1 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2 Alan Hayward <alan.hayward@arm.com>
3 David Sherwood <david.sherwood@arm.com>
5 * optabs.def (fold_left_plus_optab): New optab.
6 * doc/md.texi (fold_left_plus_@var{m}): Document.
7 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
8 * internal-fn.c (fold_left_direct): Define.
9 (expand_fold_left_optab_fn): Likewise.
10 (direct_fold_left_optab_supported_p): Likewise.
11 * fold-const-call.c (fold_const_fold_left): New function.
12 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
13 * tree-parloops.c (valid_reduction_p): New function.
14 (gather_scalar_reductions): Use it.
15 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
16 (vect_finish_replace_stmt): Declare.
17 * tree-vect-loop.c (fold_left_reduction_fn): New function.
18 (needs_fold_left_reduction_p): New function, split out from...
19 (vect_is_simple_reduction): ...here. Accept reductions that
20 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
21 (vect_force_simple_reduction): Also store the reduction type in
22 the assignment's STMT_VINFO_REDUC_TYPE.
23 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
24 (merge_with_identity): New function.
25 (vect_expand_fold_left): Likewise.
26 (vectorize_fold_left_reduction): Likewise.
27 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
28 scalar phi in place for it. Check for target support and reject
29 cases that would reassociate the operation. Defer the transform
30 phase to vectorize_fold_left_reduction.
31 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
32 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
33 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
35 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
37 * tree-if-conv.c (predicate_mem_writes): Remove redundant
40 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
41 Alan Hayward <alan.hayward@arm.com>
42 David Sherwood <david.sherwood@arm.com>
44 * target.def (legitimize_address_displacement): Take the original
46 * targhooks.h (default_legitimize_address_displacement): Update
48 * targhooks.c (default_legitimize_address_displacement): Likewise.
49 * doc/tm.texi: Regenerate.
50 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
51 as an argument, moving assert of ad->disp == ad->disp_term to...
52 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
53 Try calling targetm.legitimize_address_displacement before expanding
54 the address rather than afterwards, and adjust for the new interface.
55 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
56 Match the new hook interface. Handle SVE addresses.
57 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
60 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
62 * Makefile.in (OBJS): Add early-remat.o.
63 * target.def (select_early_remat_modes): New hook.
64 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
65 * doc/tm.texi: Regenerate.
66 * targhooks.h (default_select_early_remat_modes): Declare.
67 * targhooks.c (default_select_early_remat_modes): New function.
68 * timevar.def (TV_EARLY_REMAT): New timevar.
69 * passes.def (pass_early_remat): New pass.
70 * tree-pass.h (make_pass_early_remat): Declare.
71 * early-remat.c: New file.
72 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
74 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
76 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
77 Alan Hayward <alan.hayward@arm.com>
78 David Sherwood <david.sherwood@arm.com>
80 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
81 vfm1 with a bound_epilog parameter.
82 (vect_do_peeling): Update calls accordingly, and move the prologue
83 call earlier in the function. Treat the base bound_epilog as 0 for
84 fully-masked loops and retain vf - 1 for other loops. Add 1 to
85 this base when peeling for gaps.
86 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
87 with fully-masked loops.
88 (vect_estimate_min_profitable_iters): Handle the single peeled
89 iteration in that case.
91 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
92 Alan Hayward <alan.hayward@arm.com>
93 David Sherwood <david.sherwood@arm.com>
95 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
96 single-element interleaving even if the size is not a power of 2.
97 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
98 accesses for single-element interleaving if the group size is
101 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
102 Alan Hayward <alan.hayward@arm.com>
103 David Sherwood <david.sherwood@arm.com>
105 * doc/md.texi (fold_extract_last_@var{m}): Document.
106 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
107 * optabs.def (fold_extract_last_optab): New optab.
108 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
109 * internal-fn.c (fold_extract_direct): New macro.
110 (expand_fold_extract_optab_fn): Likewise.
111 (direct_fold_extract_optab_supported_p): Likewise.
112 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
113 * tree-vect-loop.c (vect_model_reduction_cost): Handle
114 EXTRACT_LAST_REDUCTION.
115 (get_initial_def_for_reduction): Do not create an initial vector
116 for EXTRACT_LAST_REDUCTION reductions.
117 (vectorizable_reduction): Leave the scalar phi in place for
118 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
119 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
120 epilogue code for EXTRACT_LAST_REDUCTION and defer the
121 transform phase to vectorizable_condition.
122 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
124 (vect_finish_stmt_generation): ...here.
125 (vect_finish_replace_stmt): New function.
126 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
127 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
129 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
131 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
132 Alan Hayward <alan.hayward@arm.com>
133 David Sherwood <david.sherwood@arm.com>
135 * doc/md.texi (extract_last_@var{m}): Document.
136 * optabs.def (extract_last_optab): New optab.
137 * internal-fn.def (EXTRACT_LAST): New internal function.
138 * internal-fn.c (cond_unary_direct): New macro.
139 (expand_cond_unary_optab_fn): Likewise.
140 (direct_cond_unary_optab_supported_p): Likewise.
141 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
142 loops using EXTRACT_LAST.
143 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
144 (extract_last_<mode>): ...this optab.
145 (vec_extract<mode><Vel>): Update accordingly.
147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
148 Alan Hayward <alan.hayward@arm.com>
149 David Sherwood <david.sherwood@arm.com>
151 * target.def (empty_mask_is_expensive): New hook.
152 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
153 * doc/tm.texi: Regenerate.
154 * targhooks.h (default_empty_mask_is_expensive): Declare.
155 * targhooks.c (default_empty_mask_is_expensive): New function.
156 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
157 if the target says that empty masks are expensive.
158 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
160 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
162 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
163 Alan Hayward <alan.hayward@arm.com>
164 David Sherwood <david.sherwood@arm.com>
166 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
167 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
168 (vect_use_loop_mask_for_alignment_p): New function.
169 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
170 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
171 niters_skip argument. Make sure that the first niters_skip elements
172 of the first iteration are inactive.
173 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
174 Update call to vect_set_loop_masks_directly.
175 (get_misalign_in_elems): New function, split out from...
176 (vect_gen_prolog_loop_niters): ...here.
177 (vect_update_init_of_dr): Take a code argument that specifies whether
178 the adjustment should be added or subtracted.
179 (vect_update_init_of_drs): Likewise.
180 (vect_prepare_for_masked_peels): New function.
181 (vect_do_peeling): Skip prologue peeling if we're using a mask
182 instead. Update call to vect_update_inits_of_drs.
183 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
185 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
186 alignment. Do not include the number of peeled iterations in
187 the minimum threshold in that case.
188 (vectorizable_induction): Adjust the start value down by
189 LOOP_VINFO_MASK_SKIP_NITERS iterations.
190 (vect_transform_loop): Call vect_prepare_for_masked_peels.
191 Take the number of skipped iterations into account when calculating
193 * tree-vect-stmts.c (vect_gen_while_not): New function.
195 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
196 Alan Hayward <alan.hayward@arm.com>
197 David Sherwood <david.sherwood@arm.com>
199 * doc/sourcebuild.texi (vect_fully_masked): Document.
200 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
202 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
204 (vect_analyze_loop_2): ...here. Don't check the vectorization
205 factor against the number of loop iterations if the loop is
208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
209 Alan Hayward <alan.hayward@arm.com>
210 David Sherwood <david.sherwood@arm.com>
212 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
213 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
214 (dump_groups): Update accordingly.
215 (iv_use::mem_type): New member variable.
216 (address_p): New function.
217 (record_use): Add a mem_type argument and initialize the new
219 (record_group_use): Add a mem_type argument. Use address_p.
220 Remove obsolete null checks of base_object. Update call to record_use.
221 (find_interesting_uses_op): Update call to record_group_use.
222 (find_interesting_uses_cond): Likewise.
223 (find_interesting_uses_address): Likewise.
224 (get_mem_type_for_internal_fn): New function.
225 (find_address_like_use): Likewise.
226 (find_interesting_uses_stmt): Try find_address_like_use before
227 calling find_interesting_uses_op.
228 (addr_offset_valid_p): Use the iv mem_type field as the type
229 of the addressed memory.
230 (add_autoinc_candidates): Likewise.
231 (get_address_cost): Likewise.
232 (split_small_address_groups_p): Use address_p.
233 (split_address_groups): Likewise.
234 (add_iv_candidate_for_use): Likewise.
235 (autoinc_possible_for_pair): Likewise.
236 (rewrite_groups): Likewise.
237 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
238 (determine_group_iv_cost): Update after split of USE_ADDRESS.
239 (get_alias_ptr_type_for_ptr_address): New function.
240 (rewrite_use_address): Rewrite address uses in calls that were
241 identified by find_address_like_use.
243 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
244 Alan Hayward <alan.hayward@arm.com>
245 David Sherwood <david.sherwood@arm.com>
247 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
249 * gimple-expr.h (is_gimple_addressable: Likewise.
250 * gimple-expr.c (is_gimple_address): Likewise.
251 * internal-fn.c (expand_call_mem_ref): New function.
252 (expand_mask_load_optab_fn): Use it.
253 (expand_mask_store_optab_fn): Likewise.
255 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
256 Alan Hayward <alan.hayward@arm.com>
257 David Sherwood <david.sherwood@arm.com>
259 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
260 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
261 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
262 (cond_umax@var{mode}): Document.
263 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
264 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
265 (cond_umin_optab, cond_umax_optab): New optabs.
266 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
267 (COND_IOR, COND_XOR): New internal functions.
268 * internal-fn.h (get_conditional_internal_fn): Declare.
269 * internal-fn.c (cond_binary_direct): New macro.
270 (expand_cond_binary_optab_fn): Likewise.
271 (direct_cond_binary_optab_supported_p): Likewise.
272 (get_conditional_internal_fn): New function.
273 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
274 Cope with reduction statements that are vectorized as calls rather
276 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
277 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
278 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
279 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
280 (UNSPEC_COND_EOR): New unspecs.
281 (optab): Add mappings for them.
282 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
283 (sve_int_op, sve_fp_op): New int attributes.
285 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
286 Alan Hayward <alan.hayward@arm.com>
287 David Sherwood <david.sherwood@arm.com>
289 * optabs.def (while_ult_optab): New optab.
290 * doc/md.texi (while_ult@var{m}@var{n}): Document.
291 * internal-fn.def (WHILE_ULT): New internal function.
292 * internal-fn.h (direct_internal_fn_supported_p): New override
293 that takes two types as argument.
294 * internal-fn.c (while_direct): New macro.
295 (expand_while_optab_fn): New function.
296 (convert_optab_supported_p): Likewise.
297 (direct_while_optab_supported_p): New macro.
298 * wide-int.h (wi::udiv_ceil): New function.
299 * tree-vectorizer.h (rgroup_masks): New structure.
300 (vec_loop_masks): New typedef.
301 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
303 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
304 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
305 (vect_max_vf): New function.
306 (slpeel_make_loop_iterate_ntimes): Delete.
307 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
308 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
309 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
310 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
311 internal-fn.h, stor-layout.h and optabs-query.h.
312 (vect_set_loop_mask): New function.
313 (add_preheader_seq): Likewise.
314 (add_header_seq): Likewise.
315 (interleave_supported_p): Likewise.
316 (vect_maybe_permute_loop_masks): Likewise.
317 (vect_set_loop_masks_directly): Likewise.
318 (vect_set_loop_condition_masked): Likewise.
319 (vect_set_loop_condition_unmasked): New function, split out from
320 slpeel_make_loop_iterate_ntimes.
321 (slpeel_make_loop_iterate_ntimes): Rename to..
322 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
323 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
324 (vect_do_peeling): Update call accordingly.
325 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
327 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
328 mask_compare_type, can_fully_mask_p and fully_masked_p.
329 (release_vec_loop_masks): New function.
330 (_loop_vec_info): Use it to free the loop masks.
331 (can_produce_all_loop_masks_p): New function.
332 (vect_get_max_nscalars_per_iter): Likewise.
333 (vect_verify_full_masking): Likewise.
334 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
335 retries, and free the mask rgroups before retrying. Check loop-wide
336 reasons for disallowing fully-masked loops. Make the final decision
337 about whether use a fully-masked loop or not.
338 (vect_estimate_min_profitable_iters): Do not assume that peeling
339 for the number of iterations will be needed for fully-masked loops.
340 (vectorizable_reduction): Disable fully-masked loops.
341 (vectorizable_live_operation): Likewise.
342 (vect_halve_mask_nunits): New function.
343 (vect_double_mask_nunits): Likewise.
344 (vect_record_loop_mask): Likewise.
345 (vect_get_loop_mask): Likewise.
346 (vect_transform_loop): Handle the case in which the final loop
347 iteration might handle a partial vector. Call vect_set_loop_condition
348 instead of slpeel_make_loop_iterate_ntimes.
349 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
350 (check_load_store_masking): New function.
351 (prepare_load_store_mask): Likewise.
352 (vectorizable_store): Handle fully-masked loops.
353 (vectorizable_load): Likewise.
354 (supportable_widening_operation): Use vect_halve_mask_nunits for
356 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
357 (vect_gen_while): New function.
358 * config/aarch64/aarch64.md (umax<mode>3): New expander.
359 (aarch64_uqdec<mode>): New insn.
361 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
362 Alan Hayward <alan.hayward@arm.com>
363 David Sherwood <david.sherwood@arm.com>
365 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
366 (reduc_xor_scal_optab): New optabs.
367 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
368 (reduc_xor_scal_@var{m}): Document.
369 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
370 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
372 * fold-const-call.c (fold_const_call): Handle them.
373 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
374 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
375 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
376 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
377 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
378 (UNSPEC_XORV): New unspecs.
379 (optab): Add entries for them.
380 (BITWISEV): New int iterator.
381 (bit_reduc_op): New int attributes.
383 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
384 Alan Hayward <alan.hayward@arm.com>
385 David Sherwood <david.sherwood@arm.com>
387 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
388 * internal-fn.def (VEC_SHL_INSERT): New internal function.
389 * optabs.def (vec_shl_insert_optab): New optab.
390 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
391 (duplicate_and_interleave): Likewise.
392 * tree-vect-loop.c: Include internal-fn.h.
393 (neutral_op_for_slp_reduction): New function, split out from
394 get_initial_defs_for_reduction.
395 (get_initial_def_for_reduction): Handle option 2 for variable-length
396 vectors by loading the neutral value into a vector and then shifting
397 the initial value into element 0.
398 (get_initial_defs_for_reduction): Replace the code argument with
399 the neutral value calculated by neutral_op_for_slp_reduction.
400 Use gimple_build_vector for constant-length vectors.
401 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
402 but the first group_size elements have a neutral value.
403 Use duplicate_and_interleave otherwise.
404 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
405 Update call to get_initial_defs_for_reduction. Handle SLP
406 reductions for variable-length vectors by creating one vector
407 result for each scalar result, with the elements associated
408 with other scalar results stubbed out with the neutral value.
409 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
410 Require IFN_VEC_SHL_INSERT for double reductions on
411 variable-length vectors, or SLP reductions that have
412 a neutral value. Require can_duplicate_and_interleave_p
413 support for variable-length unchained SLP reductions if there
414 is no neutral value, such as for MIN/MAX reductions. Also require
415 the number of vector elements to be a multiple of the number of
416 SLP statements when doing variable-length unchained SLP reductions.
417 Update call to vect_create_epilog_for_reduction.
418 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
419 and remove initial values.
420 (duplicate_and_interleave): Make public.
421 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
422 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
424 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
425 Alan Hayward <alan.hayward@arm.com>
426 David Sherwood <david.sherwood@arm.com>
428 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
429 (can_duplicate_and_interleave_p): New function.
430 (vect_get_and_check_slp_defs): Take the vector of statements
431 rather than just the current one. Remove excess parentheses.
