2 * Copyright (C) 2024 Mikulas Patocka
4 * This file is part of Ajla.
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
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.
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/>.
19 #define OP_SIZE_NATIVE OP_SIZE_8
20 #define OP_SIZE_ADDRESS OP_SIZE_NATIVE
22 #define JMP_LIMIT JMP_EXTRA_LONG
24 #define UNALIGNED_TRAP (!cpu_test_feature(CPU_FEATURE_unaligned))
26 #define ALU_WRITES_FLAGS(alu, im) 0
27 #define ALU1_WRITES_FLAGS(alu) 0
28 #define ROT_WRITES_FLAGS(alu, size, im) 0
29 #define COND_IS_LOGICAL(cond) 0
31 #define ARCH_PARTIAL_ALU(size) 0
32 #define ARCH_IS_3ADDRESS(alu, f) 1
33 #define ARCH_IS_3ADDRESS_IMM(alu, f) 1
34 #define ARCH_IS_3ADDRESS_ROT(alu, size) 1
35 #define ARCH_IS_3ADDRESS_ROT_IMM(alu) 1
36 #define ARCH_IS_2ADDRESS(alu) 1
37 #define ARCH_IS_3ADDRESS_FP 1
38 #define ARCH_HAS_FLAGS 0
39 #define ARCH_PREFERS_SX(size) 0
40 #define ARCH_HAS_BWX 1
41 #define ARCH_HAS_MUL 1
42 #define ARCH_HAS_DIV 1
43 #define ARCH_HAS_ANDN cpu_test_feature(CPU_FEATURE_zbb)
44 #define ARCH_HAS_SHIFTED_ADD(bits) ((bits) <= 3 && cpu_test_feature(CPU_FEATURE_zba))
45 #define ARCH_HAS_BTX(btx, size, cnst) (((size) == OP_SIZE_8 || (cnst)) && cpu_test_feature(CPU_FEATURE_zbs))
46 #define ARCH_SHIFT_SIZE OP_SIZE_4
47 #define ARCH_NEEDS_BARRIER 0
49 #define i_size(size) OP_SIZE_NATIVE
50 #define i_size_rot(size) maximum(size, OP_SIZE_4)
51 #define i_size_cmp(size) OP_SIZE_NATIVE
120 #define R_UPCALL R_S1
121 #define R_TIMESTAMP R_S2
123 #define R_SCRATCH_1 R_A0
124 #define R_SCRATCH_2 R_A1
125 #define R_SCRATCH_3 R_A2
126 #define R_SCRATCH_4 R_A3
127 #define R_SCRATCH_NA_1 R_A4
128 #define R_SCRATCH_NA_2 R_A5
129 #ifdef HAVE_BITWISE_FRAME
130 #define R_SCRATCH_NA_3 R_A6
133 #define R_SAVED_1 R_S3
134 #define R_SAVED_2 R_S4
143 #define R_OFFSET_IMM R_T0
144 #define R_CONST_IMM R_T1
145 #define R_CONST_HELPER R_T2
146 #define R_CMP_RESULT R_T3
148 #define FR_SCRATCH_1 R_FT0
149 #define FR_SCRATCH_2 R_FT1
151 #define SUPPORTED_FP 0x6
153 #define FRAME_SIZE 0x70
155 static bool reg_is_fp(unsigned reg)
157 return reg >= 0x20 && reg < 0x40;
160 static const uint8_t regs_saved[] = { R_S5, R_S6, R_S7, R_S8, R_S9, R_S10, R_S11 };
161 static const uint8_t regs_volatile[] = { R_RA,
162 #ifndef HAVE_BITWISE_FRAME
165 R_A7, R_T4, R_T5, R_T6 };
166 static const uint8_t fp_saved[] = { 0 };
167 #define n_fp_saved 0U
168 static const uint8_t fp_volatile[] = { R_FT2, R_FT3, R_FT4, R_FT5, R_FT6, R_FT7, R_FA0, R_FA1, R_FA2, R_FA3, R_FA4, R_FA5, R_FA6, R_FA7, R_FT8, R_FT9, R_FT10, R_FT11 };
169 #define reg_is_saved(r) (((r) >= R_S0 && (r) <= R_S1) || ((r) >= R_S2 && (r) <= R_S11) || ((r) >= R_FS0 && (r) <= R_FS1) || ((r) >= R_FS2 && (r) <= R_FS11))
171 static const struct {
174 } riscv_compress[] = {
175 #include "riscv-c.inc"
178 static bool attr_w gen_load_constant(struct codegen_context *ctx, unsigned reg, uint64_t c)
181 int32_t c1, c2, c3, c4;
189 c2 = (c >> 12) & 0xfffffUL;
195 c3 = (c >> 32) & 0xfffUL;
199 c += 0x100000000000UL;
206 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
209 gen_eight((uint64_t)c4 << 12);
213 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_ADD, 0);
221 gen_insn(INSN_ROT, OP_SIZE_NATIVE, ROT_SHL, 0);
229 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
230 gen_one(R_CONST_HELPER);
232 gen_eight((uint64_t)c2 << 12);
234 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_ADD, 0);
237 gen_one(R_CONST_HELPER);
239 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
242 gen_eight((uint64_t)c2 << 12);
246 if (c1 || r == R_ZERO) {
247 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_ADD, 0);
256 static bool attr_w gen_address(struct codegen_context *ctx, unsigned base, int64_t imm, unsigned purpose, unsigned size)
258 ctx->base_reg = base;
259 ctx->offset_imm = imm;
260 ctx->offset_reg = false;
262 case IMM_PURPOSE_LDR_OFFSET:
263 case IMM_PURPOSE_LDR_SX_OFFSET:
264 case IMM_PURPOSE_STR_OFFSET:
265 case IMM_PURPOSE_VLDR_VSTR_OFFSET:
266 case IMM_PURPOSE_MVI_CLI_OFFSET:
267 if (likely(imm >= -0x800) && likely(imm < 0x800))
271 internal(file_line, "gen_address: invalid purpose %u (imm %"PRIxMAX", size %u)", purpose, (uintmax_t)imm, size);
