Skip to content

Commit 1ac6faa

Browse files
committed
Merge pull request adafruit#833 from Vogtinator/arm-native
Basic native ARM emitter
2 parents 516b09e + 16ee30c commit 1ac6faa

File tree

8 files changed

+577
-4
lines changed

8 files changed

+577
-4
lines changed

py/asmarm.c

Lines changed: 327 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,327 @@
1+
/*
2+
* This file is part of the Micro Python project, http://micropython.org/
3+
*
4+
* The MIT License (MIT)
5+
*
6+
* Copyright (c) 2014 Fabian Vogt
7+
* Copyright (c) 2013, 2014 Damien P. George
8+
*
9+
* Permission is hereby granted, free of charge, to any person obtaining a copy
10+
* of this software and associated documentation files (the "Software"), to deal
11+
* in the Software without restriction, including without limitation the rights
12+
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
13+
* copies of the Software, and to permit persons to whom the Software is
14+
* furnished to do so, subject to the following conditions:
15+
*
16+
* The above copyright notice and this permission notice shall be included in
17+
* all copies or substantial portions of the Software.
18+
*
19+
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20+
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21+
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
22+
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23+
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24+
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25+
* THE SOFTWARE.
26+
*/
27+
28+
#include <stdio.h>
29+
#include <assert.h>
30+
#include <string.h>
31+
32+
#include "mpconfig.h"
33+
#include "misc.h"
34+
#include "asmarm.h"
35+
36+
// wrapper around everything in this file
37+
#if MICROPY_EMIT_ARM
38+
39+
#define SIGNED_FIT24(x) (((x) & 0xff800000) == 0) || (((x) & 0xff000000) == 0xff000000)
40+
41+
struct _asm_arm_t {
42+
uint pass;
43+
uint code_offset;
44+
uint code_size;
45+
byte *code_base;
46+
byte dummy_data[4];
47+
48+
uint max_num_labels;
49+
int *label_offsets;
50+
int num_locals;
51+
uint push_reglist;
52+
uint stack_adjust;
53+
};
54+
55+
asm_arm_t *asm_arm_new(uint max_num_labels) {
56+
asm_arm_t *as;
57+
58+
as = m_new0(asm_arm_t, 1);
59+
as->max_num_labels = max_num_labels;
60+
as->label_offsets = m_new(int, max_num_labels);
61+
62+
return as;
63+
}
64+
65+
void asm_arm_free(asm_arm_t *as, bool free_code) {
66+
if (free_code) {
67+
m_del(byte, as->code_base, as->code_size);
68+
}
69+
70+
m_del_obj(asm_arm_t, as);
71+
}
72+
73+
void asm_arm_start_pass(asm_arm_t *as, uint pass) {
74+
as->pass = pass;
75+
as->code_offset = 0;
76+
if (pass == ASM_ARM_PASS_COMPUTE) {
77+
memset(as->label_offsets, -1, as->max_num_labels * sizeof(int));
78+
}
79+
}
80+
81+
void asm_arm_end_pass(asm_arm_t *as) {
82+
if (as->pass == ASM_ARM_PASS_COMPUTE) {
83+
// calculate size of code in bytes
84+
as->code_size = as->code_offset;
85+
as->code_base = m_new(byte, as->code_size);
86+
}
87+
}
88+
89+
// all functions must go through this one to emit bytes
90+
// if as->pass < ASM_ARM_PASS_EMIT, then this function only returns a buffer of 4 bytes length
91+
STATIC byte *asm_arm_get_cur_to_write_bytes(asm_arm_t *as, int num_bytes_to_write) {
92+
if (as->pass < ASM_ARM_PASS_EMIT) {
93+
as->code_offset += num_bytes_to_write;
94+
return as->dummy_data;
95+
} else {
96+
assert(as->code_offset + num_bytes_to_write <= as->code_size);
97+
byte *c = as->code_base + as->code_offset;
98+
as->code_offset += num_bytes_to_write;
99+
return c;
100+
}
101+
}
102+
103+
uint asm_arm_get_code_size(asm_arm_t *as) {
104+
return as->code_size;
105+
}
106+
107+