432 Restriction rejectinon of vect_constant_def and vect_external_def
433 for variable-length vectors to boolean types, or types for which
434 can_duplicate_and_interleave_p is false.
435 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
436 (duplicate_and_interleave): New function.
437 (vect_get_constant_vectors): Use gimple_build_vector for
438 constant-length vectors and suitable variable-length constant
439 vectors. Use duplicate_and_interleave for other variable-length
440 vectors. Don't defer the update when inserting new statements.
442 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
443 Alan Hayward <alan.hayward@arm.com>
444 David Sherwood <david.sherwood@arm.com>
446 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
447 min_profitable_iters doesn't go negative.
449 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
450 Alan Hayward <alan.hayward@arm.com>
451 David Sherwood <david.sherwood@arm.com>
453 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
454 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
455 * optabs.def (vec_mask_load_lanes_optab): New optab.
456 (vec_mask_store_lanes_optab): Likewise.
457 * internal-fn.def (MASK_LOAD_LANES): New internal function.
458 (MASK_STORE_LANES): Likewise.
459 * internal-fn.c (mask_load_lanes_direct): New macro.
460 (mask_store_lanes_direct): Likewise.
461 (expand_mask_load_optab_fn): Handle masked operations.
462 (expand_mask_load_lanes_optab_fn): New macro.
463 (expand_mask_store_optab_fn): Handle masked operations.
464 (expand_mask_store_lanes_optab_fn): New macro.
465 (direct_mask_load_lanes_optab_supported_p): Likewise.
466 (direct_mask_store_lanes_optab_supported_p): Likewise.
467 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
469 (vect_load_lanes_supported): Likewise.
470 * tree-vect-data-refs.c (strip_conversion): New function.
471 (can_group_stmts_p): Likewise.
472 (vect_analyze_data_ref_accesses): Use it instead of checking
473 for a pair of assignments.
474 (vect_store_lanes_supported): Take a masked_p parameter.
475 (vect_load_lanes_supported): Likewise.
476 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
477 vect_store_lanes_supported and vect_load_lanes_supported.
478 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
479 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
480 parameter. Don't allow gaps for masked accesses.
481 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
482 and vect_load_lanes_supported.
483 (get_load_store_type): Take a masked_p parameter and update
484 call to get_group_load_store_type.
485 (vectorizable_store): Update call to get_load_store_type.
486 Handle IFN_MASK_STORE_LANES.
487 (vectorizable_load): Update call to get_load_store_type.
488 Handle IFN_MASK_LOAD_LANES.
490 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
491 Alan Hayward <alan.hayward@arm.com>
492 David Sherwood <david.sherwood@arm.com>
494 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
496 * config/aarch64/aarch64-protos.h
497 (aarch64_sve_struct_memory_operand_p): Declare.
498 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
499 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
500 (VPRED, vpred): Handle SVE structure modes.
501 * config/aarch64/constraints.md (Utx): New constraint.
502 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
503 (aarch64_sve_struct_nonimmediate_operand): New predicates.
504 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
505 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
506 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
507 structure modes. Split into pieces after RA.
508 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
509 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
511 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
513 (aarch64_classify_address): Likewise.
514 (sizetochar): Move earlier in file.
515 (aarch64_print_operand): Handle SVE register lists.
516 (aarch64_array_mode): New function.
517 (aarch64_sve_struct_memory_operand_p): Likewise.
518 (TARGET_ARRAY_MODE): Redefine.
520 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
521 Alan Hayward <alan.hayward@arm.com>
522 David Sherwood <david.sherwood@arm.com>
524 * target.def (array_mode): New target hook.
525 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
526 * doc/tm.texi: Regenerate.
527 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
528 * hooks.c (hook_optmode_mode_uhwi_none): New function.
529 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
531 * stor-layout.c (mode_for_array): Likewise. Support polynomial
534 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
535 Alan Hayward <alan.hayward@arm.com>
536 David Sherwood <david.sherwood@arm.com>
538 * fold-const.c (fold_binary_loc): Check the argument types
539 rather than the result type when testing for a vector operation.
541 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
543 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
544 * doc/tm.texi: Regenerate.
546 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
547 Alan Hayward <alan.hayward@arm.com>
548 David Sherwood <david.sherwood@arm.com>
550 * doc/invoke.texi (-msve-vector-bits=): Document new option.
551 (sve): Document new AArch64 extension.
552 * doc/md.texi (w): Extend the description of the AArch64
553 constraint to include SVE vectors.
554 (Upl, Upa): Document new AArch64 predicate constraints.
555 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
557 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
558 (msve-vector-bits=): New option.
559 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
560 SVE when these are disabled.
561 (sve): New extension.
562 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
563 modes. Adjust their number of units based on aarch64_sve_vg.
564 (MAX_BITSIZE_MODE_ANY_MODE): Define.
565 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
566 aarch64_addr_query_type.
567 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
568 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
569 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
570 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
571 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
572 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
573 (aarch64_simd_imm_zero_p): Delete.
574 (aarch64_check_zero_based_sve_index_immediate): Declare.
575 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
576 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
577 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
578 (aarch64_sve_float_mul_immediate_p): Likewise.
579 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
581 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
582 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
583 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
584 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
585 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
586 (aarch64_regmode_natural_size): Likewise.
587 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
588 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
590 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
591 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
592 for VG and the SVE predicate registers.
593 (V_ALIASES): Add a "z"-prefixed alias.
594 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
595 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
596 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
597 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
598 (REG_CLASS_NAMES): Add entries for them.
599 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
600 and the predicate registers.
601 (aarch64_sve_vg): Declare.
602 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
603 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
604 (REGMODE_NATURAL_SIZE): Define.
605 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
607 * config/aarch64/aarch64.c: Include cfgrtl.h.
608 (simd_immediate_info): Add a constructor for series vectors,
609 and an associated step field.
610 (aarch64_sve_vg): New variable.
611 (aarch64_dbx_register_number): Handle VG and the predicate registers.
612 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
613 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
614 (VEC_ANY_DATA, VEC_STRUCT): New constants.
615 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
616 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
617 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
618 (aarch64_get_mask_mode): New functions.
619 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
620 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
621 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
622 predicate modes and predicate registers. Explicitly restrict
623 GPRs to modes of 16 bytes or smaller. Only allow FP registers
624 to store a vector mode if it is recognized by
625 aarch64_classify_vector_mode.
626 (aarch64_regmode_natural_size): New function.
627 (aarch64_hard_regno_caller_save_mode): Return the original mode
629 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
630 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
631 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
632 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
634 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
635 does not overlap dest if the function is frame-related. Handle
637 (aarch64_split_add_offset): New function.
638 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
639 them aarch64_add_offset.
640 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
641 and update call to aarch64_sub_sp.
642 (aarch64_add_cfa_expression): New function.
643 (aarch64_expand_prologue): Pass extra temporary registers to the
644 functions above. Handle the case in which we need to emit new
645 DW_CFA_expressions for registers that were originally saved
646 relative to the stack pointer, but now have to be expressed
647 relative to the frame pointer.
648 (aarch64_output_mi_thunk): Pass extra temporary registers to the
650 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
651 IP0 and IP1 values for SVE frames.
652 (aarch64_expand_vec_series): New function.
653 (aarch64_expand_sve_widened_duplicate): Likewise.
654 (aarch64_expand_sve_const_vector): Likewise.
655 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
656 Handle SVE constants. Use emit_move_insn to move a force_const_mem
657 into the register, rather than emitting a SET directly.
658 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
659 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
660 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
661 (offset_9bit_signed_scaled_p): New functions.
662 (aarch64_replicate_bitmask_imm): New function.
663 (aarch64_bitmask_imm): Use it.
664 (aarch64_cannot_force_const_mem): Reject expressions involving
665 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
666 (aarch64_classify_index): Handle SVE indices, by requiring
667 a plain register index with a scale that matches the element size.
668 (aarch64_classify_address): Handle SVE addresses. Assert that
669 the mode of the address is VOIDmode or an integer mode.
670 Update call to aarch64_classify_symbol.
671 (aarch64_classify_symbolic_expression): Update call to
672 aarch64_classify_symbol.
673 (aarch64_const_vec_all_in_range_p): New function.
674 (aarch64_print_vector_float_operand): Likewise.
675 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
676 "vN" for FP registers with SVE modes. Handle (const ...) vectors
677 and the FP immediates 1.0 and 0.5.
678 (aarch64_print_address_internal): Handle SVE addresses.
679 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
680 (aarch64_regno_regclass): Handle predicate registers.
681 (aarch64_secondary_reload): Handle big-endian reloads of SVE
683 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
684 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
685 (aarch64_convert_sve_vector_bits): New function.
686 (aarch64_override_options): Use it to handle -msve-vector-bits=.
687 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
689 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
690 Handle SVE vector and predicate modes. Accept VL-based constants
691 that need only one temporary register, and VL offsets that require
692 no temporary registers.
693 (aarch64_conditional_register_usage): Mark the predicate registers
694 as fixed if SVE isn't available.
695 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
696 Return true for SVE vector and predicate modes.
697 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
698 rather than an unsigned int. Handle SVE modes.
699 (aarch64_preferred_simd_mode): Update call accordingly. Handle
701 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
703 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
704 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
705 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
706 (aarch64_sve_float_mul_immediate_p): New functions.
707 (aarch64_sve_valid_immediate): New function.
708 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
709 Explicitly reject structure modes. Check for INDEX constants.
710 Handle PTRUE and PFALSE constants.
711 (aarch64_check_zero_based_sve_index_immediate): New function.
712 (aarch64_simd_imm_zero_p): Delete.
713 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
714 vector modes. Accept constants in the range of CNT[BHWD].
715 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
716 ask for an Advanced SIMD mode.
717 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
718 (aarch64_simd_vector_alignment): Handle SVE predicates.
719 (aarch64_vectorize_preferred_vector_alignment): New function.
720 (aarch64_simd_vector_alignment_reachable): Use it instead of
722 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
723 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
725 (MAX_VECT_LEN): Delete.
726 (expand_vec_perm_d): Add a vec_flags field.
727 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
728 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
729 (aarch64_evpc_ext): Don't apply a big-endian lane correction
731 (aarch64_evpc_rev): Rename to...
732 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
733 (aarch64_evpc_rev_global): New function.
734 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
735 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
737 (aarch64_evpc_sve_tbl): New function.
738 (aarch64_expand_vec_perm_const_1): Update after rename of
739 aarch64_evpc_rev. Handle SVE permutes too, trying
740 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
741 than aarch64_evpc_tbl.
742 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
743 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
744 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
745 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
746 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
747 (aarch64_expand_sve_vcond): New functions.
748 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
749 of aarch64_vector_mode_p.
750 (aarch64_dwarf_poly_indeterminate_value): New function.
751 (aarch64_compute_pressure_classes): Likewise.
752 (aarch64_can_change_mode_class): Likewise.
753 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
754 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
755 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
756 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
757 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
758 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
759 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
760 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
762 (Dn, Dl, Dr): Accept const as well as const_vector.
763 (Dz): Likewise. Compare against CONST0_RTX.
764 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
765 of "vector" where appropriate.
766 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
767 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
768 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
769 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
770 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
771 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
772 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
773 (v_int_equiv): Extend to SVE modes.
774 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
776 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
777 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
778 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
779 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
780 (SVE_COND_FP_CMP): New int iterators.
781 (perm_hilo): Handle the new unpack unspecs.
782 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
784 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
785 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
786 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
787 (aarch64_equality_operator, aarch64_constant_vector_operand)
788 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
789 (aarch64_sve_nonimmediate_operand): Likewise.
790 (aarch64_sve_general_operand): Likewise.
791 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
792 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
793 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
794 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
795 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
796 (aarch64_sve_float_arith_immediate): Likewise.
797 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
798 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
799 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
800 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
801 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
802 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
803 (aarch64_sve_float_arith_operand): Likewise.
804 (aarch64_sve_float_arith_with_sub_operand): Likewise.
805 (aarch64_sve_float_mul_operand): Likewise.
806 (aarch64_sve_vec_perm_operand): Likewise.
807 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
808 (aarch64_mov_operand): Accept const_poly_int and const_vector.
809 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
810 as well as const_vector.
811 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
812 in file. Use CONST0_RTX and CONSTM1_RTX.
813 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
814 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
815 Use aarch64_simd_imm_zero.
816 * config/aarch64/aarch64-sve.md: New file.
817 * config/aarch64/aarch64.md: Include it.
818 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
819 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
820 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
821 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
822 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
823 (sve): New attribute.
824 (enabled): Disable instructions with the sve attribute unless
826 (movqi, movhi): Pass CONST_POLY_INT operaneds through
827 aarch64_expand_mov_immediate.
828 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
829 CNT[BHSD] immediates.
830 (movti): Split CONST_POLY_INT moves into two halves.
831 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
832 Split additions that need a temporary here if the destination
833 is the stack pointer.
834 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
835 (*add<mode>3_poly_1): New instruction.
836 (set_clobber_cc): New expander.
838 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
840 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
841 parameter and use it instead of GET_MODE_SIZE (innermode). Use
842 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
843 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
844 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
845 Change innermode from fixed_mode_size to machine_mode.
846 (simplify_subreg): Update call accordingly. Handle a constant-sized
847 subreg of a variable-length CONST_VECTOR.
849 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
850 Alan Hayward <alan.hayward@arm.com>
851 David Sherwood <david.sherwood@arm.com>
853 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
854 (add_offset_to_base): New function, split out from...
855 (create_mem_ref): ...here. When handling a scale other than 1,
856 check first whether the address is valid without the offset.
857 Add it into the base if so, leaving the index and scale as-is.
859 2018-01-12 Jakub Jelinek <jakub@redhat.com>
862 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
863 fold_for_warn before checking if arg2 is INTEGER_CST.
865 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
867 * config/rs6000/predicates.md (load_multiple_operation): Delete.
868 (store_multiple_operation): Delete.
869 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
870 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
871 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
872 guarded by TARGET_STRING.
873 (rs6000_output_load_multiple): Delete.
874 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
875 OPTION_MASK_STRING / TARGET_STRING handling.
876 (print_operand) <'N', 'O'>: Add comment that these are unused now.
877 (const rs6000_opt_masks) <"string">: Change mask to 0.
878 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
879 (MASK_STRING): Delete.
880 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
882 (load_multiple): Delete.
889 (store_multiple): Delete.
896 (movmemsi_8reg): Delete.
897 (corresponding unnamed define_insn): Delete.
898 (movmemsi_6reg): Delete.
899 (corresponding unnamed define_insn): Delete.
900 (movmemsi_4reg): Delete.
901 (corresponding unnamed define_insn): Delete.
902 (movmemsi_2reg): Delete.
903 (corresponding unnamed define_insn): Delete.
904 (movmemsi_1reg): Delete.
905 (corresponding unnamed define_insn): Delete.
906 * config/rs6000/rs6000.opt (mno-string): New.
907 (mstring): Replace by deprecation warning stub.
908 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
910 2018-01-12 Jakub Jelinek <jakub@redhat.com>
912 * regrename.c (regrename_do_replace): If replacing the same
913 reg multiple times, try to reuse last created gen_raw_REG.
916 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
917 main to workaround a bug in GDB.
919 2018-01-12 Tom de Vries <tom@codesourcery.com>
922 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
924 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
926 PR rtl-optimization/80481
927 * ira-color.c (get_cap_member): New function.
928 (allocnos_conflict_by_live_ranges_p): Use it.
929 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
930 (setup_slot_coalesced_allocno_live_ranges): Ditto.
932 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
935 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
936 (*saddl_se_1): Ditto.
938 (*saddl_se_1): Ditto.
940 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
942 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
943 rather than wi::to_widest for DR_INITs.
944 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
945 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
946 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
948 (vect_analyze_group_access_1): Note that here.
950 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
952 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
953 polynomial type sizes.
955 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
957 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
958 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
959 (gimple_add_tmp_var): Likewise.
