/home/travis/build/MoarVM/MoarVM/src/mast/compiler.c
Line | Count | Source (jump to first uncovered line) |
1 | | #include "moar.h" |
2 | | #include "nodes.h" |
3 | | |
4 | | /* Some constants. */ |
5 | | #define HEADER_SIZE 92 |
6 | 1.21k | #define BYTECODE_VERSION 5 |
7 | 9.05k | #define FRAME_HEADER_SIZE (11 * 4 + 3 * 2) |
8 | 431 | #define FRAME_HANDLER_SIZE (4 * 4 + 2 * 2) |
9 | 5.61k | #define FRAME_SLV_SIZE (2 * 2 + 2 * 4) |
10 | 3.41k | #define SC_DEP_SIZE 4 |
11 | 1.21k | #define EXTOP_SIZE (4 + 8) |
12 | 2.43k | #define SCDEP_HEADER_OFFSET 12 |
13 | 2.43k | #define EXTOP_HEADER_OFFSET 20 |
14 | 2.43k | #define FRAME_HEADER_OFFSET 28 |
15 | 2.43k | #define CALLSITE_HEADER_OFFSET 36 |
16 | 2.43k | #define STRING_HEADER_OFFSET 44 |
17 | 0 | #define SCDATA_HEADER_OFFSET 52 |
18 | 2.43k | #define BYTECODE_HEADER_OFFSET 60 |
19 | 2.43k | #define ANNOTATION_HEADER_OFFSET 68 |
20 | 1.21k | #define HLL_NAME_HEADER_OFFSET 76 |
21 | 3.64k | #define SPECIAL_FRAME_HEADER_OFFSET 80 |
22 | 0 | #define EXTOP_BASE 1024 |
23 | | |
24 | | /* Frame flags. */ |
25 | | #define FRAME_FLAG_EXIT_HANDLER 1 |
26 | | #define FRAME_FLAG_IS_THUNK 2 |
27 | 4.52k | #define FRAME_FLAG_HAS_CODE_OBJ 4 |
28 | | #define FRAME_FLAG_NO_INLINE 8 |
29 | 8.68k | #define FRAME_FLAG_HAS_INDEX 32768 |
30 | 4.52k | #define FRAME_FLAG_HAS_SLV 65536 |
31 | | |
32 | | typedef struct { |
33 | | /* callsite ID */ |
34 | | unsigned short callsite_id; |
35 | | |
36 | | /* the identifier for the callsite, to clean up later */ |
37 | | unsigned char *identifier; |
38 | | |
39 | | /* the uthash hash handle. */ |
40 | | UT_hash_handle hash_handle; |
41 | | } CallsiteReuseEntry; |
42 | | |
43 | | /* Information about a handler. */ |
44 | | typedef struct { |
45 | | /* Offset of start of protected region from frame start. */ |
46 | | unsigned int start_offset; |
47 | | |
48 | | /* Offset of end of protected region, exclusive, from frame start. */ |
49 | | unsigned int end_offset; |
50 | | |
51 | | /* Exception categry mask. */ |
52 | | unsigned int category_mask; |
53 | | |
54 | | /* Handler action. */ |
55 | | unsigned short action; |
56 | | |
57 | | /* Local holding block to invoke, if invokey handler. */ |
58 | | unsigned short local; |
59 | | |
60 | | /* Label, which will need resolving. */ |
61 | | MASTNode *label; |
62 | | |
63 | | /* Local holding a label in case we have a labeled loop. */ |
64 | | unsigned short label_reg; |
65 | | } FrameHandler; |
66 | | |
67 | | /* Handler actions. */ |
68 | 317 | #define HANDLER_UNWIND_GOTO 0 |
69 | 29 | #define HANDLER_UNWIND_GOTO_OBJ 1 |
70 | 431 | #define HANDLER_INVOKE 2 |
71 | | |
72 | | /* Information about a label. */ |
73 | | typedef struct { |
74 | | MAST_Label *label; |
75 | | MVMint32 offset; /* Negative if unknown. */ |
76 | | MVMuint16 num_resolve; |
77 | | MVMuint16 alloc_resolve; |
78 | | MVMuint32 *resolve; |
79 | | } LabelInfo; |
80 | | |
81 | | /* Describes the state for the frame we're currently compiling. */ |
82 | | typedef struct { |
83 | | /* Position of start of bytecode. */ |
84 | | unsigned int bytecode_start; |
85 | | |
86 | | /* Position of start of frame entry. */ |
87 | | unsigned int frame_start; |
88 | | |
89 | | /* Types of locals and lexicals, with counts. */ |
90 | | unsigned short *local_types; |
91 | | unsigned short *lexical_types; |
92 | | unsigned int num_locals; |
93 | | unsigned int num_lexicals; |
94 | | |
95 | | /* Number of annotations. */ |
96 | | unsigned int num_annotations; |
97 | | |
98 | | /* Handlers count and list. */ |
99 | | unsigned int num_handlers; |
100 | | FrameHandler *handlers; |
101 | | |
102 | | /* Labels we have so far (either through finding them or finding a need |
103 | | * to fix them up). */ |
104 | | LabelInfo *labels; |
105 | | unsigned int num_labels; |
106 | | unsigned int alloc_labels; |
107 | | |
108 | | /* Number of unresolved labels. */ |
109 | | unsigned int unresolved_labels; |
110 | | } FrameState; |
111 | | |
112 | | /* Describes the current writer state for the compilation unit as a whole. */ |
113 | | typedef struct { |
114 | | /* The set of node types. */ |
115 | | MASTNodeTypes *types; |
116 | | |
117 | | /* The current frame and frame count. */ |
118 | | FrameState *cur_frame; |
119 | | unsigned int num_frames; |
120 | | |
121 | | /* String heap and seen hash mapping known strings to indexes. */ |
122 | | MASTNode *strings; |
123 | | MASTNode *seen_strings; |
124 | | |
125 | | /* The SC dependencies segment; we know the size up front. */ |
126 | | char *scdep_seg; |
127 | | unsigned int scdep_bytes; |
128 | | |
129 | | /* The extension ops segment; we know the size ahead of time. */ |
130 | | char *extops_seg; |
131 | | unsigned int extops_bytes; |
132 | | unsigned int num_extops; |
133 | | |
134 | | /* The frame segment. */ |
135 | | char *frame_seg; |
136 | | unsigned int frame_pos; |
137 | | unsigned int frame_alloc; |
138 | | |
139 | | /* The callsite segment and number of callsites. */ |
140 | | char *callsite_seg; |
141 | | unsigned int callsite_pos; |
142 | | unsigned int callsite_alloc; |
143 | | unsigned int num_callsites; |
144 | | |
145 | | /* The bytecode segment. */ |
146 | | char *bytecode_seg; |
147 | | unsigned int bytecode_pos; |
148 | | unsigned int bytecode_alloc; |
149 | | |
150 | | /* The annotation segment. */ |
151 | | char *annotation_seg; |
152 | | unsigned int annotation_pos; |
153 | | unsigned int annotation_alloc; |
154 | | |
155 | | /* Current instruction info */ |
156 | | const MVMOpInfo *current_op_info; |
157 | | |
158 | | /* Zero-based index of current frame */ |
159 | | unsigned int current_frame_idx; |
160 | | |
161 | | /* Zero-based index of MAST instructions */ |
162 | | unsigned int current_ins_idx; |
163 | | |
164 | | /* Zero-based index of current operand */ |
165 | | unsigned int current_operand_idx; |
166 | | |
167 | | /* The compilation unit we're compiling. */ |
168 | | MAST_CompUnit *cu; |
169 | | |
170 | | /* Hash for callsite descriptor strings to callsite IDs */ |
171 | | CallsiteReuseEntry *callsite_reuse_head; |
172 | | |
173 | | /* Last Annotated node, for error reporting */ |
174 | | MAST_Annotated *last_annotated; |
175 | | } WriterState; |
176 | | |
177 | | static unsigned int umax(unsigned int a, unsigned int b); |
178 | | static void memcpy_endian(char *dest, const void *src, size_t size); |
179 | | static void write_int64(char *buffer, size_t offset, unsigned long long value); |
180 | | static void write_int32(char *buffer, size_t offset, unsigned int value); |
181 | | static void write_int16(char *buffer, size_t offset, unsigned short value); |
182 | | static void write_int8(char *buffer, size_t offset, unsigned char value); |
183 | | static void write_double(char *buffer, size_t offset, double value); |
184 | | static void ensure_space(VM, char **buffer, unsigned int *alloc, unsigned int pos, unsigned int need); |
185 | | static void cleanup_frame(VM, FrameState *fs); |
186 | | static void cleanup_all(VM, WriterState *ws); |
187 | | static unsigned int get_string_heap_index(VM, WriterState *ws, VMSTR *strval); |
188 | | static unsigned short get_frame_index(VM, WriterState *ws, MASTNode *frame); |
189 | | static unsigned short type_to_local_type(VM, WriterState *ws, MASTNode *type); |
190 | | static void compile_operand(VM, WriterState *ws, unsigned char op_flags, MASTNode *operand); |
191 | | static unsigned short get_callsite_id(VM, WriterState *ws, MASTNode *flags, MASTNode *args); |
192 | | static void compile_instruction(VM, WriterState *ws, MASTNode *node); |
193 | | static void compile_frame(VM, WriterState *ws, MASTNode *node, unsigned short idx); |
194 | | static char * form_string_heap(VM, WriterState *ws, unsigned int *string_heap_size); |
195 | | static char * form_bytecode_output(VM, WriterState *ws, unsigned int *bytecode_size); |
196 | | char * MVM_mast_compile(VM, MASTNode *node, MASTNodeTypes *types, unsigned int *size); |
197 | | |
198 | 0 | static unsigned int umax(unsigned int a, unsigned int b) { |
199 | 0 | return a > b ? a : b; |
200 | 0 | } |
201 | | |
202 | | /* copies memory dependent on endianness */ |
203 | 1.17M | static void memcpy_endian(char *dest, const void *src, size_t size) { |
204 | 1.17M | #ifdef MVM_BIGENDIAN |
205 | | size_t i; |
206 | | char *srcbytes = (char *)src; |
207 | | for (i = 0; i < size; i++) |
208 | | dest[size - i - 1] = srcbytes[i]; |
209 | | #else |
210 | 1.17M | memcpy(dest, src, size); |
211 | 1.17M | #endif |
212 | 1.17M | } |
213 | | |
214 | | /* Writes an int64 into a buffer. */ |
215 | 1.29k | static void write_int64(char *buffer, size_t offset, unsigned long long value) { |
216 | 1.29k | memcpy_endian(buffer + offset, &value, 8); |
217 | 1.29k | } |
218 | | |
219 | | /* Writes an int32 into a buffer. */ |
220 | 263k | static void write_int32(char *buffer, size_t offset, unsigned int value) { |
221 | 263k | memcpy_endian(buffer + offset, &value, 4); |
222 | 263k | } |
223 | | |
224 | | /* Writes an int16 into a buffer. */ |
225 | 909k | static void write_int16(char *buffer, size_t offset, unsigned short value) { |
226 | 909k | memcpy_endian(buffer + offset, &value, 2); |
227 | 909k | } |
228 | | |
229 | | /* Writes an int8 into a buffer. */ |
230 | 9.93k | static void write_int8(char *buffer, size_t offset, unsigned char value) { |
231 | 9.93k | memcpy(buffer + offset, &value, 1); |
232 | 9.93k | } |
233 | | |
234 | | /* Writes an double into a buffer. */ |
235 | 498 | static void write_double(char *buffer, size_t offset, double value) { |
236 | 498 | memcpy_endian(buffer + offset, &value, 8); |
237 | 498 | } |
238 | | |