273 g(gen_load_constant(ctx, R_OFFSET_IMM, imm));
274 gen_insn(INSN_ALU, OP_SIZE_ADDRESS, ALU_ADD, 0);
275 gen_one(R_OFFSET_IMM);
276 gen_one(R_OFFSET_IMM);
278 ctx->base_reg = R_OFFSET_IMM;
283 static bool is_direct_const(int64_t imm, unsigned purpose, unsigned size)
286 case IMM_PURPOSE_STORE_VALUE:
290 case IMM_PURPOSE_ADD:
291 case IMM_PURPOSE_AND:
293 case IMM_PURPOSE_XOR:
294 case IMM_PURPOSE_TEST:
295 case IMM_PURPOSE_CMP:
296 case IMM_PURPOSE_CMP_LOGICAL:
297 if (likely(imm >= -0x800) && likely(imm < 0x800))
300 case IMM_PURPOSE_SUB:
301 if (likely(imm > -0x800) && likely(imm <= 0x800))
304 case IMM_PURPOSE_ANDN:
306 case IMM_PURPOSE_JMP_2REGS:
308 case IMM_PURPOSE_MUL:
310 case IMM_PURPOSE_BITWISE:
313 internal(file_line, "is_direct_const: invalid purpose %u (imm %"PRIxMAX", size %u)", purpose, (uintmax_t)imm, size);
318 static bool attr_w gen_imm(struct codegen_context *ctx, int64_t imm, unsigned purpose, unsigned size)
320 if (is_direct_const(imm, purpose, size)) {
321 ctx->const_imm = imm;
322 ctx->const_reg = false;
324 g(gen_load_constant(ctx, R_CONST_IMM, imm));
325 ctx->const_reg = true;
330 static bool attr_w gen_entry(struct codegen_context *ctx)
332 g(gen_imm(ctx, -FRAME_SIZE, IMM_PURPOSE_ADD, OP_SIZE_NATIVE));
333 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_ADD, 0);
338 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x08, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
339 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
340 gen_address_offset();
343 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x10, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
344 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
345 gen_address_offset();
348 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x18, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
349 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
350 gen_address_offset();
353 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x20, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
354 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
355 gen_address_offset();
358 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x28, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
359 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
360 gen_address_offset();
363 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x30, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
364 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
365 gen_address_offset();
368 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x38, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
369 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
370 gen_address_offset();
373 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x40, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
374 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
375 gen_address_offset();
378 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x48, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
379 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
380 gen_address_offset();
383 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x50, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
384 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
385 gen_address_offset();
388 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x58, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
389 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
390 gen_address_offset();
393 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x60, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
394 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
395 gen_address_offset();
398 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x68, IMM_PURPOSE_STR_OFFSET, OP_SIZE_NATIVE));
399 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
400 gen_address_offset();
403 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
407 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
411 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
412 gen_one(R_TIMESTAMP);
415 gen_insn(INSN_JMP_INDIRECT, 0, 0, 0);
