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 * doc/md.texi (gather_load@var{m}): Document.
6 (mask_gather_load@var{m}): Likewise.
7 * genopinit.c (main): Add supports_vec_gather_load and
8 supports_vec_gather_load_cached to target_optabs.
9 * optabs-tree.c (init_tree_optimization_optabs): Use
10 ggc_cleared_alloc to allocate target_optabs.
11 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
12 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
14 * internal-fn.h (internal_load_fn_p): Declare.
15 (internal_gather_scatter_fn_p): Likewise.
16 (internal_fn_mask_index): Likewise.
17 (internal_gather_scatter_fn_supported_p): Likewise.
18 * internal-fn.c (gather_load_direct): New macro.
19 (expand_gather_load_optab_fn): New function.
20 (direct_gather_load_optab_supported_p): New macro.
21 (direct_internal_fn_optab): New function.
22 (internal_load_fn_p): Likewise.
23 (internal_gather_scatter_fn_p): Likewise.
24 (internal_fn_mask_index): Likewise.
25 (internal_gather_scatter_fn_supported_p): Likewise.
26 * optabs-query.c (supports_at_least_one_mode_p): New function.
27 (supports_vec_gather_load_p): Likewise.
28 * optabs-query.h (supports_vec_gather_load_p): Declare.
29 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
30 and memory_type field.
31 (NUM_PATTERNS): Bump to 15.
32 * tree-vect-data-refs.c: Include internal-fn.h.
33 (vect_gather_scatter_fn_p): New function.
34 (vect_describe_gather_scatter_call): Likewise.
35 (vect_check_gather_scatter): Try using internal functions for
36 gather loads. Recognize existing calls to a gather load function.
37 (vect_analyze_data_refs): Consider using gather loads if
38 supports_vec_gather_load_p.
39 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
40 (vect_get_gather_scatter_offset_type): Likewise.
41 (vect_convert_mask_for_vectype): Likewise.
42 (vect_add_conversion_to_patterm): Likewise.
43 (vect_try_gather_scatter_pattern): Likewise.
44 (vect_recog_gather_scatter_pattern): New pattern recognizer.
45 (vect_vect_recog_func_ptrs): Add it.
46 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
47 internal_fn_mask_index and internal_gather_scatter_fn_p.
48 (check_load_store_masking): Take the gather_scatter_info as an
49 argument and handle gather loads.
50 (vect_get_gather_scatter_ops): New function.
51 (vectorizable_call): Check internal_load_fn_p.
52 (vectorizable_load): Likewise. Handle gather load internal
54 (vectorizable_store): Update call to check_load_store_masking.
55 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
56 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
57 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
58 (aarch64_gather_scale_operand_d): New predicates.
59 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
60 (mask_gather_load<mode>): New insns.
62 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
63 Alan Hayward <alan.hayward@arm.com>
64 David Sherwood <david.sherwood@arm.com>
66 * optabs.def (fold_left_plus_optab): New optab.
67 * doc/md.texi (fold_left_plus_@var{m}): Document.
68 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
69 * internal-fn.c (fold_left_direct): Define.
70 (expand_fold_left_optab_fn): Likewise.
71 (direct_fold_left_optab_supported_p): Likewise.
72 * fold-const-call.c (fold_const_fold_left): New function.
73 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
74 * tree-parloops.c (valid_reduction_p): New function.
75 (gather_scalar_reductions): Use it.
76 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
77 (vect_finish_replace_stmt): Declare.
78 * tree-vect-loop.c (fold_left_reduction_fn): New function.
79 (needs_fold_left_reduction_p): New function, split out from...
80 (vect_is_simple_reduction): ...here. Accept reductions that
81 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
82 (vect_force_simple_reduction): Also store the reduction type in
83 the assignment's STMT_VINFO_REDUC_TYPE.
84 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
85 (merge_with_identity): New function.
86 (vect_expand_fold_left): Likewise.
87 (vectorize_fold_left_reduction): Likewise.
88 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
89 scalar phi in place for it. Check for target support and reject
90 cases that would reassociate the operation. Defer the transform
91 phase to vectorize_fold_left_reduction.
92 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
93 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
94 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
96 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
98 * tree-if-conv.c (predicate_mem_writes): Remove redundant
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 * target.def (legitimize_address_displacement): Take the original
106 offset as a poly_int.
107 * targhooks.h (default_legitimize_address_displacement): Update
109 * targhooks.c (default_legitimize_address_displacement): Likewise.
110 * doc/tm.texi: Regenerate.
111 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
112 as an argument, moving assert of ad->disp == ad->disp_term to...
113 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
114 Try calling targetm.legitimize_address_displacement before expanding
115 the address rather than afterwards, and adjust for the new interface.
116 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
117 Match the new hook interface. Handle SVE addresses.
118 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
121 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
123 * Makefile.in (OBJS): Add early-remat.o.
124 * target.def (select_early_remat_modes): New hook.
125 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
126 * doc/tm.texi: Regenerate.
127 * targhooks.h (default_select_early_remat_modes): Declare.
128 * targhooks.c (default_select_early_remat_modes): New function.
129 * timevar.def (TV_EARLY_REMAT): New timevar.
130 * passes.def (pass_early_remat): New pass.
131 * tree-pass.h (make_pass_early_remat): Declare.
132 * early-remat.c: New file.
133 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
135 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
138 Alan Hayward <alan.hayward@arm.com>
139 David Sherwood <david.sherwood@arm.com>
141 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
142 vfm1 with a bound_epilog parameter.
143 (vect_do_peeling): Update calls accordingly, and move the prologue
144 call earlier in the function. Treat the base bound_epilog as 0 for
145 fully-masked loops and retain vf - 1 for other loops. Add 1 to
146 this base when peeling for gaps.
147 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
148 with fully-masked loops.
149 (vect_estimate_min_profitable_iters): Handle the single peeled
150 iteration in that case.
152 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
153 Alan Hayward <alan.hayward@arm.com>
154 David Sherwood <david.sherwood@arm.com>
156 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
157 single-element interleaving even if the size is not a power of 2.
158 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
159 accesses for single-element interleaving if the group size is
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 * doc/md.texi (fold_extract_last_@var{m}): Document.
167 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
168 * optabs.def (fold_extract_last_optab): New optab.
169 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
170 * internal-fn.c (fold_extract_direct): New macro.
171 (expand_fold_extract_optab_fn): Likewise.
172 (direct_fold_extract_optab_supported_p): Likewise.
173 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
174 * tree-vect-loop.c (vect_model_reduction_cost): Handle
175 EXTRACT_LAST_REDUCTION.
176 (get_initial_def_for_reduction): Do not create an initial vector
177 for EXTRACT_LAST_REDUCTION reductions.
178 (vectorizable_reduction): Leave the scalar phi in place for
179 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
180 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
181 epilogue code for EXTRACT_LAST_REDUCTION and defer the
182 transform phase to vectorizable_condition.
183 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
185 (vect_finish_stmt_generation): ...here.
186 (vect_finish_replace_stmt): New function.
187 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
188 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
190 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
192 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
193 Alan Hayward <alan.hayward@arm.com>
194 David Sherwood <david.sherwood@arm.com>
196 * doc/md.texi (extract_last_@var{m}): Document.
197 * optabs.def (extract_last_optab): New optab.
198 * internal-fn.def (EXTRACT_LAST): New internal function.
199 * internal-fn.c (cond_unary_direct): New macro.
200 (expand_cond_unary_optab_fn): Likewise.
201 (direct_cond_unary_optab_supported_p): Likewise.
202 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
203 loops using EXTRACT_LAST.
204 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
205 (extract_last_<mode>): ...this optab.
206 (vec_extract<mode><Vel>): Update accordingly.
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 * target.def (empty_mask_is_expensive): New hook.
213 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
214 * doc/tm.texi: Regenerate.
215 * targhooks.h (default_empty_mask_is_expensive): Declare.
216 * targhooks.c (default_empty_mask_is_expensive): New function.
217 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
218 if the target says that empty masks are expensive.
219 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
221 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
223 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
224 Alan Hayward <alan.hayward@arm.com>
225 David Sherwood <david.sherwood@arm.com>
227 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
228 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
229 (vect_use_loop_mask_for_alignment_p): New function.
230 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
231 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
232 niters_skip argument. Make sure that the first niters_skip elements
233 of the first iteration are inactive.