961 2018-01-12 Martin Liska <mliska@suse.cz>
963 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
964 (gimple_alloc_sizes): Likewise.
965 (dump_gimple_statistics): Use PRIu64 in printf format.
966 * gimple.h: Change uint64_t to int.
968 2018-01-12 Martin Liska <mliska@suse.cz>
970 * tree-core.h: Use uint64_t instead of int.
971 * tree.c (tree_node_counts): Likewise.
972 (tree_node_sizes): Likewise.
973 (dump_tree_statistics): Use PRIu64 in printf format.
975 2018-01-12 Martin Liska <mliska@suse.cz>
977 * Makefile.in: As qsort_chk is implemented in vec.c, add
978 vec.o to linkage of gencfn-macros.
979 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
980 passing the info to record_node_allocation_statistics.
981 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
983 * ggc-common.c (struct ggc_usage): Add operator== and use
984 it in operator< and compare function.
985 * mem-stats.h (struct mem_usage): Likewise.
986 * vec.c (struct vec_usage): Remove operator< and compare
987 function. Can be simply inherited.
989 2018-01-12 Martin Jambor <mjambor@suse.cz>
992 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
993 * tree-ssa-math-opts.c: Include domwalk.h.
994 (convert_mult_to_fma_1): New function.
995 (fma_transformation_info): New type.
996 (fma_deferring_state): Likewise.
997 (cancel_fma_deferring): New function.
998 (result_of_phi): Likewise.
999 (last_fma_candidate_feeds_initial_phi): Likewise.
1000 (convert_mult_to_fma): Added deferring logic, split actual
1001 transformation to convert_mult_to_fma_1.
1002 (math_opts_dom_walker): New type.
1003 (math_opts_dom_walker::after_dom_children): New method, body moved
1004 here from pass_optimize_widening_mul::execute, added deferring logic
1006 (pass_optimize_widening_mul::execute): Moved most of code to
1007 math_opts_dom_walker::after_dom_children.
1008 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
1009 * config/i386/i386.c (ix86_option_override_internal): Added
1010 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
1012 2018-01-12 Richard Biener <rguenther@suse.de>
1015 * dwarf2out.c (gen_variable_die): Do not reset old_die for
1016 inline instance vars.
1018 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
1021 * config/rx/rx.c (rx_is_restricted_memory_address):
1024 2018-01-12 Richard Biener <rguenther@suse.de>
1026 PR tree-optimization/80846
1027 * target.def (split_reduction): New target hook.
1028 * targhooks.c (default_split_reduction): New function.
1029 * targhooks.h (default_split_reduction): Declare.
1030 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
1031 target requests first reduce vectors by combining low and high
1033 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
1034 (get_vectype_for_scalar_type_and_size): Export.
1035 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
1036 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
1037 * doc/tm.texi: Regenerate.
1038 * config/i386/i386.c (ix86_split_reduction): Implement
1039 TARGET_VECTORIZE_SPLIT_REDUCTION.
1041 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1044 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
1045 in PIC mode except for TARGET_VXWORKS_RTP.
1046 * config/sparc/sparc.c: Include cfgrtl.h.
1047 (TARGET_INIT_PIC_REG): Define.
1048 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
1049 (sparc_pic_register_p): New predicate.
1050 (sparc_legitimate_address_p): Use it.
1051 (sparc_legitimize_pic_address): Likewise.
1052 (sparc_delegitimize_address): Likewise.
1053 (sparc_mode_dependent_address_p): Likewise.
1054 (gen_load_pcrel_sym): Remove 4th parameter.
1055 (load_got_register): Adjust call to above. Remove obsolete stuff.
1056 (sparc_expand_prologue): Do not call load_got_register here.
1057 (sparc_flat_expand_prologue): Likewise.
1058 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
1059 (sparc_use_pseudo_pic_reg): New function.
1060 (sparc_init_pic_reg): Likewise.
1061 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1062 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
1064 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
1066 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
1067 Add item for branch_cost.
1069 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1071 PR rtl-optimization/83565
1072 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
1073 not extend the result to a larger mode for rotate operations.
1074 (num_sign_bit_copies1): Likewise.
1076 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1079 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
1081 Use values-Xc.o for -pedantic.
1082 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
1084 2018-01-12 Martin Liska <mliska@suse.cz>
1087 * ipa-devirt.c (final_warning_record::grow_type_warnings):
1089 (possible_polymorphic_call_targets): Use it.
1090 (ipa_devirt): Likewise.
1092 2018-01-12 Martin Liska <mliska@suse.cz>
1094 * profile-count.h (enum profile_quality): Use 0 as invalid
1095 enum value of profile_quality.
1097 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
1099 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
1100 -mext-string options.
1102 2018-01-12 Richard Biener <rguenther@suse.de>
1104 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
1105 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
1106 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
1108 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
1110 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
1112 * configure.ac (--with-long-double-format): Add support for the
1113 configuration option to change the default long double format on
1115 * config.gcc (powerpc*-linux*-*): Likewise.
1116 * configure: Regenerate.
1117 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
1118 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
1119 used without modification.
1121 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1123 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
1124 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
1125 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
1126 MISC_BUILTIN_SPEC_BARRIER.
1127 (rs6000_init_builtins): Likewise.
1128 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
1130 (speculation_barrier): New define_insn.
1131 * doc/extend.texi: Document __builtin_speculation_barrier.
1133 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1136 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
1137 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
1138 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
1140 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
1141 integral modes instead of "ss" and "sd".
1142 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
1143 vectors with 32-bit and 64-bit elements.
1144 (vecdupssescalarmodesuffix): New mode attribute.
1145 (vec_dup<mode>): Use it.
1147 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
1150 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
1151 frame if argument is passed on stack.
1153 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1156 * ree.c (combine_reaching_defs): Optimize also
1157 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
1158 reg2=any_extend(exp); reg1=reg2;, formatting fix.
1160 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1163 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
1165 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1168 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
1169 after they are computed.
1171 2018-01-11 Bin Cheng <bin.cheng@arm.com>
1173 PR tree-optimization/83695
1174 * gimple-loop-linterchange.cc
1175 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
1176 reset cached scev information after interchange.
1177 (pass_linterchange::execute): Remove call to scev_reset_htab.
1179 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1181 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
1182 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
1183 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
1184 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
1185 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
1186 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
1187 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
1188 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
1189 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
1190 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
1191 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
1192 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
1193 (V_lane_reg): Likewise.
1194 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
1196 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
1197 (vfmal_lane_low<mode>_intrinsic,
1198 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
1199 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
1200 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
1201 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
1202 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
1203 vfmsl_lane_high<mode>_intrinsic): New define_insns.
1205 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1207 * config/arm/arm-cpus.in (fp16fml): New feature.
1208 (ALL_SIMD): Add fp16fml.
1209 (armv8.2-a): Add fp16fml as an option.
1210 (armv8.3-a): Likewise.
1211 (armv8.4-a): Add fp16fml as part of fp16.
1212 * config/arm/arm.h (TARGET_FP16FML): Define.
1213 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
1215 * config/arm/arm-modes.def (V2HF): Define.
1216 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
1217 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
1218 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
1219 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
1220 vfmsl_low, vfmsl_high): New set of builtins.
1221 * config/arm/iterators.md (PLUSMINUS): New code iterator.
1222 (vfml_op): New code attribute.
1223 (VFMLHALVES): New int iterator.
1224 (VFML, VFMLSEL): New mode attributes.
1225 (V_reg): Define mapping for V2HF.
1226 (V_hi, V_lo): New mode attributes.
1227 (VF_constraint): Likewise.
1228 (vfml_half, vfml_half_selector): New int attributes.
1229 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
1231 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
1232 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
1234 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
1235 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
1236 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
1237 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
1239 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
1240 Document new effective target and option set.
1242 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1244 * config/arm/arm-cpus.in (armv8_4): New feature.
1245 (ARMv8_4a): New fgroup.
1246 (armv8.4-a): New arch.
1247 * config/arm/arm-tables.opt: Regenerate.
1248 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
1249 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
1250 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
1251 Add matching rules for -march=armv8.4-a and extensions.
1252 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
1254 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
1257 * config/rx/rx.md (BW): New mode attribute.
1258 (sync_lock_test_and_setsi): Add mode suffix to insn output.
1260 2018-01-11 Richard Biener <rguenther@suse.de>
1262 PR tree-optimization/83435
1263 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
1264 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
1265 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
1267 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1268 Alan Hayward <alan.hayward@arm.com>
1269 David Sherwood <david.sherwood@arm.com>
1271 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
1273 (aarch64_classify_address): Initialize it. Track polynomial offsets.
1274 (aarch64_print_address_internal): Use it to check for a zero offset.
1276 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1277 Alan Hayward <alan.hayward@arm.com>
1278 David Sherwood <david.sherwood@arm.com>
1280 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
1281 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
1282 Return a poly_int64 rather than a HOST_WIDE_INT.
1283 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
1284 rather than a HOST_WIDE_INT.
1285 * config/aarch64/aarch64.h (aarch64_frame): Protect with
1286 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
1287 hard_fp_offset, frame_size, initial_adjust, callee_offset and
1288 final_offset from HOST_WIDE_INT to poly_int64.
1289 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
1290 to_constant when getting the number of units in an Advanced SIMD
1292 (aarch64_builtin_vectorized_function): Check for a constant number
1294 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
1296 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
1297 attribute instead of GET_MODE_NUNITS.
1298 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
1299 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
1300 GET_MODE_SIZE for fixed-size registers.
1301 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
1302 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
1303 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
1304 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
1305 (aarch64_print_operand, aarch64_print_address_internal)
1306 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
1307 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
1308 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
1309 Handle polynomial GET_MODE_SIZE.
1310 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
1311 wider than SImode without modification.
1312 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
1313 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
1314 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
1315 passing and returning SVE modes.
1316 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
1317 rather than GEN_INT.
1318 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
1319 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
1320 (aarch64_allocate_and_probe_stack_space): Likewise.
1321 (aarch64_layout_frame): Cope with polynomial offsets.
1322 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
1323 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
1325 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
1326 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
1327 poly_int64 rather than a HOST_WIDE_INT.
1328 (aarch64_get_separate_components, aarch64_process_components)
1329 (aarch64_expand_prologue, aarch64_expand_epilogue)
1330 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
1331 (aarch64_anchor_offset): New function, split out from...
1332 (aarch64_legitimize_address): ...here.
1333 (aarch64_builtin_vectorization_cost): Handle polynomial
1334 TYPE_VECTOR_SUBPARTS.
1335 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
1337 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
1338 number of elements from the PARALLEL rather than the mode.
1339 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
1340 rather than GET_MODE_BITSIZE.
1341 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
1342 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
1343 (aarch64_expand_vec_perm_const_1): Handle polynomial
1344 d->perm.length () and d->perm elements.
1345 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
1346 Apply to_constant to d->perm elements.
1347 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
1348 polynomial CONST_VECTOR_NUNITS.
1349 (aarch64_move_pointer): Take amount as a poly_int64 rather
1351 (aarch64_progress_pointer): Avoid temporary variable.
1352 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
1353 the mode attribute instead of GET_MODE.
1355 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1356 Alan Hayward <alan.hayward@arm.com>
1357 David Sherwood <david.sherwood@arm.com>
1359 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
1360 x exists before using it.
1361 (aarch64_add_constant_internal): Rename to...
1362 (aarch64_add_offset_1): ...this. Replace regnum with separate
1363 src and dest rtxes. Handle the case in which they're different,
1364 including when the offset is zero. Replace scratchreg with an rtx.
1365 Use 2 additions if there is no spare register into which we can
1366 move a 16-bit constant.
1367 (aarch64_add_constant): Delete.
1368 (aarch64_add_offset): Replace reg with separate src and dest
1369 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
1370 Use aarch64_add_offset_1.
1371 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
1372 an rtx rather than an int. Take the delta as a poly_int64
1373 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
1374 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
1375 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
1376 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
1377 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
1379 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
1380 aarch64_add_constant.
1382 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1384 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
1385 Use scalar_float_mode.
1387 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1389 * config/aarch64/aarch64-simd.md
1390 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
1391 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
1392 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
1393 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
1394 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
1395 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
1396 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
1397 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
1398 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
1399 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
1401 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1404 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
1405 targ_options->x_arm_arch_string is non NULL.
1407 2018-01-11 Tamar Christina <tamar.christina@arm.com>
1409 * config/aarch64/aarch64.h
1410 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
1412 2018-01-11 Sudakshina Das <sudi.das@arm.com>
1415 * expmed.c (emit_store_flag_force): Swap if const op0
1416 and change VOIDmode to mode of op0.
1418 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1420 PR rtl-optimization/83761
1421 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
1422 than bytes to mode_for_size.
1424 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1427 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
1428 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
1431 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1434 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
1435 when in layout mode.
1436 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
1437 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
1440 2018-01-10 Michael Collison <michael.collison@arm.com>
1442 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
1443 * config/aarch64/aarch64-option-extension.def: Add
1444 AARCH64_OPT_EXTENSION of 'fp16fml'.
1445 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1446 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
1447 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
1448 * config/aarch64/constraints.md (Ui7): New constraint.
1449 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
1450 (VFMLA_SEL_W): Ditto.
1453 (VFMLA16_LOW): New int iterator.
1454 (VFMLA16_HIGH): Ditto.
1455 (UNSPEC_FMLAL): New unspec.
1456 (UNSPEC_FMLSL): Ditto.
1457 (UNSPEC_FMLAL2): Ditto.
1458 (UNSPEC_FMLSL2): Ditto.
1459 (f16mac): New code attribute.
1460 * config/aarch64/aarch64-simd-builtins.def
1461 (aarch64_fmlal_lowv2sf): Ditto.
1462 (aarch64_fmlsl_lowv2sf): Ditto.
1463 (aarch64_fmlalq_lowv4sf): Ditto.
1464 (aarch64_fmlslq_lowv4sf): Ditto.
1465 (aarch64_fmlal_highv2sf): Ditto.
1466 (aarch64_fmlsl_highv2sf): Ditto.
1467 (aarch64_fmlalq_highv4sf): Ditto.
1468 (aarch64_fmlslq_highv4sf): Ditto.
1469 (aarch64_fmlal_lane_lowv2sf): Ditto.
1470 (aarch64_fmlsl_lane_lowv2sf): Ditto.
1471 (aarch64_fmlal_laneq_lowv2sf): Ditto.
1472 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
1473 (aarch64_fmlalq_lane_lowv4sf): Ditto.
1474 (aarch64_fmlsl_lane_lowv4sf): Ditto.
1475 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
1476 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
1477 (aarch64_fmlal_lane_highv2sf): Ditto.
1478 (aarch64_fmlsl_lane_highv2sf): Ditto.
1479 (aarch64_fmlal_laneq_highv2sf): Ditto.
1480 (aarch64_fmlsl_laneq_highv2sf): Ditto.
1481 (aarch64_fmlalq_lane_highv4sf): Ditto.
1482 (aarch64_fmlsl_lane_highv4sf): Ditto.
1483 (aarch64_fmlalq_laneq_highv4sf): Ditto.
1484 (aarch64_fmlsl_laneq_highv4sf): Ditto.
1485 * config/aarch64/aarch64-simd.md:
1486 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
1487 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
1488 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
1489 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
1490 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
1491 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
1492 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
1493 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
1494 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
1495 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
1496 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
1497 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
1498 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
1499 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
1500 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
1501 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
1502 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
1503 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
1504 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
1505 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
1506 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
1507 (vfmlsl_low_u32): Ditto.
1508 (vfmlalq_low_u32): Ditto.
1509 (vfmlslq_low_u32): Ditto.
1510 (vfmlal_high_u32): Ditto.
1511 (vfmlsl_high_u32): Ditto.
1512 (vfmlalq_high_u32): Ditto.
