1/*
2 * RISC-V translation routines for the RV64Zfh Standard Extension.
3 *
4 * Copyright (c) 2020 Chih-Min Chao, chihmin.chao@sifive.com
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2 or later, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program.  If not, see <http://www.gnu.org/licenses/>.
17 */
18
19#define REQUIRE_ZFH(ctx) do { \
20    if (!ctx->ext_zfh) {      \
21        return false;         \
22    }                         \
23} while (0)
24
25static bool trans_flh(DisasContext *ctx, arg_flh *a)
26{
27    TCGv_i64 dest;
28    TCGv t0;
29
30    REQUIRE_FPU;
31    REQUIRE_ZFH(ctx);
32
33    t0 = get_gpr(ctx, a->rs1, EXT_NONE);
34    if (a->imm) {
35        TCGv temp = temp_new(ctx);
36        tcg_gen_addi_tl(temp, t0, a->imm);
37        t0 = temp;
38    }
39
40    dest = cpu_fpr[a->rd];
41    tcg_gen_qemu_ld_i64(dest, t0, ctx->mem_idx, MO_TEUW);
42    gen_nanbox_h(dest, dest);
43
44    mark_fs_dirty(ctx);
45    return true;
46}
47
48static bool trans_fsh(DisasContext *ctx, arg_fsh *a)
49{
50    TCGv t0;
51
52    REQUIRE_FPU;
53    REQUIRE_ZFH(ctx);
54
55    t0 = get_gpr(ctx, a->rs1, EXT_NONE);
56    if (a->imm) {
57        TCGv temp = tcg_temp_new();
58        tcg_gen_addi_tl(temp, t0, a->imm);
59        t0 = temp;
60    }
61
62    tcg_gen_qemu_st_i64(cpu_fpr[a->rs2], t0, ctx->mem_idx, MO_TEUW);
63
64    return true;
65}
66
67static bool trans_fmadd_h(DisasContext *ctx, arg_fmadd_h *a)
68{
69    REQUIRE_FPU;
70    REQUIRE_ZFH(ctx);
71
72    gen_set_rm(ctx, a->rm);
73    gen_helper_fmadd_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
74                       cpu_fpr[a->rs2], cpu_fpr[a->rs3]);
75    mark_fs_dirty(ctx);
76    return true;
77}
78
79static bool trans_fmsub_h(DisasContext *ctx, arg_fmsub_h *a)
80{
81    REQUIRE_FPU;
82    REQUIRE_ZFH(ctx);
83
84    gen_set_rm(ctx, a->rm);
85    gen_helper_fmsub_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
86                       cpu_fpr[a->rs2], cpu_fpr[a->rs3]);
87    mark_fs_dirty(ctx);
88    return true;
89}
90
91static bool trans_fnmsub_h(DisasContext *ctx, arg_fnmsub_h *a)
92{
93    REQUIRE_FPU;
94    REQUIRE_ZFH(ctx);
95
96    gen_set_rm(ctx, a->rm);
97    gen_helper_fnmsub_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
98                        cpu_fpr[a->rs2], cpu_fpr[a->rs3]);
99    mark_fs_dirty(ctx);
100    return true;
101}
102
103static bool trans_fnmadd_h(DisasContext *ctx, arg_fnmadd_h *a)
104{
105    REQUIRE_FPU;
106    REQUIRE_ZFH(ctx);
107
108    gen_set_rm(ctx, a->rm);
109    gen_helper_fnmadd_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
110                        cpu_fpr[a->rs2], cpu_fpr[a->rs3]);
111    mark_fs_dirty(ctx);
112    return true;
113}
114
115static bool trans_fadd_h(DisasContext *ctx, arg_fadd_h *a)
116{
117    REQUIRE_FPU;
118    REQUIRE_ZFH(ctx);
119
120    gen_set_rm(ctx, a->rm);
121    gen_helper_fadd_h(cpu_fpr[a->rd], cpu_env,
122                      cpu_fpr[a->rs1], cpu_fpr[a->rs2]);
123    mark_fs_dirty(ctx);
124    return true;
125}
126
127static bool trans_fsub_h(DisasContext *ctx, arg_fsub_h *a)
128{
129    REQUIRE_FPU;
130    REQUIRE_ZFH(ctx);
131
132    gen_set_rm(ctx, a->rm);
133    gen_helper_fsub_h(cpu_fpr[a->rd], cpu_env,