234 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
235 Update call to vect_set_loop_masks_directly.
236 (get_misalign_in_elems): New function, split out from...
237 (vect_gen_prolog_loop_niters): ...here.
238 (vect_update_init_of_dr): Take a code argument that specifies whether
239 the adjustment should be added or subtracted.
240 (vect_update_init_of_drs): Likewise.
241 (vect_prepare_for_masked_peels): New function.
242 (vect_do_peeling): Skip prologue peeling if we're using a mask
243 instead. Update call to vect_update_inits_of_drs.
244 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
246 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
247 alignment. Do not include the number of peeled iterations in
248 the minimum threshold in that case.
249 (vectorizable_induction): Adjust the start value down by
250 LOOP_VINFO_MASK_SKIP_NITERS iterations.
251 (vect_transform_loop): Call vect_prepare_for_masked_peels.
252 Take the number of skipped iterations into account when calculating
254 * tree-vect-stmts.c (vect_gen_while_not): New function.
256 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
257 Alan Hayward <alan.hayward@arm.com>
258 David Sherwood <david.sherwood@arm.com>
260 * doc/sourcebuild.texi (vect_fully_masked): Document.
261 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
263 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
265 (vect_analyze_loop_2): ...here. Don't check the vectorization
266 factor against the number of loop iterations if the loop is
269 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
270 Alan Hayward <alan.hayward@arm.com>
271 David Sherwood <david.sherwood@arm.com>
273 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
274 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
275 (dump_groups): Update accordingly.
276 (iv_use::mem_type): New member variable.
277 (address_p): New function.
278 (record_use): Add a mem_type argument and initialize the new
280 (record_group_use): Add a mem_type argument. Use address_p.
281 Remove obsolete null checks of base_object. Update call to record_use.
282 (find_interesting_uses_op): Update call to record_group_use.
283 (find_interesting_uses_cond): Likewise.
284 (find_interesting_uses_address): Likewise.
285 (get_mem_type_for_internal_fn): New function.
286 (find_address_like_use): Likewise.
287 (find_interesting_uses_stmt): Try find_address_like_use before
288 calling find_interesting_uses_op.
289 (addr_offset_valid_p): Use the iv mem_type field as the type
290 of the addressed memory.
291 (add_autoinc_candidates): Likewise.
292 (get_address_cost): Likewise.
293 (split_small_address_groups_p): Use address_p.
294 (split_address_groups): Likewise.
295 (add_iv_candidate_for_use): Likewise.
296 (autoinc_possible_for_pair): Likewise.
297 (rewrite_groups): Likewise.
298 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
299 (determine_group_iv_cost): Update after split of USE_ADDRESS.
300 (get_alias_ptr_type_for_ptr_address): New function.
301 (rewrite_use_address): Rewrite address uses in calls that were
302 identified by find_address_like_use.
304 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
305 Alan Hayward <alan.hayward@arm.com>
306 David Sherwood <david.sherwood@arm.com>
308 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
310 * gimple-expr.h (is_gimple_addressable: Likewise.
311 * gimple-expr.c (is_gimple_address): Likewise.
312 * internal-fn.c (expand_call_mem_ref): New function.
313 (expand_mask_load_optab_fn): Use it.
314 (expand_mask_store_optab_fn): Likewise.
316 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
317 Alan Hayward <alan.hayward@arm.com>
318 David Sherwood <david.sherwood@arm.com>
320 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
321 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
322 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
323 (cond_umax@var{mode}): Document.
324 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
325 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
326 (cond_umin_optab, cond_umax_optab): New optabs.
327 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
328 (COND_IOR, COND_XOR): New internal functions.
329 * internal-fn.h (get_conditional_internal_fn): Declare.
330 * internal-fn.c (cond_binary_direct): New macro.
331 (expand_cond_binary_optab_fn): Likewise.
332 (direct_cond_binary_optab_supported_p): Likewise.
333 (get_conditional_internal_fn): New function.
334 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
335 Cope with reduction statements that are vectorized as calls rather
337 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
338 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
339 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
340 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
341 (UNSPEC_COND_EOR): New unspecs.
342 (optab): Add mappings for them.
343 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
344 (sve_int_op, sve_fp_op): New int attributes.
346 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
347 Alan Hayward <alan.hayward@arm.com>
348 David Sherwood <david.sherwood@arm.com>
350 * optabs.def (while_ult_optab): New optab.
351 * doc/md.texi (while_ult@var{m}@var{n}): Document.
352 * internal-fn.def (WHILE_ULT): New internal function.
353 * internal-fn.h (direct_internal_fn_supported_p): New override
354 that takes two types as argument.
355 * internal-fn.c (while_direct): New macro.
356 (expand_while_optab_fn): New function.
357 (convert_optab_supported_p): Likewise.
358 (direct_while_optab_supported_p): New macro.
359 * wide-int.h (wi::udiv_ceil): New function.
360 * tree-vectorizer.h (rgroup_masks): New structure.
361 (vec_loop_masks): New typedef.
362 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
364 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
365 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
366 (vect_max_vf): New function.
367 (slpeel_make_loop_iterate_ntimes): Delete.
368 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
369 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
370 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
371 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
372 internal-fn.h, stor-layout.h and optabs-query.h.
373 (vect_set_loop_mask): New function.
374 (add_preheader_seq): Likewise.
375 (add_header_seq): Likewise.
376 (interleave_supported_p): Likewise.
377 (vect_maybe_permute_loop_masks): Likewise.
378 (vect_set_loop_masks_directly): Likewise.
379 (vect_set_loop_condition_masked): Likewise.
380 (vect_set_loop_condition_unmasked): New function, split out from
381 slpeel_make_loop_iterate_ntimes.
382 (slpeel_make_loop_iterate_ntimes): Rename to..
383 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
384 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
385 (vect_do_peeling): Update call accordingly.
386 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
388 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
389 mask_compare_type, can_fully_mask_p and fully_masked_p.
390 (release_vec_loop_masks): New function.
391 (_loop_vec_info): Use it to free the loop masks.
392 (can_produce_all_loop_masks_p): New function.
393 (vect_get_max_nscalars_per_iter): Likewise.
394 (vect_verify_full_masking): Likewise.
395 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
396 retries, and free the mask rgroups before retrying. Check loop-wide
397 reasons for disallowing fully-masked loops. Make the final decision
398 about whether use a fully-masked loop or not.
399 (vect_estimate_min_profitable_iters): Do not assume that peeling
400 for the number of iterations will be needed for fully-masked loops.
401 (vectorizable_reduction): Disable fully-masked loops.
402 (vectorizable_live_operation): Likewise.
403 (vect_halve_mask_nunits): New function.
404 (vect_double_mask_nunits): Likewise.
405 (vect_record_loop_mask): Likewise.
406 (vect_get_loop_mask): Likewise.
407 (vect_transform_loop): Handle the case in which the final loop
408 iteration might handle a partial vector. Call vect_set_loop_condition
409 instead of slpeel_make_loop_iterate_ntimes.
410 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
411 (check_load_store_masking): New function.
412 (prepare_load_store_mask): Likewise.
413 (vectorizable_store): Handle fully-masked loops.
414 (vectorizable_load): Likewise.
415 (supportable_widening_operation): Use vect_halve_mask_nunits for
417 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
418 (vect_gen_while): New function.
419 * config/aarch64/aarch64.md (umax<mode>3): New expander.
420 (aarch64_uqdec<mode>): New insn.
422 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
423 Alan Hayward <alan.hayward@arm.com>
424 David Sherwood <david.sherwood@arm.com>
426 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
427 (reduc_xor_scal_optab): New optabs.
428 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
429 (reduc_xor_scal_@var{m}): Document.
430 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
431 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
433 * fold-const-call.c (fold_const_call): Handle them.
434 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
435 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
436 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
437 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
438 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
439 (UNSPEC_XORV): New unspecs.
440 (optab): Add entries for them.
441 (BITWISEV): New int iterator.
442 (bit_reduc_op): New int attributes.
444 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
445 Alan Hayward <alan.hayward@arm.com>
446 David Sherwood <david.sherwood@arm.com>
448 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
449 * internal-fn.def (VEC_SHL_INSERT): New internal function.
450 * optabs.def (vec_shl_insert_optab): New optab.