1513 (vfmlslq_high_u32): Ditto.
1514 (vfmlal_lane_low_u32): Ditto.
1515 (vfmlsl_lane_low_u32): Ditto.
1516 (vfmlal_laneq_low_u32): Ditto.
1517 (vfmlsl_laneq_low_u32): Ditto.
1518 (vfmlalq_lane_low_u32): Ditto.
1519 (vfmlslq_lane_low_u32): Ditto.
1520 (vfmlalq_laneq_low_u32): Ditto.
1521 (vfmlslq_laneq_low_u32): Ditto.
1522 (vfmlal_lane_high_u32): Ditto.
1523 (vfmlsl_lane_high_u32): Ditto.
1524 (vfmlal_laneq_high_u32): Ditto.
1525 (vfmlsl_laneq_high_u32): Ditto.
1526 (vfmlalq_lane_high_u32): Ditto.
1527 (vfmlslq_lane_high_u32): Ditto.
1528 (vfmlalq_laneq_high_u32): Ditto.
1529 (vfmlslq_laneq_high_u32): Ditto.
1530 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
1531 (AARCH64_FL_FOR_ARCH8_4): New.
1532 (AARCH64_ISA_F16FML): New ISA flag.
1533 (TARGET_F16FML): New feature flag for fp16fml.
1534 (doc/invoke.texi): Document new fp16fml option.
1536 2018-01-10 Michael Collison <michael.collison@arm.com>
1538 * config/aarch64/aarch64-builtins.c:
1539 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
1540 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1541 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
1542 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
1543 (AARCH64_ISA_SHA3): New ISA flag.
1544 (TARGET_SHA3): New feature flag for sha3.
1545 * config/aarch64/iterators.md (sha512_op): New int attribute.
1546 (CRYPTO_SHA512): New int iterator.
1547 (UNSPEC_SHA512H): New unspec.
1548 (UNSPEC_SHA512H2): Ditto.
1549 (UNSPEC_SHA512SU0): Ditto.
1550 (UNSPEC_SHA512SU1): Ditto.
1551 * config/aarch64/aarch64-simd-builtins.def
1552 (aarch64_crypto_sha512hqv2di): New builtin.
1553 (aarch64_crypto_sha512h2qv2di): Ditto.
1554 (aarch64_crypto_sha512su0qv2di): Ditto.
1555 (aarch64_crypto_sha512su1qv2di): Ditto.
1556 (aarch64_eor3qv8hi): Ditto.
1557 (aarch64_rax1qv2di): Ditto.
1558 (aarch64_xarqv2di): Ditto.
1559 (aarch64_bcaxqv8hi): Ditto.
1560 * config/aarch64/aarch64-simd.md:
1561 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
1562 (aarch64_crypto_sha512su0qv2di): Ditto.
1563 (aarch64_crypto_sha512su1qv2di): Ditto.
1564 (aarch64_eor3qv8hi): Ditto.
1565 (aarch64_rax1qv2di): Ditto.
1566 (aarch64_xarqv2di): Ditto.
1567 (aarch64_bcaxqv8hi): Ditto.
1568 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
1569 (vsha512h2q_u64): Ditto.
1570 (vsha512su0q_u64): Ditto.
1571 (vsha512su1q_u64): Ditto.
1572 (veor3q_u16): Ditto.
1573 (vrax1q_u64): Ditto.
1575 (vbcaxq_u16): Ditto.
1576 * config/arm/types.md (crypto_sha512): New type attribute.
1577 (crypto_sha3): Ditto.
1578 (doc/invoke.texi): Document new sha3 option.
1580 2018-01-10 Michael Collison <michael.collison@arm.com>
1582 * config/aarch64/aarch64-builtins.c:
1583 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
1584 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1585 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
1586 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
1587 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
1588 (AARCH64_ISA_SM4): New ISA flag.
1589 (TARGET_SM4): New feature flag for sm4.
1590 * config/aarch64/aarch64-simd-builtins.def
1591 (aarch64_sm3ss1qv4si): Ditto.
1592 (aarch64_sm3tt1aq4si): Ditto.
1593 (aarch64_sm3tt1bq4si): Ditto.
1594 (aarch64_sm3tt2aq4si): Ditto.
1595 (aarch64_sm3tt2bq4si): Ditto.
1596 (aarch64_sm3partw1qv4si): Ditto.
1597 (aarch64_sm3partw2qv4si): Ditto.
1598 (aarch64_sm4eqv4si): Ditto.
1599 (aarch64_sm4ekeyqv4si): Ditto.
1600 * config/aarch64/aarch64-simd.md:
1601 (aarch64_sm3ss1qv4si): Ditto.
1602 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
1603 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
1604 (aarch64_sm4eqv4si): Ditto.
1605 (aarch64_sm4ekeyqv4si): Ditto.
1606 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
1607 (sm3part_op): Ditto.
1608 (CRYPTO_SM3TT): Ditto.
1609 (CRYPTO_SM3PART): Ditto.
1610 (UNSPEC_SM3SS1): New unspec.
1611 (UNSPEC_SM3TT1A): Ditto.
1612 (UNSPEC_SM3TT1B): Ditto.
1613 (UNSPEC_SM3TT2A): Ditto.
1614 (UNSPEC_SM3TT2B): Ditto.
1615 (UNSPEC_SM3PARTW1): Ditto.
1616 (UNSPEC_SM3PARTW2): Ditto.
1617 (UNSPEC_SM4E): Ditto.
1618 (UNSPEC_SM4EKEY): Ditto.
1619 * config/aarch64/constraints.md (Ui2): New constraint.
1620 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
1621 * config/arm/types.md (crypto_sm3): New type attribute.
1622 (crypto_sm4): Ditto.
1623 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
1624 (vsm3tt1aq_u32): Ditto.
1625 (vsm3tt1bq_u32): Ditto.
1626 (vsm3tt2aq_u32): Ditto.
1627 (vsm3tt2bq_u32): Ditto.
1628 (vsm3partw1q_u32): Ditto.
1629 (vsm3partw2q_u32): Ditto.
1630 (vsm4eq_u32): Ditto.
1631 (vsm4ekeyq_u32): Ditto.
1632 (doc/invoke.texi): Document new sm4 option.
1634 2018-01-10 Michael Collison <michael.collison@arm.com>
1636 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
1637 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
1638 (AARCH64_FL_FOR_ARCH8_4): New.
1639 (AARCH64_FL_V8_4): New flag.
1640 (doc/invoke.texi): Document new armv8.4-a option.
1642 2018-01-10 Michael Collison <michael.collison@arm.com>
1644 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1645 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
1646 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
1647 * config/aarch64/aarch64-option-extension.def: Add
1648 AARCH64_OPT_EXTENSION of 'sha2'.
1649 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
1650 (crypto): Disable sha2 and aes if crypto disabled.
1651 (crypto): Enable aes and sha2 if enabled.
1652 (simd): Disable sha2 and aes if simd disabled.
1653 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
1655 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
1656 (TARGET_SHA2): New feature flag for sha2.
1657 (TARGET_AES): New feature flag for aes.
1658 * config/aarch64/aarch64-simd.md:
1659 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
1660 conditional on TARGET_AES.
1661 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
1662 (aarch64_crypto_sha1hsi): Make pattern conditional
1664 (aarch64_crypto_sha1hv4si): Ditto.
1665 (aarch64_be_crypto_sha1hv4si): Ditto.
1666 (aarch64_crypto_sha1su1v4si): Ditto.
1667 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
1668 (aarch64_crypto_sha1su0v4si): Ditto.
1669 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
1670 (aarch64_crypto_sha256su0v4si): Ditto.
1671 (aarch64_crypto_sha256su1v4si): Ditto.
1672 (doc/invoke.texi): Document new aes and sha2 options.
1674 2018-01-10 Martin Sebor <msebor@redhat.com>
1676 PR tree-optimization/83781
1677 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
1680 2018-01-11 Martin Sebor <msebor@gmail.com>
1681 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1683 PR tree-optimization/83501
1684 PR tree-optimization/81703
1686 * tree-ssa-strlen.c (get_string_cst): Rename...
1687 (get_string_len): ...to this. Handle global constants.
1688 (handle_char_store): Adjust.
1690 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
1691 Jim Wilson <jimw@sifive.com>
1693 * config/riscv/riscv-protos.h (riscv_output_return): New.
1694 * config/riscv/riscv.c (struct machine_function): New naked_p field.
1695 (riscv_attribute_table, riscv_output_return),
1696 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
1697 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
1698 (riscv_compute_frame_info): Only compute frame->mask if not a naked
1700 (riscv_expand_prologue): Add early return for naked function.
1701 (riscv_expand_epilogue): Likewise.
1702 (riscv_function_ok_for_sibcall): Return false for naked function.
1703 (riscv_set_current_function): New.
1704 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
1705 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
1706 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
1707 * doc/extend.texi (RISC-V Function Attributes): New.
1709 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
1711 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
1712 check for 128-bit long double before checking TCmode.
1713 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
1714 128-bit long doubles before checking TFmode or TCmode.
1715 (FLOAT128_IBM_P): Likewise.
1717 2018-01-10 Martin Sebor <msebor@redhat.com>
1719 PR tree-optimization/83671
1720 * builtins.c (c_strlen): Unconditionally return zero for the empty
1722 Use -Warray-bounds for warnings.
1723 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
1724 for non-constant array indices with COMPONENT_REF, arrays of
1725 arrays, and pointers to arrays.
1726 (gimple_fold_builtin_strlen): Determine and set length range for
1727 non-constant character arrays.
1729 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
1732 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
1735 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
1737 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
1739 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
1742 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
1743 VECTOR_MEM_ALTIVEC_OR_VSX_P.
1744 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
1745 indexed_or_indirect_operand predicate.
1746 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
1747 (*vsx_le_perm_load_v8hi): Likewise.
1748 (*vsx_le_perm_load_v16qi): Likewise.
1749 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
1750 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
1751 (*vsx_le_perm_store_v8hi): Likewise.
1752 (*vsx_le_perm_store_v16qi): Likewise.
1753 (eight unnamed splitters): Likewise.
1755 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
1757 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
1758 * config/rs6000/emmintrin.h: Likewise.
1759 * config/rs6000/mmintrin.h: Likewise.
1760 * config/rs6000/xmmintrin.h: Likewise.
1762 2018-01-10 David Malcolm <dmalcolm@redhat.com>
1765 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
1767 * tree.c (tree_nop_conversion): Return true for location wrapper
1769 (maybe_wrap_with_location): New function.
1770 (selftest::check_strip_nops): New function.
1771 (selftest::test_location_wrappers): New function.
1772 (selftest::tree_c_tests): Call it.
1773 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
1774 (maybe_wrap_with_location): New decl.
1775 (EXPR_LOCATION_WRAPPER_P): New macro.
1776 (location_wrapper_p): New inline function.
1777 (tree_strip_any_location_wrapper): New inline function.
1779 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
1782 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
1783 stack_realign_offset for the largest alignment of stack slot
1785 (ix86_find_max_used_stack_alignment): New function.
1786 (ix86_finalize_stack_frame_flags): Use it. Set
1787 max_used_stack_alignment if we don't realign stack.
1788 * config/i386/i386.h (machine_function): Add
1789 max_used_stack_alignment.
1791 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
1793 * config/arm/arm.opt (-mbranch-cost): New option.
1794 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
1797 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
1800 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
1801 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
1803 2018-01-10 Richard Biener <rguenther@suse.de>
1806 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
1807 early out so it also covers the case where we have a non-NULL
1810 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
1812 PR tree-optimization/83753
1813 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
1814 for non-strided grouped accesses if the number of elements is 1.
1816 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1819 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
1820 * i386.h (TARGET_USE_GATHER): Define.
1821 * x86-tune.def (X86_TUNE_USE_GATHER): New.
1823 2018-01-10 Martin Liska <mliska@suse.cz>
1826 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
1827 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
1829 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
1830 CLEANUP_NO_PARTITIONING is not set.
1832 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
1834 * doc/rtl.texi: Remove documentation of (const ...) wrappers
1835 for vectors, as a partial revert of r254296.
1836 * rtl.h (const_vec_p): Delete.
1837 (const_vec_duplicate_p): Don't test for vector CONSTs.
1838 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
1839 * expmed.c (make_tree): Likewise.
1842 * common.md (E, F): Use CONSTANT_P instead of checking for
1844 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
1845 checking for CONST_VECTOR.
1847 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1850 * predict.c (force_edge_cold): Handle in more sane way edges
1853 2018-01-09 Carl Love <cel@us.ibm.com>
1855 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
1857 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
1858 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
1859 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
1860 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
1861 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
1862 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
1863 * config/rs6000/rs6000-protos.h: Add extern defition for
1864 rs6000_generate_float2_double_code.
1865 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
1867 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
1868 (float2_v2df): Add define_expand.
1870 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
1873 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
1874 op_mode in the force_to_mode call.
1876 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1878 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
1879 instead of checking each element individually.
1880 (aarch64_evpc_uzp): Likewise.
1881 (aarch64_evpc_zip): Likewise.
1882 (aarch64_evpc_ext): Likewise.
1883 (aarch64_evpc_rev): Likewise.
1884 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
1885 instead of checking each element individually. Return true without
1887 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
1888 whether all selected elements come from the same input, instead of
1889 checking each element individually. Remove calls to gen_rtx_REG,
1890 start_sequence and end_sequence and instead assert that no rtl is
1893 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1895 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
1896 order of HIGH and CONST checks.
1898 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1900 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
1901 if the destination isn't an SSA_NAME.
1903 2018-01-09 Richard Biener <rguenther@suse.de>
1905 PR tree-optimization/83668
1906 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
1908 (canonicalize_loop_form): ... here, renamed from ...
1909 (canonicalize_loop_closed_ssa_form): ... this and amended to
1910 swap successor edges for loop exit blocks to make us use
1911 the RPO order we need for initial schedule generation.
1913 2018-01-09 Joseph Myers <joseph@codesourcery.com>
1915 PR tree-optimization/64811
1916 * match.pd: When optimizing comparisons with Inf, avoid
1917 introducing or losing exceptions from comparisons with NaN.
1919 2018-01-09 Martin Liska <mliska@suse.cz>
1922 * asan.c (shadow_mem_size): Add gcc_assert.
1924 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
1926 Don't save registers in main().
1929 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
1930 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
1931 * config/avr/avr.c (avr_set_current_function): Don't error if
1932 naked, OS_task or OS_main are specified at the same time.
1933 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
1935 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
1937 * common/config/avr/avr-common.c (avr_option_optimization_table):
1938 Switch on -mmain-is-OS_task for optimizing compilations.
1940 2018-01-09 Richard Biener <rguenther@suse.de>
1942 PR tree-optimization/83572
1943 * graphite.c: Include cfganal.h.
1944 (graphite_transform_loops): Connect infinite loops to exit
1945 and remove fake edges at the end.
1947 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1949 * ipa-inline.c (edge_badness): Revert accidental checkin.
1951 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1954 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
1955 symbols; not inline clones.
1957 2018-01-09 Jakub Jelinek <jakub@redhat.com>
1960 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
1961 hard registers. Formatting fixes.
1963 PR preprocessor/83722
1964 * gcc.c (try_generate_repro): Pass
1965 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
1966 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
1969 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
1970 Kito Cheng <kito.cheng@gmail.com>
1972 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
1973 (riscv_leaf_function_p): Delete.
1974 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
1976 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1978 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
1980 (do_ifelse): New function.
1981 (do_isel): New function.
1982 (do_sub3): New function.
1983 (do_add3): New function.
1984 (do_load_mask_compare): New function.
1985 (do_overlap_load_compare): New function.
1986 (expand_compare_loop): New function.
1987 (expand_block_compare): Call expand_compare_loop() when appropriate.
1988 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
1990 (-mblock-compare-inline-loop-limit): New option.