134                      cpu_fpr[a->rs1], cpu_fpr[a->rs2]);
135    mark_fs_dirty(ctx);
136    return true;
137}
138
139static bool trans_fmul_h(DisasContext *ctx, arg_fmul_h *a)
140{
141    REQUIRE_FPU;
142    REQUIRE_ZFH(ctx);
143
144    gen_set_rm(ctx, a->rm);
145    gen_helper_fmul_h(cpu_fpr[a->rd], cpu_env,
146                      cpu_fpr[a->rs1], cpu_fpr[a->rs2]);
147    mark_fs_dirty(ctx);
148    return true;
149}
150
151static bool trans_fdiv_h(DisasContext *ctx, arg_fdiv_h *a)
152{
153    REQUIRE_FPU;
154    REQUIRE_ZFH(ctx);
155
156    gen_set_rm(ctx, a->rm);
157    gen_helper_fdiv_h(cpu_fpr[a->rd], cpu_env,
158                      cpu_fpr[a->rs1], cpu_fpr[a->rs2]);
159    mark_fs_dirty(ctx);
160    return true;
161}
162
163static bool trans_fsqrt_h(DisasContext *ctx, arg_fsqrt_h *a)
164{
165    REQUIRE_FPU;
166    REQUIRE_ZFH(ctx);
167
168    gen_set_rm(ctx, a->rm);
169    gen_helper_fsqrt_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1]);
170    mark_fs_dirty(ctx);
171    return true;
172}
173
174static bool trans_fsgnj_h(DisasContext *ctx, arg_fsgnj_h *a)
175{
176    REQUIRE_FPU;
177    REQUIRE_ZFH(ctx);
178
179    if (a->rs1 == a->rs2) { /* FMOV */
180        gen_check_nanbox_h(cpu_fpr[a->rd], cpu_fpr[a->rs1]);
181    } else {
182        TCGv_i64 rs1 = tcg_temp_new_i64();
183        TCGv_i64 rs2 = tcg_temp_new_i64();
184
185        gen_check_nanbox_h(rs1, cpu_fpr[a->rs1]);
186        gen_check_nanbox_h(rs2, cpu_fpr[a->rs2]);
187
188        /* This formulation retains the nanboxing of rs2. */
189        tcg_gen_deposit_i64(cpu_fpr[a->rd], rs2, rs1, 0, 15);
190        tcg_temp_free_i64(rs1);
191        tcg_temp_free_i64(rs2);
192    }
193
194    mark_fs_dirty(ctx);
195    return true;
196}
197
198static bool trans_fsgnjn_h(DisasContext *ctx, arg_fsgnjn_h *a)
199{
200    TCGv_i64 rs1, rs2, mask;
201
202    REQUIRE_FPU;
203    REQUIRE_ZFH(ctx);
204
205    rs1 = tcg_temp_new_i64();
206    gen_check_nanbox_h(rs1, cpu_fpr[a->rs1]);
207
208    if (a->rs1 == a->rs2) { /* FNEG */
209        tcg_gen_xori_i64(cpu_fpr[a->rd], rs1, MAKE_64BIT_MASK(15, 1));
210    } else {
211        rs2 = tcg_temp_new_i64();
212        gen_check_nanbox_h(rs2, cpu_fpr[a->rs2]);
213
214        /*
215         * Replace bit 15 in rs1 with inverse in rs2.
216         * This formulation retains the nanboxing of rs1.
217         */
218        mask = tcg_const_i64(~MAKE_64BIT_MASK(15, 1));
219        tcg_gen_not_i64(rs2, rs2);
220        tcg_gen_andc_i64(rs2, rs2, mask);
221        tcg_gen_and_i64(rs1, mask, rs1);
222        tcg_gen_or_i64(cpu_fpr[a->rd], rs1, rs2);
223
224        tcg_temp_free_i64(mask);
225        tcg_temp_free_i64(rs2);
226    }
227    mark_fs_dirty(ctx);
228    return true;
229}
230
231static bool trans_fsgnjx_h(DisasContext *ctx, arg_fsgnjx_h *a)
232{
233    TCGv_i64 rs1, rs2;
234
235    REQUIRE_FPU;
236    REQUIRE_ZFH(ctx);
237
238    rs1 = tcg_temp_new_i64();
239    gen_check_nanbox_s(rs1, cpu_fpr[a->rs1]);
240
241    if (a->rs1 == a->rs2) { /* FABS */
242        tcg_gen_andi_i64(cpu_fpr[a->rd], rs1, ~MAKE_64BIT_MASK(15, 1));
243    } else {
244        rs2 = tcg_temp_new_i64();
245        gen_check_nanbox_s(rs2, cpu_fpr[a->rs2]);
246
247        /*
248         * Xor bit 15 in rs1 with that in rs2.
