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 _(mul4zc,NIL, NIL, NIL, P128, NIL, P128, P128, NIL, CARRY) \
84 _(mla4, P128, CARRY, NIL, P128, NIL, P128, P128, NIL, CARRY) \
85 _(mla4zc,P128, NIL, NIL, P128, NIL, P128, P128, NIL, CARRY) \
86 _(mmul4, NIL, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
87 _(mmla4, P128, NIL, P128, P128, P128, P128, P128, X128, CARRY) \
88 _(mont4, P128, NIL, NIL, P128, NIL, P128, P128, X128, CARRY)
90 /*----- Assembler test interface ------------------------------------------*/
94 #define PARAM(v, ty) ty *v,
95 #define PARAM_NIL(v, q)
96 #define PARAM_P128(v, q) PARAM(v, q p128)
97 #define PARAM_X128(v, q) PARAM(v, q x128)
98 #define PARAM_CARRY(v, q) PARAM(v, q carry)
100 #define DECLSTUB(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
101 extern void test_##fn(PARAM_##tyzz(zz, EMPTY) PARAM_##tycc(cc, EMPTY) \
102 PARAM_##tyyy(yy, EMPTY) \
103 PARAM_##tyu(u, const) PARAM_##tyx(x, const) \
104 PARAM_##tyv(v, const) PARAM_##tyy(y, const) \
105 unsigned n, unsigned long long *cyv);
108 /*----- Conversion functions ----------------------------------------------*/
110 #define DEFTYPE(ty, ld, st, nby) \
112 static void cvt_##ty(const char *buf, dstr *d) \
114 dstr dd = DSTR_INIT; \
119 type_hex.cvt(buf, &dd); \
120 if (dd.len != N(x->w)*nby) die(1, "invalid length for " #ty); \
121 dstr_ensure(d, sizeof(*x)); d->len = sizeof(*x); \
122 x = (ty *)d->buf; p = (const octet *)dd.buf; \
123 for (i = 0; i < N(x->w); i++) { x->w[i] = ld(p); p += nby; } \
127 static void dump_##ty(dstr *d, FILE *fp) \
129 dstr dd = DSTR_INIT; \
131 const ty *x = (const ty *)d->buf; \
134 dstr_ensure(&dd, N(x->w)*nby); p = (octet *)dd.buf; \
135 for (i = 0; i < N(x->w); i++) { st(p, x->w[i]); p += nby; } \
136 dd.len = N(x->w)*nby; \
137 type_hex.dump(&dd, fp); \
141 static int eq_##ty(const ty *x, const ty *y) \
145 for (i = 0; i < N(x->w); i++) \
146 if (x->w[i] != y->w[i]) return (0); \
150 static const struct test_type type_##ty = { cvt_##ty, dump_##ty };
152 DEFTYPE(p128
, LDW
, STW
, NWBY
)
153 DEFTYPE(x128
, LDW
, STW
, NWBY
)
154 DEFTYPE(carry
, LDD
, STD
, NDBY
)
156 /*----- Test functions ----------------------------------------------------*/
158 #define DECL_IN(v, ty) \
159 dstr *d_##v = dp++; const ty *v = (const ty *)d_##v->buf;
160 #define DECL_IN_NIL(v)
161 #define DECL_IN_P128(v) DECL_IN(v, p128)
162 #define DECL_IN_X128(v) DECL_IN(v, x128)
163 #define DECL_IN_CARRY(v) DECL_IN(v, carry)
165 #define DECL_OUT(v, ty) \
166 dstr dd_##v = DSTR_INIT, *d_##v = &dd_##v; ty *v;
167 #define DECL_OUT_NIL(v)
168 #define DECL_OUT_P128(v) DECL_OUT(v, p128)
169 #define DECL_OUT_X128(v) DECL_OUT(v, x128)
170 #define DECL_OUT_CARRY(v) DECL_OUT(v, carry)
172 #define INIT_OUT(v, ty) \
173 dstr_ensure(d_##v, sizeof(ty)); v = (ty *)d_##v->buf;