1992 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1995 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
1996 Reverse order of second and third operands in first alternative.
1997 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
1998 of first and second elements in UNSPEC_VPERMR vector.
1999 (altivec_expand_vec_perm_le): Likewise.
2001 2017-01-08 Jeff Law <law@redhat.com>
2003 PR rtl-optimizatin/81308
2004 * tree-switch-conversion.c (cfg_altered): New file scoped static.
2005 (process_switch): If group_case_labels makes a change, then set
2007 (pass_convert_switch::execute): If a switch is converted, then
2008 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
2010 PR rtl-optimization/81308
2011 * recog.c (split_all_insns): Conditionally cleanup the CFG after
2014 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
2016 PR target/83663 - Revert r255946
2017 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
2018 generation for cases where splatting a value is not useful.
2019 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
2020 across a vec_duplicate and a paradoxical subreg forming a vector
2021 mode to a vec_concat.
2023 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2025 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
2026 -march=armv8.3-a variants.
2027 * config/arm/t-multilib: Likewise.
2028 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
2030 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2032 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
2034 (cceq_ior_compare_complement): Give it a name so I can use it, and
2035 change boolean_or_operator predicate to boolean_operator so it can
2036 be used to generate a crand.
2037 (eqne): New code iterator.
2038 (bd/bd_neg): New code_attrs.
2039 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
2040 a single define_insn.
2041 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
2042 decrement (bdnzt/bdnzf/bdzt/bdzf).
2043 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
2044 with the new names of the branch decrement patterns, and added the
2045 names of the branch decrement conditional patterns.
2047 2018-01-08 Richard Biener <rguenther@suse.de>
2049 PR tree-optimization/83563
2050 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
2053 2018-01-08 Richard Biener <rguenther@suse.de>
2056 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
2058 2018-01-08 Richard Biener <rguenther@suse.de>
2060 PR tree-optimization/83685
2061 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
2062 references to abnormals.
2064 2018-01-08 Richard Biener <rguenther@suse.de>
2067 * dwarf2out.c (output_indirect_strings): Handle empty
2068 skeleton_debug_str_hash.
2069 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
2071 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2073 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
2074 (emit_store_direct): Likewise.
2075 (arc_trampoline_adjust_address): Likewise.
2076 (arc_asm_trampoline_template): New function.
2077 (arc_initialize_trampoline): Use asm_trampoline_template.
2078 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
2079 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
2080 * config/arc/arc.md (flush_icache): Delete pattern.
2082 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2084 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
2085 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
2088 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2091 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
2092 by not USED_FOR_TARGET.
2093 (make_pass_resolve_sw_modes): Likewise.
2095 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2097 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
2100 2018-01-08 Richard Biener <rguenther@suse.de>
2103 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
2105 2018-01-08 Richard Biener <rguenther@suse.de>
2108 * match.pd ((t * 2) / 2) -> t): Add missing :c.
2110 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
2113 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
2114 basic blocks with a small number of successors.
2115 (convert_control_dep_chain_into_preds): Improve handling of
2117 (dump_predicates): Split apart into...
2118 (dump_pred_chain): ...here...
2119 (dump_pred_info): ...and here.
2120 (can_one_predicate_be_invalidated_p): Add debugging printfs.
2121 (can_chain_union_be_invalidated_p): Improve check for invalidation
2123 (uninit_uses_cannot_happen): Avoid unnecessary if
2124 convert_control_dep_chain_into_preds yielded nothing.
2126 2018-01-06 Martin Sebor <msebor@redhat.com>
2128 PR tree-optimization/83640
2129 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
2130 subtracting negative offset from size.
2131 (builtin_access::overlap): Adjust offset bounds of the access to fall
2132 within the size of the object if possible.
2134 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
2136 PR rtl-optimization/83699
2137 * expmed.c (extract_bit_field_1): Restrict the vector usage of
2138 extract_bit_field_as_subreg to cases in which the extracted
2139 value is also a vector.
2141 * lra-constraints.c (process_alt_operands): Test for the equivalence
2142 substitutions when detecting a possible reload cycle.
2144 2018-01-06 Jakub Jelinek <jakub@redhat.com>
2147 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
2148 by default if flag_selective_schedling{,2}. Formatting fixes.
2150 PR rtl-optimization/83682
2151 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
2152 if it has non-VECTOR_MODE element mode.
2153 (vec_duplicate_p): Likewise.
2156 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
2157 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
2159 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2162 * config/i386/i386-builtin.def
2163 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
2164 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
2165 Require also OPTION_MASK_ISA_AVX512F in addition to
2166 OPTION_MASK_ISA_GFNI.
2167 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
2168 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
2169 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
2170 to OPTION_MASK_ISA_GFNI.
2171 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
2172 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
2173 OPTION_MASK_ISA_AVX512BW.
2174 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
2175 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
2176 addition to OPTION_MASK_ISA_GFNI.
2177 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
2178 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
2179 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
2180 to OPTION_MASK_ISA_GFNI.
2181 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
2182 a requirement for all ISAs rather than any of them with a few
2184 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
2186 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
2187 bitmasks to be enabled with 3 exceptions, instead of requiring any
2188 enabled ISA with lots of exceptions.
2189 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
2190 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
2191 Change avx512bw in isa attribute to avx512f.
2192 * config/i386/sgxintrin.h: Add license boilerplate.
2193 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
2194 to __AVX512F__ and __AVX512VL to __AVX512VL__.
2195 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
2196 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
2198 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
2199 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
2200 temporarily sse2 rather than sse if not enabled already.
2203 * config/i386/sse.md (VI248_VLBW): Rename to ...
2204 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
2205 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
2206 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
2207 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
2208 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
2209 mode iterator instead of VI248_VLBW.
2211 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
2213 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
2214 (record_modified): Skip clobbers; add debug output.
2215 (param_change_prob): Use sreal frequencies.
2217 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2219 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
2220 punt for user-aligned variables.
2222 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2224 * tree-chrec.c (chrec_contains_symbols): Return true for
2227 2018-01-05 Sudakshina Das <sudi.das@arm.com>
2230 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
2231 of (x|y) == x for BICS pattern.
2233 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2235 PR tree-optimization/83605
2236 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
2237 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
2240 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
2242 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
2243 * config/epiphany/rtems.h: New file.
2245 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2246 Uros Bizjak <ubizjak@gmail.com>
2249 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
2250 QIreg_operand instead of register_operand predicate.
2251 * config/i386/i386.c (ix86_rop_should_change_byte_p,
2252 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
2253 comments instead of -fmitigate[-_]rop.
2255 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2258 * cgraphunit.c (symbol_table::compile): Switch to text_section
2259 before calling assembly_start debug hook.
2260 * run-rtl-passes.c (run_rtl_passes): Likewise.
2263 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2265 * tree-vrp.c (extract_range_from_binary_expr_1): Check
2266 range_int_cst_p rather than !symbolic_range_p before calling
2267 extract_range_from_multiplicative_op_1.
2269 2017-01-04 Jeff Law <law@redhat.com>
2271 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
2272 redundant test in assertion.
2274 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2276 * doc/rtl.texi: Document machine_mode wrapper classes.
2278 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2280 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
2283 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2285 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
2286 the VEC_PERM_EXPR fold to fail.
2288 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2291 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
2292 to switched_sections.
2294 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2297 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
2300 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
2303 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
2304 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
2306 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2309 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
2310 is BLKmode and bitpos not zero or mode change is needed.
2312 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2315 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
2318 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
2321 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
2322 instead of MULT rtx. Update all corresponding splitters.
2324 (*ssub<modesuffix>): Ditto.
2326 (*cmp_sadd_di): Update split patterns.
2327 (*cmp_sadd_si): Ditto.
2328 (*cmp_sadd_sidi): Ditto.
2329 (*cmp_ssub_di): Ditto.
2330 (*cmp_ssub_si): Ditto.
2331 (*cmp_ssub_sidi): Ditto.
2332 * config/alpha/predicates.md (const23_operand): New predicate.
2333 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
2334 Look for ASHIFT, not MULT inner operand.
2335 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
2337 2018-01-04 Martin Liska <mliska@suse.cz>
2339 PR gcov-profile/83669
2340 * gcov.c (output_intermediate_file): Add version to intermediate
2342 * doc/gcov.texi: Document new field 'version' in intermediate
2343 file format. Fix location of '-k' option of gcov command.
2345 2018-01-04 Martin Liska <mliska@suse.cz>
2348 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
2350 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2352 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
2354 2018-01-03 Martin Sebor <msebor@redhat.com>
2356 PR tree-optimization/83655
2357 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
2358 checking calls with invalid arguments.
2360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2362 * tree-vect-stmts.c (vect_get_store_rhs): New function.
2363 (vectorizable_mask_load_store): Delete.
2364 (vectorizable_call): Return false for masked loads and stores.
2365 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
2366 instead of gimple_assign_rhs1.
2367 (vectorizable_load): Handle IFN_MASK_LOAD.
2368 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
2370 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2372 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
2374 (vectorizable_mask_load_store): ...here.
2375 (vectorizable_load): ...and here.
2377 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2379 * tree-vect-stmts.c (vect_build_all_ones_mask)
2380 (vect_build_zero_merge_argument): New functions, split out from...
2381 (vectorizable_load): ...here.
2383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2385 * tree-vect-stmts.c (vect_check_store_rhs): New function,
2387 (vectorizable_mask_load_store): ...here.
2388 (vectorizable_store): ...and here.
2390 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2392 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
2394 (vectorizable_mask_load_store): ...here.
2396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2398 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
2399 (vect_model_store_cost): Take a vec_load_store_type instead of a
2401 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
2402 (vect_model_store_cost): Take a vec_load_store_type instead of a
2404 (vectorizable_mask_load_store): Update accordingly.
2405 (vectorizable_store): Likewise.
2406 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
2408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2410 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
2411 IFN_MASK_LOAD calls here rather than...
2412 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
2414 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2415 Alan Hayward <alan.hayward@arm.com>
2416 David Sherwood <david.sherwood@arm.com>
2418 * expmed.c (extract_bit_field_1): For vector extracts,
2419 fall back to extract_bit_field_as_subreg if vec_extract
2422 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2423 Alan Hayward <alan.hayward@arm.com>
2424 David Sherwood <david.sherwood@arm.com>
2426 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
2427 they are variable or constant sized.
2428 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
2429 slots for constant-sized data.
2431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2432 Alan Hayward <alan.hayward@arm.com>
2433 David Sherwood <david.sherwood@arm.com>
2435 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
2436 handling COND_EXPRs with boolean comparisons, try to find a better
2437 basis for the mask type than the boolean itself.
2439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2441 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
2442 is calculated and how it can be overridden.
2443 * genmodes.c (max_bitsize_mode_any_mode): New variable.
2444 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
2446 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
2449 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2450 Alan Hayward <alan.hayward@arm.com>
2451 David Sherwood <david.sherwood@arm.com>
2453 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
2454 Remove the mode argument.
2455 (aarch64_simd_valid_immediate): Remove the mode and inverse
2457 * config/aarch64/iterators.md (bitsize): New iterator.
2458 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
2459 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
2460 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
2461 aarch64_simd_valid_immediate.
2462 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
2463 (aarch64_reg_or_bic_imm): Likewise.
2464 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
2465 with an insn_type enum and msl with a modifier_type enum.
2466 Replace element_width with a scalar_mode. Change the shift
2467 to unsigned int. Add constructors for scalar_float_mode and
2468 scalar_int_mode elements.
2469 (aarch64_vect_float_const_representable_p): Delete.
2470 (aarch64_can_const_movi_rtx_p)
2471 (aarch64_simd_scalar_immediate_valid_for_move)
2472 (aarch64_simd_make_constant): Update call to
2473 aarch64_simd_valid_immediate.
2474 (aarch64_advsimd_valid_immediate_hs): New function.
2475 (aarch64_advsimd_valid_immediate): Likewise.
2476 (aarch64_simd_valid_immediate): Remove mode and inverse
2477 arguments. Rewrite to use the above. Use const_vec_duplicate_p
2478 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
2479 and aarch64_float_const_representable_p on the result.
2480 (aarch64_output_simd_mov_immediate): Remove mode argument.
2481 Update call to aarch64_simd_valid_immediate and use of
2482 simd_immediate_info.
2483 (aarch64_output_scalar_simd_mov_immediate): Update call
2486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2487 Alan Hayward <alan.hayward@arm.com>
2488 David Sherwood <david.sherwood@arm.com>
2490 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
2491 (mode_nunits): Likewise CONST_MODE_NUNITS.
2492 * machmode.def (ADJUST_NUNITS): Document.
2493 * genmodes.c (mode_data::need_nunits_adj): New field.
2494 (blank_mode): Update accordingly.
2495 (adj_nunits): New variable.
2496 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
2498 (emit_mode_size_inline): Set need_bytesize_adj for all modes
2499 listed in adj_nunits.
2500 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
2501 listed in adj_nunits. Don't emit case statements for such modes.
2502 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
2503 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
2504 nothing if adj_nunits is nonnull.
2505 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
2506 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
2507 (emit_mode_fbit): Update use of print_maybe_const_decl.
2508 (emit_move_size): Likewise. Treat the array as non-const
2510 (emit_mode_adjustments): Handle adj_nunits.
2512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2514 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
2515 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
2516 (VECTOR_MODES): Use it.
2517 (make_vector_modes): Take the prefix as an argument.
2519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2520 Alan Hayward <alan.hayward@arm.com>
2521 David Sherwood <david.sherwood@arm.com>
2523 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
2524 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
2525 for MODE_VECTOR_BOOL.
2526 * machmode.def (VECTOR_BOOL_MODE): Document.
2527 * genmodes.c (VECTOR_BOOL_MODE): New macro.
2528 (make_vector_bool_mode): New function.
2529 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
2531 * lto-streamer-in.c (lto_input_mode_table): Likewise.
2532 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
2534 * stor-layout.c (int_mode_for_mode): Likewise.
2535 * tree.c (build_vector_type_for_mode): Likewise.
2536 * varasm.c (output_constant_pool_2): Likewise.
2537 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
2538 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
2539 for MODE_VECTOR_BOOL.
2540 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
2541 of mode class checks.
2542 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
2543 instead of a list of mode class checks.
2544 (expand_vector_scalar_condition): Likewise.
2545 (type_for_widest_vector_mode): Handle BImode as an inner mode.
2547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2548 Alan Hayward <alan.hayward@arm.com>
2549 David Sherwood <david.sherwood@arm.com>
2551 * machmode.h (mode_size): Change from unsigned short to
2553 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
2554 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
2555 or if measurement_type is not polynomial.
2556 (fixed_size_mode::includes_p): Check for constant-sized modes.
2557 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
2558 return a poly_uint16 rather than an unsigned short.
2559 (emit_mode_size): Change the type of mode_size from unsigned short
2560 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
2561 (emit_mode_adjustments): Cope with polynomial vector sizes.
2562 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2564 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2566 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
2567 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
2568 * caller-save.c (setup_save_areas): Likewise.
2569 (replace_reg_with_saved_mem): Likewise.
2570 * calls.c (emit_library_call_value_1): Likewise.
2571 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
2572 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
2573 (gen_lowpart_for_combine): Likewise.
2574 * convert.c (convert_to_integer_1): Likewise.
2575 * cse.c (equiv_constant, cse_insn): Likewise.
2576 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
2577 (cselib_subst_to_values): Likewise.
2578 * dce.c (word_dce_process_block): Likewise.
2579 * df-problems.c (df_word_lr_mark_ref): Likewise.
2580 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
2581 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
2582 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
2583 (rtl_for_decl_location): Likewise.
2584 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
2585 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
2586 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
2587 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
2588 (expand_expr_real_1): Likewise.
2589 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
2590 (pad_below): Likewise.
2591 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
2592 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
2593 * ira.c (get_subreg_tracking_sizes): Likewise.
