85a153271c559fc17185718be784af0635e075d8
3 * Testing optimized 128-bit multipliers
5 * (c) 2017 Straylight/Edgeware
8 /*----- Licensing notice --------------------------------------------------*
10 * This file is part of Catacomb.
12 * Catacomb is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU Library General Public License as
14 * published by the Free Software Foundation; either version 2 of the
15 * License, or (at your option) any later version.
17 * Catacomb is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU Library General Public License for more details.
22 * You should have received a copy of the GNU Library General Public
23 * License along with Catacomb; if not, write to the Free
24 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
28 /*----- Header files ------------------------------------------------------*/
32 #ifdef ENABLE_ASM_DEBUG
41 #include <mLib/macros.h>
42 #include <mLib/report.h>
43 #include <mLib/testrig.h>
50 /*----- CPU feature detection ---------------------------------------------*/
53 # define VARIANT _x86_sse2
55 static int cpu_features_p(void) { return (cpu_feature_p(CPUFEAT_X86_SSE2
)); }
59 # define VARIANT _amd64_sse2
61 static int cpu_features_p(void) { return (cpu_feature_p(CPUFEAT_X86_SSE2
)); }
65 # error "Unsupported CPU family."
75 typedef struct { mpw w
[4]; } p128
;
76 typedef struct { mpw w
[8]; } x128
;
77 typedef struct { mpd w
[6]; } carry
;
80 /*----- Test operation table ----------------------------------------------*/
83 /* a c u x v y z' y' c' */ \
84 _(dmul4, NIL, CARRY, P128, P128, P128, P128, P128, NIL, CARRY) \
85 _(dmla4, P128, CARRY, P128, P128, P128, P128, P128, NIL, CARRY) \
86 _(mul4, NIL, CARRY, NIL, P128, NIL, P128, P128, NIL, CARRY) \
87 _(mul4zc,NIL, NIL, NIL, P128, NIL, P128, P128, NIL, CARRY) \
88 _(mla4, P128, CARRY, NIL, P128, NIL, P128, P128, NIL, CARRY) \
89 _(mla4zc,P128, NIL, NIL, P128, NIL, P128, P128, NIL, CARRY) \
90 _(mmul4, NIL, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
91 _(mmla4, P128, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
92 _(mont4, P128, NIL, NIL, P128, NIL, P128, P128, X128, CARRY)
94 /*----- Assembler test interface ------------------------------------------*/
98 #define PARAM(v, ty) ty *v,
99 #define PARAM_NIL(v, q)
100 #define PARAM_P128(v, q) PARAM(v, q p128)
101 #define PARAM_X128(v, q) PARAM(v, q x128)
102 #define PARAM_CARRY(v, q) PARAM(v, q carry)
104 #define DECLSTUB(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
105 extern void test_##fn(PARAM_##tyzz(zz, EMPTY) PARAM_##tycc(cc, EMPTY) \
106 PARAM_##tyyy(yy, EMPTY) \
107 PARAM_##tyu(u, const) PARAM_##tyx(x, const) \
108 PARAM_##tyv(v, const) PARAM_##tyy(y, const) \
109 unsigned n, unsigned long long *cyv);
