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 ------------------------------------------------------*/
37 #include <mLib/macros.h>
38 #include <mLib/report.h>
39 #include <mLib/testrig.h>
46 /*----- CPU feature detection ---------------------------------------------*/
49 # define VARIANT _x86_sse2
51 static int cpu_features_p(void) { return (cpu_feature_p(CPUFEAT_X86_SSE2)); }
55 # define VARIANT _amd64_sse2
57 static int cpu_features_p(void) { return (cpu_feature_p(CPUFEAT_X86_SSE2)); }
61 # error "Unsupported CPU family."
71 typedef struct { mpw w[4]; } p128;
72 typedef struct { mpw w[8]; } x128;
73 typedef struct { mpd w[6]; } carry;
76 /*----- Test operation table ----------------------------------------------*/
79 /* a c u x v y z' y' c' */ \
80 _(dmul4, NIL, CARRY, P128, P128, P128, P128, P128, NIL, CARRY) \
81 _(dmla4, P128, CARRY, P128, P128, P128, P128, P128, NIL, CARRY) \
82 _(mul4, NIL, CARRY, NIL, P128, NIL, P128, P128, NIL, CARRY) \
83 _(mla4, P128, CARRY, NIL, P128, NIL, P128, P128, NIL, CARRY) \
84 _(mmul4, NIL, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
85 _(mmla4, P128, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
86 _(mont4, P128, NIL, NIL, P128, NIL, P128, P128, X128, CARRY)
88 /*----- Assembler test interface ------------------------------------------*/
92 #define PARAM(v, ty) ty *v,
93 #define PARAM_NIL(v, q)
94 #define PARAM_P128(v, q) PARAM(v, q p128)
95 #define PARAM_X128(v, q) PARAM(v, q x128)
96 #define PARAM_CARRY(v, q) PARAM(v, q carry)
98 #define DECLSTUB(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
99 extern void test_##fn(PARAM_##tyzz(zz, EMPTY) PARAM_##tycc(cc, EMPTY) \
100 PARAM_##tyyy(yy, EMPTY) \
101 PARAM_##tyu(u, const) PARAM_##tyx(x, const) \
102 PARAM_##tyv(v, const) PARAM_##tyy(y, const) \
103 unsigned n, unsigned long long *cyv);
106 /*----- Conversion functions ----------------------------------------------*/
108 #define DEFTYPE(ty, ld, st, nby) \
110 static void cvt_##ty(const char *buf, dstr *d) \
112 dstr dd = DSTR_INIT; \
117 type_hex.cvt(buf, &dd); \
118 if (dd.len != N(x->w)*nby) die(1, "invalid length for " #ty); \
119 dstr_ensure(d, sizeof(*x)); d->len = sizeof(*x); \
120 x = (ty *)d->buf; p = (const octet *)dd.buf; \
121 for (i = 0; i < N(x->w); i++) { x->w[i] = ld(p); p += nby; } \
125 static void dump_##ty(dstr *d, FILE *fp) \
127 dstr dd = DSTR_INIT; \
129 const ty *x = (const ty *)d->buf; \
132 dstr_ensure(&dd, N(x->w)*nby); p = (octet *)dd.buf; \
133 for (i = 0; i < N(x->w); i++) { st(p, x->w[i]); p += nby; } \
134 dd.len = N(x->w)*nby; \
135 type_hex.dump(&dd, fp); \
139 static int eq_##ty(const ty *x, const ty *y) \
143 for (i = 0; i < N(x->w); i++) \
144 if (x->w[i] != y->w[i]) return (0); \
148 static const struct test_type type_##ty = { cvt_##ty, dump_##ty };
150 DEFTYPE(p128, LDW, STW, NWBY)
151 DEFTYPE(x128, LDW, STW, NWBY)
152 DEFTYPE(carry, LDD, STD, NDBY)
154 /*----- Test functions ----------------------------------------------------*/
156 #define DECL_IN(v, ty) \
157 dstr *d_##v = dp++; const ty *v = (const ty *)d_##v->buf;
158 #define DECL_IN_NIL(v)
159 #define DECL_IN_P128(v) DECL_IN(v, p128)
160 #define DECL_IN_X128(v) DECL_IN(v, x128)