2594 * ira-build.c (ira_create_allocno_objects): Likewise.
2595 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
2596 (ira_sort_regnos_for_alter_reg): Likewise.
2597 * ira-costs.c (record_operand_costs): Likewise.
2598 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
2599 (resolve_simple_move): Likewise.
2600 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
2601 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
2602 (lra_constraints): Likewise.
2603 (CONST_POOL_OK_P): Reject variable-sized modes.
2604 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
2605 (add_pseudo_to_slot, lra_spill): Likewise.
2606 * omp-low.c (omp_clause_aligned_alignment): Likewise.
2607 * optabs-query.c (get_best_extraction_insn): Likewise.
2608 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
2609 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
2610 (expand_mult_highpart, valid_multiword_target_p): Likewise.
2611 * recog.c (offsettable_address_addr_space_p): Likewise.
2612 * regcprop.c (maybe_mode_change): Likewise.
2613 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
2614 * regrename.c (build_def_use): Likewise.
2615 * regstat.c (dump_reg_info): Likewise.
2616 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
2617 (find_reloads, find_reloads_subreg_address): Likewise.
2618 * reload1.c (eliminate_regs_1): Likewise.
2619 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
2620 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
2621 (simplify_binary_operation_1, simplify_subreg): Likewise.
2622 * targhooks.c (default_function_arg_padding): Likewise.
2623 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
2624 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
2625 (verify_gimple_assign_ternary): Likewise.
2626 * tree-inline.c (estimate_move_cost): Likewise.
2627 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2628 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
2629 (get_address_cost_ainc): Likewise.
2630 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
2631 (vect_supportable_dr_alignment): Likewise.
2632 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
2633 (vectorizable_reduction): Likewise.
2634 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
2635 (vectorizable_operation, vectorizable_load): Likewise.
2636 * tree.c (build_same_sized_truth_vector_type): Likewise.
2637 * valtrack.c (cleanup_auto_inc_dec): Likewise.
2638 * var-tracking.c (emit_note_insn_var_location): Likewise.
2639 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
2640 (ADDR_VEC_ALIGN): Likewise.
2642 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2643 Alan Hayward <alan.hayward@arm.com>
2644 David Sherwood <david.sherwood@arm.com>
2646 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
2648 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
2649 or if measurement_type is polynomial.
2650 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
2651 * combine.c (make_extraction): Likewise.
2652 * dse.c (find_shift_sequence): Likewise.
2653 * dwarf2out.c (mem_loc_descriptor): Likewise.
2654 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
2655 (extract_bit_field, extract_low_bits): Likewise.
2656 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
2657 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
2658 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
2659 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
2660 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
2661 * reload.c (find_reloads): Likewise.
2662 * reload1.c (alter_reg): Likewise.
2663 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
2664 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
2665 * tree-if-conv.c (predicate_mem_writes): Likewise.
2666 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
2667 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
2668 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
2669 * valtrack.c (dead_debug_insert_temp): Likewise.
2670 * varasm.c (mergeable_constant_section): Likewise.
2671 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
2673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2674 Alan Hayward <alan.hayward@arm.com>
2675 David Sherwood <david.sherwood@arm.com>
2677 * expr.c (expand_assignment): Cope with polynomial mode sizes
2678 when assigning to a CONCAT.
2680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2681 Alan Hayward <alan.hayward@arm.com>
2682 David Sherwood <david.sherwood@arm.com>
2684 * machmode.h (mode_precision): Change from unsigned short to
2686 (mode_to_precision): Return a poly_uint16 rather than an unsigned
2688 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
2689 or if measurement_type is not polynomial.
2690 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
2691 in which the mode is already known to be a scalar_int_mode.
2692 * genmodes.c (emit_mode_precision): Change the type of mode_precision
2693 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
2695 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2696 for GET_MODE_PRECISION.
2697 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2698 for GET_MODE_PRECISION.
2699 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
2701 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
2702 (expand_field_assignment, make_extraction): Likewise.
2703 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
2704 (get_last_value): Likewise.
2705 * convert.c (convert_to_integer_1): Likewise.
2706 * cse.c (cse_insn): Likewise.
2707 * expr.c (expand_expr_real_1): Likewise.
2708 * lra-constraints.c (simplify_operand_subreg): Likewise.
2709 * optabs-query.c (can_atomic_load_p): Likewise.
2710 * optabs.c (expand_atomic_load): Likewise.
2711 (expand_atomic_store): Likewise.
2712 * ree.c (combine_reaching_defs): Likewise.
2713 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
2714 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
2715 * tree.h (type_has_mode_precision_p): Likewise.
2716 * ubsan.c (instrument_si_overflow): Likewise.
2718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2719 Alan Hayward <alan.hayward@arm.com>
2720 David Sherwood <david.sherwood@arm.com>
2722 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
2723 polynomial numbers of units.
2724 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
2725 (valid_vector_subparts_p): New function.
2726 (build_vector_type): Remove temporary shim and take the number
2727 of units as a poly_uint64 rather than an int.
2728 (build_opaque_vector_type): Take the number of units as a
2729 poly_uint64 rather than an int.
2730 * tree.c (build_vector_from_ctor): Handle polynomial
2731 TYPE_VECTOR_SUBPARTS.
2732 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
2733 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
2734 (build_vector_from_val): If the number of units is variable,
2735 use build_vec_duplicate_cst for constant operands and
2736 VEC_DUPLICATE_EXPR otherwise.
2737 (make_vector_type): Remove temporary is_constant ().
2738 (build_vector_type, build_opaque_vector_type): Take the number of
2739 units as a poly_uint64 rather than an int.
2740 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
2742 * cfgexpand.c (expand_debug_expr): Likewise.
2743 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
2744 (store_constructor, expand_expr_real_1): Likewise.
2745 (const_scalar_mask_from_tree): Likewise.
2746 * fold-const-call.c (fold_const_reduction): Likewise.
2747 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
2748 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
2749 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
2750 (fold_relational_const): Likewise.
2751 (native_interpret_vector): Likewise. Change the size from an
2752 int to an unsigned int.
2753 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
2754 TYPE_VECTOR_SUBPARTS.
2755 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
2756 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
2757 duplicating a non-constant operand into a variable-length vector.
2758 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
2759 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
2760 * ipa-icf.c (sem_variable::equals): Likewise.
2761 * match.pd: Likewise.
2762 * omp-simd-clone.c (simd_clone_subparts): Likewise.
2763 * print-tree.c (print_node): Likewise.
2764 * stor-layout.c (layout_type): Likewise.
2765 * targhooks.c (default_builtin_vectorization_cost): Likewise.
2766 * tree-cfg.c (verify_gimple_comparison): Likewise.
2767 (verify_gimple_assign_binary): Likewise.
2768 (verify_gimple_assign_ternary): Likewise.
2769 (verify_gimple_assign_single): Likewise.
2770 * tree-pretty-print.c (dump_generic_node): Likewise.
2771 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2772 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
2773 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
2774 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
2775 (vect_shift_permute_load_chain): Likewise.
2776 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
2777 (expand_vector_condition, optimize_vector_constructor): Likewise.
2778 (lower_vec_perm, get_compute_type): Likewise.
2779 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
2780 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
2781 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
2782 (vect_recog_mask_conversion_pattern): Likewise.
2783 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
2784 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
2785 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2786 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
2787 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
2788 (vectorizable_shift, vectorizable_operation, vectorizable_store)
2789 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
2790 (supportable_widening_operation): Likewise.
2791 (supportable_narrowing_operation): Likewise.
2792 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
2794 * varasm.c (output_constant): Likewise.
2796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2797 Alan Hayward <alan.hayward@arm.com>
2798 David Sherwood <david.sherwood@arm.com>
2800 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
2801 so that both the length == 3 and length != 3 cases set up their
2802 own permute vectors. Add comments explaining why we know the
2803 number of elements is constant.
2804 (vect_permute_load_chain): Likewise.
2806 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2807 Alan Hayward <alan.hayward@arm.com>
2808 David Sherwood <david.sherwood@arm.com>
2810 * machmode.h (mode_nunits): Change from unsigned char to
2812 (ONLY_FIXED_SIZE_MODES): New macro.
2813 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
2814 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
2815 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
2817 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
2818 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
2819 or if measurement_type is not polynomial.
2820 * genmodes.c (ZERO_COEFFS): New macro.
2821 (emit_mode_nunits_inline): Make mode_nunits_inline return a
2823 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
2824 Use ZERO_COEFFS when emitting initializers.
2825 * data-streamer.h (bp_pack_poly_value): New function.
2826 (bp_unpack_poly_value): Likewise.
2827 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2828 for GET_MODE_NUNITS.
2829 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2830 for GET_MODE_NUNITS.
2831 * tree.c (make_vector_type): Remove temporary shim and make
2832 the real function take the number of units as a poly_uint64
2834 (build_vector_type_for_mode): Handle polynomial nunits.
2835 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
2836 * emit-rtl.c (const_vec_series_p_1): Likewise.
2837 (gen_rtx_CONST_VECTOR): Likewise.
2838 * fold-const.c (test_vec_duplicate_folding): Likewise.
2839 * genrecog.c (validate_pattern): Likewise.
2840 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
2841 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
2842 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
2843 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
2844 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
2845 * rtlanal.c (subreg_get_info): Likewise.
2846 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2847 (vect_grouped_load_supported): Likewise.
2848 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
2849 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
2850 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
2851 (simplify_const_unary_operation, simplify_binary_operation_1)
2852 (simplify_const_binary_operation, simplify_ternary_operation)
2853 (test_vector_ops_duplicate, test_vector_ops): Likewise.
2854 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
2855 instead of CONST_VECTOR_NUNITS.
2856 * varasm.c (output_constant_pool_2): Likewise.
2857 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
2858 explicit-encoded elements in the XVEC for variable-length vectors.
2860 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2862 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
2864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2865 Alan Hayward <alan.hayward@arm.com>
2866 David Sherwood <david.sherwood@arm.com>
2868 * coretypes.h (fixed_size_mode): Declare.
2869 (fixed_size_mode_pod): New typedef.
2870 * builtins.h (target_builtins::x_apply_args_mode)
2871 (target_builtins::x_apply_result_mode): Change type to
2872 fixed_size_mode_pod.
2873 * builtins.c (apply_args_size, apply_result_size, result_vector)
2874 (expand_builtin_apply_args_1, expand_builtin_apply)
2875 (expand_builtin_return): Update accordingly.
2877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2879 * cse.c (hash_rtx_cb): Hash only the encoded elements.
2880 * cselib.c (cselib_hash_rtx): Likewise.
2881 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
2882 CONST_VECTOR encoding.
2884 2017-01-03 Jakub Jelinek <jakub@redhat.com>
2885 Jeff Law <law@redhat.com>
2888 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
2889 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
2890 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
2891 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
2894 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
2895 explicitly probe *sp in a noreturn function if there were any callee
2896 register saves or frame pointer is needed.
2898 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2901 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
2902 BLKmode for ternary, binary or unary expressions.
2905 * var-tracking.c (delete_vta_debug_insn): New inline function.
2906 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
2907 insns from get_insns () to NULL instead of each bb separately.
2908 Use delete_vta_debug_insn. No longer static.
2909 (vt_debug_insns_local, variable_tracking_main_1): Adjust
2910 delete_vta_debug_insns callers.
2911 * rtl.h (delete_vta_debug_insns): Declare.
2912 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
2913 instead of variable_tracking_main.
2915 2018-01-03 Martin Sebor <msebor@redhat.com>
2917 PR tree-optimization/83603
2918 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
2919 arguments past the endof the argument list in functions declared
2920 without a prototype.
2921 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
2922 Avoid checking when arguments are null.
2924 2018-01-03 Martin Sebor <msebor@redhat.com>
2927 * doc/extend.texi (attribute const): Fix a typo.
2928 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
2929 issuing -Wsuggest-attribute for void functions.
2931 2018-01-03 Martin Sebor <msebor@redhat.com>
2933 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
2934 offset_int::from instead of wide_int::to_shwi.
2935 (maybe_diag_overlap): Remove assertion.
2936 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
2937 * gimple-ssa-sprintf.c (format_directive): Same.
2938 (parse_directive): Same.
2939 (sprintf_dom_walker::compute_format_length): Same.
2940 (try_substitute_return_value): Same.
2942 2017-01-03 Jeff Law <law@redhat.com>
2945 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
2946 non-constant residual for zero at runtime and avoid probing in
2947 that case. Reorganize code for trailing problem to mirror handling
2950 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2952 PR tree-optimization/83501
2953 * tree-ssa-strlen.c (get_string_cst): New.
2954 (handle_char_store): Call get_string_cst.
2956 2018-01-03 Martin Liska <mliska@suse.cz>
2958 PR tree-optimization/83593
2959 * tree-ssa-strlen.c: Include tree-cfg.h.
2960 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
2961 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
2962 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
2964 (strlen_dom_walker::before_dom_children): Call
2965 gimple_purge_dead_eh_edges. Dump tranformation with details
2967 (strlen_dom_walker::before_dom_children): Update call by adding
2968 new argument cleanup_eh.
2969 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
2971 2018-01-03 Martin Liska <mliska@suse.cz>
2974 * cif-code.def (VARIADIC_THUNK): New enum value.
2975 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
2978 2018-01-03 Jan Beulich <jbeulich@suse.com>
2980 * sse.md (mov<mode>_internal): Tighten condition for when to use
2981 vmovdqu<ssescalarsize> for TI and OI modes.
2983 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2985 Update copyright years.
2987 2018-01-03 Martin Liska <mliska@suse.cz>
2990 * ipa-visibility.c (function_and_variable_visibility): Skip
2991 functions with noipa attribure.
2993 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2995 * gcc.c (process_command): Update copyright notice dates.
2996 * gcov-dump.c (print_version): Ditto.
2997 * gcov.c (print_version): Ditto.
2998 * gcov-tool.c (print_version): Ditto.
2999 * gengtype.c (create_file): Ditto.
3000 * doc/cpp.texi: Bump @copying's copyright year.
3001 * doc/cppinternals.texi: Ditto.
3002 * doc/gcc.texi: Ditto.
3003 * doc/gccint.texi: Ditto.
3004 * doc/gcov.texi: Ditto.
3005 * doc/install.texi: Ditto.
3006 * doc/invoke.texi: Ditto.
3008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3010 * vector-builder.h (vector_builder::m_full_nelts): Change from
3011 unsigned int to poly_uint64.
3012 (vector_builder::full_nelts): Update prototype accordingly.
3013 (vector_builder::new_vector): Likewise.
3014 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
3015 (vector_builder::operator ==): Likewise.
3016 (vector_builder::finalize): Likewise.
3017 * int-vector-builder.h (int_vector_builder::int_vector_builder):
3018 Take the number of elements as a poly_uint64 rather than an
3020 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
3021 from unsigned int to poly_uint64.
3022 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
3023 (vec_perm_indices::new_vector): Likewise.
3024 (vec_perm_indices::length): Likewise.
3025 (vec_perm_indices::nelts_per_input): Likewise.
3026 (vec_perm_indices::input_nelts): Likewise.
3027 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
3028 number of elements per input as a poly_uint64 rather than an
3029 unsigned int. Use the original encoding for variable-length
3030 vectors, rather than clamping each individual element.
3031 For the second and subsequent elements in each pattern,
3032 clamp the step and base before clamping their sum.
3033 (vec_perm_indices::series_p): Handle polynomial element counts.
3034 (vec_perm_indices::all_in_range_p): Likewise.
3035 (vec_perm_indices_to_tree): Likewise.
3036 (vec_perm_indices_to_rtx): Likewise.
3037 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
3038 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
3039 (tree_vector_builder::new_binary_operation): Handle polynomial
3040 element counts. Return false if we need to know the number
3041 of elements at compile time.
3042 * fold-const.c (fold_vec_perm): Punt if the number of elements
3043 isn't known at compile time.