239 | | /* Ensures the specified buffer has enough space and expands it if so. */ |
240 | 879k | static void ensure_space(VM, char **buffer, unsigned int *alloc, unsigned int pos, unsigned int need) { |
241 | 879k | if (pos + need > *alloc) { |
242 | 2.44k | do { *alloc = *alloc * 2; } while (pos + need > *alloc); |
243 | 2.44k | *buffer = (char *)MVM_realloc(*buffer, *alloc); |
244 | 2.44k | } |
245 | 879k | } |
246 | | |
247 | | /* Cleans up all allocated memory related to a frame. */ |
248 | 4.52k | static void cleanup_frame(VM, FrameState *fs) { |
249 | 4.52k | if (fs->local_types) |
250 | 4.52k | MVM_free(fs->local_types); |
251 | 4.52k | if (fs->lexical_types) |
252 | 4.52k | MVM_free(fs->lexical_types); |
253 | 4.52k | if (fs->handlers) |
254 | 124 | MVM_free(fs->handlers); |
255 | 4.52k | if (fs->labels) { |
256 | 1.79k | MVMuint32 i; |
257 | 21.4k | for (i = 0; i < fs->num_labels; i++) |
258 | 19.6k | if (fs->labels[i].alloc_resolve) |
259 | 0 | MVM_free(fs->labels[i].resolve); |
260 | 1.79k | MVM_free(fs->labels); |
261 | 1.79k | } |
262 | 4.52k | MVM_free(fs); |
263 | 4.52k | } |
264 | | |
265 | | /* Cleans up all allocated memory related to this compilation. */ |
266 | 1.21k | static void cleanup_all(VM, WriterState *ws) { |
267 | 1.21k | CallsiteReuseEntry *current, *tmp; |
268 | 1.21k | unsigned bucket_tmp; |
269 | 1.21k | if (ws->cur_frame) |
270 | 0 | cleanup_frame(vm, ws->cur_frame); |
271 | 1.21k | if (ws->scdep_seg) |
272 | 1.12k | MVM_free(ws->scdep_seg); |
273 | 1.21k | if (ws->extops_seg) |
274 | 1.21k | MVM_free(ws->extops_seg); |
275 | 1.21k | if (ws->frame_seg) |
276 | 1.21k | MVM_free(ws->frame_seg); |
277 | 1.21k | if (ws->callsite_seg) |
278 | 1.21k | MVM_free(ws->callsite_seg); |
279 | 1.21k | if (ws->bytecode_seg) |
280 | 1.21k | MVM_free(ws->bytecode_seg); |
281 | 1.21k | if (ws->annotation_seg) |
282 | 1.21k | MVM_free(ws->annotation_seg); |
283 | 3.59k | HASH_ITER(hash_handle, ws->callsite_reuse_head, current, tmp, bucket_tmp) { |
284 | 3.59k | MVM_free(current->identifier); |
285 | 3.59k | } |
286 | 1.21k | MVM_HASH_DESTROY(tc, hash_handle, CallsiteReuseEntry, ws->callsite_reuse_head); |
287 | 1.21k | MVM_free(ws); |
288 | 1.21k | } |
289 | | |
290 | | /* Gets the index of a string already in the string heap, or |
291 | | * adds it to the heap if it's not already there. */ |
292 | 72.4k | static unsigned int get_string_heap_index(VM, WriterState *ws, VMSTR *strval) { |
293 | 72.4k | if (EXISTSKEY(vm, ws->seen_strings, strval)) { |
294 | 38.3k | return (unsigned int)ATKEY_I(vm, ws->seen_strings, strval); |
295 | 38.3k | } |
296 | 34.0k | else { |
297 | 34.0k | unsigned int index = (unsigned int)ELEMS(vm, ws->strings); |
298 | 34.0k | if (index >= 0x7FFFFFFF) { |
299 | 0 | cleanup_all(vm, ws); |
300 | 0 | DIE(vm, "Too many strings in compilation unit"); |
301 | 0 | } |
302 | 34.0k | BINDPOS_S(vm, ws->strings, index, strval); |
303 | 34.0k | BINDKEY_I(vm, ws->seen_strings, strval, index); |
304 | 34.0k | return index; |
305 | 34.0k | } |
306 | 72.4k | } |
307 | | |
308 | | /* Locates the index of a frame. */ |
309 | 6.73k | static unsigned short get_frame_index(VM, WriterState *ws, MASTNode *frame) { |
310 | 6.73k | if (((MAST_Frame *)frame)->flags & FRAME_FLAG_HAS_INDEX) { |
311 | 6.73k | return (short)((MAST_Frame *)frame)->index; |
312 | 6.73k | } |
313 | 0 | else { |
314 | 0 | int num_frames = ELEMS(vm, ws->cu->frames); |
315 | 0 | unsigned short i; |
316 | 0 | for (i = 0; i < num_frames; i++) |
317 | 0 | if (ATPOS(vm, ws->cu->frames, i) == frame) |
318 | 0 | return i; |
319 | 0 | cleanup_all(vm, ws); |
320 | 0 | DIE(vm, "MAST::Frame passed for code ref not found in compilation unit"); |
321 | 0 | } |
322 | 6.73k | } |
323 | | |
324 | | /* Takes a 6model object type and turns it into a local/lexical type flag. */ |
325 | 61.4k | static unsigned short type_to_local_type(VM, WriterState *ws, MASTNode *type) { |
326 | 61.4k | const MVMStorageSpec *ss; |
327 | 61.4k | if (VM_OBJ_IS_NULL(type)) |
328 | 0 | return MVM_reg_obj; |
329 | 61.4k | ss = REPR(type)->get_storage_spec(vm, STABLE(type)); |
330 | 61.4k | if (ss->inlineable) { |
331 | 25.0k | switch (ss->boxed_primitive) { |
332 | 20.3k | case MVM_STORAGE_SPEC_BP_INT: |
333 | 20.3k | if (ss->is_unsigned) { |
334 | 0 | switch (ss->bits) { |
335 | 0 | case 8: |
336 | 0 | return MVM_reg_uint8; |
337 | 0 | case 16: |
338 | 0 | return MVM_reg_uint16; |
339 | 0 | case 32: |
340 | 0 | return MVM_reg_uint32; |
341 | 0 | case 64: |
342 | 0 | return MVM_reg_uint64; |
343 | 0 | default: |
344 | 0 | cleanup_all(vm, ws); |
345 | 0 | DIE(vm, "Invalid int size for local/lexical"); |
346 | 0 | } |
347 | 0 | } |
348 | 20.3k | else { |
349 | 20.3k | switch (ss->bits) { |
350 | 2 | case 8: |
351 | 2 | return MVM_reg_int8; |
352 | 2 | case 16: |
353 | 2 | return MVM_reg_int16; |
354 | 2 | case 32: |
355 | 2 | return MVM_reg_int32; |
356 | 20.3k | case 64: |
357 | 20.3k | return MVM_reg_int64; |
358 | 0 | default: |
359 | 0 | cleanup_all(vm, ws); |
360 | 0 | DIE(vm, "Invalid int size for local/lexical"); |
361 | 20.3k | } |
362 | 20.3k | } |
363 | 0 | break; |
364 | 759 | case MVM_STORAGE_SPEC_BP_NUM: |
365 | 759 | switch (ss->bits) { |
366 | 0 | case 32: |
367 | 0 | return MVM_reg_num32; |
368 | 759 | case 64: |
369 | 759 | return MVM_reg_num64; |
370 | 0 | default: |
371 | 0 | cleanup_all(vm, ws); |
372 | 0 | DIE(vm, "Invalid num size for local/lexical"); |
373 | 759 | } |
374 | 0 | break; |
375 | 3.95k | case MVM_STORAGE_SPEC_BP_STR: |
376 | 3.95k | return MVM_reg_str; |
377 | 0 | default: |
378 | 0 | cleanup_all(vm, ws); |
379 | 0 | DIE(vm, "Type used for local/lexical has invalid boxed primitive in storage spec"); |
380 | 25.0k | } |
381 | 25.0k | } |
382 | 36.3k | else { |
383 | 36.3k | return MVM_reg_obj; |
384 | 36.3k | } |
385 | 61.4k | } |
386 | | |
387 | | /* Grows label storage. */ |
388 | 19.6k | static void add_label(VM, FrameState *fs, MAST_Label *l, MVMint32 offset) { |
389 | 19.6k | if (fs->num_labels == fs->alloc_labels) { |
390 | 3.43k | if (fs->alloc_labels) |
391 | 1.64k | fs->alloc_labels *= 2; |
392 | 3.43k | else |
393 | 1.79k | fs->alloc_labels = 8; |
394 | 3.43k | fs->labels = MVM_realloc(fs->labels, fs->alloc_labels * sizeof(LabelInfo)); |
395 | 3.43k | } |
396 | 19.6k | fs->labels[fs->num_labels].label = l; |
397 | 19.6k | fs->labels[fs->num_labels].offset = offset; |
398 | 19.6k | fs->labels[fs->num_labels].resolve = NULL; |
399 | 19.6k | fs->labels[fs->num_labels].num_resolve = 0; |
400 | 19.6k | fs->labels[fs->num_labels].alloc_resolve = 0; |
401 | 19.6k | fs->num_labels++; |
402 | 19.6k | } |
403 | | |
404 | | /* Takes a label and either writes its offset if we already saw it, or writes |
405 | | * a zero and records that a fixups is needed. */ |
406 | 31.1k | static void write_label_or_add_fixup(VM, WriterState *ws, MAST_Label *l) { |
407 | 31.1k | FrameState *fs = ws->cur_frame; |
408 | 31.1k | LabelInfo *info = NULL; |
409 | 31.1k | MVMuint32 i; |
410 | 31.1k | |
411 | 31.1k | /* Ensure we've space to write an offset. */ |
412 | 31.1k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 4); |
413 | 31.1k | |
414 | 31.1k | /* Look for the label. */ |
415 | 248k | for (i = 0; i < fs->num_labels; i++) { |
416 | 233k | if (fs->labels[i].label == l) { |
417 | 16.2k | /* Found it. If we know its offset, write and we're done. */ |
418 | 16.2k | MVMint32 offset = fs->labels[i].offset; |
419 | 16.2k | if (offset >= 0) { |
420 | 6.72k | write_int32(ws->bytecode_seg, ws->bytecode_pos, offset); |
421 | 6.72k | ws->bytecode_pos += 4; |
422 | 6.72k | return; |
423 | 6.72k | } |
424 | 16.2k | |
425 | 16.2k | /* Otherwise, note this label to add the resolve need to. */ |
426 | 9.57k | info = &(fs->labels[i]); |
427 | 9.57k | break; |
428 | 16.2k | } |
429 | 233k | } |
430 | 31.1k | |
431 | 31.1k | /* If we don't have an entry for this label yet, add it. */ |
432 | 24.4k | if (!info) { |
433 | 14.8k | add_label(vm, fs, l, -1); |
434 | 14.8k | info = &(fs->labels[fs->num_labels - 1]); |
435 | 14.8k | } |
436 | 24.4k | if (info->num_resolve == info->alloc_resolve) { |
437 | 14.9k | if (info->alloc_resolve) |
438 | 109 | info->alloc_resolve *= 2; |
439 | 14.9k | else |
440 | 14.8k | info->alloc_resolve = 8; |
441 | 14.9k | info->resolve = MVM_realloc(info->resolve, info->alloc_resolve * sizeof(MVMuint32)); |
442 | 14.9k | } |
443 | 24.4k | info->resolve[info->num_resolve] = ws->bytecode_pos; |
444 | 24.4k | info->num_resolve++; |
445 | 24.4k | fs->unresolved_labels++; |
446 | 24.4k | |
447 | 24.4k | /* Write zero, to be fixed up later. */ |
448 | 24.4k | write_int32(ws->bytecode_seg, ws->bytecode_pos, 0); |
449 | 24.4k | ws->bytecode_pos += 4; |
450 | 24.4k | } |
451 | | |
452 | | /* Takes a label, and either adds it to the labels collection or, if it's been |
453 | | * seen already, resolves its fixups. */ |
454 | 19.6k | static void add_label_and_resolve_fixups(VM, WriterState *ws, MAST_Label *l) { |
455 | 19.6k | FrameState *fs = ws->cur_frame; |
456 | 19.6k | MVMuint32 offset = ws->bytecode_pos - ws->cur_frame->bytecode_start; |
457 | 19.6k | MVMuint32 i, j; |
458 | 19.6k | |
459 | 19.6k | /* See if it has an existing entry. */ |
460 | 179k | for (i = 0; i < fs->num_labels; i++) { |
461 | 175k | if (fs->labels[i].label == l) { |
462 | 14.8k | /* Found it. Must not already have an offset, or it's a dupe. */ |
463 | 14.8k | if (fs->labels[i].offset < 0) { |
464 | 14.8k | /* Fix up existing usages. */ |
465 | 14.8k | MVMuint32 *resolve = fs->labels[i].resolve; |
466 | 14.8k | MVMuint32 nr = fs->labels[i].num_resolve; |
467 | 39.2k | for (j = 0; j < nr; j++) |