451 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
452 (duplicate_and_interleave): Likewise.
453 * tree-vect-loop.c: Include internal-fn.h.
454 (neutral_op_for_slp_reduction): New function, split out from
455 get_initial_defs_for_reduction.
456 (get_initial_def_for_reduction): Handle option 2 for variable-length
457 vectors by loading the neutral value into a vector and then shifting
458 the initial value into element 0.
459 (get_initial_defs_for_reduction): Replace the code argument with
460 the neutral value calculated by neutral_op_for_slp_reduction.
461 Use gimple_build_vector for constant-length vectors.
462 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
463 but the first group_size elements have a neutral value.
464 Use duplicate_and_interleave otherwise.
465 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
466 Update call to get_initial_defs_for_reduction. Handle SLP
467 reductions for variable-length vectors by creating one vector
468 result for each scalar result, with the elements associated
469 with other scalar results stubbed out with the neutral value.
470 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
471 Require IFN_VEC_SHL_INSERT for double reductions on
472 variable-length vectors, or SLP reductions that have
473 a neutral value. Require can_duplicate_and_interleave_p
474 support for variable-length unchained SLP reductions if there
475 is no neutral value, such as for MIN/MAX reductions. Also require
476 the number of vector elements to be a multiple of the number of
477 SLP statements when doing variable-length unchained SLP reductions.
478 Update call to vect_create_epilog_for_reduction.
479 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
480 and remove initial values.
481 (duplicate_and_interleave): Make public.
482 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
483 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
485 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
486 Alan Hayward <alan.hayward@arm.com>
487 David Sherwood <david.sherwood@arm.com>
489 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
490 (can_duplicate_and_interleave_p): New function.
491 (vect_get_and_check_slp_defs): Take the vector of statements
492 rather than just the current one. Remove excess parentheses.
493 Restriction rejectinon of vect_constant_def and vect_external_def
494 for variable-length vectors to boolean types, or types for which
495 can_duplicate_and_interleave_p is false.
496 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
497 (duplicate_and_interleave): New function.
498 (vect_get_constant_vectors): Use gimple_build_vector for
499 constant-length vectors and suitable variable-length constant
500 vectors. Use duplicate_and_interleave for other variable-length
501 vectors. Don't defer the update when inserting new statements.
503 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
504 Alan Hayward <alan.hayward@arm.com>
505 David Sherwood <david.sherwood@arm.com>
507 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
508 min_profitable_iters doesn't go negative.
510 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
511 Alan Hayward <alan.hayward@arm.com>
512 David Sherwood <david.sherwood@arm.com>
514 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
515 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
516 * optabs.def (vec_mask_load_lanes_optab): New optab.
517 (vec_mask_store_lanes_optab): Likewise.
518 * internal-fn.def (MASK_LOAD_LANES): New internal function.
519 (MASK_STORE_LANES): Likewise.
520 * internal-fn.c (mask_load_lanes_direct): New macro.
521 (mask_store_lanes_direct): Likewise.
522 (expand_mask_load_optab_fn): Handle masked operations.
523 (expand_mask_load_lanes_optab_fn): New macro.
524 (expand_mask_store_optab_fn): Handle masked operations.
525 (expand_mask_store_lanes_optab_fn): New macro.
526 (direct_mask_load_lanes_optab_supported_p): Likewise.
527 (direct_mask_store_lanes_optab_supported_p): Likewise.
528 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
530 (vect_load_lanes_supported): Likewise.
531 * tree-vect-data-refs.c (strip_conversion): New function.
532 (can_group_stmts_p): Likewise.
533 (vect_analyze_data_ref_accesses): Use it instead of checking
534 for a pair of assignments.
535 (vect_store_lanes_supported): Take a masked_p parameter.
536 (vect_load_lanes_supported): Likewise.
537 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
538 vect_store_lanes_supported and vect_load_lanes_supported.
539 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
540 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
541 parameter. Don't allow gaps for masked accesses.
542 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
543 and vect_load_lanes_supported.
544 (get_load_store_type): Take a masked_p parameter and update
545 call to get_group_load_store_type.
546 (vectorizable_store): Update call to get_load_store_type.
547 Handle IFN_MASK_STORE_LANES.
548 (vectorizable_load): Update call to get_load_store_type.
549 Handle IFN_MASK_LOAD_LANES.
551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
552 Alan Hayward <alan.hayward@arm.com>
553 David Sherwood <david.sherwood@arm.com>
555 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
557 * config/aarch64/aarch64-protos.h
558 (aarch64_sve_struct_memory_operand_p): Declare.
559 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
560 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
561 (VPRED, vpred): Handle SVE structure modes.
562 * config/aarch64/constraints.md (Utx): New constraint.
563 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
564 (aarch64_sve_struct_nonimmediate_operand): New predicates.
565 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
566 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
567 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
568 structure modes. Split into pieces after RA.
569 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
570 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
572 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
574 (aarch64_classify_address): Likewise.
575 (sizetochar): Move earlier in file.
576 (aarch64_print_operand): Handle SVE register lists.
577 (aarch64_array_mode): New function.
578 (aarch64_sve_struct_memory_operand_p): Likewise.
579 (TARGET_ARRAY_MODE): Redefine.
581 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
582 Alan Hayward <alan.hayward@arm.com>
583 David Sherwood <david.sherwood@arm.com>
585 * target.def (array_mode): New target hook.
586 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
587 * doc/tm.texi: Regenerate.
588 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
589 * hooks.c (hook_optmode_mode_uhwi_none): New function.
590 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
592 * stor-layout.c (mode_for_array): Likewise. Support polynomial
595 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
596 Alan Hayward <alan.hayward@arm.com>
597 David Sherwood <david.sherwood@arm.com>
599 * fold-const.c (fold_binary_loc): Check the argument types
600 rather than the result type when testing for a vector operation.
602 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
604 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
605 * doc/tm.texi: Regenerate.
607 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
608 Alan Hayward <alan.hayward@arm.com>
609 David Sherwood <david.sherwood@arm.com>
611 * doc/invoke.texi (-msve-vector-bits=): Document new option.
612 (sve): Document new AArch64 extension.
613 * doc/md.texi (w): Extend the description of the AArch64
614 constraint to include SVE vectors.
615 (Upl, Upa): Document new AArch64 predicate constraints.
616 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
618 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
619 (msve-vector-bits=): New option.
620 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
621 SVE when these are disabled.
622 (sve): New extension.
623 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
624 modes. Adjust their number of units based on aarch64_sve_vg.
625 (MAX_BITSIZE_MODE_ANY_MODE): Define.
626 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
627 aarch64_addr_query_type.
628 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
629 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
630 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
631 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
632 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
633 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
634 (aarch64_simd_imm_zero_p): Delete.
635 (aarch64_check_zero_based_sve_index_immediate): Declare.
636 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
637 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
638 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
639 (aarch64_sve_float_mul_immediate_p): Likewise.
640 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
642 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
643 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
644 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
645 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
646 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
647 (aarch64_regmode_natural_size): Likewise.
648 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
649 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
651 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
652 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
653 for VG and the SVE predicate registers.
654 (V_ALIASES): Add a "z"-prefixed alias.
655 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
656 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
657 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
658 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
659 (REG_CLASS_NAMES): Add entries for them.
660 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
661 and the predicate registers.
662 (aarch64_sve_vg): Declare.
663 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
664 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
665 (REGMODE_NATURAL_SIZE): Define.
666 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
668 * config/aarch64/aarch64.c: Include cfgrtl.h.
669 (simd_immediate_info): Add a constructor for series vectors,
670 and an associated step field.
671 (aarch64_sve_vg): New variable.
672 (aarch64_dbx_register_number): Handle VG and the predicate registers.
673 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
674 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
675 (VEC_ANY_DATA, VEC_STRUCT): New constants.
676 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
677 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
678 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
679 (aarch64_get_mask_mode): New functions.
680 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
681 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
682 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
683 predicate modes and predicate registers. Explicitly restrict
684 GPRs to modes of 16 bytes or smaller. Only allow FP registers
685 to store a vector mode if it is recognized by
686 aarch64_classify_vector_mode.
687 (aarch64_regmode_natural_size): New function.
688 (aarch64_hard_regno_caller_save_mode): Return the original mode
690 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
691 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
692 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
693 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
695 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
696 does not overlap dest if the function is frame-related. Handle
698 (aarch64_split_add_offset): New function.