249         * This formulation retains the nanboxing of rs1.
250         */
251        tcg_gen_andi_i64(rs2, rs2, MAKE_64BIT_MASK(15, 1));
252        tcg_gen_xor_i64(cpu_fpr[a->rd], rs1, rs2);
253
254        tcg_temp_free_i64(rs2);
255    }
256
257    mark_fs_dirty(ctx);
258    return true;
259}
260
261static bool trans_fmin_h(DisasContext *ctx, arg_fmin_h *a)
262{
263    REQUIRE_FPU;
264    REQUIRE_ZFH(ctx);
265
266    gen_helper_fmin_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
267                      cpu_fpr[a->rs2]);
268    mark_fs_dirty(ctx);
269    return true;
270}
271
272static bool trans_fmax_h(DisasContext *ctx, arg_fmax_h *a)
273{
274    REQUIRE_FPU;
275    REQUIRE_ZFH(ctx);
276
277    gen_helper_fmax_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1],
278                      cpu_fpr[a->rs2]);
279    mark_fs_dirty(ctx);
280    return true;
281}
282
283static bool trans_fcvt_s_h(DisasContext *ctx, arg_fcvt_s_h *a)
284{
285    REQUIRE_FPU;
286    REQUIRE_ZFH(ctx);
287
288    gen_set_rm(ctx, a->rm);
289    gen_helper_fcvt_s_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1]);
290
291    mark_fs_dirty(ctx);
292
293    return true;
294}
295
296static bool trans_fcvt_d_h(DisasContext *ctx, arg_fcvt_d_h *a)
297{
298    REQUIRE_FPU;
299    REQUIRE_ZFH(ctx);
300    REQUIRE_EXT(ctx, RVD);
301
302    gen_set_rm(ctx, a->rm);
303    gen_helper_fcvt_d_h(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1]);
304
305    mark_fs_dirty(ctx);
306
307
308    return true;
309}
310
311static bool trans_fcvt_h_s(DisasContext *ctx, arg_fcvt_h_s *a)
312{
313    REQUIRE_FPU;
314    REQUIRE_ZFH(ctx);
315
316    gen_set_rm(ctx, a->rm);
317    gen_helper_fcvt_h_s(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1]);
318
319    mark_fs_dirty(ctx);
320
321    return true;
322}
323
324static bool trans_fcvt_h_d(DisasContext *ctx, arg_fcvt_h_d *a)
325{
326    REQUIRE_FPU;
327    REQUIRE_ZFH(ctx);
328    REQUIRE_EXT(ctx, RVD);
329
330    gen_set_rm(ctx, a->rm);
331    gen_helper_fcvt_h_d(cpu_fpr[a->rd], cpu_env, cpu_fpr[a->rs1]);
332
333    mark_fs_dirty(ctx);
334
335    return true;
336}
337
338static bool trans_fcvt_w_h(DisasContext *ctx, arg_fcvt_w_h *a)
339{
340    REQUIRE_FPU;
341    REQUIRE_ZFH(ctx);
342
343    TCGv dest = dest_gpr(ctx, a->rd);
344
345    gen_set_rm(ctx, a->rm);
346    gen_helper_fcvt_w_h(dest, cpu_env, cpu_fpr[a->rs1]);
347    gen_set_gpr(ctx, a->rd, dest);
348    return true;
349}
350
351static bool trans_fcvt_wu_h(DisasContext *ctx, arg_fcvt_wu_h *a)
352{
353    REQUIRE_FPU;
354    REQUIRE_ZFH(ctx);
355
356    TCGv dest = dest_gpr(ctx, a->rd);
357
358    gen_set_rm(ctx, a->rm);
359    gen_helper_fcvt_wu_h(dest, cpu_env, cpu_fpr[a->rs1]);
360    gen_set_gpr(ctx, a->rd, dest);
361    return true;
362}
363
364static bool trans_fcvt_h_w(DisasContext *ctx, arg_fcvt_h_w *a)