468 | 24.4k | write_int32(ws->bytecode_seg, resolve[j], offset); |
469 | 14.8k | fs->labels[i].offset = offset; |
470 | 14.8k | fs->labels[i].alloc_resolve = 0; |
471 | 14.8k | fs->labels[i].num_resolve = 0; |
472 | 14.8k | fs->unresolved_labels -= nr; |
473 | 14.8k | MVM_free(fs->labels[i].resolve); |
474 | 14.8k | } |
475 | 0 | else { |
476 | 0 | cleanup_all(vm, ws); |
477 | 0 | DIE(vm, "Duplicate label"); |
478 | 0 | } |
479 | 14.8k | return; |
480 | 14.8k | } |
481 | 175k | } |
482 | 19.6k | |
483 | 19.6k | /* If we get here, no entry; create one. */ |
484 | 4.78k | add_label(vm, fs, l, offset); |
485 | 4.78k | } |
486 | | |
487 | | /* Rreturns a label's offset, dying if it's not possible. */ |
488 | | static MVMuint32 demand_label_offset(VM, WriterState *ws, MAST_Label *l, |
489 | 431 | const char *error) { |
490 | 431 | FrameState *fs = ws->cur_frame; |
491 | 431 | MVMuint32 nl = fs->num_labels; |
492 | 431 | MVMuint32 i; |
493 | 7.44k | for (i = 0; i < nl; i++) { |
494 | 7.44k | if (fs->labels[i].label == l) { |
495 | 431 | if (fs->labels[i].offset >= 0) |
496 | 431 | return fs->labels[i].offset; |
497 | 0 | break; |
498 | 431 | } |
499 | 7.44k | } |
500 | 0 | cleanup_all(vm, ws); |
501 | 0 | DIE(vm, "%s", error); |
502 | 0 | } |
503 | | |
504 | | /* Compiles the operand to an instruction; this involves checking |
505 | | * that we have a node of the correct type for it and writing out |
506 | | * the appropriate thing to the bytecode stream. */ |
507 | 545k | static void compile_operand(VM, WriterState *ws, unsigned char op_flags, MASTNode *operand) { |
508 | 545k | unsigned char op_rw = op_flags & MVM_operand_rw_mask; |
509 | 545k | unsigned char op_type = op_flags & MVM_operand_type_mask; |
510 | 545k | unsigned short int local_type; |
511 | 545k | if (op_rw == MVM_operand_literal) { |
512 | 119k | /* Literal; go by type. */ |
513 | 119k | switch (op_type) { |
514 | 1.29k | case MVM_operand_int64: { |
515 | 1.29k | if (ISTYPE(vm, operand, ws->types->IVal)) { |
516 | 1.29k | MAST_IVal *iv = GET_IVal(operand); |
517 | 1.29k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 8); |
518 | 1.29k | write_int64(ws->bytecode_seg, ws->bytecode_pos, iv->value); |
519 | 1.29k | ws->bytecode_pos += 8; |
520 | 1.29k | } |
521 | 0 | else { |
522 | 0 | cleanup_all(vm, ws); |
523 | 0 | DIE(vm, "Expected MAST::IVal, but didn't get one"); |
524 | 0 | } |
525 | 1.29k | break; |
526 | 1.29k | } |
527 | 38.7k | case MVM_operand_int16: { |
528 | 38.7k | if (ISTYPE(vm, operand, ws->types->IVal)) { |
529 | 38.7k | MAST_IVal *iv = GET_IVal(operand); |
530 | 38.7k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
531 | 38.7k | if (iv->value > 32767 || iv->value < -32768) { |
532 | 0 | cleanup_all(vm, ws); |
533 | 0 | DIE(vm, "Value outside range of 16-bit MAST::IVal"); |
534 | 0 | } |
535 | 38.7k | write_int16(ws->bytecode_seg, ws->bytecode_pos, (short)iv->value); |
536 | 38.7k | ws->bytecode_pos += 2; |
537 | 38.7k | } |
538 | 0 | else { |
539 | 0 | cleanup_all(vm, ws); |
540 | 0 | DIE(vm, "Expected MAST::IVal, but didn't get one"); |
541 | 0 | } |
542 | 38.7k | break; |
543 | 1.29k | } |
544 | 498 | case MVM_operand_num64: { |
545 | 498 | if (ISTYPE(vm, operand, ws->types->NVal)) { |
546 | 498 | MAST_NVal *nv = GET_NVal(operand); |
547 | 498 | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 8); |
548 | 498 | write_double(ws->bytecode_seg, ws->bytecode_pos, nv->value); |
549 | 498 | ws->bytecode_pos += 8; |
550 | 498 | } |
551 | 0 | else { |
552 | 0 | cleanup_all(vm, ws); |
553 | 0 | DIE(vm, "Expected MAST::NVal, but didn't get one"); |
554 | 0 | } |
555 | 498 | break; |
556 | 1.29k | } |
557 | 42.1k | case MVM_operand_str: { |
558 | 42.1k | if (ISTYPE(vm, operand, ws->types->SVal)) { |
559 | 42.1k | MAST_SVal *sv = GET_SVal(operand); |
560 | 42.1k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 4); |
561 | 42.1k | write_int32(ws->bytecode_seg, ws->bytecode_pos, |
562 | 42.1k | get_string_heap_index(vm, ws, sv->value)); |
563 | 42.1k | ws->bytecode_pos += 4; |
564 | 42.1k | } |
565 | 0 | else { |
566 | 0 | cleanup_all(vm, ws); |
567 | 0 | DIE(vm, "Expected MAST::SVal, but didn't get one"); |
568 | 0 | } |
569 | 42.1k | break; |
570 | 1.29k | } |
571 | 31.1k | case MVM_operand_ins: { |
572 | 31.1k | if (ISTYPE(vm, operand, ws->types->Label)) { |
573 | 31.1k | write_label_or_add_fixup(vm, ws, GET_Label(operand)); |
574 | 31.1k | } |
575 | 0 | else { |
576 | 0 | cleanup_all(vm, ws); |
577 | 0 | DIE(vm, "Expected MAST::Label, but didn't get one"); |
578 | 0 | } |
579 | 31.1k | break; |
580 | 1.29k | } |
581 | 5.36k | case MVM_operand_coderef: { |
582 | 5.36k | if (ISTYPE(vm, operand, ws->types->Frame)) { |
583 | 5.36k | /* Find the frame index in the compilation unit. */ |
584 | 5.36k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
585 | 5.36k | write_int16(ws->bytecode_seg, ws->bytecode_pos, |
586 | 5.36k | get_frame_index(vm, ws, operand)); |
587 | 5.36k | ws->bytecode_pos += 2; |
588 | 5.36k | } |
589 | 0 | else { |
590 | 0 | cleanup_all(vm, ws); |
591 | 0 | DIE(vm, "Expected MAST::Frame, but didn't get one"); |
592 | 0 | } |
593 | 5.36k | break; |
594 | 1.29k | } |
595 | 0 | default: |
596 | 0 | cleanup_all(vm, ws); |
597 | 0 | DIE(vm, "Unhandled literal type in MAST compiler"); |
598 | 119k | } |
599 | 119k | } |
600 | 426k | else if (op_rw == MVM_operand_read_reg || op_rw == MVM_operand_write_reg) { |
601 | 422k | /* The operand node had best be a MAST::Local. */ |
602 | 422k | if (ISTYPE(vm, operand, ws->types->Local)) { |
603 | 422k | MAST_Local *l = GET_Local(operand); |
604 | 422k | |
605 | 422k | /* Ensure it's within the set of known locals. */ |
606 | 422k | if (l->index >= ws->cur_frame->num_locals) { |
607 | 0 | cleanup_all(vm, ws); |
608 | 0 | DIE(vm, "MAST::Local index out of range"); |
609 | 0 | } |
610 | 422k | |
611 | 422k | /* Check the type matches. */ |
612 | 422k | local_type = ws->cur_frame->local_types[l->index]; |
613 | 422k | if (op_type != local_type << 3 && op_type != MVM_operand_type_var) { |
614 | 0 | unsigned int current_frame_idx = ws->current_frame_idx; |
615 | 0 | unsigned int current_ins_idx = ws->current_ins_idx; |
616 | 0 | const char *name = ws->current_op_info->name; |
617 | 0 | unsigned int current_operand_idx = ws->current_operand_idx; |
618 | 0 | cleanup_all(vm, ws); |
619 | 0 | DIE(vm, "At Frame %u, Instruction %u, op '%s', operand %u, " |
620 | 0 | "MAST::Local of wrong type (%u) specified; expected %u", |
621 | 0 | current_frame_idx, current_ins_idx, |
622 | 0 | name, current_operand_idx, |
623 | 0 | local_type, (op_type >> 3)); |
624 | 0 | } |
625 | 422k | |
626 | 422k | /* Write the operand type. */ |
627 | 422k | if (l->index < 0 || l->index > 32768) |
628 | 0 | DIE(vm, "Frame %u local access out of range", ws->current_frame_idx); |
629 | 422k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
630 | 422k | write_int16(ws->bytecode_seg, ws->bytecode_pos, (unsigned short)l->index); |
631 | 422k | ws->bytecode_pos += 2; |
632 | 422k | } |
633 | 0 | else { |
634 | 0 | unsigned int current_frame_idx = ws->current_frame_idx; |
635 | 0 | unsigned int current_ins_idx = ws->current_ins_idx; |
636 | 0 | const char *name = ws->current_op_info->name; |
637 | 0 | unsigned int current_operand_idx = ws->current_operand_idx; |
638 | 0 | cleanup_all(vm, ws); |
639 | 0 | DIE(vm, "At Frame %u, Instruction %u, op '%s', operand %u, expected MAST::Local, but didn't get one", |
640 | 0 | current_frame_idx, current_ins_idx, name, current_operand_idx); |
641 | 0 | } |
642 | 422k | } |
643 | 4.26k | else if (op_rw == MVM_operand_read_lex || op_rw == MVM_operand_write_lex) { |
644 | 4.26k | /* The operand node should be a MAST::Lexical. */ |
645 | 4.26k | if (ISTYPE(vm, operand, ws->types->Lexical)) { |
646 | 4.26k | MAST_Lexical *l = GET_Lexical(operand); |
647 | 4.26k | |
648 | 4.26k | /* Write the index, then the frame count. */ |
649 | 4.26k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 4); |
650 | 4.26k | write_int16(ws->bytecode_seg, ws->bytecode_pos, (unsigned short)l->index); |
651 | 4.26k | ws->bytecode_pos += 2; |
652 | 4.26k | write_int16(ws->bytecode_seg, ws->bytecode_pos, (unsigned short)l->frames_out); |
653 | 4.26k | ws->bytecode_pos += 2; |
654 | 4.26k | } |
655 | 0 | else { |
656 | 0 | cleanup_all(vm, ws); |
657 | 0 | DIE(vm, "Expected MAST::Lexical, but didn't get one"); |
658 | 0 | } |
659 | 4.26k | } |
660 | 0 | else { |
661 | 0 | cleanup_all(vm, ws); |
662 | 0 | DIE(vm, "Unknown operand type cannot be compiled"); |
663 | 0 | } |
664 | 545k | ws->current_operand_idx++; |
665 | 545k | } |
666 | | |
667 | | /* Takes a set of flags describing a callsite. Writes out a callsite |
668 | | * descriptor and returns the index of it. */ |
669 | 14.9k | static unsigned short get_callsite_id(VM, WriterState *ws, MASTNode *flag_node, MASTNode *args) { |
670 | 14.9k | unsigned int num_nameds = 0; |
671 | 14.9k | unsigned short i, identifier_len; |
672 | 14.9k | unsigned char *flags, *identifier; |
673 | 14.9k | unsigned int *named_idxs; |
674 | 14.9k | CallsiteReuseEntry *entry = NULL; |
675 | 14.9k | |
676 | 14.9k | /* Get callsite elements and work out if a padding byte will be needed. */ |
677 | 14.9k | unsigned short elems = (unsigned short)ELEMS(vm, flag_node); |
678 | 14.9k | unsigned short align = elems % 2; |
679 | 14.9k | |
680 | 14.9k | /* See if the callsite has any named args, and get string pool entries |
681 | 14.9k | * for them if so. */ |
682 | 14.9k | flags = (unsigned char *)MVM_malloc(elems); |
683 | 14.9k | named_idxs = (unsigned int *)MVM_malloc(elems * sizeof(int)); |