421 static bool attr_w gen_escape_arg(struct codegen_context *ctx, ip_t ip, uint32_t escape_label)
423 g(gen_load_constant(ctx, R_RET1, (int32_t)ip));
425 gen_insn(INSN_JMP, 0, 0, 0);
426 gen_four(escape_label);
431 static bool attr_w gen_escape(struct codegen_context *ctx)
433 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
437 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x08, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
438 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
440 gen_address_offset();
442 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x10, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
443 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
445 gen_address_offset();
447 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x18, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
448 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
450 gen_address_offset();
452 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x20, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
453 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
455 gen_address_offset();
457 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x28, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
458 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
460 gen_address_offset();
462 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x30, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
463 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
465 gen_address_offset();
467 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x38, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
468 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
470 gen_address_offset();
472 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x40, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
473 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
475 gen_address_offset();
477 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x48, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
478 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
480 gen_address_offset();
482 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x50, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
483 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
485 gen_address_offset();
487 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x58, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
488 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
490 gen_address_offset();
492 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x60, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
493 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
495 gen_address_offset();
497 g(gen_address(ctx, R_SP, FRAME_SIZE - 0x68, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_NATIVE));
498 gen_insn(INSN_MOV, OP_SIZE_NATIVE, 0, 0);
500 gen_address_offset();
502 g(gen_imm(ctx, FRAME_SIZE, IMM_PURPOSE_ADD, OP_SIZE_NATIVE));
503 gen_insn(INSN_ALU, OP_SIZE_NATIVE, ALU_ADD, 0);
508 gen_insn(INSN_RET, 0, 0, 0);
513 static bool attr_w gen_upcall_argument(struct codegen_context attr_unused *ctx, unsigned attr_unused arg)
518 static bool attr_w gen_upcall(struct codegen_context *ctx, unsigned offset, unsigned n_args)
520 g(gen_address(ctx, R_UPCALL, offset, IMM_PURPOSE_LDR_OFFSET, OP_SIZE_ADDRESS));
521 gen_insn(INSN_MOV, OP_SIZE_ADDRESS, 0, 0);
522 gen_one(R_SCRATCH_NA_1);
523 gen_address_offset();
525 gen_insn(INSN_CALL_INDIRECT, OP_SIZE_ADDRESS, 0, 0);
526 gen_one(R_SCRATCH_NA_1);
528 g(gen_upcall_end(ctx, n_args));
533 static bool attr_w gen_cmp_test_jmp(struct codegen_context *ctx, unsigned insn, unsigned op_size, unsigned reg1, unsigned reg2, unsigned cond, uint32_t label);
535 static bool attr_w gen_timestamp_test(struct codegen_context *ctx, uint32_t escape_label)
537 g(gen_address(ctx, R_UPCALL, offsetof(struct cg_upcall_vector_s, ts), IMM_PURPOSE_LDR_SX_OFFSET, OP_SIZE_4));
538 gen_insn(INSN_MOVSX, OP_SIZE_4, 0, 0);
539 gen_one(R_SCRATCH_1);
540 gen_address_offset();
542 g(gen_cmp_test_jmp(ctx, INSN_CMP, OP_SIZE_NATIVE, R_SCRATCH_1, R_TIMESTAMP, COND_NE, escape_label));