3045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3047 * vec-perm-indices.h (vec_perm_builder): Change element type
3048 from HOST_WIDE_INT to poly_int64.
3049 (vec_perm_indices::element_type): Update accordingly.
3050 (vec_perm_indices::clamp): Handle polynomial element_types.
3051 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
3052 (vec_perm_indices::all_in_range_p): Likewise.
3053 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
3055 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
3056 polynomial vec_perm_indices element types.
3057 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
3058 * fold-const.c (fold_vec_perm): Likewise.
3059 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
3060 * tree-vect-generic.c (lower_vec_perm): Likewise.
3061 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
3062 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
3063 element type to HOST_WIDE_INT.
3065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3066 Alan Hayward <alan.hayward@arm.com>
3067 David Sherwood <david.sherwood@arm.com>
3069 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
3070 rather than an int. Use plus_constant.
3071 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
3072 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
3074 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3075 Alan Hayward <alan.hayward@arm.com>
3076 David Sherwood <david.sherwood@arm.com>
3078 * calls.c (emit_call_1, expand_call): Change struct_value_size from
3079 a HOST_WIDE_INT to a poly_int64.
3081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3082 Alan Hayward <alan.hayward@arm.com>
3083 David Sherwood <david.sherwood@arm.com>
3085 * calls.c (load_register_parameters): Cope with polynomial
3086 mode sizes. Require a constant size for BLKmode parameters
3087 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
3088 forces a parameter to be padded at the lsb end in order to
3089 fill a complete number of words, require the parameter size
3090 to be ordered wrt UNITS_PER_WORD.
3092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3093 Alan Hayward <alan.hayward@arm.com>
3094 David Sherwood <david.sherwood@arm.com>
3096 * reload1.c (spill_stack_slot_width): Change element type
3097 from unsigned int to poly_uint64_pod.
3098 (alter_reg): Treat mode sizes as polynomial.
3100 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3101 Alan Hayward <alan.hayward@arm.com>
3102 David Sherwood <david.sherwood@arm.com>
3104 * reload.c (complex_word_subreg_p): New function.
3105 (reload_inner_reg_of_subreg, push_reload): Use it.
3107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3108 Alan Hayward <alan.hayward@arm.com>
3109 David Sherwood <david.sherwood@arm.com>
3111 * lra-constraints.c (process_alt_operands): Reject matched
3112 operands whose sizes aren't ordered.
3113 (match_reload): Refer to this check here.
3115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3116 Alan Hayward <alan.hayward@arm.com>
3117 David Sherwood <david.sherwood@arm.com>
3119 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
3120 that the mode size is in the set {1, 2, 4, 8, 16}.
3122 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3123 Alan Hayward <alan.hayward@arm.com>
3124 David Sherwood <david.sherwood@arm.com>
3126 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
3127 Use plus_constant instead of gen_rtx_PLUS.
3129 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3130 Alan Hayward <alan.hayward@arm.com>
3131 David Sherwood <david.sherwood@arm.com>
3133 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
3134 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
3135 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
3136 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
3137 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
3138 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
3139 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
3140 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
3141 * config/i386/i386.c (ix86_push_rounding): ...this new function.
3142 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
3144 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
3145 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
3146 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
3147 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
3148 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
3149 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
3150 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
3151 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
3152 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
3153 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
3155 * expr.c (emit_move_resolve_push): Treat the input and result
3156 of PUSH_ROUNDING as a poly_int64.
3157 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
3158 (emit_push_insn): Likewise.
3159 * lra-eliminations.c (mark_not_eliminable): Likewise.
3160 * recog.c (push_operand): Likewise.
3161 * reload1.c (elimination_effects): Likewise.
3162 * rtlanal.c (nonzero_bits1): Likewise.
3163 * calls.c (store_one_arg): Likewise. Require the padding to be
3164 known at compile time.
3166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3167 Alan Hayward <alan.hayward@arm.com>
3168 David Sherwood <david.sherwood@arm.com>
3170 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
3171 Use plus_constant instead of gen_rtx_PLUS.
3173 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3174 Alan Hayward <alan.hayward@arm.com>
3175 David Sherwood <david.sherwood@arm.com>
3177 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
3180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3181 Alan Hayward <alan.hayward@arm.com>
3182 David Sherwood <david.sherwood@arm.com>
3184 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
3185 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
3186 via stack temporaries. Treat the mode size as polynomial too.
3188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3189 Alan Hayward <alan.hayward@arm.com>
3190 David Sherwood <david.sherwood@arm.com>
3192 * expr.c (expand_expr_real_2): When handling conversions involving
3193 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
3194 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
3195 as a poly_uint64 too.
3197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3198 Alan Hayward <alan.hayward@arm.com>
3199 David Sherwood <david.sherwood@arm.com>
3201 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
3203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3204 Alan Hayward <alan.hayward@arm.com>
3205 David Sherwood <david.sherwood@arm.com>
3207 * combine.c (can_change_dest_mode): Handle polynomial
3208 REGMODE_NATURAL_SIZE.
3209 * expmed.c (store_bit_field_1): Likewise.
3210 * expr.c (store_constructor): Likewise.
3211 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
3212 and polynomial REGMODE_NATURAL_SIZE.
3213 (gen_lowpart_common): Likewise.
3214 * reginfo.c (record_subregs_of_mode): Likewise.
3215 * rtlanal.c (read_modify_subreg_p): Likewise.
3217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3218 Alan Hayward <alan.hayward@arm.com>
3219 David Sherwood <david.sherwood@arm.com>
3221 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
3222 numbers of elements.
3224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3225 Alan Hayward <alan.hayward@arm.com>
3226 David Sherwood <david.sherwood@arm.com>
3228 * match.pd: Cope with polynomial numbers of vector elements.
3230 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3231 Alan Hayward <alan.hayward@arm.com>
3232 David Sherwood <david.sherwood@arm.com>
3234 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
3235 in a POINTER_PLUS_EXPR.
3237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3238 Alan Hayward <alan.hayward@arm.com>
3239 David Sherwood <david.sherwood@arm.com>
3241 * omp-simd-clone.c (simd_clone_subparts): New function.
3242 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
3243 (ipa_simd_modify_function_body): Likewise.
3245 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3246 Alan Hayward <alan.hayward@arm.com>
3247 David Sherwood <david.sherwood@arm.com>
3249 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
3250 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
3251 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
3252 (expand_vector_condition, vector_element): Likewise.
3253 (subparts_gt): New function.
3254 (get_compute_type): Use subparts_gt.
3255 (count_type_subparts): Delete.
3256 (expand_vector_operations_1): Use subparts_gt instead of
3257 count_type_subparts.
3259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3260 Alan Hayward <alan.hayward@arm.com>
3261 David Sherwood <david.sherwood@arm.com>
3263 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
3264 (vect_compile_time_alias): ...this new function. Do the calculation
3265 on poly_ints rather than trees.
3266 (vect_prune_runtime_alias_test_list): Update call accordingly.
3268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3269 Alan Hayward <alan.hayward@arm.com>
3270 David Sherwood <david.sherwood@arm.com>
3272 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
3274 (vect_schedule_slp_instance): Likewise.
3276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3277 Alan Hayward <alan.hayward@arm.com>
3278 David Sherwood <david.sherwood@arm.com>
3280 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
3281 constant and extern definitions for variable-length vectors.
3282 (vect_get_constant_vectors): Note that the number of units
3283 is known to be constant.
3285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3286 Alan Hayward <alan.hayward@arm.com>
3287 David Sherwood <david.sherwood@arm.com>
3289 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
3290 of units as polynomial. Choose between WIDE and NARROW based
3293 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3294 Alan Hayward <alan.hayward@arm.com>
3295 David Sherwood <david.sherwood@arm.com>
3297 * tree-vect-stmts.c (simd_clone_subparts): New function.
3298 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
3300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3301 Alan Hayward <alan.hayward@arm.com>
3302 David Sherwood <david.sherwood@arm.com>
3304 * tree-vect-stmts.c (vectorizable_call): Treat the number of
3305 vectors as polynomial. Use build_index_vector for
3308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3309 Alan Hayward <alan.hayward@arm.com>
3310 David Sherwood <david.sherwood@arm.com>
3312 * tree-vect-stmts.c (get_load_store_type): Treat the number of
3313 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
3314 for variable-length vectors.
3315 (vectorizable_mask_load_store): Treat the number of units as
3316 polynomial, asserting that it is constant if the condition has
3317 already been enforced.
3318 (vectorizable_store, vectorizable_load): Likewise.
3320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3321 Alan Hayward <alan.hayward@arm.com>
3322 David Sherwood <david.sherwood@arm.com>
3324 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
3325 of units as polynomial. Punt if we can't tell at compile time
3326 which vector contains the final result.
3328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3329 Alan Hayward <alan.hayward@arm.com>
3330 David Sherwood <david.sherwood@arm.com>
3332 * tree-vect-loop.c (vectorizable_induction): Treat the number
3333 of units as polynomial. Punt on SLP inductions. Use an integer
3334 VEC_SERIES_EXPR for variable-length integer reductions. Use a
3335 cast of such a series for variable-length floating-point
3338 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3339 Alan Hayward <alan.hayward@arm.com>
3340 David Sherwood <david.sherwood@arm.com>
3342 * tree.h (build_index_vector): Declare.
3343 * tree.c (build_index_vector): New function.
3344 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
3345 of units as polynomial, forcibly converting it to a constant if
3346 vectorizable_reduction has already enforced the condition.
3347 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
3348 to create a {1,2,3,...} vector.
3349 (vectorizable_reduction): Treat the number of units as polynomial.
3350 Choose vectype_in based on the largest scalar element size rather
3351 than the smallest number of units. Enforce the restrictions
3354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3355 Alan Hayward <alan.hayward@arm.com>
3356 David Sherwood <david.sherwood@arm.com>
3358 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
3359 number of units as polynomial.
3361 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3362 Alan Hayward <alan.hayward@arm.com>
3363 David Sherwood <david.sherwood@arm.com>
3365 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
3366 * target.def (autovectorize_vector_sizes): Return the vector sizes
3367 by pointer, using vector_sizes rather than a bitmask.
3368 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
3369 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
3370 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
3372 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
3373 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
3374 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
3375 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
3376 * omp-general.c (omp_max_vf): Likewise.
3377 * omp-low.c (omp_clause_aligned_alignment): Likewise.
3378 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
3379 * tree-vect-loop.c (vect_analyze_loop): Likewise.
3380 * tree-vect-slp.c (vect_slp_bb): Likewise.
3381 * doc/tm.texi: Regenerate.
3382 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
3384 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
3385 the vector size as a poly_uint64 rather than an unsigned int.
3386 (current_vector_size): Change from an unsigned int to a poly_uint64.
3387 (get_vectype_for_scalar_type): Update accordingly.
3388 * tree.h (build_truth_vector_type): Take the size and number of
3389 units as a poly_uint64 rather than an unsigned int.
3390 (build_vector_type): Add a temporary overload that takes
3391 the number of units as a poly_uint64 rather than an unsigned int.
3392 * tree.c (make_vector_type): Likewise.
3393 (build_truth_vector_type): Take the number of units as a poly_uint64
3394 rather than an unsigned int.
3396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3397 Alan Hayward <alan.hayward@arm.com>
3398 David Sherwood <david.sherwood@arm.com>
3400 * target.def (get_mask_mode): Take the number of units and length
3401 as poly_uint64s rather than unsigned ints.
3402 * targhooks.h (default_get_mask_mode): Update accordingly.
3403 * targhooks.c (default_get_mask_mode): Likewise.
3404 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
3405 * doc/tm.texi: Regenerate.
3407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3408 Alan Hayward <alan.hayward@arm.com>
3409 David Sherwood <david.sherwood@arm.com>
3411 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
3412 * omp-general.c (omp_max_vf): Likewise.
3413 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
3414 (expand_omp_simd): Handle polynomial safelen.
3415 * omp-low.c (omplow_simd_context): Add a default constructor.
3416 (omplow_simd_context::max_vf): Change from int to poly_uint64.
3417 (lower_rec_simd_input_clauses): Update accordingly.
3418 (lower_rec_input_clauses): Likewise.
3420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3421 Alan Hayward <alan.hayward@arm.com>
3422 David Sherwood <david.sherwood@arm.com>
3424 * tree-vectorizer.h (vect_nunits_for_cost): New function.
3425 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
3426 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
3427 (vect_analyze_slp_cost): Likewise.
3428 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
3429 (vect_model_load_cost): Likewise.
3431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3432 Alan Hayward <alan.hayward@arm.com>
3433 David Sherwood <david.sherwood@arm.com>
3435 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
3436 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
3437 from an unsigned int * to a poly_uint64_pod *.
3438 (calculate_unrolling_factor): New function.
3439 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
3441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3442 Alan Hayward <alan.hayward@arm.com>
3443 David Sherwood <david.sherwood@arm.com>
3445 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
3446 from an unsigned int to a poly_uint64.
3447 (_loop_vec_info::slp_unrolling_factor): Likewise.
3448 (_loop_vec_info::vectorization_factor): Change from an int
3450 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
3451 (vect_get_num_vectors): New function.
3452 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
3453 (vect_get_num_copies): Use vect_get_num_vectors.
3454 (vect_analyze_data_ref_dependences): Change max_vf from an int *
3455 to an unsigned int *.
3456 (vect_analyze_data_refs): Change min_vf from an int * to a
3458 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
3459 than an unsigned HOST_WIDE_INT.
3460 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
3461 (vect_analyze_data_ref_dependence): Change max_vf from an int *
3462 to an unsigned int *.
3463 (vect_analyze_data_ref_dependences): Likewise.
3464 (vect_compute_data_ref_alignment): Handle polynomial vf.
3465 (vect_enhance_data_refs_alignment): Likewise.
3466 (vect_prune_runtime_alias_test_list): Likewise.
3467 (vect_shift_permute_load_chain): Likewise.
3468 (vect_supportable_dr_alignment): Likewise.
3469 (dependence_distance_ge_vf): Take the vectorization factor as a
3470 poly_uint64 rather than an unsigned HOST_WIDE_INT.
3471 (vect_analyze_data_refs): Change min_vf from an int * to a
3473 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
3474 vfm1 as a poly_uint64 rather than an int. Make the same change
3475 for the returned bound_scalar.
3476 (vect_gen_vector_loop_niters): Handle polynomial vf.
3477 (vect_do_peeling): Likewise. Update call to
3478 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
3479 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
3481 * tree-vect-loop.c (vect_determine_vectorization_factor)
3482 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
3483 (vect_get_known_peeling_cost): Likewise.
3484 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
3485 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
3486 (vect_transform_loop): Likewise. Use the lowest possible VF when
3487 updating the upper bounds of the loop.
3488 (vect_min_worthwhile_factor): Make static. Return an unsigned int
3490 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
3491 polynomial unroll factors.
3492 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
3493 (vect_make_slp_decision): Likewise.
3494 (vect_supported_load_permutation_p): Likewise, and polynomial
3496 (vect_analyze_slp_cost): Handle polynomial vf.
3497 (vect_slp_analyze_node_operations): Likewise.
3498 (vect_slp_analyze_bb_1): Likewise.
3499 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
3500 than an unsigned HOST_WIDE_INT.
3501 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
3502 (vectorizable_load): Handle polynomial vf.
3503 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
3505 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
3507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3508 Alan Hayward <alan.hayward@arm.com>
3509 David Sherwood <david.sherwood@arm.com>
3511 * match.pd: Handle bit operations involving three constants
3512 and try to fold one pair.
3514 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3516 * tree-vect-loop-manip.c: Include gimple-fold.h.
3517 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
3518 niters_maybe_zero parameters. Handle other cases besides a step of 1.