684 | 41.8k | for (i = 0; i < elems; i++) { |
685 | 26.9k | flags[i] = (unsigned char)ATPOS_I_C(vm, flag_node, i); |
686 | 26.9k | if (flags[i] & (MVM_CALLSITE_ARG_NAMED)) { |
687 | 3.12k | MASTNode *argname = ATPOS(vm, args, i + num_nameds); |
688 | 3.12k | if (ISTYPE(vm, argname, ws->types->SVal)) { |
689 | 3.12k | named_idxs[num_nameds] = get_string_heap_index(vm, ws, |
690 | 3.12k | ((MAST_SVal *)argname)->value); |
691 | 3.12k | num_nameds++; |
692 | 3.12k | } |
693 | 0 | else { |
694 | 0 | DIE(vm, "Malformed callsite args: missing MAST::SVal for argument name"); |
695 | 0 | } |
696 | 3.12k | } |
697 | 26.9k | } |
698 | 14.9k | |
699 | 14.9k | /* See if we already know this callsite. */ |
700 | 14.9k | identifier_len = elems + num_nameds * sizeof(int); |
701 | 14.9k | identifier = MVM_malloc(identifier_len); |
702 | 14.9k | memcpy(identifier, flags, elems); |
703 | 14.9k | memcpy(identifier + elems, named_idxs, identifier_len - elems); |
704 | 14.9k | HASH_FIND(hash_handle, ws->callsite_reuse_head, identifier, identifier_len, entry); |
705 | 14.9k | if (entry) { |
706 | 11.3k | MVM_free(flags); |
707 | 11.3k | MVM_free(named_idxs); |
708 | 11.3k | MVM_free(identifier); |
709 | 11.3k | return entry->callsite_id; |
710 | 11.3k | } |
711 | 3.59k | entry = (CallsiteReuseEntry *)MVM_malloc(sizeof(CallsiteReuseEntry)); |
712 | 3.59k | entry->callsite_id = (unsigned short)ws->num_callsites; |
713 | 3.59k | entry->identifier = identifier; |
714 | 3.59k | HASH_ADD_KEYPTR(hash_handle, ws->callsite_reuse_head, identifier, identifier_len, entry); |
715 | 3.59k | |
716 | 3.59k | /* Emit callsite; be sure to pad if there's uneven number of flags. */ |
717 | 3.59k | ensure_space(vm, &ws->callsite_seg, &ws->callsite_alloc, ws->callsite_pos, |
718 | 3.59k | 2 + elems + align); |
719 | 3.59k | write_int16(ws->callsite_seg, ws->callsite_pos, elems); |
720 | 3.59k | ws->callsite_pos += 2; |
721 | 10.7k | for (i = 0; i < elems; i++) |
722 | 7.14k | write_int8(ws->callsite_seg, ws->callsite_pos++, flags[i]); |
723 | 3.59k | if (align) |
724 | 2.78k | write_int8(ws->callsite_seg, ws->callsite_pos++, 0); |
725 | 3.59k | |
726 | 3.59k | /* Emit any nameds. */ |
727 | 3.59k | if (num_nameds) { |
728 | 1.12k | ensure_space(vm, &ws->callsite_seg, &ws->callsite_alloc, ws->callsite_pos, |
729 | 1.12k | 4 * num_nameds); |
730 | 2.38k | for (i = 0; i < num_nameds; i++) { |
731 | 1.25k | write_int32(ws->callsite_seg, ws->callsite_pos, named_idxs[i]); |
732 | 1.25k | ws->callsite_pos += 4; |
733 | 1.25k | } |
734 | 1.12k | } |
735 | 3.59k | |
736 | 3.59k | MVM_free(flags); |
737 | 3.59k | MVM_free(named_idxs); |
738 | 3.59k | |
739 | 3.59k | return (unsigned short)ws->num_callsites++; |
740 | 3.59k | } |
741 | | |
742 | 19.8k | #define OVERRIDE_WITH_32 1 |
743 | 39.3k | #define OVERRIDE_WITH_16 2 |
744 | | |
745 | | /* Compiles an instruction (which may actaully be any of the |
746 | | * nodes valid directly in a Frame's instruction list, which |
747 | | * means labels are valid too). */ |
748 | 274k | static void compile_instruction(VM, WriterState *ws, MASTNode *node) { |
749 | 274k | if (ISTYPE(vm, node, ws->types->Op)) { |
750 | 228k | MAST_Op *o = GET_Op(node); |
751 | 228k | const MVMOpInfo *info; |
752 | 228k | int i; |
753 | 228k | unsigned char override_second_argument = 0; |
754 | 228k | |
755 | 228k | /* Look up opcode and get argument info. */ |
756 | 228k | unsigned short op = o->op; |
757 | 228k | info = MVM_op_get_op(op); |
758 | 228k | if (!info) |
759 | 0 | DIE(vm, "Invalid op specified in instruction %d", op); |
760 | 228k | ws->current_op_info = info; |
761 | 228k | ws->current_operand_idx = 0; |
762 | 228k | |
763 | 228k | /* Ensure argument count matches up. */ |
764 | 228k | if (info->num_operands != 0 && ELEMS(vm, o->operands) != info->num_operands) { |
765 | 0 | unsigned int current_frame_idx = ws->current_frame_idx; |
766 | 0 | unsigned int current_ins_idx = ws->current_ins_idx; |
767 | 0 | const char *name = ws->current_op_info->name; |
768 | 0 | cleanup_all(vm, ws); |
769 | 0 | DIE(vm, "At Frame %u, Instruction %u, op '%s' has invalid number (%u) of operands; needs %u.", |
770 | 0 | current_frame_idx, current_ins_idx, name, |
771 | 0 | ELEMS(vm, o->operands), info->num_operands); |
772 | 0 | } |
773 | 228k | |
774 | 228k | /* If we're outputting a const_i64 instruction, we may want to */ |
775 | 228k | /* turn it into a const_i64_32 or const_i64_16 instead if it fits */ |
776 | 228k | if (op == MVM_OP_const_i64) { |
777 | 19.7k | MASTNode *operand = ATPOS(vm, o->operands, 1); |
778 | 19.7k | MAST_IVal *iv = GET_IVal(operand); |
779 | 19.7k | if (INT16_MIN <= iv->value && iv->value <= INT16_MAX) { |
780 | 19.6k | override_second_argument = OVERRIDE_WITH_16; |
781 | 86 | } else if (INT32_MIN <= iv->value && iv->value <= INT32_MAX) { |
782 | 86 | override_second_argument = OVERRIDE_WITH_32; |
783 | 86 | } |
784 | 19.7k | } |
785 | 228k | |
786 | 228k | /* Write opcode. */ |
787 | 228k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
788 | 228k | if (override_second_argument == 0) |
789 | 208k | write_int16(ws->bytecode_seg, ws->bytecode_pos, op); |
790 | 19.7k | else if (override_second_argument == OVERRIDE_WITH_16) |
791 | 19.6k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_const_i64_16); |
792 | 86 | else if (override_second_argument == OVERRIDE_WITH_32) |
793 | 86 | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_const_i64_32); |
794 | 228k | ws->bytecode_pos += 2; |
795 | 228k | |
796 | 228k | /* Write operands. */ |
797 | 735k | for (i = 0; i < info->num_operands; i++) { |
798 | 507k | if (i == 1 && override_second_argument) { |
799 | 19.7k | MASTNode *operand = ATPOS(vm, o->operands, 1); |
800 | 19.7k | MAST_IVal *iv = GET_IVal(operand); |
801 | 19.7k | if (override_second_argument == OVERRIDE_WITH_32) { |
802 | 86 | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 4); |
803 | 86 | write_int32(ws->bytecode_seg, ws->bytecode_pos, (MVMint32)iv->value); |
804 | 86 | ws->bytecode_pos += 4; |
805 | 19.6k | } else { |
806 | 19.6k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
807 | 19.6k | write_int16(ws->bytecode_seg, ws->bytecode_pos, (MVMint16)iv->value); |
808 | 19.6k | ws->bytecode_pos += 2; |
809 | 19.6k | } |
810 | 487k | } else { |
811 | 487k | compile_operand(vm, ws, info->operands[i], ATPOS(vm, o->operands, i)); |
812 | 487k | } |
813 | 507k | } |
814 | 228k | } |
815 | 46.4k | else if (ISTYPE(vm, node, ws->types->ExtOp)) { |
816 | 0 | MAST_ExtOp *o = GET_ExtOp(node); |
817 | 0 | MASTNode *operands; |
818 | 0 | int i, num_operands; |
819 | 0 |
|
820 | 0 | /* Look up opcode and get argument info. */ |
821 | 0 | unsigned short op = o->op; |
822 | 0 | if (op < EXTOP_BASE || (op - EXTOP_BASE) >= ELEMS(vm, ws->cu->extop_sigs)) |
823 | 0 | DIE(vm, "Invalid extension op %d specified", op); |
824 | 0 | operands = ATPOS(vm, ws->cu->extop_sigs, op - EXTOP_BASE); |
825 | 0 | if (VM_OBJ_IS_NULL(operands)) |
826 | 0 | DIE(vm, "Missing extension op operand array for instruction %d", op); |
827 | 0 | ws->current_op_info = NULL; |
828 | 0 | ws->current_operand_idx = 0; |
829 | 0 |
|
830 | 0 | /* Ensure argument count matches up. */ |
831 | 0 | num_operands = ELEMS(vm, operands); |
832 | 0 | if (ELEMS(vm, o->operands) != num_operands) { |
833 | 0 | unsigned int current_frame_idx = ws->current_frame_idx; |
834 | 0 | unsigned int current_ins_idx = ws->current_ins_idx; |
835 | 0 | char *c_name = VM_STRING_TO_C_STRING(vm, o->name); |
836 | 0 | char *waste[] = { c_name, NULL }; |
837 | 0 | cleanup_all(vm, ws); |
838 | 0 | DIE_FREE(vm, waste, "At Frame %u, Instruction %u, op '%s' has invalid number (%u) of operands; needs %u.", |
839 | 0 | current_frame_idx, current_ins_idx, |
840 | 0 | c_name, |
841 | 0 | ELEMS(vm, o->operands), num_operands); |
842 | 0 | } |
843 | 0 |
|
844 | 0 | /* Write opcode. */ |
845 | 0 | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
846 | 0 | write_int16(ws->bytecode_seg, ws->bytecode_pos, op); |
847 | 0 | ws->bytecode_pos += 2; |
848 | 0 |
|
849 | 0 | /* Write operands. */ |
850 | 0 | for (i = 0; i < num_operands; i++) |
851 | 0 | compile_operand(vm, ws, ATPOS_I(vm, operands, i), ATPOS(vm, o->operands, i)); |
852 | 0 | } |
853 | 46.4k | else if (ISTYPE(vm, node, ws->types->Label)) { |
854 | 19.6k | add_label_and_resolve_fixups(vm, ws, GET_Label(node)); |
855 | 19.6k | } |
856 | 26.8k | else if (ISTYPE(vm, node, ws->types->Call)) { |
857 | 14.9k | MAST_Call *c = GET_Call(node); |
858 | 14.9k | unsigned char res_type = 0; |
859 | 14.9k | unsigned short num_flags, flag_pos, arg_pos, arg_out_pos, callsite_id; |
860 | 14.9k | unsigned short call_op; |
861 | 14.9k | switch (c->op) { |
862 | 0 | case 1: call_op = MVM_OP_nativeinvoke_v; break; |
863 | 16 | case 2: call_op = MVM_OP_speshresolve; break; |
864 | 14.9k | default: call_op = MVM_OP_invoke_v; break; |
865 | 14.9k | } |
866 | 14.9k | |
867 | 14.9k | /* Emit callsite (may re-use existing one) and emit loading of it. */ |
868 | 14.9k | callsite_id = get_callsite_id(vm, ws, c->flags, c->args); |
869 | 14.9k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 4); |
870 | 14.9k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_prepargs); |
871 | 14.9k | ws->bytecode_pos += 2; |
872 | 14.9k | write_int16(ws->bytecode_seg, ws->bytecode_pos, callsite_id); |
873 | 14.9k | ws->bytecode_pos += 2; |
874 | 14.9k | |
875 | 14.9k | /* for errors */ |
876 | 14.9k | ws->current_op_info = MVM_op_get_op(MVM_OP_prepargs); |
877 | 14.9k | ws->current_operand_idx = 0; |
878 | 14.9k | |
879 | 14.9k | /* Set up args. */ |
