| blob | b9ead55d07bee36918e07a9616351db90d7c76d9 |
1 /*
2 * Copyright (C) 2024 Mikulas Patocka
3 *
4 * This file is part of Ajla.
5 *
6 * Ajla is free software: you can redistribute it and/or modify it under the
7 * terms of the GNU General Public License as published by the Free Software
8 * Foundation, either version 3 of the License, or (at your option) any later
9 * version.
10 *
11 * Ajla is distributed in the hope that it will be useful, but WITHOUT ANY
12 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
13 * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * Ajla. If not, see <https://www.gnu.org/licenses/>.
17 */
19 #if !defined(ARCH_SPARC64)
20 #define SPARC_9 cpu_test_feature(CPU_FEATURE_sparc9)
21 #define FRAME_SIZE 0x60
22 #define OP_SIZE_NATIVE (SPARC_9 ? OP_SIZE_8 : OP_SIZE_4)
23 #define OP_SIZE_ADDRESS OP_SIZE_4
24 #else
25 #define SPARC_9 1
26 #define FRAME_SIZE 0xb0
27 #define OP_SIZE_NATIVE OP_SIZE_8
28 #define OP_SIZE_ADDRESS OP_SIZE_8
29 #endif
31 #define JMP_LIMIT JMP_LONG
33 #define UNALIGNED_TRAP 1
35 #define ALU_WRITES_FLAGS(size, alu, is_mem, is_imm, imm) 0
36 #define ALU1_WRITES_FLAGS(alu) 0
37 #define ROT_WRITES_FLAGS(alu, size, im) 0
38 #define COND_IS_LOGICAL(cond) 0
40 #define ARCH_PARTIAL_ALU(size) 0
41 #define ARCH_IS_3ADDRESS(alu, f) 1
42 #define ARCH_IS_3ADDRESS_IMM(alu, f) 1
43 #define ARCH_IS_3ADDRESS_ROT(alu, size) 1
44 #define ARCH_IS_3ADDRESS_ROT_IMM(alu) 1
45 #define ARCH_IS_2ADDRESS(alu) 1
46 #define ARCH_IS_3ADDRESS_FP 1
47 #define ARCH_HAS_JMP_2REGS(cond) 0
48 #define ARCH_HAS_FLAGS 1
49 #define ARCH_PREFERS_SX(size) 0
50 #define ARCH_HAS_BWX 1
51 #define ARCH_HAS_MUL SPARC_9
52 #define ARCH_HAS_DIV SPARC_9
53 #define ARCH_HAS_ANDN 1
54 #define ARCH_HAS_SHIFTED_ADD(bits) 0
55 #define ARCH_HAS_BTX(btx, size, cnst) 0
56 #define ARCH_SHIFT_SIZE OP_SIZE_4
57 #define ARCH_BOOL_SIZE OP_SIZE_NATIVE
58 #define ARCH_HAS_FP_GP_MOV 0
59 #define ARCH_NEEDS_BARRIER 0
61 #define i_size(size) OP_SIZE_NATIVE
62 #define i_size_rot(size) maximum(size, OP_SIZE_4)
63 #define i_size_cmp(size) maximum(size, OP_SIZE_4)
65 /*#define SUPPORT_QUAD_PRECISION*/
67 #define R_G0 0x00
68 #define R_G1 0x01
69 #define R_G2 0x02
70 #define R_G3 0x03
71 #define R_G4 0x04
72 #define R_G5 0x05
73 #define R_G6 0x06
74 #define R_G7 0x07
75 #define R_O0 0x08
76 #define R_O1 0x09
77 #define R_O2 0x0a
78 #define R_O3 0x0b
79 #define R_O4 0x0c
80 #define R_O5 0x0d
81 #define R_O6 0x0e
82 #define R_O7 0x0f
83 #define R_L0 0x10
84 #define R_L1 0x11
85 #define R_L2 0x12
86 #define R_L3 0x13
87 #define R_L4 0x14
88 #define R_L5 0x15
89 #define R_L6 0x16
90 #define R_L7 0x17
91 #define R_I0 0x18
92 #define R_I1 0x19
93 #define R_I2 0x1a
94 #define R_I3 0x1b
95 #define R_I4 0x1c
96 #define R_I5 0x1d
97 #define R_I6 0x1e
98 #define R_I7 0x1f
100 #define FSR_0 0x20
101 #define FSR_1 0x21
102 #define FSR_2 0x22
103 #define FSR_3 0x23
104 #define FSR_4 0x24
105 #define FSR_5 0x25
106 #define FSR_6 0x26
107 #define FSR_7 0x27