174 #define INIT_OUT_NIL(v)
175 #define INIT_OUT_P128(v) INIT_OUT(v, p128)
176 #define INIT_OUT_X128(v) INIT_OUT(v, x128)
177 #define INIT_OUT_CARRY(v) INIT_OUT(v, carry)
181 #define ARG_P128(v) ARG(v)
182 #define ARG_X128(v) ARG(v)
183 #define ARG_CARRY(v) ARG(v)
185 #define CHECK(v, vv, ty) if (!eq_##ty(v, vv)) ok = 0;
186 #define CHECK_NIL(v, vv)
187 #define CHECK_P128(v, vv) CHECK(v, vv, p128)
188 #define CHECK_X128(v, vv) CHECK(v, vv, x128)
189 #define CHECK_CARRY(v, vv) CHECK(v, vv, carry)
191 #define DUMP(v, ty) \
192 fprintf(stderr, "\n\t%-6s = ", #v); dump_##ty(d_##v, stderr);
194 #define DUMP_P128(v) DUMP(v, p128)
195 #define DUMP_X128(v) DUMP(v, x128)
196 #define DUMP_CARRY(v) DUMP(v, carry)
198 #define COPY(v, vv, ty) *v = *vv;
199 #define COPY_NIL(v, vv)
200 #define COPY_P128(v, vv) COPY(v, vv, p128)
201 #define COPY_X128(v, vv) COPY(v, vv, x128)
202 #define COPY_CARRY(v, vv) COPY(v, vv, carry)
204 #define FREE_OUT(v, ty) dstr_destroy(d_##v);
205 #define FREE_OUT_NIL(v)
206 #define FREE_OUT_P128(v) FREE_OUT(v, p128)
207 #define FREE_OUT_X128(v) FREE_OUT(v, x128)
208 #define FREE_OUT_CARRY(v) FREE_OUT(v, carry)
210 #define DEFTESTFN(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
212 static int v##fn(dstr dv[]) \
221 DECL_IN_##tyzz(zz_exp) \
222 DECL_IN_##tyyy(yy_exp) \
223 DECL_IN_##tycc(cc_exp) \
224 DECL_OUT_##tyzz(zz_out) \
225 DECL_OUT_##tyyy(yy_out) \
226 DECL_OUT_##tycc(cc_out) \
227 unsigned long long cyv[1]; \
230 INIT_OUT_##tyzz(zz_out) \
231 INIT_OUT_##tyyy(yy_out) \
232 INIT_OUT_##tycc(cc_out) \
234 COPY_##tya(zz_out, a); \
235 COPY_##tyc(cc_out, c); \
236 test_##fn(zz_out ARG_##tycc(cc_out) ARG_##tyyy(yy_out) \
237 ARG_##tyu(u) ARG_##tyx(x) ARG_##tyv(v) ARG_##tyy(y), \
239 CHECK_##tyzz(zz_exp, zz_out) \
240 CHECK_##tyyy(yy_exp, yy_out) \
241 CHECK_##tycc(cc_exp, cc_out) \
244 fputs(#fn " failed", stderr); \
251 DUMP_##tyzz(zz_exp) \
252 DUMP_##tyzz(zz_out) \
253 DUMP_##tyyy(yy_exp) \
254 DUMP_##tyyy(yy_out) \
255 DUMP_##tycc(cc_exp) \
256 DUMP_##tycc(cc_out) \
257 fputc('\n', stderr); \
260 FREE_OUT_##tyzz(zz_out); \
261 FREE_OUT_##tyyy(yy_out); \
262 FREE_OUT_##tycc(cc_out); \
269 /*----- Main code ---------------------------------------------------------*/
272 #define P128 &type_p128,
273 #define X128 &type_x128,
274 #define CARRY &type_carry,
276 static test_chunk tests
[] = {
277 #define DEFCHUNK(fn, tya, tyc, tyu, tyx, tyv, tyy, tyzz, tyyy, tycc) \
278 { #fn, v##fn, { tya tyc tyu tyx tyv tyy tyzz tyyy tycc } },
284 int main(int argc
, char *argv
[])
287 if (!cpu_features_p())
288 { fprintf(stderr
, "required cpu feature not available\n"); exit(77); }
289 test_run(argc
, argv
, tests
, SRCDIR
"/t/mpx-mul4");
293 /*----- That's all, folks -------------------------------------------------*/