880 | 14.9k | num_flags = (unsigned short)ELEMS(vm, c->flags); |
881 | 14.9k | arg_pos = c->op == 1 ? 1 : 0; |
882 | 14.9k | arg_out_pos = 0; |
883 | 41.8k | for (flag_pos = 0; flag_pos < num_flags; flag_pos++) { |
884 | 26.9k | /* Handle any special flags. */ |
885 | 26.9k | unsigned char flag = (unsigned char)ATPOS_I_C(vm, c->flags, flag_pos); |
886 | 26.9k | if (flag & MVM_CALLSITE_ARG_NAMED) { |
887 | 3.12k | if (c->op == 2) { |
888 | 0 | cleanup_all(vm, ws); |
889 | 0 | DIE(vm, "At Frame %u, Instruction %u, illegal named arg to speshresolve.", |
890 | 0 | ws->current_frame_idx, ws->current_ins_idx); |
891 | 0 | } |
892 | 3.12k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 6); |
893 | 3.12k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_argconst_s); |
894 | 3.12k | ws->bytecode_pos += 2; |
895 | 3.12k | write_int16(ws->bytecode_seg, ws->bytecode_pos, arg_out_pos); |
896 | 3.12k | ws->bytecode_pos += 2; |
897 | 3.12k | compile_operand(vm, ws, MVM_operand_str, ATPOS(vm, c->args, arg_pos)); |
898 | 3.12k | arg_pos++; |
899 | 3.12k | arg_out_pos++; |
900 | 3.12k | } |
901 | 23.7k | else if (flag & MVM_CALLSITE_ARG_FLAT) { |
902 | 158 | if (c->op == 2) { |
903 | 0 | cleanup_all(vm, ws); |
904 | 0 | DIE(vm, "At Frame %u, Instruction %u, illegal flat arg to speshresolve.", |
905 | 0 | ws->current_frame_idx, ws->current_ins_idx); |
906 | 0 | } |
907 | 158 | /* don't need to do anything special */ |
908 | 158 | } |
909 | 26.9k | |
910 | 26.9k | /* Now go by flag type. */ |
911 | 26.9k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 6); |
912 | 26.9k | if (c->op == 2 && !(flag & MVM_CALLSITE_ARG_OBJ)) { |
913 | 0 | cleanup_all(vm, ws); |
914 | 0 | DIE(vm, "At Frame %u, Instruction %u, illegal non-object arg to speshresolve.", |
915 | 0 | ws->current_frame_idx, ws->current_ins_idx); |
916 | 0 | } |
917 | 26.9k | if (flag & MVM_CALLSITE_ARG_OBJ) { |
918 | 13.4k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_arg_o); |
919 | 13.4k | ws->bytecode_pos += 2; |
920 | 13.4k | write_int16(ws->bytecode_seg, ws->bytecode_pos, arg_out_pos); |
921 | 13.4k | ws->bytecode_pos += 2; |
922 | 13.4k | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_obj, |
923 | 13.4k | ATPOS(vm, c->args, arg_pos)); |
924 | 13.4k | } |
925 | 13.4k | else if (flag & MVM_CALLSITE_ARG_STR) { |
926 | 7.49k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_arg_s); |
927 | 7.49k | ws->bytecode_pos += 2; |
928 | 7.49k | write_int16(ws->bytecode_seg, ws->bytecode_pos, arg_out_pos); |
929 | 7.49k | ws->bytecode_pos += 2; |
930 | 7.49k | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_str, |
931 | 7.49k | ATPOS(vm, c->args, arg_pos)); |
932 | 7.49k | } |
933 | 5.96k | else if (flag & MVM_CALLSITE_ARG_INT) { |
934 | 5.75k | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_arg_i); |
935 | 5.75k | ws->bytecode_pos += 2; |
936 | 5.75k | write_int16(ws->bytecode_seg, ws->bytecode_pos, arg_out_pos); |
937 | 5.75k | ws->bytecode_pos += 2; |
938 | 5.75k | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_int64, |
939 | 5.75k | ATPOS(vm, c->args, arg_pos)); |
940 | 5.75k | } |
941 | 210 | else if (flag & MVM_CALLSITE_ARG_NUM) { |
942 | 210 | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_arg_n); |
943 | 210 | ws->bytecode_pos += 2; |
944 | 210 | write_int16(ws->bytecode_seg, ws->bytecode_pos, arg_out_pos); |
945 | 210 | ws->bytecode_pos += 2; |
946 | 210 | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_num64, |
947 | 210 | ATPOS(vm, c->args, arg_pos)); |
948 | 210 | } |
949 | 0 | else { |
950 | 0 | unsigned int current_frame_idx = ws->current_frame_idx; |
951 | 0 | unsigned int current_ins_idx = ws->current_ins_idx; |
952 | 0 | const char *name = ws->current_op_info->name; |
953 | 0 | cleanup_all(vm, ws); |
954 | 0 | /* |
955 | 0 | DIE(vm, "At Frame %u, Instruction %u, op '%s', " |
956 | 0 | "file %s, line %u, unhandled arg type %u.", |
957 | 0 | current_frame_idx, current_ins_idx, name, |
958 | 0 | ws->last_annotated ? VM_STRING_TO_C_STRING(vm, ws->last_annotated->file) : "", |
959 | 0 | ws->last_annotated ? ws->last_annotated->line : 0, |
960 | 0 | flag); |
961 | 0 | */ |
962 | 0 | DIE(vm, "At Frame %u, Instruction %u, op '%s', unhandled arg type %u.", |
963 | 0 | current_frame_idx, current_ins_idx, name, flag); |
964 | 0 | } |
965 | 26.9k | |
966 | 26.9k | arg_pos++; |
967 | 26.9k | arg_out_pos++; |
968 | 26.9k | } |
969 | 14.9k | |
970 | 14.9k | /* Select operation based on return type. */ |
971 | 14.9k | if (c->op == 2) { |
972 | 16 | if (ISTYPE(vm, c->result, ws->types->Local)) { |
973 | 16 | MAST_Local *l = GET_Local(c->result); |
974 | 16 | if (l->index >= ws->cur_frame->num_locals) { |
975 | 0 | cleanup_all(vm, ws); |
976 | 0 | DIE(vm, "MAST::Local index out of range"); |
977 | 0 | } |
978 | 16 | if (ws->cur_frame->local_types[l->index] != MVM_reg_obj) { |
979 | 0 | cleanup_all(vm, ws); |
980 | 0 | DIE(vm, "At Frame %u, Instruction %u, speshresolve must have an object result.", |
981 | 0 | ws->current_frame_idx, ws->current_ins_idx); |
982 | 0 | } |
983 | 16 | res_type = MVM_operand_obj; |
984 | 16 | } |
985 | 0 | else { |
986 | 0 | cleanup_all(vm, ws); |
987 | 0 | DIE(vm, "At Frame %u, Instruction %u, speshresolve must have a result.", |
988 | 0 | ws->current_frame_idx, ws->current_ins_idx); |
989 | 0 | } |
990 | 16 | } |
991 | 14.9k | else if (ISTYPE(vm, c->result, ws->types->Local)) { |
992 | 13.2k | MAST_Local *l = GET_Local(c->result); |
993 | 13.2k | |
994 | 13.2k | /* Ensure it's within the set of known locals. */ |
995 | 13.2k | if (l->index >= ws->cur_frame->num_locals) { |
996 | 0 | cleanup_all(vm, ws); |
997 | 0 | DIE(vm, "MAST::Local index out of range"); |
998 | 0 | } |
999 | 13.2k | |
1000 | 13.2k | /* Go by type. */ |
1001 | 13.2k | switch (ws->cur_frame->local_types[l->index]) { |
1002 | 1 | case MVM_reg_int64: |
1003 | 1 | call_op = c->op == 0 ? MVM_OP_invoke_i : MVM_OP_nativeinvoke_i; |
1004 | 1 | res_type = MVM_operand_int64; |
1005 | 1 | break; |
1006 | 0 | case MVM_reg_num64: |
1007 | 0 | call_op = c->op == 0 ? MVM_OP_invoke_n : MVM_OP_nativeinvoke_n; |
1008 | 0 | res_type = MVM_operand_num64; |
1009 | 0 | break; |
1010 | 288 | case MVM_reg_str: |
1011 | 288 | call_op = c->op == 0 ? MVM_OP_invoke_s : MVM_OP_nativeinvoke_s; |
1012 | 288 | res_type = MVM_operand_str; |
1013 | 288 | break; |
1014 | 12.9k | case MVM_reg_obj: |
1015 | 12.9k | call_op = c->op == 0 ? MVM_OP_invoke_o : MVM_OP_nativeinvoke_o; |
1016 | 12.9k | res_type = MVM_operand_obj; |
1017 | 12.9k | break; |
1018 | 0 | default: |
1019 | 0 | cleanup_all(vm, ws); |
1020 | 0 | DIE(vm, "Invalid MAST::Local type for return value"); |
1021 | 13.2k | } |
1022 | 13.2k | } |
1023 | 14.9k | |
1024 | 14.9k | /* Emit the invocation op. */ |
1025 | 14.9k | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, c->op == 0 ? 6 : 8); |
1026 | 14.9k | write_int16(ws->bytecode_seg, ws->bytecode_pos, call_op); |
1027 | 14.9k | ws->bytecode_pos += 2; |
1028 | 14.9k | if (call_op != MVM_OP_invoke_v && call_op != MVM_OP_nativeinvoke_v) |
1029 | 13.2k | compile_operand(vm, ws, MVM_operand_read_reg | res_type, c->result); |
1030 | 14.9k | if (c->op == 2) |
1031 | 16 | compile_operand(vm, ws, MVM_operand_str, c->target); |
1032 | 14.9k | else |
1033 | 14.9k | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_obj, c->target); |
1034 | 14.9k | if (c->op == 1) |
1035 | 0 | compile_operand(vm, ws, MVM_operand_read_reg | MVM_operand_obj, ATPOS(vm, c->args, 0)); |
1036 | 14.9k | } |
1037 | 11.8k | else if (ISTYPE(vm, node, ws->types->Annotated)) { |
1038 | 11.4k | MAST_Annotated *a = GET_Annotated(node); |
1039 | 11.4k | unsigned int i; |
1040 | 11.4k | unsigned int num_ins = ELEMS(vm, a->instructions); |
1041 | 11.4k | unsigned int offset = ws->bytecode_pos - ws->cur_frame->bytecode_start; |
1042 | 11.4k | |
1043 | 11.4k | ws->last_annotated = a; |
1044 | 11.4k | ensure_space(vm, &ws->annotation_seg, &ws->annotation_alloc, ws->annotation_pos, 12); |
1045 | 11.4k | write_int32(ws->annotation_seg, ws->annotation_pos, offset); |
1046 | 11.4k | write_int32(ws->annotation_seg, ws->annotation_pos + 4, get_string_heap_index(vm, ws, a->file)); |
1047 | 11.4k | write_int32(ws->annotation_seg, ws->annotation_pos + 8, (unsigned int)a->line); |
1048 | 11.4k | ws->annotation_pos += 12; |
1049 | 11.4k | ws->cur_frame->num_annotations++; |
1050 | 11.4k | |
1051 | 96.3k | for (i = 0; i < num_ins; i++) |
1052 | 84.8k | compile_instruction(vm, ws, ATPOS(vm, a->instructions, i)); |
1053 | 11.4k | } |
1054 | 431 | else if (ISTYPE(vm, node, ws->types->HandlerScope)) { |
1055 | 431 | MAST_HandlerScope *hs = GET_HandlerScope(node); |
1056 | 431 | unsigned int i; |
1057 | 431 | unsigned int num_ins = ELEMS(vm, hs->instructions); |
1058 | 431 | unsigned int start = ws->bytecode_pos - ws->cur_frame->bytecode_start; |
1059 | 431 | unsigned int end; |
1060 | 431 | |
1061 | 2.90k | for (i = 0; i < num_ins; i++) |
1062 | 2.47k | compile_instruction(vm, ws, ATPOS(vm, hs->instructions, i)); |
1063 | 431 | end = ws->bytecode_pos - ws->cur_frame->bytecode_start; |
1064 | 431 | |
1065 | 431 | ws->cur_frame->num_handlers++; |
1066 | 431 | if (ws->cur_frame->handlers) |
1067 | 307 | ws->cur_frame->handlers = (FrameHandler *)MVM_realloc(ws->cur_frame->handlers, |
1068 | 307 | ws->cur_frame->num_handlers * sizeof(FrameHandler)); |
1069 | 431 | else |
1070 | 124 | ws->cur_frame->handlers = (FrameHandler *)MVM_malloc( |
1071 | 124 | ws->cur_frame->num_handlers * sizeof(FrameHandler)); |
1072 | 431 | |
1073 | 431 | i = ws->cur_frame->num_handlers - 1; |
1074 | 431 | ws->cur_frame->handlers[i].start_offset = start; |
1075 | 431 | ws->cur_frame->handlers[i].end_offset = end; |
1076 | 431 | ws->cur_frame->handlers[i].category_mask = (unsigned int)hs->category_mask; |
1077 | 431 | ws->cur_frame->handlers[i].action = (unsigned short)hs->action; |
1078 | 431 | if (ws->cur_frame->handlers[i].category_mask & MVM_EX_CAT_LABELED) { |
1079 | 31 | if (ISTYPE(vm, hs->label_local, ws->types->Local)) { |
1080 | 31 | MAST_Local *l = GET_Local(hs->label_local); |
1081 | 31 | |
1082 | 31 | /* Ensure it's within the set of known locals and an object. */ |
1083 | 31 | if (l->index >= ws->cur_frame->num_locals) { |
1084 | 0 | cleanup_all(vm, ws); |
1085 | 0 | DIE(vm, "MAST::Local index out of range in HandlerScope"); |
1086 | 0 | } |
1087 | 31 | if (ws->cur_frame->local_types[l->index] != MVM_reg_obj) { |
1088 | 0 | cleanup_all(vm, ws); |
1089 | 0 | DIE(vm, "MAST::Local for HandlerScope must be an object"); |
1090 | 0 | } |
1091 | 31 | |
1092 | 31 | /* Stash local index. */ |
1093 | 31 | ws->cur_frame->handlers[i].label_reg = (unsigned short)l->index; |
1094 | 31 | } |
1095 | 0 | else { |
1096 | 0 | cleanup_all(vm, ws); |
1097 | 0 | DIE(vm, "MAST::Local required for HandlerScope with loop label"); |
1098 | 0 | } |
1099 | 31 | } |
1100 | 431 | |
1101 | 431 | /* Ensure we have a label. */ |
1102 | 431 | if (ISTYPE(vm, hs->goto_label, ws->types->Label)) { |
1103 | 431 | ws->cur_frame->handlers[i].label = hs->goto_label; |
1104 | 431 | } |
1105 | 0 | else { |
1106 | 0 | cleanup_all(vm, ws); |
1107 | 0 | DIE(vm, "MAST::Label required for HandlerScope goto"); |
1108 | 0 | } |
1109 | 431 | |
1110 | 431 | /* May need a block also. */ |
1111 | 431 | if (hs->action == HANDLER_INVOKE) { |
1112 | 114 | if (ISTYPE(vm, hs->block_local, ws->types->Local)) { |
1113 | 114 | MAST_Local *l = GET_Local(hs->block_local); |
1114 | 114 | |
1115 | 114 | /* Ensure it's within the set of known locals and an object. */ |
1116 | 114 | if (l->index >= ws->cur_frame->num_locals) { |
1117 | 0 | cleanup_all(vm, ws); |
1118 | 0 | DIE(vm, "MAST::Local index out of range in HandlerScope"); |
1119 | 0 | } |
1120 | 114 | if (ws->cur_frame->local_types[l->index] != MVM_reg_obj) { |
1121 | 0 | cleanup_all(vm, ws); |
1122 | 0 | DIE(vm, "MAST::Local for HandlerScope must be an object"); |
1123 | 0 | } |
1124 | 114 | |
1125 | 114 | /* Stash local index. */ |
1126 | 114 | ws->cur_frame->handlers[i].local = (unsigned short)l->index; |
1127 | 114 | } |
1128 | 0 | else { |
1129 | 0 | cleanup_all(vm, ws); |
1130 | 0 | DIE(vm, "MAST::Local required for HandlerScope invoke action"); |
1131 | 0 | } |
1132 | 114 | } |
1133 | 317 | else if (hs->action == HANDLER_UNWIND_GOTO || hs->action == HANDLER_UNWIND_GOTO_OBJ) { |
1134 | 317 | ws->cur_frame->handlers[i].local = 0; |
1135 | 317 | } |
1136 | 0 | else { |
1137 | 0 | cleanup_all(vm, ws); |
1138 | 0 | DIE(vm, "Invalid action code for handler scope"); |
1139 | 0 | } |
1140 | 431 | } |
1141 | 0 | else { |
1142 | 0 | cleanup_all(vm, ws); |
1143 | 0 | DIE(vm, "Invalid MAST node in instruction list (must be Op, ExtOp, Call, Label, or Annotated)"); |
1144 | 0 | } |
1145 | 274k | ws->current_ins_idx++; |
1146 | 274k | } |
1147 | | |
1148 | | /* Compiles a frame. */ |
1149 | 4.52k | static void compile_frame(VM, WriterState *ws, MASTNode *node, unsigned short idx) { |
1150 | 4.52k | MAST_Frame *f; |
1151 | 4.52k | FrameState *fs; |
1152 | 4.52k | unsigned int i, num_ins, instructions_start; |
1153 | 4.52k | MASTNode *last_inst = NULL; |
1154 | 4.52k | MVMuint16 num_slvs; |
1155 | 4.52k | |
1156 | 4.52k | /* Ensure we have a node of the right type. */ |
1157 | 4.52k | if (!ISTYPE(vm, node, ws->types->Frame)) { |
1158 | 0 | cleanup_all(vm, ws); |
1159 | 0 | DIE(vm, "Child of CompUnit must be a Frame"); |
1160 | 0 | } |
1161 | 4.52k | f = GET_Frame(node); |
1162 | 4.52k | |
1163 | 4.52k | /* Allocate frame state. */ |
1164 | 4.52k | fs = ws->cur_frame = (FrameState *)MVM_malloc(sizeof(FrameState)); |
1165 | 4.52k | fs->bytecode_start = ws->bytecode_pos; |
1166 | 4.52k | fs->frame_start = ws->frame_pos; |
1167 | 4.52k | fs->labels = NULL; |
1168 | 4.52k | fs->num_labels = 0; |
1169 | 4.52k | fs->alloc_labels = 0; |
1170 | 4.52k | fs->unresolved_labels = 0; |
1171 | 4.52k | |
1172 | 4.52k | /* Count locals and lexicals. */ |
1173 | 4.52k | fs->num_locals = ELEMS(vm, f->local_types); |
1174 | 4.52k | fs->num_lexicals = ELEMS(vm, f->lexical_types); |
1175 | 4.52k | |
1176 | 4.52k | if (fs->num_locals > (1 << 16)) { |
1177 | 0 | cleanup_all(vm, ws); |
1178 | 0 | DIE(vm, "Too many locals in this frame."); |
1179 | 0 | } |
1180 | 4.52k | |
1181 | 4.52k | if (ELEMS(vm, f->lexical_names) != fs->num_lexicals) { |
1182 | 0 | cleanup_all(vm, ws); |
1183 | 0 | DIE(vm, "Lexical types list and lexical names list have unequal length"); |
1184 | 0 | } |
1185 | 4.52k | |
1186 | 4.52k | /* initialize number of annotation */ |
1187 | 4.52k | fs->num_annotations = 0; |
1188 | 4.52k | |
1189 | 4.52k | /* initialize number of handlers and handlers pointer */ |
1190 | 4.52k | fs->num_handlers = 0; |
1191 | 4.52k | fs->handlers = NULL; |
1192 | 4.52k | |
1193 | 4.52k | /* Ensure space is available to write frame entry, and write the |
1194 | 4.52k | * header, apart from the bytecode length, which we'll fill in |
1195 | 4.52k | * later. */ |
1196 | 4.52k | ensure_space(vm, &ws->frame_seg, &ws->frame_alloc, ws->frame_pos, |
1197 | 4.52k | FRAME_HEADER_SIZE + fs->num_locals * 2 + fs->num_lexicals * 6); |
1198 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos, fs->bytecode_start); |
1199 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 4, 0); /* Filled in later. */ |
1200 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 8, fs->num_locals); |
1201 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 12, fs->num_lexicals); |
1202 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 16, |
1203 | 4.52k | get_string_heap_index(vm, ws, f->cuuid)); |
1204 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 20, |
1205 | 4.52k | get_string_heap_index(vm, ws, f->name)); |
1206 | 4.52k | |
1207 | 4.52k | /* Handle outer. The current index means "no outer". */ |
1208 | 4.52k | if (ISTYPE(vm, f->outer, ws->types->Frame)) { |
1209 | 1.94k | /* First, see if we have the index cached. If not, go hunting. */ |
1210 | 1.94k | if (((MAST_Frame *)f->outer)->flags & FRAME_FLAG_HAS_INDEX) { |
1211 | 1.94k | write_int16(ws->frame_seg, ws->frame_pos + 24, |
1212 | 1.94k | ((MAST_Frame *)f->outer)->index); |
1213 | 1.94k | } |
1214 | 0 | else { |
1215 | 0 | unsigned short j, found, num_frames; |
1216 | 0 | found = 0; |
1217 | 0 | num_frames = (unsigned short)ELEMS(vm, ws->cu->frames); |
1218 | 0 | for (j = 0; j < num_frames; j++) { |
1219 | 0 | if (ATPOS(vm, ws->cu->frames, j) == f->outer) { |
1220 | 0 | write_int16(ws->frame_seg, ws->frame_pos + 24, j); |
1221 | 0 | found = 1; |
1222 | 0 | break; |
1223 | 0 | } |
1224 | 0 | } |
1225 | 0 | if (!found) { |
1226 | 0 | cleanup_all(vm, ws); |
1227 | 0 | DIE(vm, "Could not locate outer frame in frame list"); |
1228 | 0 | } |
1229 | 0 | } |
1230 | 1.94k | } |
1231 | 2.58k | else { |
1232 | 2.58k | write_int16(ws->frame_seg, ws->frame_pos + 24, idx); |
1233 | 2.58k | } |
1234 | 4.52k | |
1235 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 26, ws->annotation_pos); |
1236 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 30, 0); /* number of annotation; fill in later */ |
1237 | 4.52k | write_int32(ws->frame_seg, ws->frame_pos + 34, 0); /* number of handlers; fill in later */ |
1238 | 4.52k | write_int16(ws->frame_seg, ws->frame_pos + 38, (MVMint16)f->flags); |
1239 | 4.52k | num_slvs = f->flags & FRAME_FLAG_HAS_SLV |
1240 | 415 | ? (MVMuint16)ELEMS(vm, f->static_lex_values) / 4 |
1241 | 4.11k | : 0; |
1242 | 4.52k | write_int16(ws->frame_seg, ws->frame_pos + 40, num_slvs); |
1243 | 4.52k | |
1244 | 4.52k | if (f->flags & FRAME_FLAG_HAS_CODE_OBJ) { |
1245 | 441 | write_int32(ws->frame_seg, ws->frame_pos + 42, f->code_obj_sc_dep_idx + 1); |
1246 | 441 | write_int32(ws->frame_seg, ws->frame_pos + 46, f->code_obj_sc_idx); |
1247 | 441 | } |
1248 | 4.08k | else { |
1249 | 4.08k | write_int32(ws->frame_seg, ws->frame_pos + 42, 0); |
1250 | 4.08k | write_int32(ws->frame_seg, ws->frame_pos + 46, 0); |
1251 | 4.08k | } |
1252 | 4.52k | |
1253 | 4.52k | ws->frame_pos += FRAME_HEADER_SIZE; |
1254 | 4.52k | |
1255 | 4.52k | /* Write locals, as well as collecting our own array of type info. */ |
1256 | 4.52k | fs->local_types = (short unsigned int *)MVM_malloc(sizeof(unsigned short) * fs->num_locals); |
1257 | 62.6k | for (i = 0; i < fs->num_locals; i++) { |
1258 | 58.1k | unsigned short local_type = type_to_local_type(vm, ws, ATPOS(vm, f->local_types, i)); |
1259 | 58.1k | fs->local_types[i] = local_type; |
1260 | 58.1k | write_int16(ws->frame_seg, ws->frame_pos, local_type); |
1261 | 58.1k | ws->frame_pos += 2; |
1262 | 58.1k | } |
1263 | 4.52k | |
1264 | 4.52k | /* Write lexicals. */ |
1265 | 4.52k | fs->lexical_types = (short unsigned int *)MVM_malloc(sizeof(unsigned short) * fs->num_lexicals); |
1266 | 7.76k | for (i = 0; i < fs->num_lexicals; i++) { |
1267 | 3.24k | unsigned short lexical_type = type_to_local_type(vm, ws, ATPOS(vm, f->lexical_types, i)); |
1268 | 3.24k | fs->lexical_types[i] = lexical_type; |
1269 | 3.24k | write_int16(ws->frame_seg, ws->frame_pos, lexical_type); |
1270 | 3.24k | ws->frame_pos += 2; |
1271 | 3.24k | write_int32(ws->frame_seg, ws->frame_pos, |
1272 | 3.24k | get_string_heap_index(vm, ws, ATPOS_S_C(vm, f->lexical_names, i))); |
1273 | 3.24k | ws->frame_pos += 4; |
1274 | 3.24k | } |
1275 | 4.52k | |