108 #define FSR_8 0x28
109 #define FSR_9 0x29
110 #define FSR_10 0x2a
111 #define FSR_11 0x2b
112 #define FSR_12 0x2c
113 #define FSR_13 0x2d
114 #define FSR_14 0x2e
115 #define FSR_15 0x2f
116 #define FSR_16 0x30
117 #define FSR_17 0x31
118 #define FSR_18 0x32
119 #define FSR_19 0x33
120 #define FSR_20 0x34
121 #define FSR_21 0x35
122 #define FSR_22 0x36
123 #define FSR_23 0x37
124 #define FSR_24 0x38
125 #define FSR_25 0x39
126 #define FSR_26 0x3a
127 #define FSR_27 0x3b
128 #define FSR_28 0x3c
129 #define FSR_29 0x3d
130 #define FSR_30 0x3e
131 #define FSR_31 0x3f
133 #define FDR_0 0x20
134 #define FDR_2 0x22
135 #define FDR_4 0x24
136 #define FDR_6 0x26
137 #define FDR_8 0x28
138 #define FDR_10 0x2a
139 #define FDR_12 0x2c
140 #define FDR_14 0x2e
141 #define FDR_16 0x30
142 #define FDR_18 0x32
143 #define FDR_20 0x34
144 #define FDR_22 0x36
145 #define FDR_24 0x38
146 #define FDR_26 0x3a
147 #define FDR_28 0x3c
148 #define FDR_30 0x3e
149 #define FDR_32 0x21
150 #define FDR_34 0x23
151 #define FDR_36 0x25
152 #define FDR_38 0x27
153 #define FDR_40 0x29
154 #define FDR_42 0x2b
155 #define FDR_44 0x2d
156 #define FDR_46 0x2f
157 #define FDR_48 0x31
158 #define FDR_50 0x33
159 #define FDR_52 0x35
160 #define FDR_54 0x37
161 #define FDR_56 0x39
162 #define FDR_58 0x3b
163 #define FDR_60 0x3d
164 #define FDR_62 0x3f
166 #define FQR_0 0x20
167 #define FQR_4 0x24
168 #define FQR_8 0x28
169 #define FQR_12 0x2c
170 #define FQR_16 0x30
171 #define FQR_20 0x34
172 #define FQR_24 0x38
173 #define FQR_28 0x3c
174 #define FQR_32 0x21
175 #define FQR_36 0x25
176 #define FQR_40 0x29
177 #define FQR_44 0x2d
178 #define FQR_48 0x31
179 #define FQR_52 0x35
180 #define FQR_56 0x39
181 #define FQR_60 0x3d
183 #define R_ZERO 0x00
184 #define R_SP R_O6
186 #define R_FRAME R_I0
187 #define R_UPCALL R_I1
188 #define R_TIMESTAMP R_I2
190 #define R_SCRATCH_1 R_O1
191 #define R_SCRATCH_2 R_O0
192 #define R_SCRATCH_3 R_O3
193 #define R_SCRATCH_4 R_O2
195 #define R_SCRATCH_NA_1 R_O4
196 #define R_SCRATCH_NA_2 R_O5
197 #ifdef HAVE_BITWISE_FRAME
198 #define R_SCRATCH_NA_3 R_O7
199 #endif
201 #define R_SAVED_1 R_L0
202 #define R_SAVED_2 R_L1
204 #define R_ARG0 R_O0
205 #define R_ARG1 R_O1
206 #define R_ARG2 R_O2
207 #define R_ARG3 R_O3
208 #define R_RET0 R_O0
210 #define R_OFFSET_IMM R_G1
211 #define R_CONST_IMM R_G2
212 #define R_CONST_HELPER R_G3
214 #ifdef SUPPORT_QUAD_PRECISION
215 #define FR_SCRATCH_1 FDR_0
216 #define FR_SCRATCH_2 FDR_4
217 #define SUPPORTED_FP 0x16
218 #else
219 #define FR_SCRATCH_1 FDR_0
220 #define FR_SCRATCH_2 FDR_2
221 #define SUPPORTED_FP 0x6
222 #endif
224 static bool reg_is_fp(unsigned reg)
225 {
226 return reg >= 0x20 && reg < 0x40;
227 }
229 static const uint8_t regs_saved[] = { R_L2, R_L3, R_L4, R_L5, R_L6, R_L7, R_I3, R_I4, R_I5 };
230 static const uint8_t regs_volatile[] = { R_G4, R_G5,
231 #ifndef HAVE_BITWISE_FRAME
232 R_O7,
233 #endif
234 };
235 static const uint8_t fp_saved[] = { 0 };
236 #define n_fp_saved 0U