365{
366    REQUIRE_FPU;
367    REQUIRE_ZFH(ctx);
368
369    TCGv t0 = get_gpr(ctx, a->rs1, EXT_SIGN);
370
371    gen_set_rm(ctx, a->rm);
372    gen_helper_fcvt_h_w(cpu_fpr[a->rd], cpu_env, t0);
373
374    mark_fs_dirty(ctx);
375    return true;
376}
377
378static bool trans_fcvt_h_wu(DisasContext *ctx, arg_fcvt_h_wu *a)
379{
380    REQUIRE_FPU;
381    REQUIRE_ZFH(ctx);
382
383    TCGv t0 = get_gpr(ctx, a->rs1, EXT_SIGN);
384
385    gen_set_rm(ctx, a->rm);
386    gen_helper_fcvt_h_wu(cpu_fpr[a->rd], cpu_env, t0);
387
388    mark_fs_dirty(ctx);
389    return true;
390}
391
392static bool trans_fmv_x_h(DisasContext *ctx, arg_fmv_x_h *a)
393{
394    REQUIRE_FPU;
395    REQUIRE_ZFH(ctx);
396
397    TCGv dest = dest_gpr(ctx, a->rd);
398
399#if defined(TARGET_RISCV64)
400    /* 16 bits -> 64 bits */
401    tcg_gen_ext16s_tl(dest, cpu_fpr[a->rs1]);
402#else
403    /* 16 bits -> 32 bits */
404    tcg_gen_extrl_i64_i32(dest, cpu_fpr[a->rs1]);
405    tcg_gen_ext16s_tl(dest, dest);
406#endif
407
408    gen_set_gpr(ctx, a->rd, dest);
409    return true;
410}
411
412static bool trans_fmv_h_x(DisasContext *ctx, arg_fmv_h_x *a)
413{
414    REQUIRE_FPU;
415    REQUIRE_ZFH(ctx);
416
417    TCGv t0 = get_gpr(ctx, a->rs1, EXT_ZERO);
418
419    tcg_gen_extu_tl_i64(cpu_fpr[a->rd], t0);
420    gen_nanbox_h(cpu_fpr[a->rd], cpu_fpr[a->rd]);
421
422    mark_fs_dirty(ctx);
423    return true;
424}
425
426static bool trans_fcvt_l_h(DisasContext *ctx, arg_fcvt_l_h *a)
427{
428    REQUIRE_64BIT(ctx);
429    REQUIRE_FPU;
430    REQUIRE_ZFH(ctx);
431
432    TCGv dest = dest_gpr(ctx, a->rd);
433
434    gen_set_rm(ctx, a->rm);
435    gen_helper_fcvt_l_h(dest, cpu_env, cpu_fpr[a->rs1]);
436    gen_set_gpr(ctx, a->rd, dest);
437    return true;
438}
439
440static bool trans_fcvt_lu_h(DisasContext *ctx, arg_fcvt_lu_h *a)
441{
442    REQUIRE_64BIT(ctx);
443    REQUIRE_FPU;
444    REQUIRE_ZFH(ctx);
445
446    TCGv dest = dest_gpr(ctx, a->rd);
447
448    gen_set_rm(ctx, a->rm);
449    gen_helper_fcvt_lu_h(dest, cpu_env, cpu_fpr[a->rs1]);
450    gen_set_gpr(ctx, a->rd, dest);
451    return true;
452}
453
454static bool trans_fcvt_h_l(DisasContext *ctx, arg_fcvt_h_l *a)
455{
456    REQUIRE_64BIT(ctx);
457    REQUIRE_FPU;
458    REQUIRE_ZFH(ctx);
459
460    TCGv t0 = get_gpr(ctx, a->rs1, EXT_SIGN);
461
462    gen_set_rm(ctx, a->rm);
463    gen_helper_fcvt_h_l(cpu_fpr[a->rd], cpu_env, t0);
464
465    mark_fs_dirty(ctx);
466    return true;
467}
468
469static bool trans_fcvt_h_lu(DisasContext *ctx, arg_fcvt_h_lu *a)
470{
471    REQUIRE_64BIT(ctx);
472    REQUIRE_FPU;
473    REQUIRE_ZFH(ctx);
474
475    TCGv t0 = get_gpr(ctx, a->rs1, EXT_SIGN);
476
477    gen_set_rm(ctx, a->rm);
478    gen_helper_fcvt_h_lu(cpu_fpr[a->rd], cpu_env, t0);
479
480    mark_fs_dirty(ctx);
481    return true;
482}
483