void *asm_arm_get_code(asm_arm_t *as) {
108+
return as->code_base;
109+
}
110+
111+
// Insert word into instruction flow
112+
STATIC void emit(asm_arm_t *as, uint op) {
113+
*(uint*)asm_arm_get_cur_to_write_bytes(as, 4) = op;
114+
}
115+
116+
// Insert word into instruction flow, add "ALWAYS" condition code
117+
STATIC void emit_al(asm_arm_t *as, uint op) {
118+
emit(as, op | ARM_CC_AL);
119+
}
120+
121+
// Basic instructions without condition code
122+
STATIC uint asm_arm_op_push(uint reglist) {
123+
// stmfd sp!, {reglist}
124+
return 0x92d0000 | (reglist & 0xFFFF);
125+
}
126+
127+
STATIC uint asm_arm_op_pop(uint reglist) {
128+
// ldmfd sp!, {reglist}
129+
return 0x8bd0000 | (reglist & 0xFFFF);
130+
}
131+
132+
STATIC uint asm_arm_op_mov_reg(uint rd, uint rn) {
133+
// mov rd, rn
134+
return 0x1a00000 | (rd << 12) | rn;
135+
}
136+
137+
STATIC uint asm_arm_op_mov_imm(uint rd, uint imm) {
138+
// mov rd, #imm
139+
return 0x3a00000 | (rd << 12) | imm;
140+
}
141+
142+
STATIC uint asm_arm_op_mvn_imm(uint rd, uint imm) {
143+
// mvn rd, #imm
144+
return 0x3e00000 | (rd << 12) | imm;
145+
}
146+
147+
STATIC uint asm_arm_op_add_imm(uint rd, uint rn, uint imm) {
148+
// add rd, rn, #imm
149+
return 0x2800000 | (rn << 16) | (rd << 12) | (imm & 0xFF);
150+
}
151+
152+
STATIC uint asm_arm_op_add_reg(uint rd, uint rn, uint rm) {
153+
// add rd, rn, rm
154+
return 0x0800000 | (rn << 16) | (rd << 12) | rm;
155+
}
156+
157+
STATIC uint asm_arm_op_sub_imm(uint rd, uint rn, uint imm) {
158+
// sub rd, rn, #imm
159+
return 0x2400000 | (rn << 16) | (rd << 12) | (imm & 0xFF);
160+
}
161+
162+
void asm_arm_bkpt(asm_arm_t *as) {
163+
// bkpt #0
164+
emit_al(as, 0x1200070);
165+
}
166+
167+
// locals:
168+
// - stored on the stack in ascending order
169+
// - numbered 0 through as->num_locals-1
170+
// - SP points to first local
171+
//
172+
// | SP
173+
// v
174+
// l0 l1 l2 ... l(n-1)
175+
// ^ ^
176+
// | low address | high address in RAM
177+
178+
void asm_arm_entry(asm_arm_t *as, int num_locals) {
179+
180+
if (num_locals < 0) {
181+
num_locals = 0;
182+
}
183+
184+
as->stack_adjust = 0;
185+
as->num_locals = num_locals;
186+
as->push_reglist = 1 << REG_R1 | 1 << REG_R2 | 1 << REG_R3 | 1 << REG_R4
187+
| 1 << REG_R5 | 1 << REG_R6 | 1 << REG_R7 | 1 << REG_R8;
188+
189+
// Only adjust the stack if there are more locals than usable registers
190+
if(num_locals > 3) {
191+
as->stack_adjust = num_locals * 4;
192+
// Align stack to 8 bytes
193+
if(as->num_locals & 1)
194+
as->stack_adjust += 4;
195+
}
196+
197+
emit_al(as, asm_arm_op_push(as->push_reglist | 1 << REG_LR));
198+
if (as->stack_adjust > 0) {
199+
emit_al(as, asm_arm_op_sub_imm(REG_SP, REG_SP, as->stack_adjust));
200+
}
201+
}
202+
203+
void asm_arm_exit(asm_arm_t *as) {
204+
if (as->stack_adjust > 0) {
205+
emit_al(as, asm_arm_op_add_imm(REG_SP, REG_SP, as->stack_adjust));
206+
}
207+
208+
emit_al(as, asm_arm_op_pop(as->push_reglist | (1 << REG_PC)));
209+
}
210+
211+
void asm_arm_label_assign(asm_arm_t *as, uint label) {
212+
assert(label < as->max_num_labels);
213+
if (as->pass < ASM_ARM_PASS_EMIT) {
214+
// assign label offset
215+
assert(as->label_offsets[label] == -1);
216+
as->label_offsets[label] = as->code_offset;
217+
} else {
218+
// ensure label offset has not changed from PASS_COMPUTE to PASS_EMIT
219+
assert(as->label_offsets[label] == as->code_offset);
220+
}
221+
}
222+
223+
void asm_arm_align(asm_arm_t* as, uint align) {
224+
// TODO fill unused data with NOPs?
225+
as->code_offset = (as->code_offset + align - 1) & (~(align - 1));
226+
}
227+
228+
void asm_arm_data(asm_arm_t* as, uint bytesize, uint val) {
229+