112 /*----- Conversion functions ----------------------------------------------*/
114 static mp
*combine_mpw(mp
*d
, const mpw
*v
, size_t n
, unsigned off
)
122 for (i
= 0, o
= 0; i
< n
; i
++, o
+= off
) {
123 w
[0] = v
[i
]; mp_build(&m
, w
, w
+ 1);
124 t
= mp_lsl(t
, &m
, o
); d
= mp_add(d
, d
, t
);
126 mp_drop(t
); return (d
);
129 static mp
*combine_mpd(mp
*d
, const mpd
*v
, size_t n
, unsigned off
)
137 for (i
= 0, o
= 0; i
< n
; i
++, o
+= off
) {
138 w
[0] = MPW(v
[i
]); w
[1] = MPW(v
[i
] >> MPW_BITS
); mp_build(&m
, w
, w
+ 2);
139 t
= mp_lsl(t
, &m
, o
); d
= mp_add(d
, d
, t
);
141 mp_drop(t
); return (d
);
144 #define DEFTYPE(ty, ld, st, nby, combfn, off) \
146 static void cvt_##ty(const char *buf, dstr *d) \
148 dstr dd = DSTR_INIT; \
153 type_hex.cvt(buf, &dd); \
154 if (dd.len != N(x->w)*nby) die(1, "invalid length for " #ty); \
155 dstr_ensure(d, sizeof(*x)); d->len = sizeof(*x); \
156 x = (ty *)d->buf; p = (const octet *)dd.buf; \
157 for (i = 0; i < N(x->w); i++) { x->w[i] = ld(p); p += nby; } \
161 static void dump_##ty(dstr *d, FILE *fp) \
163 dstr dd = DSTR_INIT; \
165 const ty *x = (const ty *)d->buf; \
166 mp *xx = combfn(MP_NEW, x->w, N(x->w), off); \
169 dstr_ensure(&dd, N(x->w)*nby); p = (octet *)dd.buf; \
170 for (i = 0; i < N(x->w); i++) { st(p, x->w[i]); p += nby; } \
171 dd.len = N(x->w)*nby; \
172 type_hex.dump(&dd, fp); \
175 fputs(" = 0x", fp); mp_writefile(xx, fp, 16); \
176 fputs(" = ", fp); mp_writefile(xx, fp, 10); \
180 static int eq_##ty(const ty *x, const ty *y) \
182 mp *xx = combfn(MP_NEW, x->w, N(x->w), off), \
183 *yy = combfn(MP_NEW, y->w, N(y->w), off); \
184 int rc = MP_EQ(xx, yy); \
185 MP_DROP(xx); MP_DROP(yy); \
189 static const struct test_type type_##ty = { cvt_##ty, dump_##ty };
191 DEFTYPE(p128
, LDW
, STW
, NWBY
, combine_mpw
, MPW_BITS
)
192 DEFTYPE(x128
, LDW
, STW
, NWBY
, combine_mpw
, MPW_BITS
/2)
193 DEFTYPE(carry
, LDD
, STD
, NDBY
, combine_mpd
, MPW_BITS
/2)
195 /*----- Test functions ----------------------------------------------------*/
197 #define DECL_IN(v, ty) \
198 dstr *d_##v = dp++; const ty *v = (const ty *)d_##v->buf;
199 #define DECL_IN_NIL(v)
200 #define DECL_IN_P128(v) DECL_IN(v, p128)
201 #define DECL_IN_X128(v) DECL_IN(v, x128)
202 #define DECL_IN_CARRY(v) DECL_IN(v, carry)
204 #define DECL_OUT(v, ty) \
205 dstr dd_##v = DSTR_INIT, *d_##v = &dd_##v; ty *v;
206 #define DECL_OUT_NIL(v)
207 #define DECL_OUT_P128(v) DECL_OUT(v, p128)
208 #define DECL_OUT_X128(v) DECL_OUT(v, x128)
209 #define DECL_OUT_CARRY(v) DECL_OUT(v, carry)
211 #define INIT_OUT(v, ty) \
212 dstr_ensure(d_##v, sizeof(ty)); v = (ty *)d_##v->buf;
213 #define INIT_OUT_NIL(v)
214 #define INIT_OUT_P128(v) INIT_OUT(v, p128)