699 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
700 them aarch64_add_offset.
701 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
702 and update call to aarch64_sub_sp.
703 (aarch64_add_cfa_expression): New function.
704 (aarch64_expand_prologue): Pass extra temporary registers to the
705 functions above. Handle the case in which we need to emit new
706 DW_CFA_expressions for registers that were originally saved
707 relative to the stack pointer, but now have to be expressed
708 relative to the frame pointer.
709 (aarch64_output_mi_thunk): Pass extra temporary registers to the
711 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
712 IP0 and IP1 values for SVE frames.
713 (aarch64_expand_vec_series): New function.
714 (aarch64_expand_sve_widened_duplicate): Likewise.
715 (aarch64_expand_sve_const_vector): Likewise.
716 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
717 Handle SVE constants. Use emit_move_insn to move a force_const_mem
718 into the register, rather than emitting a SET directly.
719 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
720 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
721 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
722 (offset_9bit_signed_scaled_p): New functions.
723 (aarch64_replicate_bitmask_imm): New function.
724 (aarch64_bitmask_imm): Use it.
725 (aarch64_cannot_force_const_mem): Reject expressions involving
726 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
727 (aarch64_classify_index): Handle SVE indices, by requiring
728 a plain register index with a scale that matches the element size.
729 (aarch64_classify_address): Handle SVE addresses. Assert that
730 the mode of the address is VOIDmode or an integer mode.
731 Update call to aarch64_classify_symbol.
732 (aarch64_classify_symbolic_expression): Update call to
733 aarch64_classify_symbol.
734 (aarch64_const_vec_all_in_range_p): New function.
735 (aarch64_print_vector_float_operand): Likewise.
736 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
737 "vN" for FP registers with SVE modes. Handle (const ...) vectors
738 and the FP immediates 1.0 and 0.5.
739 (aarch64_print_address_internal): Handle SVE addresses.
740 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
741 (aarch64_regno_regclass): Handle predicate registers.
742 (aarch64_secondary_reload): Handle big-endian reloads of SVE
744 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
745 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
746 (aarch64_convert_sve_vector_bits): New function.
747 (aarch64_override_options): Use it to handle -msve-vector-bits=.
748 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
750 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
751 Handle SVE vector and predicate modes. Accept VL-based constants
752 that need only one temporary register, and VL offsets that require
753 no temporary registers.
754 (aarch64_conditional_register_usage): Mark the predicate registers
755 as fixed if SVE isn't available.
756 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
757 Return true for SVE vector and predicate modes.
758 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
759 rather than an unsigned int. Handle SVE modes.
760 (aarch64_preferred_simd_mode): Update call accordingly. Handle
762 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
764 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
765 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
766 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
767 (aarch64_sve_float_mul_immediate_p): New functions.
768 (aarch64_sve_valid_immediate): New function.
769 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
770 Explicitly reject structure modes. Check for INDEX constants.
771 Handle PTRUE and PFALSE constants.
772 (aarch64_check_zero_based_sve_index_immediate): New function.
773 (aarch64_simd_imm_zero_p): Delete.
774 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
775 vector modes. Accept constants in the range of CNT[BHWD].
776 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
777 ask for an Advanced SIMD mode.
778 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
779 (aarch64_simd_vector_alignment): Handle SVE predicates.
780 (aarch64_vectorize_preferred_vector_alignment): New function.
781 (aarch64_simd_vector_alignment_reachable): Use it instead of
783 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
784 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
786 (MAX_VECT_LEN): Delete.
787 (expand_vec_perm_d): Add a vec_flags field.
788 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
789 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
790 (aarch64_evpc_ext): Don't apply a big-endian lane correction
792 (aarch64_evpc_rev): Rename to...
793 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
794 (aarch64_evpc_rev_global): New function.
795 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
796 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
798 (aarch64_evpc_sve_tbl): New function.
799 (aarch64_expand_vec_perm_const_1): Update after rename of
800 aarch64_evpc_rev. Handle SVE permutes too, trying
801 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
802 than aarch64_evpc_tbl.
803 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
804 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
805 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
806 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
807 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
808 (aarch64_expand_sve_vcond): New functions.
809 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
810 of aarch64_vector_mode_p.
811 (aarch64_dwarf_poly_indeterminate_value): New function.
812 (aarch64_compute_pressure_classes): Likewise.
813 (aarch64_can_change_mode_class): Likewise.
814 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
815 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
816 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
817 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
818 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
819 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
820 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
821 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
823 (Dn, Dl, Dr): Accept const as well as const_vector.
824 (Dz): Likewise. Compare against CONST0_RTX.
825 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
826 of "vector" where appropriate.
827 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
828 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
829 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
830 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
831 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
832 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
833 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
834 (v_int_equiv): Extend to SVE modes.
835 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
837 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
838 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
839 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
840 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
841 (SVE_COND_FP_CMP): New int iterators.
842 (perm_hilo): Handle the new unpack unspecs.
843 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
845 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
846 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
847 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
848 (aarch64_equality_operator, aarch64_constant_vector_operand)
849 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
850 (aarch64_sve_nonimmediate_operand): Likewise.
851 (aarch64_sve_general_operand): Likewise.
852 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
853 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
854 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
855 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
856 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
857 (aarch64_sve_float_arith_immediate): Likewise.
858 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
859 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
860 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
861 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
862 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
863 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
864 (aarch64_sve_float_arith_operand): Likewise.
865 (aarch64_sve_float_arith_with_sub_operand): Likewise.
866 (aarch64_sve_float_mul_operand): Likewise.
867 (aarch64_sve_vec_perm_operand): Likewise.
868 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
869 (aarch64_mov_operand): Accept const_poly_int and const_vector.
870 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
871 as well as const_vector.
872 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
873 in file. Use CONST0_RTX and CONSTM1_RTX.
874 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
875 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
876 Use aarch64_simd_imm_zero.
877 * config/aarch64/aarch64-sve.md: New file.
878 * config/aarch64/aarch64.md: Include it.
879 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
880 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
881 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
882 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
883 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
884 (sve): New attribute.
885 (enabled): Disable instructions with the sve attribute unless
887 (movqi, movhi): Pass CONST_POLY_INT operaneds through
888 aarch64_expand_mov_immediate.
889 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
890 CNT[BHSD] immediates.
891 (movti): Split CONST_POLY_INT moves into two halves.
892 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
893 Split additions that need a temporary here if the destination
894 is the stack pointer.
895 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
896 (*add<mode>3_poly_1): New instruction.
897 (set_clobber_cc): New expander.
899 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
901 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
902 parameter and use it instead of GET_MODE_SIZE (innermode). Use
903 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
904 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
905 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
906 Change innermode from fixed_mode_size to machine_mode.
907 (simplify_subreg): Update call accordingly. Handle a constant-sized
908 subreg of a variable-length CONST_VECTOR.
910 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
911 Alan Hayward <alan.hayward@arm.com>
912 David Sherwood <david.sherwood@arm.com>
914 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
915 (add_offset_to_base): New function, split out from...
916 (create_mem_ref): ...here. When handling a scale other than 1,
917 check first whether the address is valid without the offset.
918 Add it into the base if so, leaving the index and scale as-is.
920 2018-01-12 Jakub Jelinek <jakub@redhat.com>
923 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
924 fold_for_warn before checking if arg2 is INTEGER_CST.
926 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
928 * config/rs6000/predicates.md (load_multiple_operation): Delete.
929 (store_multiple_operation): Delete.
930 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
931 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
932 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
933 guarded by TARGET_STRING.
934 (rs6000_output_load_multiple): Delete.
935 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
936 OPTION_MASK_STRING / TARGET_STRING handling.
937 (print_operand) <'N', 'O'>: Add comment that these are unused now.
938 (const rs6000_opt_masks) <"string">: Change mask to 0.
939 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
940 (MASK_STRING): Delete.
941 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
943 (load_multiple): Delete.
950 (store_multiple): Delete.
957 (movmemsi_8reg): Delete.
958 (corresponding unnamed define_insn): Delete.
959 (movmemsi_6reg): Delete.
960 (corresponding unnamed define_insn): Delete.
961 (movmemsi_4reg): Delete.
962 (corresponding unnamed define_insn): Delete.