237 #ifdef SUPPORT_QUAD_PRECISION
238 static const uint8_t fp_volatile[] = { FDR_8, FDR_12, FDR_16, FDR_20, FDR_24, FDR_28 };
239 #else
240 static const uint8_t fp_volatile[] = { FDR_4, FDR_6, FDR_8, FDR_10, FDR_12, FDR_14, FDR_16, FDR_18, FDR_20, FDR_22, FDR_24, FDR_26, FDR_28, FDR_30 };
241 #endif
242 #define reg_is_saved(r) ((r) >= R_L0 && (r) <= R_I7)
244 static bool attr_w gen_load_constant(struct codegen_context *ctx, unsigned reg, uint64_t c)
245 {
246 int32_t cl;
247 if (SPARC_9) {
248 if (c >= (uint64_t)-4096) {
249 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
250 gen_one(reg);
251 gen_one(R_ZERO);
252 gen_one(ARG_IMM);
253 gen_eight(c);
254 return true;
255 }
256 if (c >= 0x100000000ULL) {
257 int32_t cu = c >> 32;
258 if (cu < -4096 || cu >= 4096) {
259 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
260 gen_one(R_CONST_HELPER);
261 gen_one(ARG_IMM);
262 gen_eight(cu & 0xFFFFFC00UL);
263 cu &= 0x3FFU;
264 if (cu) {
265 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
266 gen_one(R_CONST_HELPER);
267 gen_one(R_CONST_HELPER);
268 gen_one(ARG_IMM);
269 gen_eight(cu);
270 }
271 } else {
272 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
273 gen_one(R_CONST_HELPER);
274 gen_one(R_ZERO);
275 gen_one(ARG_IMM);
276 gen_eight(cu);
277 }
278 if (!(c & 0xFFFFFFFFULL)) {
279 gen_insn(INSN_ROT, OP_SIZE_NATIVE, ROT_SHL, 0);
280 gen_one(reg);
281 gen_one(R_CONST_HELPER);
282 gen_one(ARG_IMM);
283 gen_eight(32);
284 return true;
285 }
286 gen_insn(INSN_ROT, OP_SIZE_NATIVE, ROT_SHL, 0);
287 gen_one(R_CONST_HELPER);
288 gen_one(R_CONST_HELPER);
289 gen_one(ARG_IMM);
290 gen_eight(32);
291 }
292 }
293 cl = c;
294 if (cl < 0 || cl >= 4096) {
295 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
296 gen_one(reg);
297 gen_one(ARG_IMM);
298 gen_eight(cl & 0xFFFFFC00UL);
299 cl &= 0x3FFU;
300 if (cl) {
301 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
302 gen_one(reg);
303 gen_one(reg);
304 gen_one(ARG_IMM);
305 gen_eight(cl);
306 }
307 } else {
308 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
309 gen_one(reg);
310 gen_one(R_ZERO);
311 gen_one(ARG_IMM);
312 gen_eight(cl);
313 }
314 if (SPARC_9) {
315 if (c >= 0x100000000ULL) {
316 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_OR, 0);
317 gen_one(reg);
318 gen_one(reg);
319 gen_one(R_CONST_HELPER);
320 }
321 }
322 return true;
323 }
325 static bool attr_w gen_address(struct codegen_context *ctx, unsigned base, int64_t imm, unsigned purpose, unsigned size)
326 {
327 ctx->base_reg = base;
328 ctx->offset_imm = imm;
329 ctx->offset_reg = false;
330 switch (purpose) {
331 case IMM_PURPOSE_LDR_OFFSET:
332 case IMM_PURPOSE_LDR_SX_OFFSET:
333 case IMM_PURPOSE_STR_OFFSET:
334 case IMM_PURPOSE_LDP_STP_OFFSET:
335 case IMM_PURPOSE_VLDR_VSTR_OFFSET:
336 case IMM_PURPOSE_MVI_CLI_OFFSET:
337 if (likely(imm >= -4096) && likely(imm < 4096))
338 return true;
339 break;
340 default:
341 internal(file_line, "gen_address: invalid purpose %u (imm %"PRIxMAX", size %u)", purpose, (uintmax_t)imm, size);