161 #define DECL_IN_CARRY(v) DECL_IN(v, carry)
163 #define DECL_OUT(v, ty) \
164 dstr dd_##v = DSTR_INIT, *d_##v = &dd_##v; ty *v;
165 #define DECL_OUT_NIL(v)
166 #define DECL_OUT_P128(v) DECL_OUT(v, p128)
167 #define DECL_OUT_X128(v) DECL_OUT(v, x128)
168 #define DECL_OUT_CARRY(v) DECL_OUT(v, carry)
170 #define INIT_OUT(v, ty) \
171 dstr_ensure(d_##v, sizeof(ty)); v = (ty *)d_##v->buf;
172 #define INIT_OUT_NIL(v)
173 #define INIT_OUT_P128(v) INIT_OUT(v, p128)
174 #define INIT_OUT_X128(v) INIT_OUT(v, x128)
175 #define INIT_OUT_CARRY(v) INIT_OUT(v, carry)
179 #define ARG_P128(v) ARG(v)
180 #define ARG_X128(v) ARG(v)
181 #define ARG_CARRY(v) ARG(v)
183 #define CHECK(v, vv, ty) if (!eq_##ty(v, vv)) ok = 0;
184 #define CHECK_NIL(v, vv)
185 #define CHECK_P128(v, vv) CHECK(v, vv, p128)
186 #define CHECK_X128(v, vv) CHECK(v, vv, x128)
187 #define CHECK_CARRY(v, vv) CHECK(v, vv, carry)
189 #define DUMP(v, ty) \
190 fprintf(stderr, "\n\t%-6s = ", #v); dump_##ty(d_##v, stderr);
192 #define DUMP_P128(v) DUMP(v, p128)
193 #define DUMP_X128(v) DUMP(v, x128)
194 #define DUMP_CARRY(v) DUMP(v, carry)
196 #define COPY(v, vv, ty) *v = *vv;
197 #define COPY_NIL(v, vv)
198 #define COPY_P128(v, vv) COPY(v, vv, p128)
199 #define COPY_X128(v, vv) COPY(v, vv, x128)
200 #define COPY_CARRY(v, vv) COPY(v, vv, carry)
202 #define FREE_OUT(v, ty) dstr_destroy(d_##v);
203 #define FREE_OUT_NIL(v)
204 #define FREE_OUT_P128(v) FREE_OUT(v, p128)
205 #define FREE_OUT_X128(v) FREE_OUT(v, x128)
206 #define FREE_OUT_CARRY(v) FREE_OUT(v, carry)
208 #define DEFTESTFN(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
210 static int v##fn(dstr dv[]) \
219 DECL_IN_##tyzz(zz_exp) \
220 DECL_IN_##tyyy(yy_exp) \
221 DECL_IN_##tycc(cc_exp) \
222 DECL_OUT_##tyzz(zz_out) \
223 DECL_OUT_##tyyy(yy_out) \
224 DECL_OUT_##tycc(cc_out) \
225 unsigned long long cyv[1]; \
228 INIT_OUT_##tyzz(zz_out) \
229 INIT_OUT_##tyyy(yy_out) \
230 INIT_OUT_##tycc(cc_out) \
232 COPY_##tya(zz_out, a); \
233 COPY_##tyc(cc_out, c); \
234 test_##fn(zz_out ARG_##tycc(cc_out) ARG_##tyyy(yy_out) \
235 ARG_##tyu(u) ARG_##tyx(x) ARG_##tyv(v) ARG_##tyy(y), \
237 CHECK_##tyzz(zz_exp, zz_out) \
238 CHECK_##tyyy(yy_exp, yy_out) \
239 CHECK_##tycc(cc_exp, cc_out) \
242 fputs(#fn " failed", stderr); \
249 DUMP_##tyzz(zz_exp) \
250 DUMP_##tyzz(zz_out) \
251 DUMP_##tyyy(yy_exp) \
252 DUMP_##tyyy(yy_out) \
253 DUMP_##tycc(cc_exp) \
254 DUMP_##tycc(cc_out) \
255 fputc('\n', stderr); \
258 FREE_OUT_##tyzz(zz_out); \
259 FREE_OUT_##tyyy(yy_out); \
260 FREE_OUT_##tycc(cc_out); \
267 /*----- Main code ---------------------------------------------------------*/
270 #define P128 &type_p128,
271 #define X128 &type_x128,
272 #define CARRY &type_carry,
274 static test_chunk tests[] = {
275 #define DEFCHUNK(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
276 { #fn, v##fn, { tya tyc tyu tyx tyv tyy tyzz tyyy tycc } },
282 int main(int argc, char *argv[])
285 if (!cpu_features_p())
286 { fprintf(stderr, "required cpu feature not available\n"); exit(77); }
287 test_run(argc, argv, tests, SRCDIR "/t/mpx-mul4");
291 /*----- That's all, folks -------------------------------------------------*/