963 (movmemsi_2reg): Delete.
964 (corresponding unnamed define_insn): Delete.
965 (movmemsi_1reg): Delete.
966 (corresponding unnamed define_insn): Delete.
967 * config/rs6000/rs6000.opt (mno-string): New.
968 (mstring): Replace by deprecation warning stub.
969 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
971 2018-01-12 Jakub Jelinek <jakub@redhat.com>
973 * regrename.c (regrename_do_replace): If replacing the same
974 reg multiple times, try to reuse last created gen_raw_REG.
977 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
978 main to workaround a bug in GDB.
980 2018-01-12 Tom de Vries <tom@codesourcery.com>
983 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
985 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
987 PR rtl-optimization/80481
988 * ira-color.c (get_cap_member): New function.
989 (allocnos_conflict_by_live_ranges_p): Use it.
990 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
991 (setup_slot_coalesced_allocno_live_ranges): Ditto.
993 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
996 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
997 (*saddl_se_1): Ditto.
999 (*saddl_se_1): Ditto.
1001 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1003 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
1004 rather than wi::to_widest for DR_INITs.
1005 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
1006 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
1007 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
1009 (vect_analyze_group_access_1): Note that here.
1011 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1013 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
1014 polynomial type sizes.
1016 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1018 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
1019 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
1020 (gimple_add_tmp_var): Likewise.
1022 2018-01-12 Martin Liska <mliska@suse.cz>
1024 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
1025 (gimple_alloc_sizes): Likewise.
1026 (dump_gimple_statistics): Use PRIu64 in printf format.
1027 * gimple.h: Change uint64_t to int.
1029 2018-01-12 Martin Liska <mliska@suse.cz>
1031 * tree-core.h: Use uint64_t instead of int.
1032 * tree.c (tree_node_counts): Likewise.
1033 (tree_node_sizes): Likewise.
1034 (dump_tree_statistics): Use PRIu64 in printf format.
1036 2018-01-12 Martin Liska <mliska@suse.cz>
1038 * Makefile.in: As qsort_chk is implemented in vec.c, add
1039 vec.o to linkage of gencfn-macros.
1040 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
1041 passing the info to record_node_allocation_statistics.
1042 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
1044 * ggc-common.c (struct ggc_usage): Add operator== and use
1045 it in operator< and compare function.
1046 * mem-stats.h (struct mem_usage): Likewise.
1047 * vec.c (struct vec_usage): Remove operator< and compare
1048 function. Can be simply inherited.
1050 2018-01-12 Martin Jambor <mjambor@suse.cz>
1053 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
1054 * tree-ssa-math-opts.c: Include domwalk.h.
1055 (convert_mult_to_fma_1): New function.
1056 (fma_transformation_info): New type.
1057 (fma_deferring_state): Likewise.
1058 (cancel_fma_deferring): New function.
1059 (result_of_phi): Likewise.
1060 (last_fma_candidate_feeds_initial_phi): Likewise.
1061 (convert_mult_to_fma): Added deferring logic, split actual
1062 transformation to convert_mult_to_fma_1.
1063 (math_opts_dom_walker): New type.
1064 (math_opts_dom_walker::after_dom_children): New method, body moved
1065 here from pass_optimize_widening_mul::execute, added deferring logic
1067 (pass_optimize_widening_mul::execute): Moved most of code to
1068 math_opts_dom_walker::after_dom_children.
1069 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
1070 * config/i386/i386.c (ix86_option_override_internal): Added
1071 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
1073 2018-01-12 Richard Biener <rguenther@suse.de>
1076 * dwarf2out.c (gen_variable_die): Do not reset old_die for
1077 inline instance vars.
1079 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
1082 * config/rx/rx.c (rx_is_restricted_memory_address):
1085 2018-01-12 Richard Biener <rguenther@suse.de>
1087 PR tree-optimization/80846
1088 * target.def (split_reduction): New target hook.
1089 * targhooks.c (default_split_reduction): New function.
1090 * targhooks.h (default_split_reduction): Declare.
1091 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
1092 target requests first reduce vectors by combining low and high
1094 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
1095 (get_vectype_for_scalar_type_and_size): Export.
1096 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
1097 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
1098 * doc/tm.texi: Regenerate.
1099 * config/i386/i386.c (ix86_split_reduction): Implement
1100 TARGET_VECTORIZE_SPLIT_REDUCTION.
1102 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1105 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
1106 in PIC mode except for TARGET_VXWORKS_RTP.
1107 * config/sparc/sparc.c: Include cfgrtl.h.
1108 (TARGET_INIT_PIC_REG): Define.
1109 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
1110 (sparc_pic_register_p): New predicate.
1111 (sparc_legitimate_address_p): Use it.
1112 (sparc_legitimize_pic_address): Likewise.
1113 (sparc_delegitimize_address): Likewise.
1114 (sparc_mode_dependent_address_p): Likewise.
1115 (gen_load_pcrel_sym): Remove 4th parameter.
1116 (load_got_register): Adjust call to above. Remove obsolete stuff.
1117 (sparc_expand_prologue): Do not call load_got_register here.
1118 (sparc_flat_expand_prologue): Likewise.
1119 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
1120 (sparc_use_pseudo_pic_reg): New function.
1121 (sparc_init_pic_reg): Likewise.
1122 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1123 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
1125 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
1127 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
1128 Add item for branch_cost.
1130 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1132 PR rtl-optimization/83565
1133 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
1134 not extend the result to a larger mode for rotate operations.
1135 (num_sign_bit_copies1): Likewise.
1137 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1140 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
1142 Use values-Xc.o for -pedantic.
1143 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
1145 2018-01-12 Martin Liska <mliska@suse.cz>
1148 * ipa-devirt.c (final_warning_record::grow_type_warnings):
1150 (possible_polymorphic_call_targets): Use it.
1151 (ipa_devirt): Likewise.
1153 2018-01-12 Martin Liska <mliska@suse.cz>
1155 * profile-count.h (enum profile_quality): Use 0 as invalid
1156 enum value of profile_quality.
1158 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
1160 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
1161 -mext-string options.
1163 2018-01-12 Richard Biener <rguenther@suse.de>
1165 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
1166 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
1167 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
1169 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
1171 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
1173 * configure.ac (--with-long-double-format): Add support for the
1174 configuration option to change the default long double format on
1176 * config.gcc (powerpc*-linux*-*): Likewise.
1177 * configure: Regenerate.
1178 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
1179 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
1180 used without modification.
1182 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1184 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
1185 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
1186 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
1187 MISC_BUILTIN_SPEC_BARRIER.
1188 (rs6000_init_builtins): Likewise.
1189 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
1191 (speculation_barrier): New define_insn.
1192 * doc/extend.texi: Document __builtin_speculation_barrier.
1194 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1197 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
1198 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
1199 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
1201 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
1202 integral modes instead of "ss" and "sd".
1203 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
1204 vectors with 32-bit and 64-bit elements.
1205 (vecdupssescalarmodesuffix): New mode attribute.
1206 (vec_dup<mode>): Use it.
1208 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
1211 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
1212 frame if argument is passed on stack.
1214 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1217 * ree.c (combine_reaching_defs): Optimize also
1218 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
1219 reg2=any_extend(exp); reg1=reg2;, formatting fix.
1221 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1224 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
1226 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1229 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
1230 after they are computed.
1232 2018-01-11 Bin Cheng <bin.cheng@arm.com>
1234 PR tree-optimization/83695
1235 * gimple-loop-linterchange.cc
1236 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
1237 reset cached scev information after interchange.
1238 (pass_linterchange::execute): Remove call to scev_reset_htab.
1240 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1242 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
1243 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
1244 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
1245 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
1246 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
1247 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
1248 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
1249 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
1250 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
1251 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
1252 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
1253 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
1254 (V_lane_reg): Likewise.
1255 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
1257 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
1258 (vfmal_lane_low<mode>_intrinsic,
1259 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
1260 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
1261 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
1262 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
1263 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
1264 vfmsl_lane_high<mode>_intrinsic): New define_insns.
1266 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1268 * config/arm/arm-cpus.in (fp16fml): New feature.
1269 (ALL_SIMD): Add fp16fml.
1270 (armv8.2-a): Add fp16fml as an option.
1271 (armv8.3-a): Likewise.
1272 (armv8.4-a): Add fp16fml as part of fp16.
1273 * config/arm/arm.h (TARGET_FP16FML): Define.
1274 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
1276 * config/arm/arm-modes.def (V2HF): Define.
1277 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
1278 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
1279 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
1280 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
1281 vfmsl_low, vfmsl_high): New set of builtins.
1282 * config/arm/iterators.md (PLUSMINUS): New code iterator.
1283 (vfml_op): New code attribute.
1284 (VFMLHALVES): New int iterator.
1285 (VFML, VFMLSEL): New mode attributes.
1286 (V_reg): Define mapping for V2HF.
1287 (V_hi, V_lo): New mode attributes.
1288 (VF_constraint): Likewise.
1289 (vfml_half, vfml_half_selector): New int attributes.
1290 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
1292 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
1293 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
1295 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
1296 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
1297 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
1298 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
1300 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
1301 Document new effective target and option set.
1303 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1305 * config/arm/arm-cpus.in (armv8_4): New feature.
1306 (ARMv8_4a): New fgroup.
1307 (armv8.4-a): New arch.
1308 * config/arm/arm-tables.opt: Regenerate.
1309 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
1310 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
1311 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
1312 Add matching rules for -march=armv8.4-a and extensions.
1313 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
1315 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
1318 * config/rx/rx.md (BW): New mode attribute.
1319 (sync_lock_test_and_setsi): Add mode suffix to insn output.
1321 2018-01-11 Richard Biener <rguenther@suse.de>
1323 PR tree-optimization/83435
1324 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
1325 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
1326 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
1328 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1329 Alan Hayward <alan.hayward@arm.com>
1330 David Sherwood <david.sherwood@arm.com>
1332 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
1334 (aarch64_classify_address): Initialize it. Track polynomial offsets.
1335 (aarch64_print_address_internal): Use it to check for a zero offset.
1337 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1338 Alan Hayward <alan.hayward@arm.com>
1339 David Sherwood <david.sherwood@arm.com>
1341 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
1342 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
1343 Return a poly_int64 rather than a HOST_WIDE_INT.
1344 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
1345 rather than a HOST_WIDE_INT.
1346 * config/aarch64/aarch64.h (aarch64_frame): Protect with
1347 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
1348 hard_fp_offset, frame_size, initial_adjust, callee_offset and
1349 final_offset from HOST_WIDE_INT to poly_int64.
1350 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
1351 to_constant when getting the number of units in an Advanced SIMD
1353 (aarch64_builtin_vectorized_function): Check for a constant number
1355 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
1357 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
1358 attribute instead of GET_MODE_NUNITS.
1359 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
1360 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
1361 GET_MODE_SIZE for fixed-size registers.
1362 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
1363 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
1364 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
1365 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
1366 (aarch64_print_operand, aarch64_print_address_internal)
1367 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
1368 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
1369 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
1370 Handle polynomial GET_MODE_SIZE.
1371 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
1372 wider than SImode without modification.
1373 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
1374 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
1375 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
1376 passing and returning SVE modes.
1377 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
1378 rather than GEN_INT.
1379 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
1380 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
1381 (aarch64_allocate_and_probe_stack_space): Likewise.
1382 (aarch64_layout_frame): Cope with polynomial offsets.
1383 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
1384 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
1386 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
1387 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
1388 poly_int64 rather than a HOST_WIDE_INT.
1389 (aarch64_get_separate_components, aarch64_process_components)
1390 (aarch64_expand_prologue, aarch64_expand_epilogue)
1391 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
1392 (aarch64_anchor_offset): New function, split out from...
1393 (aarch64_legitimize_address): ...here.
1394 (aarch64_builtin_vectorization_cost): Handle polynomial
1395 TYPE_VECTOR_SUBPARTS.
1396 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
1398 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
1399 number of elements from the PARALLEL rather than the mode.
1400 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
1401 rather than GET_MODE_BITSIZE.
1402 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
1403 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
1404 (aarch64_expand_vec_perm_const_1): Handle polynomial
1405 d->perm.length () and d->perm elements.
1406 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
1407 Apply to_constant to d->perm elements.
1408 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
1409 polynomial CONST_VECTOR_NUNITS.
1410 (aarch64_move_pointer): Take amount as a poly_int64 rather
1412 (aarch64_progress_pointer): Avoid temporary variable.
1413 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
1414 the mode attribute instead of GET_MODE.
1416 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1417 Alan Hayward <alan.hayward@arm.com>
1418 David Sherwood <david.sherwood@arm.com>
1420 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
1421 x exists before using it.
1422 (aarch64_add_constant_internal): Rename to...
1423 (aarch64_add_offset_1): ...this. Replace regnum with separate
1424 src and dest rtxes. Handle the case in which they're different,
1425 including when the offset is zero. Replace scratchreg with an rtx.
1426 Use 2 additions if there is no spare register into which we can
1427 move a 16-bit constant.
1428 (aarch64_add_constant): Delete.
1429 (aarch64_add_offset): Replace reg with separate src and dest
1430 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
1431 Use aarch64_add_offset_1.
1432 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
1433 an rtx rather than an int. Take the delta as a poly_int64
1434 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
1435 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
1436 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
1437 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
1438 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
1440 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
1441 aarch64_add_constant.
1443 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1445 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
1446 Use scalar_float_mode.
1448 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1450 * config/aarch64/aarch64-simd.md
1451 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
1452 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
1453 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
1454 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
1455 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
1456 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
1457 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
1458 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
1459 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
1460 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
1462 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1465 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
1466 targ_options->x_arm_arch_string is non NULL.
1468 2018-01-11 Tamar Christina <tamar.christina@arm.com>
1470 * config/aarch64/aarch64.h
1471 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
1473 2018-01-11 Sudakshina Das <sudi.das@arm.com>
1476 * expmed.c (emit_store_flag_force): Swap if const op0
1477 and change VOIDmode to mode of op0.
1479 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1481 PR rtl-optimization/83761
1482 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
1483 than bytes to mode_for_size.
1485 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1488 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
1489 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
1492 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1495 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
1496 when in layout mode.
1497 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
1498 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
1501 2018-01-10 Michael Collison <michael.collison@arm.com>
1503 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
1504 * config/aarch64/aarch64-option-extension.def: Add
1505 AARCH64_OPT_EXTENSION of 'fp16fml'.
1506 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1507 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
1508 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
1509 * config/aarch64/constraints.md (Ui7): New constraint.
1510 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
1511 (VFMLA_SEL_W): Ditto.
1514 (VFMLA16_LOW): New int iterator.
1515 (VFMLA16_HIGH): Ditto.
1516 (UNSPEC_FMLAL): New unspec.
1517 (UNSPEC_FMLSL): Ditto.
1518 (UNSPEC_FMLAL2): Ditto.
1519 (UNSPEC_FMLSL2): Ditto.
1520 (f16mac): New code attribute.
1521 * config/aarch64/aarch64-simd-builtins.def
1522 (aarch64_fmlal_lowv2sf): Ditto.
1523 (aarch64_fmlsl_lowv2sf): Ditto.
1524 (aarch64_fmlalq_lowv4sf): Ditto.
1525 (aarch64_fmlslq_lowv4sf): Ditto.
1526 (aarch64_fmlal_highv2sf): Ditto.
1527 (aarch64_fmlsl_highv2sf): Ditto.
1528 (aarch64_fmlalq_highv4sf): Ditto.
1529 (aarch64_fmlslq_highv4sf): Ditto.
1530 (aarch64_fmlal_lane_lowv2sf): Ditto.
1531 (aarch64_fmlsl_lane_lowv2sf): Ditto.
1532 (aarch64_fmlal_laneq_lowv2sf): Ditto.
1533 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
1534 (aarch64_fmlalq_lane_lowv4sf): Ditto.
1535 (aarch64_fmlsl_lane_lowv4sf): Ditto.
1536 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
1537 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
1538 (aarch64_fmlal_lane_highv2sf): Ditto.
1539 (aarch64_fmlsl_lane_highv2sf): Ditto.
1540 (aarch64_fmlal_laneq_highv2sf): Ditto.
1541 (aarch64_fmlsl_laneq_highv2sf): Ditto.
1542 (aarch64_fmlalq_lane_highv4sf): Ditto.
1543 (aarch64_fmlsl_lane_highv4sf): Ditto.