byte *c = asm_arm_get_cur_to_write_bytes(as, bytesize);
230+
// only write to the buffer in the emit pass (otherwise we overflow dummy_data)
231+
if (as->pass == ASM_ARM_PASS_EMIT) {
232+
// little endian
233+
for (uint i = 0; i < bytesize; i++) {
234+
*c++ = val;
235+
val >>= 8;
236+
}
237+
}
238+
}
239+
240+
void asm_arm_mov_reg_reg(asm_arm_t *as, uint reg_dest, uint reg_src) {
241+
emit_al(as, asm_arm_op_mov_reg(reg_dest, reg_src));
242+
}
243+
244+
void asm_arm_mov_reg_i32(asm_arm_t *as, uint rd, int imm) {
245+
// TODO: There are more variants of immediate values
246+
if ((imm & 0xFF) == imm) {
247+
emit_al(as, asm_arm_op_mov_imm(rd, imm));
248+
} else if (imm < 0 && ((-imm) & 0xFF) == -imm) {
249+
emit_al(as, asm_arm_op_mvn_imm(rd, -imm));
250+
} else {
251+
//Insert immediate into code and jump over it
252+
emit_al(as, 0x59f0000 | (rd << 12)); // ldr rd, [pc]
253+
emit_al(as, 0xa000000); // b pc
254+
emit(as, imm);
255+
}
256+
}
257+
258+
void asm_arm_mov_local_reg(asm_arm_t *as, int local_num, uint rd) {
259+
// str rd, [sp, #local_num*4]
260+
emit_al(as, 0x58d0000 | (rd << 12) | (local_num << 2));
261+
}
262+
263+
void asm_arm_mov_reg_local(asm_arm_t *as, uint rd, int local_num) {
264+
// ldr rd, [sp, #local_num*4]
265+
emit_al(as, 0x59d0000 | (rd << 12) | (local_num << 2));
266+
}
267+
268+
void asm_arm_cmp_reg_i8(asm_arm_t *as, uint rd, int imm) {
269+
// cmp rd, #imm
270+
emit_al(as, 0x3500000 | (rd << 16) | (imm & 0xFF));
271+
}
272+
273+
void asm_arm_cmp_reg_reg(asm_arm_t *as, uint rd, uint rn) {
274+
// cmp rd, rn
275+
emit_al(as, 0x1500000 | (rd << 16) | rn);
276+
}
277+
278+
void asm_arm_less_op(asm_arm_t *as, uint rd, uint rn) {
279+
asm_arm_cmp_reg_reg(as, rd, rn); // cmp rd, rn
280+
emit(as, asm_arm_op_mov_imm(REG_RET, 1) | ARM_CC_LT); // movlt REG_RET, #1
281+
emit(as, asm_arm_op_mov_imm(REG_RET, 0) | ARM_CC_GE); // movge REG_RET, #0
282+
}
283+
284+
void asm_arm_add_reg(asm_arm_t *as, uint rd, uint rn, uint rm) {
285+
// add rd, rn, rm
286+
emit_al(as, asm_arm_op_add_reg(rd, rn, rm));
287+
}
288+
289+
void asm_arm_mov_reg_local_addr(asm_arm_t *as, uint rd, int local_num) {
290+
// add rd, sp, #local_num*4
291+
emit_al(as, asm_arm_op_add_imm(rd, REG_SP, local_num << 2));
292+
}
293+
294+
void asm_arm_bcc_label(asm_arm_t *as, int cond, uint label) {
295+
assert(label < as->max_num_labels);
296+
int dest = as->label_offsets[label];
297+
int rel = dest - as->code_offset;
298+
rel -= 8; // account for instruction prefetch, PC is 8 bytes ahead of this instruction
299+
rel >>= 2; // in ARM mode the branch target is 32-bit aligned, so the 2 LSB are omitted
300+
301+
if (SIGNED_FIT24(rel)) {
302+
emit(as, cond | 0xa000000 | (rel & 0xffffff));
303+
} else {
304+
printf("asm_arm_bcc: branch does not fit in 24 bits\n");
305+
}
306+
}
307+
308+
void asm_arm_b_label(asm_arm_t *as, uint label) {
309+
asm_arm_bcc_label(as, ARM_CC_AL, label);
310+
}
311+
312+
void asm_arm_bl_ind(asm_arm_t *as, void *fun_ptr, uint fun_id, uint reg_temp) {
313+
// If the table offset fits into the ldr instruction
314+
if(fun_id < (0x1000 / 4)) {
315+
emit_al(as, asm_arm_op_mov_reg(REG_LR, REG_PC)); // mov lr, pc
316+
emit_al(as, 0x597f000 | (fun_id << 2)); // ldr pc, [r7, #fun_id*4]
317+
return;
318+
}
319+
320+
emit_al(as, 0x59f0004 | (reg_temp << 12)); // ldr rd, [pc, #4]
321+
// Set lr after fun_ptr
322+
emit_al(as, asm_arm_op_add_imm(REG_LR, REG_PC, 4)); // add lr, pc, #4
323+
emit_al(as, asm_arm_op_mov_reg(REG_PC, reg_temp)); // mov pc, reg_temp
324+
emit(as, (uint) fun_ptr);
325+
}
326+
327+
#endif // MICROPY_EMIT_ARM

0 commit comments

Comments
 (0)
pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy