1 /* dfltcc_deflate.c - IBM Z DEFLATE CONVERSION CALL compression support. */
2
3 /*
4 Use the following commands to build zlib-ng with DFLTCC compression support:
5
6 $ ./configure --with-dfltcc-deflate
7 or
8
9 $ cmake -DWITH_DFLTCC_DEFLATE=1 .
10
11 and then
12
13 $ make
14 */
15
16 #include "../../zbuild.h"
17 #include "../../zutil.h"
18 #include "../../deflate.h"
19 #include "../../trees_emit.h"
20 #include "dfltcc_deflate.h"
21 #include "dfltcc_detail.h"
22
dfltcc_can_deflate_with_params(PREFIX3 (streamp)strm,int level,uInt window_bits,int strategy,int reproducible)23 static inline int dfltcc_can_deflate_with_params(PREFIX3(streamp) strm, int level, uInt window_bits, int strategy,
24 int reproducible) {
25 deflate_state *state = (deflate_state *)strm->state;
26 struct dfltcc_state *dfltcc_state = GET_DFLTCC_STATE(state);
27
28 /* Unsupported compression settings */
29 if ((dfltcc_state->level_mask & (1 << level)) == 0)
30 return 0;
31 if (window_bits != HB_BITS)
32 return 0;
33 if (strategy != Z_FIXED && strategy != Z_DEFAULT_STRATEGY)
34 return 0;
35 if (reproducible)
36 return 0;
37
38 /* Unsupported hardware */
39 if (!is_bit_set(dfltcc_state->af.fns, DFLTCC_GDHT) ||
40 !is_bit_set(dfltcc_state->af.fns, DFLTCC_CMPR) ||
41 !is_bit_set(dfltcc_state->af.fmts, DFLTCC_FMT0))
42 return 0;
43
44 return 1;
45 }
46
dfltcc_can_deflate(PREFIX3 (streamp)strm)47 int Z_INTERNAL dfltcc_can_deflate(PREFIX3(streamp) strm) {
48 deflate_state *state = (deflate_state *)strm->state;
49
50 return dfltcc_can_deflate_with_params(strm, state->level, state->w_bits, state->strategy, state->reproducible);
51 }
52
dfltcc_gdht(PREFIX3 (streamp)strm)53 static inline void dfltcc_gdht(PREFIX3(streamp) strm) {
54 deflate_state *state = (deflate_state *)strm->state;
55 struct dfltcc_param_v0 *param = &GET_DFLTCC_STATE(state)->param;
56 size_t avail_in = strm->avail_in;
57
58 dfltcc(DFLTCC_GDHT, param, NULL, NULL, &strm->next_in, &avail_in, NULL);
59 }
60
dfltcc_cmpr(PREFIX3 (streamp)strm)61 static inline dfltcc_cc dfltcc_cmpr(PREFIX3(streamp) strm) {
62 deflate_state *state = (deflate_state *)strm->state;
63 struct dfltcc_param_v0 *param = &GET_DFLTCC_STATE(state)->param;
64 size_t avail_in = strm->avail_in;
65 size_t avail_out = strm->avail_out;
66 dfltcc_cc cc;
67
68 cc = dfltcc(DFLTCC_CMPR | HBT_CIRCULAR,
69 param, &strm->next_out, &avail_out,
70 &strm->next_in, &avail_in, state->window);
71 strm->total_in += (strm->avail_in - avail_in);
72 strm->total_out += (strm->avail_out - avail_out);
73 strm->avail_in = avail_in;
74 strm->avail_out = avail_out;
75 return cc;
76 }
77
send_eobs(PREFIX3 (streamp)strm,const struct dfltcc_param_v0 * param)78 static inline void send_eobs(PREFIX3(streamp) strm, const struct dfltcc_param_v0 *param) {
79 deflate_state *state = (deflate_state *)strm->state;
80
81 send_bits(state, bi_reverse(param->eobs >> (15 - param->eobl), param->eobl), param->eobl, state->bi_buf, state->bi_valid);
82 flush_pending(strm);
83 if (state->pending != 0) {
84 /* The remaining data is located in pending_out[0:pending]. If someone
85 * calls put_byte() - this might happen in deflate() - the byte will be
86 * placed into pending_buf[pending], which is incorrect. Move the
87 * remaining data to the beginning of pending_buf so that put_byte() is
88 * usable again.
89 */
90 memmove(state->pending_buf, state->pending_out, state->pending);
91 state->pending_out = state->pending_buf;
92 }
93 #ifdef ZLIB_DEBUG
94 state->compressed_len += param->eobl;
95 #endif
96 }
97
dfltcc_deflate(PREFIX3 (streamp)strm,int flush,block_state * result)98 int Z_INTERNAL dfltcc_deflate(PREFIX3(streamp) strm, int flush, block_state *result) {
99 deflate_state *state = (deflate_state *)strm->state;
100 struct dfltcc_state *dfltcc_state = GET_DFLTCC_STATE(state);
101 struct dfltcc_param_v0 *param = &dfltcc_state->param;
102 uInt masked_avail_in;
103 dfltcc_cc cc;
104 int need_empty_block;
105 int soft_bcc;
106 int no_flush;
107
108 if (!dfltcc_can_deflate(strm)) {
109 /* Clear history. */
110 if (flush == Z_FULL_FLUSH)
111 param->hl = 0;
112 return 0;
113 }
114
115 again:
116 masked_avail_in = 0;
117 soft_bcc = 0;
118 no_flush = flush == Z_NO_FLUSH;
119
120 /* No input data. Return, except when Continuation Flag is set, which means
121 * that DFLTCC has buffered some output in the parameter block and needs to
122 * be called again in order to flush it.
123 */
124 if (strm->avail_in == 0 && !param->cf) {
125 /* A block is still open, and the hardware does not support closing
126 * blocks without adding data. Thus, close it manually.
127 */
128 if (!no_flush && param->bcf) {
129 send_eobs(strm, param);
130 param->bcf = 0;
131 }
132 /* Let one of deflate_* functions write a trailing empty block. */
133 if (flush == Z_FINISH)
134 return 0;
135 /* Clear history. */
136 if (flush == Z_FULL_FLUSH)
137 param->hl = 0;
138 /* Trigger block post-processing if necessary. */
139 *result = no_flush ? need_more : block_done;
140 return 1;
141 }
142
143 /* There is an open non-BFINAL block, we are not going to close it just
144 * yet, we have compressed more than DFLTCC_BLOCK_SIZE bytes and we see
145 * more than DFLTCC_DHT_MIN_SAMPLE_SIZE bytes. Open a new block with a new
146 * DHT in order to adapt to a possibly changed input data distribution.
147 */
148 if (param->bcf && no_flush &&
149 strm->total_in > dfltcc_state->block_threshold &&
150 strm->avail_in >= dfltcc_state->dht_threshold) {
151 if (param->cf) {
152 /* We need to flush the DFLTCC buffer before writing the
153 * End-of-block Symbol. Mask the input data and proceed as usual.
154 */
155 masked_avail_in += strm->avail_in;
156 strm->avail_in = 0;
157 no_flush = 0;
158 } else {
159 /* DFLTCC buffer is empty, so we can manually write the
160 * End-of-block Symbol right away.
161 */
162 send_eobs(strm, param);
163 param->bcf = 0;
164 dfltcc_state->block_threshold = strm->total_in + dfltcc_state->block_size;
165 }
166 }
167
168 /* No space for compressed data. If we proceed, dfltcc_cmpr() will return
169 * DFLTCC_CC_OP1_TOO_SHORT without buffering header bits, but we will still
170 * set BCF=1, which is wrong. Avoid complications and return early.
171 */
172 if (strm->avail_out == 0) {
173 *result = need_more;
174 return 1;
175 }
176
177 /* The caller gave us too much data. Pass only one block worth of
178 * uncompressed data to DFLTCC and mask the rest, so that on the next
179 * iteration we start a new block.
180 */
181 if (no_flush && strm->avail_in > dfltcc_state->block_size) {
182 masked_avail_in += (strm->avail_in - dfltcc_state->block_size);
183 strm->avail_in = dfltcc_state->block_size;
184 }
185
186 /* When we have an open non-BFINAL deflate block and caller indicates that
187 * the stream is ending, we need to close an open deflate block and open a
188 * BFINAL one.
189 */
190 need_empty_block = flush == Z_FINISH && param->bcf && !param->bhf;
191
192 /* Translate stream to parameter block */
193 param->cvt = state->wrap == 2 ? CVT_CRC32 : CVT_ADLER32;
194 if (!no_flush)
195 /* We need to close a block. Always do this in software - when there is
196 * no input data, the hardware will not honor BCC. */
197 soft_bcc = 1;
198 if (flush == Z_FINISH && !param->bcf)
199 /* We are about to open a BFINAL block, set Block Header Final bit
200 * until the stream ends.
201 */
202 param->bhf = 1;
203 /* DFLTCC-CMPR will write to next_out, so make sure that buffers with
204 * higher precedence are empty.
205 */
206 Assert(state->pending == 0, "There must be no pending bytes");
207 Assert(state->bi_valid < 8, "There must be less than 8 pending bits");
208 param->sbb = (unsigned int)state->bi_valid;
209 if (param->sbb > 0)
210 *strm->next_out = (unsigned char)state->bi_buf;
211 /* Honor history and check value */
212 param->nt = 0;
213 param->cv = state->wrap == 2 ? ZSWAP32(strm->adler) : strm->adler;
214
215 /* When opening a block, choose a Huffman-Table Type */
216 if (!param->bcf) {
217 if (state->strategy == Z_FIXED || (strm->total_in == 0 && dfltcc_state->block_threshold > 0))
218 param->htt = HTT_FIXED;
219 else {
220 param->htt = HTT_DYNAMIC;
221 dfltcc_gdht(strm);
222 }
223 }
224
225 /* Deflate */
226 do {
227 cc = dfltcc_cmpr(strm);
228 if (strm->avail_in < 4096 && masked_avail_in > 0)
229 /* We are about to call DFLTCC with a small input buffer, which is
230 * inefficient. Since there is masked data, there will be at least
231 * one more DFLTCC call, so skip the current one and make the next
232 * one handle more data.
233 */
234 break;
235 } while (cc == DFLTCC_CC_AGAIN);
236
237 /* Translate parameter block to stream */
238 strm->msg = oesc_msg(dfltcc_state->msg, param->oesc);
239 state->bi_valid = param->sbb;
240 if (state->bi_valid == 0)
241 state->bi_buf = 0; /* Avoid accessing next_out */
242 else
243 state->bi_buf = *strm->next_out & ((1 << state->bi_valid) - 1);
244 strm->adler = state->wrap == 2 ? ZSWAP32(param->cv) : param->cv;
245
246 /* Unmask the input data */
247 strm->avail_in += masked_avail_in;
248 masked_avail_in = 0;
249
250 /* If we encounter an error, it means there is a bug in DFLTCC call */
251 Assert(cc != DFLTCC_CC_OP2_CORRUPT || param->oesc == 0, "BUG");
252
253 /* Update Block-Continuation Flag. It will be used to check whether to call
254 * GDHT the next time.
255 */
256 if (cc == DFLTCC_CC_OK) {
257 if (soft_bcc) {
258 send_eobs(strm, param);
259 param->bcf = 0;
260 dfltcc_state->block_threshold = strm->total_in + dfltcc_state->block_size;
261 } else
262 param->bcf = 1;
263 if (flush == Z_FINISH) {
264 if (need_empty_block)
265 /* Make the current deflate() call also close the stream */
266 return 0;
267 else {
268 bi_windup(state);
269 *result = finish_done;
270 }
271 } else {
272 if (flush == Z_FULL_FLUSH)
273 param->hl = 0; /* Clear history */
274 *result = flush == Z_NO_FLUSH ? need_more : block_done;
275 }
276 } else {
277 param->bcf = 1;
278 *result = need_more;
279 }
280 if (strm->avail_in != 0 && strm->avail_out != 0)
281 goto again; /* deflate() must use all input or all output */
282 return 1;
283 }
284
285 /*
286 Switching between hardware and software compression.
287
288 DFLTCC does not support all zlib settings, e.g. generation of non-compressed
289 blocks or alternative window sizes. When such settings are applied on the
290 fly with deflateParams, we need to convert between hardware and software
291 window formats.
292 */
dfltcc_was_deflate_used(PREFIX3 (streamp)strm)293 static int dfltcc_was_deflate_used(PREFIX3(streamp) strm) {
294 deflate_state *state = (deflate_state *)strm->state;
295 struct dfltcc_param_v0 *param = &GET_DFLTCC_STATE(state)->param;
296
297 return strm->total_in > 0 || param->nt == 0 || param->hl > 0;
298 }
299
dfltcc_deflate_params(PREFIX3 (streamp)strm,int level,int strategy,int * flush)300 int Z_INTERNAL dfltcc_deflate_params(PREFIX3(streamp) strm, int level, int strategy, int *flush) {
301 deflate_state *state = (deflate_state *)strm->state;
302 int could_deflate = dfltcc_can_deflate(strm);
303 int can_deflate = dfltcc_can_deflate_with_params(strm, level, state->w_bits, strategy, state->reproducible);
304
305 if (can_deflate == could_deflate)
306 /* We continue to work in the same mode - no changes needed */
307 return Z_OK;
308
309 if (!dfltcc_was_deflate_used(strm))
310 /* DFLTCC was not used yet - no changes needed */
311 return Z_OK;
312
313 /* For now, do not convert between window formats - simply get rid of the old data instead */
314 *flush = Z_FULL_FLUSH;
315 return Z_OK;
316 }
317
dfltcc_deflate_done(PREFIX3 (streamp)strm,int flush)318 int Z_INTERNAL dfltcc_deflate_done(PREFIX3(streamp) strm, int flush) {
319 deflate_state *state = (deflate_state *)strm->state;
320 struct dfltcc_state *dfltcc_state = GET_DFLTCC_STATE(state);
321 struct dfltcc_param_v0 *param = &dfltcc_state->param;
322
323 /* When deflate(Z_FULL_FLUSH) is called with small avail_out, it might
324 * close the block without resetting the compression state. Detect this
325 * situation and return that deflation is not done.