1544 (aarch64_fmlalq_laneq_highv4sf): Ditto.
1545 (aarch64_fmlsl_laneq_highv4sf): Ditto.
1546 * config/aarch64/aarch64-simd.md:
1547 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
1548 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
1549 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
1550 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
1551 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
1552 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
1553 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
1554 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
1555 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
1556 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
1557 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
1558 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
1559 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
1560 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
1561 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
1562 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
1563 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
1564 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
1565 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
1566 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
1567 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
1568 (vfmlsl_low_u32): Ditto.
1569 (vfmlalq_low_u32): Ditto.
1570 (vfmlslq_low_u32): Ditto.
1571 (vfmlal_high_u32): Ditto.
1572 (vfmlsl_high_u32): Ditto.
1573 (vfmlalq_high_u32): Ditto.
1574 (vfmlslq_high_u32): Ditto.
1575 (vfmlal_lane_low_u32): Ditto.
1576 (vfmlsl_lane_low_u32): Ditto.
1577 (vfmlal_laneq_low_u32): Ditto.
1578 (vfmlsl_laneq_low_u32): Ditto.
1579 (vfmlalq_lane_low_u32): Ditto.
1580 (vfmlslq_lane_low_u32): Ditto.
1581 (vfmlalq_laneq_low_u32): Ditto.
1582 (vfmlslq_laneq_low_u32): Ditto.
1583 (vfmlal_lane_high_u32): Ditto.
1584 (vfmlsl_lane_high_u32): Ditto.
1585 (vfmlal_laneq_high_u32): Ditto.
1586 (vfmlsl_laneq_high_u32): Ditto.
1587 (vfmlalq_lane_high_u32): Ditto.
1588 (vfmlslq_lane_high_u32): Ditto.
1589 (vfmlalq_laneq_high_u32): Ditto.
1590 (vfmlslq_laneq_high_u32): Ditto.
1591 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
1592 (AARCH64_FL_FOR_ARCH8_4): New.
1593 (AARCH64_ISA_F16FML): New ISA flag.
1594 (TARGET_F16FML): New feature flag for fp16fml.
1595 (doc/invoke.texi): Document new fp16fml option.
1597 2018-01-10 Michael Collison <michael.collison@arm.com>
1599 * config/aarch64/aarch64-builtins.c:
1600 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
1601 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1602 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
1603 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
1604 (AARCH64_ISA_SHA3): New ISA flag.
1605 (TARGET_SHA3): New feature flag for sha3.
1606 * config/aarch64/iterators.md (sha512_op): New int attribute.
1607 (CRYPTO_SHA512): New int iterator.
1608 (UNSPEC_SHA512H): New unspec.
1609 (UNSPEC_SHA512H2): Ditto.
1610 (UNSPEC_SHA512SU0): Ditto.
1611 (UNSPEC_SHA512SU1): Ditto.
1612 * config/aarch64/aarch64-simd-builtins.def
1613 (aarch64_crypto_sha512hqv2di): New builtin.
1614 (aarch64_crypto_sha512h2qv2di): Ditto.
1615 (aarch64_crypto_sha512su0qv2di): Ditto.
1616 (aarch64_crypto_sha512su1qv2di): Ditto.
1617 (aarch64_eor3qv8hi): Ditto.
1618 (aarch64_rax1qv2di): Ditto.
1619 (aarch64_xarqv2di): Ditto.
1620 (aarch64_bcaxqv8hi): Ditto.
1621 * config/aarch64/aarch64-simd.md:
1622 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
1623 (aarch64_crypto_sha512su0qv2di): Ditto.
1624 (aarch64_crypto_sha512su1qv2di): Ditto.
1625 (aarch64_eor3qv8hi): Ditto.
1626 (aarch64_rax1qv2di): Ditto.
1627 (aarch64_xarqv2di): Ditto.
1628 (aarch64_bcaxqv8hi): Ditto.
1629 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
1630 (vsha512h2q_u64): Ditto.
1631 (vsha512su0q_u64): Ditto.
1632 (vsha512su1q_u64): Ditto.
1633 (veor3q_u16): Ditto.
1634 (vrax1q_u64): Ditto.
1636 (vbcaxq_u16): Ditto.
1637 * config/arm/types.md (crypto_sha512): New type attribute.
1638 (crypto_sha3): Ditto.
1639 (doc/invoke.texi): Document new sha3 option.
1641 2018-01-10 Michael Collison <michael.collison@arm.com>
1643 * config/aarch64/aarch64-builtins.c:
1644 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
1645 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1646 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
1647 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
1648 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
1649 (AARCH64_ISA_SM4): New ISA flag.
1650 (TARGET_SM4): New feature flag for sm4.
1651 * config/aarch64/aarch64-simd-builtins.def
1652 (aarch64_sm3ss1qv4si): Ditto.
1653 (aarch64_sm3tt1aq4si): Ditto.
1654 (aarch64_sm3tt1bq4si): Ditto.
1655 (aarch64_sm3tt2aq4si): Ditto.
1656 (aarch64_sm3tt2bq4si): Ditto.
1657 (aarch64_sm3partw1qv4si): Ditto.
1658 (aarch64_sm3partw2qv4si): Ditto.
1659 (aarch64_sm4eqv4si): Ditto.
1660 (aarch64_sm4ekeyqv4si): Ditto.
1661 * config/aarch64/aarch64-simd.md:
1662 (aarch64_sm3ss1qv4si): Ditto.
1663 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
1664 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
1665 (aarch64_sm4eqv4si): Ditto.
1666 (aarch64_sm4ekeyqv4si): Ditto.
1667 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
1668 (sm3part_op): Ditto.
1669 (CRYPTO_SM3TT): Ditto.
1670 (CRYPTO_SM3PART): Ditto.
1671 (UNSPEC_SM3SS1): New unspec.
1672 (UNSPEC_SM3TT1A): Ditto.
1673 (UNSPEC_SM3TT1B): Ditto.
1674 (UNSPEC_SM3TT2A): Ditto.
1675 (UNSPEC_SM3TT2B): Ditto.
1676 (UNSPEC_SM3PARTW1): Ditto.
1677 (UNSPEC_SM3PARTW2): Ditto.
1678 (UNSPEC_SM4E): Ditto.
1679 (UNSPEC_SM4EKEY): Ditto.
1680 * config/aarch64/constraints.md (Ui2): New constraint.
1681 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
1682 * config/arm/types.md (crypto_sm3): New type attribute.
1683 (crypto_sm4): Ditto.
1684 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
1685 (vsm3tt1aq_u32): Ditto.
1686 (vsm3tt1bq_u32): Ditto.
1687 (vsm3tt2aq_u32): Ditto.
1688 (vsm3tt2bq_u32): Ditto.
1689 (vsm3partw1q_u32): Ditto.
1690 (vsm3partw2q_u32): Ditto.
1691 (vsm4eq_u32): Ditto.
1692 (vsm4ekeyq_u32): Ditto.
1693 (doc/invoke.texi): Document new sm4 option.
1695 2018-01-10 Michael Collison <michael.collison@arm.com>
1697 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
1698 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
1699 (AARCH64_FL_FOR_ARCH8_4): New.
1700 (AARCH64_FL_V8_4): New flag.
1701 (doc/invoke.texi): Document new armv8.4-a option.
1703 2018-01-10 Michael Collison <michael.collison@arm.com>
1705 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
1706 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
1707 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
1708 * config/aarch64/aarch64-option-extension.def: Add
1709 AARCH64_OPT_EXTENSION of 'sha2'.
1710 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
1711 (crypto): Disable sha2 and aes if crypto disabled.
1712 (crypto): Enable aes and sha2 if enabled.
1713 (simd): Disable sha2 and aes if simd disabled.
1714 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
1716 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
1717 (TARGET_SHA2): New feature flag for sha2.
1718 (TARGET_AES): New feature flag for aes.
1719 * config/aarch64/aarch64-simd.md:
1720 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
1721 conditional on TARGET_AES.
1722 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
1723 (aarch64_crypto_sha1hsi): Make pattern conditional
1725 (aarch64_crypto_sha1hv4si): Ditto.
1726 (aarch64_be_crypto_sha1hv4si): Ditto.
1727 (aarch64_crypto_sha1su1v4si): Ditto.
1728 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
1729 (aarch64_crypto_sha1su0v4si): Ditto.