215 #define INIT_OUT_X128(v) INIT_OUT(v, x128)
216 #define INIT_OUT_CARRY(v) INIT_OUT(v, carry)
220 #define ARG_P128(v) ARG(v)
221 #define ARG_X128(v) ARG(v)
222 #define ARG_CARRY(v) ARG(v)
224 #define CHECK(v, vv, ty) if (!eq_##ty(v, vv)) ok = 0;
225 #define CHECK_NIL(v, vv)
226 #define CHECK_P128(v, vv) CHECK(v, vv, p128)
227 #define CHECK_X128(v, vv) CHECK(v, vv, x128)
228 #define CHECK_CARRY(v, vv) CHECK(v, vv, carry)
230 #define DUMP(v, ty) \
231 fprintf(stderr, "\n\t%-6s = ", #v); dump_##ty(d_##v, stderr);
233 #define DUMP_P128(v) DUMP(v, p128)
234 #define DUMP_X128(v) DUMP(v, x128)
235 #define DUMP_CARRY(v) DUMP(v, carry)
237 #define COPY(v, vv, ty) *v = *vv;
238 #define COPY_NIL(v, vv)
239 #define COPY_P128(v, vv) COPY(v, vv, p128)
240 #define COPY_X128(v, vv) COPY(v, vv, x128)
241 #define COPY_CARRY(v, vv) COPY(v, vv, carry)
243 #define FREE_OUT(v, ty) dstr_destroy(d_##v);
244 #define FREE_OUT_NIL(v)
245 #define FREE_OUT_P128(v) FREE_OUT(v, p128)
246 #define FREE_OUT_X128(v) FREE_OUT(v, x128)
247 #define FREE_OUT_CARRY(v) FREE_OUT(v, carry)
249 #define DEFTESTFN(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
251 static int v##fn(dstr dv[]) \
260 DECL_IN_##tyzz(zz_exp) \
261 DECL_IN_##tyyy(yy_exp) \
262 DECL_IN_##tycc(cc_exp) \
263 DECL_OUT_##tyzz(zz_out) \
264 DECL_OUT_##tyyy(yy_out) \
265 DECL_OUT_##tycc(cc_out) \
266 unsigned long long cyv[1]; \
269 INIT_OUT_##tyzz(zz_out) \
270 INIT_OUT_##tyyy(yy_out) \
271 INIT_OUT_##tycc(cc_out) \
273 COPY_##tya(zz_out, a); \
274 COPY_##tyc(cc_out, c); \
275 test_##fn(zz_out ARG_##tycc(cc_out) ARG_##tyyy(yy_out) \
276 ARG_##tyu(u) ARG_##tyx(x) ARG_##tyv(v) ARG_##tyy(y), \
278 CHECK_##tyzz(zz_exp, zz_out) \
279 CHECK_##tyyy(yy_exp, yy_out) \
280 CHECK_##tycc(cc_exp, cc_out) \
283 fputs(#fn " failed", stderr); \
290 DUMP_##tyzz(zz_exp) \
291 DUMP_##tyzz(zz_out) \
292 DUMP_##tyyy(yy_exp) \
293 DUMP_##tyyy(yy_out) \
294 DUMP_##tycc(cc_exp) \
295 DUMP_##tycc(cc_out) \
296 fputc('\n', stderr); \
299 FREE_OUT_##tyzz(zz_out); \
300 FREE_OUT_##tyyy(yy_out); \
301 FREE_OUT_##tycc(cc_out); \
308 /*----- Main code ---------------------------------------------------------*/
311 #define P128 &type_p128,
312 #define X128 &type_x128,
313 #define CARRY &type_carry,
315 static test_chunk tests
[] = {
316 #define DEFCHUNK(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
317 { #fn, v##fn, { tya tyc tyu tyx tyv tyy tyzz tyyy tycc } },
323 int main(int argc
, char *argv
[])
326 #ifdef ENABLE_ASM_DEBUG
329 if (!cpu_features_p())
330 { fprintf(stderr
, "required cpu feature not available\n"); exit(77); }
331 test_run(argc
, argv
, tests
, SRCDIR
"/t/mpx-mul4");
335 /*----- That's all, folks -------------------------------------------------*/