326 */
327 if (flush == Z_FULL_FLUSH && strm->avail_out == 0)
328 return 0;
329
330 /* Return that deflation is not done if DFLTCC is used and either it
331 * buffered some data (Continuation Flag is set), or has not written EOBS
332 * yet (Block-Continuation Flag is set).
333 */
334 return !dfltcc_can_deflate(strm) || (!param->cf && !param->bcf);
335 }
336
dfltcc_can_set_reproducible(PREFIX3 (streamp)strm,int reproducible)337 int Z_INTERNAL dfltcc_can_set_reproducible(PREFIX3(streamp) strm, int reproducible) {
338 deflate_state *state = (deflate_state *)strm->state;
339
340 return reproducible != state->reproducible && !dfltcc_was_deflate_used(strm);
341 }
342
343 /*
344 Preloading history.
345 */
append_history(struct dfltcc_param_v0 * param,unsigned char * history,const unsigned char * buf,uInt count)346 static void append_history(struct dfltcc_param_v0 *param, unsigned char *history, const unsigned char *buf, uInt count) {
347 size_t offset;
348 size_t n;
349
350 /* Do not use more than 32K */
351 if (count > HB_SIZE) {
352 buf += count - HB_SIZE;
353 count = HB_SIZE;
354 }
355 offset = (param->ho + param->hl) % HB_SIZE;
356 if (offset + count <= HB_SIZE)
357 /* Circular history buffer does not wrap - copy one chunk */
358 memcpy(history + offset, buf, count);
359 else {
360 /* Circular history buffer wraps - copy two chunks */
361 n = HB_SIZE - offset;
362 memcpy(history + offset, buf, n);
363 memcpy(history, buf + n, count - n);
364 }
365 n = param->hl + count;
366 if (n <= HB_SIZE)
367 /* All history fits into buffer - no need to discard anything */
368 param->hl = n;
369 else {
370 /* History does not fit into buffer - discard extra bytes */
371 param->ho = (param->ho + (n - HB_SIZE)) % HB_SIZE;
372 param->hl = HB_SIZE;
373 }
374 }
375
dfltcc_deflate_set_dictionary(PREFIX3 (streamp)strm,const unsigned char * dictionary,uInt dict_length)376 int Z_INTERNAL dfltcc_deflate_set_dictionary(PREFIX3(streamp) strm,
377 const unsigned char *dictionary, uInt dict_length) {
378 deflate_state *state = (deflate_state *)strm->state;
379 struct dfltcc_state *dfltcc_state = GET_DFLTCC_STATE(state);
380 struct dfltcc_param_v0 *param = &dfltcc_state->param;
381
382 append_history(param, state->window, dictionary, dict_length);
383 state->strstart = 1; /* Add FDICT to zlib header */
384 state->block_start = state->strstart; /* Make deflate_stored happy */
385 return Z_OK;
386 }
387
dfltcc_deflate_get_dictionary(PREFIX3 (streamp)strm,unsigned char * dictionary,uInt * dict_length)388 int Z_INTERNAL dfltcc_deflate_get_dictionary(PREFIX3(streamp) strm, unsigned char *dictionary, uInt *dict_length) {
389 deflate_state *state = (deflate_state *)strm->state;
390 struct dfltcc_state *dfltcc_state = GET_DFLTCC_STATE(state);
391 struct dfltcc_param_v0 *param = &dfltcc_state->param;
392
393 if (dictionary) {
394 if (param->ho + param->hl <= HB_SIZE)
395 /* Circular history buffer does not wrap - copy one chunk */
396 memcpy(dictionary, state->window + param->ho, param->hl);
397 else {
398 /* Circular history buffer wraps - copy two chunks */
399 memcpy(dictionary, state->window + param->ho, HB_SIZE - param->ho);
400 memcpy(dictionary + HB_SIZE - param->ho, state->window, param->ho + param->hl - HB_SIZE);
401 }
402 }
403 if (dict_length)
404 *dict_length = param->hl;
405 return Z_OK;
406 }
407