw_tdata[i][0], div_dw_tdata[i][1], (signed long long) r17);
1670          } else {
1671             printf("0x%08x / 0x%08x = 0x%08x;",
1672                    div_w_tdata[i][0], div_w_tdata[i][1], (unsigned int) r17);
1673          }
1674          printf(" CR=%x; XER=%x\n", condreg, div_xer);
1675       }
1676       printf("\n");
1677       if (repeat) {
1678          repeat = 0;
1679          do_dot = True;
1680          goto again;
1681       }
1682       k++;
1683       printf( "\n" );
1684    }
1685 
1686 }
1687 
test_fct_ops(void)1688 static void test_fct_ops(void)
1689 {
1690    test_func_t func;
1691    int k;
1692    k = 0;
1693 
1694    while ((func = fct_tests[k].test_func)) {
1695       int i, repeat = 1;
1696       simple_test_t test_group = fct_tests[k];
1697       do_dot = False;
1698 
1699 again:
1700       for (i = 0; i < nb_special_fargs; i++) {
1701          double result;
1702 #define SINGLE_MASK 0x00000000FFFFFFFFULL
1703 
1704          f14 = spec_fargs[i];
1705          // execute test insn
1706          SET_FPSCR_ZERO;
1707          (*func)();
1708          result = f17;
1709          printf("#%d: %s%s: ", i, test_group.name, do_dot ? "." : "");
1710          printf("0x%016llx (%e) ==> 0x%016llx\n",
1711                 *((unsigned long long *)(&spec_fargs[i])), spec_fargs[i],
1712                 test_group.precision == SINGLE_TEST ? (SINGLE_MASK &
1713                          *((unsigned long long *)(&result))) :
1714                          *((unsigned long long *)(&result)));
1715       }
1716       printf("\n");
1717       if (repeat) {
1718          repeat = 0;
1719          do_dot = True;
1720          goto again;
1721       }
1722       k++;
1723       printf( "\n" );
1724    }
1725 }
1726 
1727 #ifdef __powerpc64__
test_stdbrx(void)1728 void test_stdbrx(void)
1729 {
1730    unsigned long long store, val = 0xdeadbacf12345678ULL;
1731    printf("stdbrx: 0x%llx ==> ", val);
1732    r17 = (HWord_t)val;
1733    r14 = (HWord_t)&store;
1734    __asm__ __volatile__ ("stdbrx %0, 0, %1" : : "r"(r17), "r"(r14));
1735    printf("0x%llx\n", store);
1736    printf( "\n" );
1737 }
1738 #endif
1739 
1740 static test_table_t
1741          all_tests[] =
1742 {
1743                     { &test_vx_vector_one_fp_arg,
1744                       "Test VSX vector single arg instructions"},
1745                     { &test_vx_vector_fp_ops,
1746                       "Test VSX floating point compare and basic arithmetic instructions" },
1747 #ifdef __powerpc64__
1748                      { &test_bpermd,
1749                        "Test bit permute double"},
1750 #endif
1751                      { &test_xxsel,
1752                          "Test xxsel instruction" },
1753                      { &test_xxspltw,
1754                          "Test xxspltw instruction" },
1755                      { &test_div_extensions,
1756                        "Test div extensions" },
1757                      { &test_fct_ops,
1758                        "Test floating point convert [word | doubleword] unsigned, with round toward zero" },
1759 #ifdef __powerpc64__
1760                      { &test_stdbrx,
1761                       "Test stdbrx instruction"},
1762 #endif
1763                      { &test_vx_aORm_fp_ops,
1764                       "Test floating point arithmetic instructions -- with a{d|s}p or m{d|s}p"},
1765                      { &test_vx_simple_scalar_fp_ops,
1766                       "Test scalar floating point arithmetic instructions"},
1767                      { NULL, NULL }
1768 };
1769 #endif // HAS_VSX
1770 
main(int argc,char * argv[])1771 int main(int argc, char *argv[])
1772 {
1773 #ifdef HAS_VSX
1774 
1775    test_table_t aTest;
1776    test_func_t func;
1777    int i = 0;
1778 
1779    while ((func = all_tests[i].test_category)) {
1780       aTest = all_tests[i];
1781       printf( "%s\n", aTest.name );
1782       (*func)();
1783       i++;
1784    }
1785    if (spec_fargs)
1786      free(spec_fargs);
1787    if (spec_sp_fargs)
1788      free(spec_sp_fargs);
1789 
1790 #endif // HAS _VSX
1791 
1792    return 0;
1793 }
1794