1276 | 4.52k | /* Save the location of the start of instructions */ |
1277 | 4.52k | instructions_start = ws->bytecode_pos; |
1278 | 4.52k | |
1279 | 4.52k | /* Compile the instructions. */ |
1280 | 4.52k | ws->current_ins_idx = 0; |
1281 | 4.52k | num_ins = ELEMS(vm, f->instructions); |
1282 | 192k | for (i = 0; i < num_ins; i++) |
1283 | 187k | compile_instruction(vm, ws, last_inst = ATPOS(vm, f->instructions, i)); |
1284 | 4.52k | |
1285 | 4.52k | /* Fixup frames that don't have a return instruction, so |
1286 | 4.52k | * we don't have to check against bytecode length every |
1287 | 4.52k | * time through the runloop. */ |
1288 | 4.52k | if (!last_inst || !ISTYPE(vm, last_inst, ws->types->Op) |
1289 | 4.52k | || ( GET_Op(last_inst)->op != MVM_OP_return |
1290 | 4.19k | && GET_Op(last_inst)->op != MVM_OP_return_i |
1291 | 4.04k | && GET_Op(last_inst)->op != MVM_OP_return_n |
1292 | 4.00k | && GET_Op(last_inst)->op != MVM_OP_return_s |
1293 | 3.83k | && GET_Op(last_inst)->op != MVM_OP_return_o |
1294 | 0 | )) { |
1295 | 0 | ensure_space(vm, &ws->bytecode_seg, &ws->bytecode_alloc, ws->bytecode_pos, 2); |
1296 | 0 | write_int16(ws->bytecode_seg, ws->bytecode_pos, MVM_OP_return); |
1297 | 0 | ws->bytecode_pos += 2; |
1298 | 0 | } |
1299 | 4.52k | |
1300 | 4.52k | /* Fill in bytecode length. */ |
1301 | 4.52k | write_int32(ws->frame_seg, fs->frame_start + 4, ws->bytecode_pos - instructions_start); |
1302 | 4.52k | |
1303 | 4.52k | /* Fill in number of annotations. */ |
1304 | 4.52k | write_int32(ws->frame_seg, fs->frame_start + 30, fs->num_annotations); |
1305 | 4.52k | |
1306 | 4.52k | /* Fill in number of handlers. */ |
1307 | 4.52k | write_int32(ws->frame_seg, fs->frame_start + 34, fs->num_handlers); |
1308 | 4.52k | |
1309 | 4.52k | /* Write handlers. */ |
1310 | 4.95k | for (i = 0; i < fs->num_handlers; i++) { |
1311 | 431 | ensure_space(vm, &ws->frame_seg, &ws->frame_alloc, ws->frame_pos, |
1312 | 431 | FRAME_HANDLER_SIZE); |
1313 | 431 | write_int32(ws->frame_seg, ws->frame_pos, fs->handlers[i].start_offset); |
1314 | 431 | ws->frame_pos += 4; |
1315 | 431 | write_int32(ws->frame_seg, ws->frame_pos, fs->handlers[i].end_offset); |
1316 | 431 | ws->frame_pos += 4; |
1317 | 431 | write_int32(ws->frame_seg, ws->frame_pos, fs->handlers[i].category_mask); |
1318 | 431 | ws->frame_pos += 4; |
1319 | 431 | write_int16(ws->frame_seg, ws->frame_pos, fs->handlers[i].action); |
1320 | 431 | ws->frame_pos += 2; |
1321 | 431 | write_int16(ws->frame_seg, ws->frame_pos, fs->handlers[i].local); |
1322 | 431 | ws->frame_pos += 2; |
1323 | 431 | if (ws->cur_frame->handlers[i].label) |
1324 | 431 | write_int32(ws->frame_seg, ws->frame_pos, |
1325 | 431 | demand_label_offset(vm, ws, GET_Label(fs->handlers[i].label), |
1326 | 431 | "HandlerScope uses unresolved label")); |
1327 | 431 | else |
1328 | 0 | write_int32(ws->frame_seg, ws->frame_pos, 0); |
1329 | 431 | ws->frame_pos += 4; |
1330 | 431 | if (fs->handlers[i].category_mask & MVM_EX_CAT_LABELED) { |
1331 | 31 | ensure_space(vm, &ws->frame_seg, &ws->frame_alloc, ws->frame_pos, 2); |
1332 | 31 | write_int16(ws->frame_seg, ws->frame_pos, fs->handlers[i].label_reg); |
1333 | 31 | ws->frame_pos += 2; |
1334 | 31 | } |
1335 | 431 | } |
1336 | 4.52k | |
1337 | 4.52k | /* Write static lex values. */ |
1338 | 4.52k | ensure_space(vm, &ws->frame_seg, &ws->frame_alloc, ws->frame_pos, |
1339 | 4.52k | FRAME_SLV_SIZE * num_slvs); |
1340 | 5.61k | for (i = 0; i < num_slvs; i++) { |
1341 | 1.08k | write_int16(ws->frame_seg, ws->frame_pos, |
1342 | 1.08k | (MVMuint16)ATPOS_I(vm, f->static_lex_values, 4 * i)); |
1343 | 1.08k | write_int16(ws->frame_seg, ws->frame_pos + 2, |
1344 | 1.08k | (MVMuint16)ATPOS_I(vm, f->static_lex_values, 4 * i + 1)); |
1345 | 1.08k | write_int32(ws->frame_seg, ws->frame_pos + 4, |
1346 | 1.08k | (MVMuint32)ATPOS_I(vm, f->static_lex_values, 4 * i + 2)); |
1347 | 1.08k | write_int32(ws->frame_seg, ws->frame_pos + 8, |
1348 | 1.08k | (MVMuint32)ATPOS_I(vm, f->static_lex_values, 4 * i + 3)); |
1349 | 1.08k | ws->frame_pos += FRAME_SLV_SIZE; |
1350 | 1.08k | } |
1351 | 4.52k | |
1352 | 4.52k | /* Any leftover labels? */ |
1353 | 4.52k | if (fs->unresolved_labels) { |
1354 | 0 | cleanup_all(vm, ws); |
1355 | 0 | DIE(vm, "Frame has %u unresolved labels", fs->unresolved_labels); |
1356 | 0 | } |
1357 | 4.52k | |
1358 | 4.52k | /* Free the frame state. */ |
1359 | 4.52k | cleanup_frame(vm, fs); |
1360 | 4.52k | ws->cur_frame = NULL; |
1361 | 4.52k | |
1362 | 4.52k | /* Increment frame count. */ |
1363 | 4.52k | ws->num_frames++; |
1364 | 4.52k | } |
1365 | | |
1366 | | /* Takes all of the strings and joins them into a heap, encoding them as |
1367 | | * UTF-8. */ |
1368 | 1.21k | static char * form_string_heap(VM, WriterState *ws, unsigned int *string_heap_size) { |
1369 | 1.21k | char *heap; |
1370 | 1.21k | unsigned int i, num_strings, heap_size, heap_alloc; |
1371 | 1.21k | |
1372 | 1.21k | /* If we've nothing to do, just return immediately. */ |
1373 | 1.21k | num_strings = ELEMS(vm, ws->strings); |
1374 | 1.21k | if (num_strings == 0) { |
1375 | 0 | *string_heap_size = 0; |
1376 | 0 | return NULL; |
1377 | 0 | } |
1378 | 1.21k | |
1379 | 1.21k | /* Allocate heap starting point (just a guess). */ |
1380 | 1.21k | heap_size = 0; |
1381 | 1.21k | heap_alloc = num_strings * 32; |
1382 | 1.21k | heap = (char *)MVM_malloc(heap_alloc); |
1383 | 1.21k | |
1384 | 1.21k | /* Add each string to the heap. */ |
1385 | 35.2k | for (i = 0; i < num_strings; i++) { |
1386 | 34.0k | MVMuint64 bytelen; |
1387 | 34.0k | char *encoded; |
1388 | 34.0k | MVMGraphemeIter gi; |
1389 | 34.0k | unsigned short align; |
1390 | 34.0k | unsigned int need; |
1391 | 34.0k | |
1392 | 34.0k | /* Decide if we can get away with Latin-1 with an assumption of the |
1393 | 34.0k | * string already being in NFG. Latin-1 is except \r, which we also |
1394 | 34.0k | * check for here. */ |
1395 | 34.0k | MVMint32 need_utf8 = 0; |
1396 | 34.0k | MVMString *str = ATPOS_S(vm, ws->strings, i); |
1397 | 34.0k | MVM_string_gi_init(tc, &gi, str); |
1398 | 412k | while (MVM_string_gi_has_more(tc, &gi)) { |
1399 | 379k | MVMGrapheme32 g = MVM_string_gi_get_grapheme(tc, &gi); |
1400 | 379k | if (g < 0 || g >= 0xFF || g == 0x0D) { |
1401 | 97 | need_utf8 = 1; |
1402 | 97 | break; |
1403 | 97 | } |
1404 | 379k | } |
1405 | 34.0k | |
1406 | 34.0k | /* Encode it with the chosen algorithm. */ |
1407 | 34.0k | encoded = need_utf8 |
1408 | 97 | ? MVM_string_utf8_encode(tc, str, &bytelen, 0) |
1409 | 33.9k | : MVM_string_latin1_encode(tc, str, &bytelen, 0); |
1410 | 34.0k | if (bytelen > 0x3FFFFFFF) { |
1411 | 0 | cleanup_all(vm, ws); |
1412 | 0 | DIE(vm, "String too long for string constants segment"); |
1413 | 0 | } |
1414 | 34.0k | |
1415 | 34.0k | /* Ensure we have space. */ |
1416 | 24.5k | align = bytelen & 3 ? 4 - (bytelen & 3) : 0; |
1417 | 34.0k | need = 4 + bytelen + align; |
1418 | 34.0k | if (heap_size + need >= heap_alloc) { |
1419 | 0 | heap_alloc = umax(heap_alloc * 2, heap_size + need); |
1420 | 0 | heap = (char *)MVM_realloc(heap, heap_alloc); |
1421 | 0 | } |
1422 | 34.0k | |
1423 | 34.0k | /* Write byte length and UTF-8 flag into heap. */ |
1424 | 34.0k | write_int32(heap, heap_size, (bytelen << 1) | need_utf8); |
1425 | 34.0k | heap_size += 4; |
1426 | 34.0k | |
1427 | 34.0k | /* Write string. */ |
1428 | 34.0k | memcpy(heap + heap_size, encoded, bytelen); |
1429 | 34.0k | MVM_free(encoded); |
1430 | 34.0k | heap_size += bytelen; |
1431 | 34.0k | |
1432 | 34.0k | /* Add alignment. Whilst we never read this memory, it's useful to |
1433 | 34.0k | ensure it is initialised, otherwise valgrind (and similar tools) |
1434 | 34.0k | will rightly complain that we're writing garbage to disk. */ |
1435 | 34.0k | if (align) { |
1436 | 24.5k | memset(heap + heap_size, 0, align); |
1437 | 24.5k | heap_size += align; |
1438 | 24.5k | } |
1439 | 34.0k | } |
1440 | 1.21k | |
1441 | 1.21k | *string_heap_size = heap_size; |
1442 | 1.21k | return heap; |
1443 | 1.21k | } |
1444 | | |
1445 | | /* Takes all the pieces and forms the bytecode output. */ |
1446 | 1.21k | static char * form_bytecode_output(VM, WriterState *ws, unsigned int *bytecode_size) { |
1447 | 1.21k | MVMuint32 size = 0; |
1448 | 1.21k | MVMuint32 pos = 0; |
1449 | 1.21k | char *output; |
1450 | 1.21k | unsigned int string_heap_size; |
1451 | 1.21k | char *string_heap; |
1452 | 1.21k | unsigned int hll_str_idx; |
1453 | 1.21k | |
1454 | 1.21k | /* Store HLL name string, if any. */ |
1455 | 1.21k | if (!VM_STRING_IS_NULL(ws->cu->hll)) |
1456 | 1.11k | hll_str_idx = get_string_heap_index(vm, ws, ws->cu->hll); |
1457 | 1.21k | else |
1458 | 97 | hll_str_idx = get_string_heap_index(vm, ws, EMPTY_STRING(vm)); |
1459 | 1.21k | |
1460 | 1.21k | /* Build string heap. */ |
1461 | 1.21k | string_heap = form_string_heap(vm, ws, &string_heap_size); |
1462 | 1.21k | |
1463 | 1.21k | /* Work out total size. */ |
1464 | 1.21k | size += MVM_ALIGN_SECTION(HEADER_SIZE); |
1465 | 1.21k | size += MVM_ALIGN_SECTION(string_heap_size); |
1466 | 1.21k | size += MVM_ALIGN_SECTION(ws->scdep_bytes); |
1467 | 1.21k | size += MVM_ALIGN_SECTION(ws->extops_bytes); |
1468 | 1.21k | size += MVM_ALIGN_SECTION(ws->frame_pos); |
1469 | 1.21k | size += MVM_ALIGN_SECTION(ws->callsite_pos); |
1470 | 1.21k | size += MVM_ALIGN_SECTION(ws->bytecode_pos); |
1471 | 1.21k | size += MVM_ALIGN_SECTION(ws->annotation_pos); |
1472 | 1.21k | if (vm->serialized) |
1473 | 0 | size += MVM_ALIGN_SECTION(vm->serialized_size); |
1474 | 1.21k | |
1475 | 1.21k | /* Allocate space for the bytecode output. */ |
1476 | 1.21k | output = (char *)MVM_calloc(1, size); |