3519 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
3520 Add a path that uses a step of VF instead of 1, but disable it
3522 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
3523 and niters_no_overflow parameters. Update calls to
3524 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
3525 Create a new SSA name if the latter choses to use a ste other
3526 than zero, and return it via niters_vector_mult_vf_var.
3527 * tree-vect-loop.c (vect_transform_loop): Update calls to
3528 vect_do_peeling, vect_gen_vector_loop_niters and
3529 slpeel_make_loop_iterate_ntimes.
3530 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
3531 (vect_gen_vector_loop_niters): Update declarations after above changes.
3533 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
3535 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
3536 128-bit round to integer instructions.
3537 (ceil<mode>2): Likewise.
3538 (btrunc<mode>2): Likewise.
3539 (round<mode>2): Likewise.
3541 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3543 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
3544 unaligned VSX load/store on P8/P9.
3545 (expand_block_clear): Allow the use of unaligned VSX
3546 load/store on P8/P9.
3548 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3550 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
3552 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
3553 swap associated with both a load and a store.
3555 2018-01-02 Andrew Waterman <andrew@sifive.com>
3557 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
3558 * config/riscv/riscv.md (clear_cache): Use it.
3560 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
3562 * web.c: Remove out-of-date comment.
3564 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3566 * expr.c (fixup_args_size_notes): Check that any existing
3567 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
3568 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
3569 (emit_single_push_insn): ...here.
3571 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3573 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
3574 (const_vector_encoded_nelts): New function.
3575 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
3576 (const_vector_int_elt, const_vector_elt): Declare.
3577 * emit-rtl.c (const_vector_int_elt_1): New function.
3578 (const_vector_elt): Likewise.
3579 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
3580 of CONST_VECTOR_ELT.
3582 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3584 * expr.c: Include rtx-vector-builder.h.
3585 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
3586 directly on the tree encoding.
3587 (const_vector_from_tree): Likewise.
3588 * optabs.c: Include rtx-vector-builder.h.
3589 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
3590 sequence of "u" values.
3591 * vec-perm-indices.c: Include rtx-vector-builder.h.
3592 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
3593 directly on the vec_perm_indices encoding.
3595 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3597 * doc/rtl.texi (const_vector): Describe new encoding scheme.
3598 * Makefile.in (OBJS): Add rtx-vector-builder.o.
3599 * rtx-vector-builder.h: New file.
3600 * rtx-vector-builder.c: Likewise.
3601 * rtl.h (rtx_def::u2): Add a const_vector field.
3602 (CONST_VECTOR_NPATTERNS): New macro.
3603 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
3604 (CONST_VECTOR_DUPLICATE_P): Likewise.
3605 (CONST_VECTOR_STEPPED_P): Likewise.
3606 (CONST_VECTOR_ENCODED_ELT): Likewise.
3607 (const_vec_duplicate_p): Check for a duplicated vector encoding.
3608 (unwrap_const_vec_duplicate): Likewise.
3609 (const_vec_series_p): Check for a non-duplicated vector encoding.
3610 Say that the function only returns true for integer vectors.
3611 * emit-rtl.c: Include rtx-vector-builder.h.
3612 (gen_const_vec_duplicate_1): Delete.
3613 (gen_const_vector): Call gen_const_vec_duplicate instead of
3614 gen_const_vec_duplicate_1.
3615 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
3616 (gen_const_vec_duplicate): Use rtx_vector_builder.
3617 (gen_const_vec_series): Likewise.
3618 (gen_rtx_CONST_VECTOR): Likewise.
3619 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
3620 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
3621 Build a new vector rather than modifying a CONST_VECTOR in-place.
3622 (handle_special_swappables): Update call accordingly.
3623 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
3624 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
3625 Build a new vector rather than modifying a CONST_VECTOR in-place.
3626 (handle_special_swappables): Update call accordingly.
3628 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3630 * simplify-rtx.c (simplify_const_binary_operation): Use
3631 CONST_VECTOR_ELT instead of XVECEXP.
3633 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3635 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
3636 the selector elements to be different from the data elements
3637 if the selector is a VECTOR_CST.
3638 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
3639 ssizetype for the selector.
3641 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3643 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
3644 before testing each element individually.
3645 * tree-vect-generic.c (lower_vec_perm): Likewise.
3647 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3649 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
3650 * selftest-run-tests.c (selftest::run_tests): Call it.
3651 * vector-builder.h (vector_builder::operator ==): New function.
3652 (vector_builder::operator !=): Likewise.
3653 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
3654 (vec_perm_indices::all_from_input_p): New function.
3655 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
3656 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
3657 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
3658 instead of reading the VECTOR_CST directly. Detect whether both
3659 vector inputs are the same before constructing the vec_perm_indices,
3660 and update the number of inputs argument accordingly. Use the
3661 utility functions added above. Only construct sel2 if we need to.
3663 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3665 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
3666 the broadcast of the low byte.
3667 (expand_mult_highpart): Use an explicit encoding for the permutes.
3668 * optabs-query.c (can_mult_highpart_p): Likewise.
3669 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
3670 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3671 (vectorizable_bswap): Likewise.
3672 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
3673 explicit encoding for the power-of-2 permutes.
3674 (vect_permute_store_chain): Likewise.
3675 (vect_grouped_load_supported): Likewise.
3676 (vect_permute_load_chain): Likewise.
3678 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3680 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
3681 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
3682 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
3683 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
3684 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
3685 (vect_gen_perm_mask_any): Likewise.
3687 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3689 * int-vector-builder.h: New file.
3690 * vec-perm-indices.h: Include int-vector-builder.h.
3691 (vec_perm_indices): Redefine as an int_vector_builder.
3692 (auto_vec_perm_indices): Delete.
3693 (vec_perm_builder): Redefine as a stand-alone class.
3694 (vec_perm_indices::vec_perm_indices): New function.
3695 (vec_perm_indices::clamp): Likewise.
3696 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
3697 (vec_perm_indices::new_vector): New function.
3698 (vec_perm_indices::new_expanded_vector): Update for new
3699 vec_perm_indices class.
3700 (vec_perm_indices::rotate_inputs): New function.
3701 (vec_perm_indices::all_in_range_p): Operate directly on the
3702 encoded form, without computing elided elements.
3703 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
3704 encoding. Update for new vec_perm_indices class.
3705 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
3706 the given vec_perm_builder.
3707 (expand_vec_perm_var): Update vec_perm_builder constructor.
3708 (expand_mult_highpart): Use vec_perm_builder instead of
3709 auto_vec_perm_indices.
3710 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
3711 vec_perm_indices instead of auto_vec_perm_indices. Use a single
3712 or double series encoding as appropriate.
3713 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
3714 vec_perm_indices instead of auto_vec_perm_indices.
3715 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3716 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3717 (vect_permute_store_chain): Likewise.
3718 (vect_grouped_load_supported): Likewise.
3719 (vect_permute_load_chain): Likewise.
3720 (vect_shift_permute_load_chain): Likewise.
3721 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
3722 (vect_transform_slp_perm_load): Likewise.
3723 (vect_schedule_slp_instance): Likewise.
3724 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3725 (vectorizable_mask_load_store): Likewise.
3726 (vectorizable_bswap): Likewise.
3727 (vectorizable_store): Likewise.
3728 (vectorizable_load): Likewise.
3729 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
3730 vec_perm_indices instead of auto_vec_perm_indices. Use
3731 tree_to_vec_perm_builder to read the vector from a tree.
3732 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
3733 vec_perm_builder instead of a vec_perm_indices.
3734 (have_whole_vector_shift): Use vec_perm_builder and
3735 vec_perm_indices instead of auto_vec_perm_indices. Leave the
3736 truncation to calc_vec_perm_mask_for_shift.
3737 (vect_create_epilog_for_reduction): Likewise.
3738 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
3739 from auto_vec_perm_indices to vec_perm_indices.
3740 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
3741 instead of changing individual elements.
3742 (aarch64_vectorize_vec_perm_const): Use new_vector to install
3743 the vector in d.perm.
3744 * config/arm/arm.c (expand_vec_perm_d::perm): Change
3745 from auto_vec_perm_indices to vec_perm_indices.
3746 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
3747 instead of changing individual elements.
3748 (arm_vectorize_vec_perm_const): Use new_vector to install
3749 the vector in d.perm.
3750 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
3751 Update vec_perm_builder constructor.
3752 (rs6000_expand_interleave): Likewise.
3753 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
3754 (rs6000_expand_interleave): Likewise.
3756 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3758 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
3759 to qimode could truncate the indices.
3760 * optabs.c (expand_vec_perm_var): Likewise.
3762 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3764 * Makefile.in (OBJS): Add vec-perm-indices.o.
3765 * vec-perm-indices.h: New file.
3766 * vec-perm-indices.c: Likewise.
3767 * target.h (vec_perm_indices): Replace with a forward class
3769 (auto_vec_perm_indices): Move to vec-perm-indices.h.
3770 * optabs.h: Include vec-perm-indices.h.
3771 (expand_vec_perm): Delete.
3772 (selector_fits_mode_p, expand_vec_perm_var): Declare.
3773 (expand_vec_perm_const): Declare.
3774 * target.def (vec_perm_const_ok): Replace with...
3775 (vec_perm_const): ...this new hook.
3776 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
3777 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
3778 * doc/tm.texi: Regenerate.
3779 * optabs.def (vec_perm_const): Delete.
3780 * doc/md.texi (vec_perm_const): Likewise.
3781 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
3782 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
3783 expand_vec_perm for constant permutation vectors. Assert that
3784 the mode of variable permutation vectors is the integer equivalent
3785 of the mode that is being permuted.
3786 * optabs-query.h (selector_fits_mode_p): Declare.
3787 * optabs-query.c: Include vec-perm-indices.h.
3788 (selector_fits_mode_p): New function.
3789 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
3790 is defined, instead of checking whether the vec_perm_const_optab
3791 exists. Use targetm.vectorize.vec_perm_const instead of
3792 targetm.vectorize.vec_perm_const_ok. Check whether the indices
3793 fit in the vector mode before using a variable permute.
3794 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
3795 vec_perm_indices instead of an rtx.
3796 (expand_vec_perm): Replace with...
3797 (expand_vec_perm_const): ...this new function. Take the selector
3798 as a vec_perm_indices rather than an rtx. Also take the mode of
3799 the selector. Update call to shift_amt_for_vec_perm_mask.
3800 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
3801 Use vec_perm_indices::new_expanded_vector to expand the original
3802 selector into bytes. Check whether the indices fit in the vector
3803 mode before using a variable permute.
3804 (expand_vec_perm_var): Make global.
3805 (expand_mult_highpart): Use expand_vec_perm_const.
3806 * fold-const.c: Includes vec-perm-indices.h.
3807 * tree-ssa-forwprop.c: Likewise.
3808 * tree-vect-data-refs.c: Likewise.
3809 * tree-vect-generic.c: Likewise.
3810 * tree-vect-loop.c: Likewise.
3811 * tree-vect-slp.c: Likewise.
3812 * tree-vect-stmts.c: Likewise.
3813 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
3815 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
3816 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
3817 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
3818 (aarch64_vectorize_vec_perm_const): ...this new function.
3819 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3820 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3821 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
3822 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
3823 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3824 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3825 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
3827 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
3828 check for NEON modes.
3829 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
3830 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
3831 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
3832 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
3834 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
3835 the old VEC_PERM_CONST conditions.
3836 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
3837 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
3838 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
3839 (ia64_vectorize_vec_perm_const_ok): Merge into...
3840 (ia64_vectorize_vec_perm_const): ...this new function.
3841 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
3842 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
3843 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
3844 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
3845 * config/mips/mips.c (mips_expand_vec_perm_const)
3846 (mips_vectorize_vec_perm_const_ok): Merge into...
3847 (mips_vectorize_vec_perm_const): ...this new function.
3848 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
3849 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
3850 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
3851 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
3852 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
3853 (rs6000_expand_vec_perm_const): Delete.
3854 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
3856 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3857 (altivec_expand_vec_perm_const_le): Take each operand individually.
3858 Operate on constant selectors rather than rtxes.
3859 (altivec_expand_vec_perm_const): Likewise. Update call to
3860 altivec_expand_vec_perm_const_le.
3861 (rs6000_expand_vec_perm_const): Delete.
3862 (rs6000_vectorize_vec_perm_const_ok): Delete.
3863 (rs6000_vectorize_vec_perm_const): New function.
3864 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3865 an element count and rtx array.
3866 (rs6000_expand_extract_even): Update call accordingly.
3867 (rs6000_expand_interleave): Likewise.
3868 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
3869 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
3870 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
3871 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
3872 (rs6000_expand_vec_perm_const): Delete.
3873 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3874 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3875 (altivec_expand_vec_perm_const_le): Take each operand individually.
3876 Operate on constant selectors rather than rtxes.
3877 (altivec_expand_vec_perm_const): Likewise. Update call to
3878 altivec_expand_vec_perm_const_le.
3879 (rs6000_expand_vec_perm_const): Delete.
3880 (rs6000_vectorize_vec_perm_const_ok): Delete.
3881 (rs6000_vectorize_vec_perm_const): New function. Remove stray
3882 reference to the SPE evmerge intructions.
3883 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3884 an element count and rtx array.
3885 (rs6000_expand_extract_even): Update call accordingly.
3886 (rs6000_expand_interleave): Likewise.
3887 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
3888 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
3890 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3892 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3894 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
3895 vector mode and that that mode matches the mode of the data
3897 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
3898 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
3899 directly using expand_vec_perm_1 when forcing selectors into
3901 (expand_vec_perm_var): New function, split out from expand_vec_perm.
3903 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3905 * optabs-query.h (can_vec_perm_p): Delete.
3906 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
3907 * optabs-query.c (can_vec_perm_p): Split into...
3908 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
3909 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
3910 particular selector is valid.
3911 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3912 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3913 (vect_grouped_load_supported): Likewise.
3914 (vect_shift_permute_load_chain): Likewise.
3915 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
3916 (vect_transform_slp_perm_load): Likewise.
3917 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3918 (vectorizable_bswap): Likewise.
3919 (vect_gen_perm_mask_checked): Likewise.
3920 * fold-const.c (fold_ternary_loc): Likewise. Don't take
3921 implementations of variable permutation vectors into account
3922 when deciding which selector to use.
3923 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
3924 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
3925 with a false third argument.
3926 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
3927 to test whether the constant selector is valid and can_vec_perm_var_p
3928 to test whether a variable selector is valid.
3930 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3932 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
3933 * optabs-query.c (can_vec_perm_p): Likewise.
3934 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
3935 instead of vec_perm_indices.
3936 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
3937 (vect_gen_perm_mask_checked): Likewise,
3938 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
3939 (vect_gen_perm_mask_checked): Likewise,
3941 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3943 * optabs-query.h (qimode_for_vec_perm): Declare.
3944 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
3945 (qimode_for_vec_perm): ...this new function.
3946 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
3948 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3950 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
3951 does not have a conditional at the top.
3953 2018-01-02 Richard Biener <rguenther@suse.de>
3955 * ipa-inline.c (big_speedup_p): Fix expression.
3957 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
3960 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
3963 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
3967 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
3968 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
3969 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
3970 cond_taken_branch_cost 3->4.
3972 2018-01-01 Jakub Jelinek <jakub@redhat.com>
3974 PR tree-optimization/83581
3975 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
3976 TODO_cleanup_cfg if any changes have been made.
3979 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
3980 convert_modes if target mode has the right side, but different mode
3984 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
3985 last argument when extracting from CONCAT. If either from_real or
3986 from_imag is NULL, use expansion through memory. If result is not
3987 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
3988 the parts directly to inner mode, if even that fails, use expansion
3992 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
3993 check for bswap in mode rather than HImode and use that in expand_unop
3996 Copyright (C) 2018 Free Software Foundation, Inc.
3998 Copying and distribution of this file, with or without modification,
3999 are permitted in any medium without royalty provided the copyright
4000 notice and this notice are preserved.