342 }
344 g(gen_load_constant(ctx, R_OFFSET_IMM, imm));
345 ctx->offset_reg = true;
346 return true;
347 }
349 static bool is_direct_const(int64_t imm, unsigned purpose, unsigned size)
350 {
351 switch (purpose) {
352 case IMM_PURPOSE_STORE_VALUE:
353 if (!imm)
354 return true;
355 break;
356 case IMM_PURPOSE_ADD:
357 case IMM_PURPOSE_SUB:
358 case IMM_PURPOSE_CMP:
359 case IMM_PURPOSE_CMP_LOGICAL:
360 case IMM_PURPOSE_AND:
361 case IMM_PURPOSE_OR:
362 case IMM_PURPOSE_XOR:
363 case IMM_PURPOSE_ANDN:
364 case IMM_PURPOSE_TEST:
365 case IMM_PURPOSE_MUL:
366 if (likely(imm >= -4096) && likely(imm < 4096))
367 return true;
368 break;
369 case IMM_PURPOSE_CMOV:
370 if (likely(imm >= -1024) && likely(imm < 1024))
371 return true;
372 break;
373 case IMM_PURPOSE_MOVR:
374 if (likely(imm >= -512) && likely(imm < 512))
375 return true;
376 break;
377 default:
378 internal(file_line, "is_direct_const: invalid purpose %u (imm %"PRIxMAX", size %u)", purpose, (uintmax_t)imm, size);
379 }
380 return false;
381 }
383 static bool attr_w gen_entry(struct codegen_context *ctx)
384 {
385 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_SAVE, 0);
386 gen_one(R_SP);
387 gen_one(R_SP);
388 gen_one(ARG_IMM);
389 gen_eight(-FRAME_SIZE);
391 gen_insn(INSN_JMP_INDIRECT, 0, 0, 0);
392 gen_one(R_I3);
394 return true;
395 }
397 static bool attr_w gen_escape_arg(struct codegen_context *ctx, ip_t ip, uint32_t escape_label)
398 {
399 g(gen_load_constant(ctx, R_I1, ip));
401 gen_insn(INSN_JMP, 0, 0, 0);
402 gen_four(escape_label);
404 return true;
405 }
407 static bool attr_w gen_escape(struct codegen_context *ctx)
408 {
409 gen_insn(INSN_RET, 0, 0, 0);
411 return true;
412 }
414 static bool attr_w gen_upcall_argument(struct codegen_context attr_unused *ctx, unsigned attr_unused arg)
415 {
416 return true;
417 }
419 static bool attr_w gen_get_upcall_pointer(struct codegen_context *ctx, unsigned offset, unsigned reg)
420 {
421 g(gen_address(ctx, R_UPCALL, offset, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_ADDRESS));
422 gen_insn(INSN_MOV, OP_SIZE_ADDRESS, 0, 0);
423 gen_one(reg);
424 gen_address_offset();
426 return true;
427 }
429 static bool attr_w gen_upcall(struct codegen_context *ctx, unsigned offset, unsigned n_args)
430 {
431 g(gen_get_upcall_pointer(ctx, offset, R_SCRATCH_NA_1));
433 gen_insn(INSN_CALL_INDIRECT, OP_SIZE_NATIVE, 0, 0);
434 gen_one(R_SCRATCH_NA_1);
436 g(gen_upcall_end(ctx, n_args));
438 return true;
439 }
441 static bool attr_w gen_timestamp_test(struct codegen_context *ctx, uint32_t escape_label)
442 {
443 g(gen_address(ctx, R_UPCALL, offsetof(struct cg_upcall_vector_s, ts), IMM_PURPOSE_LDR_OFFSET, OP_SIZE_4));
444 gen_insn(INSN_MOV, OP_SIZE_4, 0, 0);
445 gen_one(R_SCRATCH_1);
446 gen_address_offset();
448 gen_insn(INSN_CMP, OP_SIZE_4, 0, 1);
449 gen_one(R_SCRATCH_1);
450 gen_one(R_TIMESTAMP);
452 gen_insn(INSN_JMP_COND, OP_SIZE_4, COND_NE, 0);
453 gen_four(escape_label);
455 return true;
456 }