1477 | 1.21k | |
1478 | 1.21k | /* Generate start of header. */ |
1479 | 1.21k | memcpy(output, "MOARVM\r\n", 8); |
1480 | 1.21k | write_int32(output, 8, BYTECODE_VERSION); |
1481 | 1.21k | pos += MVM_ALIGN_SECTION(HEADER_SIZE); |
1482 | 1.21k | |
1483 | 1.21k | /* Add SC dependencies section and its header entries. */ |
1484 | 1.21k | write_int32(output, SCDEP_HEADER_OFFSET, pos); |
1485 | 1.21k | write_int32(output, SCDEP_HEADER_OFFSET + 4, ELEMS(vm, ws->cu->sc_handles)); |
1486 | 1.21k | memcpy(output + pos, ws->scdep_seg, ws->scdep_bytes); |
1487 | 1.21k | pos += MVM_ALIGN_SECTION(ws->scdep_bytes); |
1488 | 1.21k | |
1489 | 1.21k | /* Add extension ops section and its header entries. */ |
1490 | 1.21k | write_int32(output, EXTOP_HEADER_OFFSET, pos); |
1491 | 1.21k | write_int32(output, EXTOP_HEADER_OFFSET + 4, ws->num_extops); |
1492 | 1.21k | memcpy(output + pos, ws->extops_seg, ws->extops_bytes); |
1493 | 1.21k | pos += MVM_ALIGN_SECTION(ws->extops_bytes); |
1494 | 1.21k | |
1495 | 1.21k | /* Add frames section and its header entries. */ |
1496 | 1.21k | write_int32(output, FRAME_HEADER_OFFSET, pos); |
1497 | 1.21k | write_int32(output, FRAME_HEADER_OFFSET + 4, ws->num_frames); |
1498 | 1.21k | memcpy(output + pos, ws->frame_seg, ws->frame_pos); |
1499 | 1.21k | pos += MVM_ALIGN_SECTION(ws->frame_pos); |
1500 | 1.21k | |
1501 | 1.21k | /* Add callsites section and its header entries. */ |
1502 | 1.21k | write_int32(output, CALLSITE_HEADER_OFFSET, pos); |
1503 | 1.21k | write_int32(output, CALLSITE_HEADER_OFFSET + 4, ws->num_callsites); |
1504 | 1.21k | memcpy(output + pos, ws->callsite_seg, ws->callsite_pos); |
1505 | 1.21k | pos += MVM_ALIGN_SECTION(ws->callsite_pos); |
1506 | 1.21k | |
1507 | 1.21k | /* Add strings heap section and its header entries. */ |
1508 | 1.21k | write_int32(output, STRING_HEADER_OFFSET, pos); |
1509 | 1.21k | write_int32(output, STRING_HEADER_OFFSET + 4, ELEMS(vm, ws->strings)); |
1510 | 1.21k | memcpy(output + pos, string_heap, string_heap_size); |
1511 | 1.21k | pos += MVM_ALIGN_SECTION(string_heap_size); |
1512 | 1.21k | if (string_heap) { |
1513 | 1.21k | MVM_free(string_heap); |
1514 | 1.21k | string_heap = NULL; |
1515 | 1.21k | } |
1516 | 1.21k | |
1517 | 1.21k | /* SC data. Write it if we have it. */ |
1518 | 1.21k | if (vm->serialized) { |
1519 | 0 | write_int32(output, SCDATA_HEADER_OFFSET, pos); |
1520 | 0 | write_int32(output, SCDATA_HEADER_OFFSET + 4, vm->serialized_size); |
1521 | 0 | memcpy(output + pos, vm->serialized, vm->serialized_size); |
1522 | 0 | pos += MVM_ALIGN_SECTION(vm->serialized_size); |
1523 | 0 | MVM_free(vm->serialized); |
1524 | 0 | vm->serialized = NULL; |
1525 | 0 | vm->serialized_size = 0; |
1526 | 0 | } |
1527 | 1.21k | |
1528 | 1.21k | /* Add bytecode section and its header entries (offset, length). */ |
1529 | 1.21k | write_int32(output, BYTECODE_HEADER_OFFSET, pos); |
1530 | 1.21k | write_int32(output, BYTECODE_HEADER_OFFSET + 4, ws->bytecode_pos); |
1531 | 1.21k | memcpy(output + pos, ws->bytecode_seg, ws->bytecode_pos); |
1532 | 1.21k | pos += MVM_ALIGN_SECTION(ws->bytecode_pos); |
1533 | 1.21k | |
1534 | 1.21k | /* Add annotation section and its header entries (offset, length). */ |
1535 | 1.21k | write_int32(output, ANNOTATION_HEADER_OFFSET, pos); |
1536 | 1.21k | write_int32(output, ANNOTATION_HEADER_OFFSET + 4, ws->annotation_pos); |
1537 | 1.21k | memcpy(output + pos, ws->annotation_seg, ws->annotation_pos); |
1538 | 1.21k | pos += MVM_ALIGN_SECTION(ws->annotation_pos); |
1539 | 1.21k | |
1540 | 1.21k | /* Add HLL and special frame indexes. */ |
1541 | 1.21k | write_int32(output, HLL_NAME_HEADER_OFFSET, hll_str_idx); |
1542 | 1.21k | if (VM_OBJ_IS_NULL(ws->cu->main_frame)) |
1543 | 1.07k | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET, 0); |
1544 | 1.21k | else |
1545 | 143 | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET, 1 + get_frame_index(vm, ws, ws->cu->main_frame)); |
1546 | 1.21k | if (VM_OBJ_IS_NULL(ws->cu->load_frame)) |
1547 | 1.07k | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET + 4, 0); |
1548 | 1.21k | else |
1549 | 143 | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET + 4, 1 + get_frame_index(vm, ws, ws->cu->load_frame)); |
1550 | 1.21k | if (VM_OBJ_IS_NULL(ws->cu->deserialize_frame)) |
1551 | 130 | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET + 8, 0); |
1552 | 1.21k | else |
1553 | 1.08k | write_int32(output, SPECIAL_FRAME_HEADER_OFFSET + 8, 1 + get_frame_index(vm, ws, ws->cu->deserialize_frame)); |
1554 | 1.21k | |
1555 | 1.21k | /* Sanity...should never fail. */ |
1556 | 1.21k | if (pos != size) |
1557 | 0 | DIE(vm, "Bytecode generated did not match expected size"); |
1558 | 1.21k | |
1559 | 1.21k | *bytecode_size = size; |
1560 | 1.21k | return output; |
1561 | 1.21k | } |
1562 | | |
1563 | | /* Main entry point to the MAST to bytecode compiler. */ |
1564 | 1.21k | char * MVM_mast_compile(VM, MASTNode *node, MASTNodeTypes *types, unsigned int *size) { |
1565 | 1.21k | MAST_CompUnit *cu; |
1566 | 1.21k | WriterState *ws; |
1567 | 1.21k | char *bytecode; |
1568 | 1.21k | unsigned short i, num_depscs, num_frames; |
1569 | 1.21k | unsigned int bytecode_size; |
1570 | 1.21k | |
1571 | 1.21k | /* Ensure we have a compilation unit. */ |
1572 | 1.21k | if (!ISTYPE(vm, node, types->CompUnit)) |
1573 | 0 | DIE(vm, "Top-level MAST node must be a CompUnit"); |
1574 | 1.21k | cu = GET_CompUnit(node); |
1575 | 1.21k | |
1576 | 1.21k | /* Initialize the writer state structure. */ |
1577 | 1.21k | ws = (WriterState *)MVM_malloc(sizeof(WriterState)); |
1578 | 1.21k | ws->types = types; |
1579 | 1.21k | ws->strings = NEWLIST_S(vm); |
1580 | 1.21k | ws->seen_strings = NEWHASH(vm); |
1581 | 1.21k | ws->cur_frame = NULL; |
1582 | 1.21k | ws->scdep_bytes = ELEMS(vm, cu->sc_handles) * SC_DEP_SIZE; |
1583 | 1.12k | ws->scdep_seg = ws->scdep_bytes ? (char *)MVM_malloc(ws->scdep_bytes) : NULL; |
1584 | 1.21k | ws->num_extops = ELEMS(vm, cu->extop_names); |
1585 | 1.21k | ws->extops_bytes = ws->num_extops * EXTOP_SIZE; |
1586 | 1.21k | ws->extops_seg = (char *)MVM_malloc(ws->extops_bytes); |
1587 | 1.21k | ws->frame_pos = 0; |
1588 | 1.21k | ws->frame_alloc = 192 * ELEMS(vm, cu->frames); |
1589 | 1.21k | ws->frame_seg = (char *)MVM_malloc(ws->frame_alloc); |
1590 | 1.21k | ws->num_frames = 0; |
1591 | 1.21k | ws->callsite_pos = 0; |
1592 | 1.21k | ws->callsite_alloc = 4096; |
1593 | 1.21k | ws->callsite_seg = (char *)MVM_malloc(ws->callsite_alloc); |
1594 | 1.21k | ws->num_callsites = 0; |
1595 | 1.21k | ws->bytecode_pos = 0; |
1596 | 1.21k | ws->bytecode_alloc = 128 * ELEMS(vm, cu->frames); |
1597 | 1.21k | ws->bytecode_seg = (char *)MVM_malloc(ws->bytecode_alloc); |
1598 | 1.21k | ws->annotation_pos = 0; |
1599 | 1.21k | ws->annotation_alloc = 64 * ELEMS(vm, cu->frames); |
1600 | 1.21k | ws->annotation_seg = (char *)MVM_malloc(ws->annotation_alloc); |
1601 | 1.21k | ws->cu = cu; |
1602 | 1.21k | ws->current_frame_idx= 0; |
1603 | 1.21k | |
1604 | 1.21k | /* If we have any strings from serializing, then we'll seed our own string |
1605 | 1.21k | * heap with them. This means the compilation unit string heap will align |
1606 | 1.21k | * perfectly with what the serialization blob needs, and thus we can use |
1607 | 1.21k | * it in deserialization. Note we use get_string_heap_index for its side |
1608 | 1.21k | * effects only here. Start from 1, as 0 means NULL string. */ |
1609 | 1.21k | if (vm->serialized_string_heap) { |
1610 | 0 | MVMint64 elems = ELEMS(vm, vm->serialized_string_heap); |
1611 | 0 | for (i = 1; i < elems; i++) |
1612 | 0 | (void)get_string_heap_index(vm, ws, ATPOS_S(vm, vm->serialized_string_heap, i)); |
1613 | 0 | vm->serialized_string_heap = NULL; |
1614 | 0 | } |
1615 | 1.21k | |
1616 | 1.21k | /* Initialize callsite reuse cache */ |
1617 | 1.21k | ws->callsite_reuse_head = NULL; |
1618 | 1.21k | |
1619 | 1.21k | /* Store each of the dependent SCs. */ |
1620 | 1.21k | num_depscs = ELEMS(vm, ws->cu->sc_handles); |
1621 | 3.41k | for (i = 0; i < num_depscs; i++) |
1622 | 2.20k | write_int32(ws->scdep_seg, i * SC_DEP_SIZE, |
1623 | 2.20k | get_string_heap_index(vm, ws, |
1624 | 2.20k | ATPOS_S_C(vm, ws->cu->sc_handles, i))); |
1625 | 1.21k | |
1626 | 1.21k | /* Store each of the extop names and signatures. */ |
1627 | 1.21k | for (i = 0; i < ws->num_extops; i++) { |
1628 | 0 | MASTNode *sig_array; |
1629 | 0 | int num_operands, j; |
1630 | 0 |
|
1631 | 0 | write_int32(ws->extops_seg, i * EXTOP_SIZE, |
1632 | 0 | get_string_heap_index(vm, ws, |
1633 | 0 | ATPOS_S_C(vm, ws->cu->extop_names, i))); |
1634 | 0 |
|
1635 | 0 | sig_array = ATPOS(vm, ws->cu->extop_sigs, i); |
1636 | 0 | num_operands = ELEMS(vm, sig_array); |
1637 | 0 | for (j = 0; j < 8; j++) |
1638 | 0 | write_int8(ws->extops_seg, i * EXTOP_SIZE + 4 + j, |
1639 | 0 | j < num_operands |
1640 | 0 | ? ATPOS_I(vm, sig_array, j) |
1641 | 0 | : 0); |
1642 | 0 | } |
1643 | 1.21k | |
1644 | 1.21k | /* Visit and compile each of the frames. */ |
1645 | 1.21k | num_frames = (unsigned short)ELEMS(vm, cu->frames); |
1646 | 5.74k | for (i = 0; i < num_frames; i++) |
1647 | 4.52k | compile_frame(vm, ws, ATPOS(vm, cu->frames, i), ws->current_frame_idx = i); |
1648 | 1.21k | |
1649 | 1.21k | /* Join all the pieces into a bytecode file. */ |
1650 | 1.21k | bytecode = form_bytecode_output(vm, ws, &bytecode_size); |
1651 | 1.21k | |
1652 | 1.21k | /* Cleanup and hand back result. */ |
1653 | 1.21k | cleanup_all(vm, ws); |
1654 | 1.21k | |
1655 | 1.21k | *size = bytecode_size; |
1656 | 1.21k | return bytecode; |
1657 | 1.21k | } |