1730 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
1731 (aarch64_crypto_sha256su0v4si): Ditto.
1732 (aarch64_crypto_sha256su1v4si): Ditto.
1733 (doc/invoke.texi): Document new aes and sha2 options.
1735 2018-01-10 Martin Sebor <msebor@redhat.com>
1737 PR tree-optimization/83781
1738 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
1741 2018-01-11 Martin Sebor <msebor@gmail.com>
1742 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1744 PR tree-optimization/83501
1745 PR tree-optimization/81703
1747 * tree-ssa-strlen.c (get_string_cst): Rename...
1748 (get_string_len): ...to this. Handle global constants.
1749 (handle_char_store): Adjust.
1751 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
1752 Jim Wilson <jimw@sifive.com>
1754 * config/riscv/riscv-protos.h (riscv_output_return): New.
1755 * config/riscv/riscv.c (struct machine_function): New naked_p field.
1756 (riscv_attribute_table, riscv_output_return),
1757 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
1758 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
1759 (riscv_compute_frame_info): Only compute frame->mask if not a naked
1761 (riscv_expand_prologue): Add early return for naked function.
1762 (riscv_expand_epilogue): Likewise.
1763 (riscv_function_ok_for_sibcall): Return false for naked function.
1764 (riscv_set_current_function): New.
1765 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
1766 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
1767 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
1768 * doc/extend.texi (RISC-V Function Attributes): New.
1770 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
1772 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
1773 check for 128-bit long double before checking TCmode.
1774 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
1775 128-bit long doubles before checking TFmode or TCmode.
1776 (FLOAT128_IBM_P): Likewise.
1778 2018-01-10 Martin Sebor <msebor@redhat.com>
1780 PR tree-optimization/83671
1781 * builtins.c (c_strlen): Unconditionally return zero for the empty
1783 Use -Warray-bounds for warnings.
1784 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
1785 for non-constant array indices with COMPONENT_REF, arrays of
1786 arrays, and pointers to arrays.
1787 (gimple_fold_builtin_strlen): Determine and set length range for
1788 non-constant character arrays.
1790 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
1793 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
1796 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
1798 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
1800 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
1803 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
1804 VECTOR_MEM_ALTIVEC_OR_VSX_P.
1805 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
1806 indexed_or_indirect_operand predicate.
1807 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
1808 (*vsx_le_perm_load_v8hi): Likewise.
1809 (*vsx_le_perm_load_v16qi): Likewise.
1810 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
1811 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
1812 (*vsx_le_perm_store_v8hi): Likewise.
1813 (*vsx_le_perm_store_v16qi): Likewise.
1814 (eight unnamed splitters): Likewise.
1816 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
1818 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
1819 * config/rs6000/emmintrin.h: Likewise.
1820 * config/rs6000/mmintrin.h: Likewise.
1821 * config/rs6000/xmmintrin.h: Likewise.
1823 2018-01-10 David Malcolm <dmalcolm@redhat.com>
1826 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
1828 * tree.c (tree_nop_conversion): Return true for location wrapper
1830 (maybe_wrap_with_location): New function.
1831 (selftest::check_strip_nops): New function.
1832 (selftest::test_location_wrappers): New function.
1833 (selftest::tree_c_tests): Call it.
1834 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
1835 (maybe_wrap_with_location): New decl.
1836 (EXPR_LOCATION_WRAPPER_P): New macro.
1837 (location_wrapper_p): New inline function.
1838 (tree_strip_any_location_wrapper): New inline function.
1840 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
1843 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
1844 stack_realign_offset for the largest alignment of stack slot
1846 (ix86_find_max_used_stack_alignment): New function.
1847 (ix86_finalize_stack_frame_flags): Use it. Set
1848 max_used_stack_alignment if we don't realign stack.
1849 * config/i386/i386.h (machine_function): Add
1850 max_used_stack_alignment.
1852 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
1854 * config/arm/arm.opt (-mbranch-cost): New option.
1855 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
1858 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
1861 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
1862 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
1864 2018-01-10 Richard Biener <rguenther@suse.de>
1867 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
1868 early out so it also covers the case where we have a non-NULL
1871 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
1873 PR tree-optimization/83753
1874 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
1875 for non-strided grouped accesses if the number of elements is 1.
1877 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
1880 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
1881 * i386.h (TARGET_USE_GATHER): Define.
1882 * x86-tune.def (X86_TUNE_USE_GATHER): New.
1884 2018-01-10 Martin Liska <mliska@suse.cz>
1887 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
1888 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
1890 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
1891 CLEANUP_NO_PARTITIONING is not set.
1893 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
1895 * doc/rtl.texi: Remove documentation of (const ...) wrappers
1896 for vectors, as a partial revert of r254296.
1897 * rtl.h (const_vec_p): Delete.
1898 (const_vec_duplicate_p): Don't test for vector CONSTs.
1899 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
1900 * expmed.c (make_tree): Likewise.
1903 * common.md (E, F): Use CONSTANT_P instead of checking for
1905 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
1906 checking for CONST_VECTOR.
1908 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1911 * predict.c (force_edge_cold): Handle in more sane way edges
1914 2018-01-09 Carl Love <cel@us.ibm.com>
1916 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
1918 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
1919 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
1920 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
1921 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
1922 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
1923 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
1924 * config/rs6000/rs6000-protos.h: Add extern defition for
1925 rs6000_generate_float2_double_code.
1926 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
1928 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
1929 (float2_v2df): Add define_expand.
1931 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
1934 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
1935 op_mode in the force_to_mode call.
1937 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1939 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
1940 instead of checking each element individually.
1941 (aarch64_evpc_uzp): Likewise.
1942 (aarch64_evpc_zip): Likewise.
1943 (aarch64_evpc_ext): Likewise.
1944 (aarch64_evpc_rev): Likewise.
1945 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
1946 instead of checking each element individually. Return true without
1948 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
1949 whether all selected elements come from the same input, instead of
1950 checking each element individually. Remove calls to gen_rtx_REG,
1951 start_sequence and end_sequence and instead assert that no rtl is
1954 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1956 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
1957 order of HIGH and CONST checks.
1959 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1961 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
1962 if the destination isn't an SSA_NAME.
1964 2018-01-09 Richard Biener <rguenther@suse.de>
1966 PR tree-optimization/83668
1967 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
1969 (canonicalize_loop_form): ... here, renamed from ...
1970 (canonicalize_loop_closed_ssa_form): ... this and amended to
1971 swap successor edges for loop exit blocks to make us use
1972 the RPO order we need for initial schedule generation.
1974 2018-01-09 Joseph Myers <joseph@codesourcery.com>
1976 PR tree-optimization/64811
1977 * match.pd: When optimizing comparisons with Inf, avoid
1978 introducing or losing exceptions from comparisons with NaN.
1980 2018-01-09 Martin Liska <mliska@suse.cz>
1983 * asan.c (shadow_mem_size): Add gcc_assert.
1985 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
1987 Don't save registers in main().
1990 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
1991 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
1992 * config/avr/avr.c (avr_set_current_function): Don't error if
1993 naked, OS_task or OS_main are specified at the same time.
1994 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
1996 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
1998 * common/config/avr/avr-common.c (avr_option_optimization_table):
1999 Switch on -mmain-is-OS_task for optimizing compilations.
2001 2018-01-09 Richard Biener <rguenther@suse.de>
2003 PR tree-optimization/83572
2004 * graphite.c: Include cfganal.h.
2005 (graphite_transform_loops): Connect infinite loops to exit
2006 and remove fake edges at the end.
2008 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
2010 * ipa-inline.c (edge_badness): Revert accidental checkin.
2012 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
2015 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
2016 symbols; not inline clones.
2018 2018-01-09 Jakub Jelinek <jakub@redhat.com>
2021 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
2022 hard registers. Formatting fixes.
2024 PR preprocessor/83722
2025 * gcc.c (try_generate_repro): Pass
2026 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
2027 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
2030 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
2031 Kito Cheng <kito.cheng@gmail.com>
2033 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
2034 (riscv_leaf_function_p): Delete.
2035 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
2037 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2039 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
2041 (do_ifelse): New function.
2042 (do_isel): New function.
2043 (do_sub3): New function.
2044 (do_add3): New function.
2045 (do_load_mask_compare): New function.
2046 (do_overlap_load_compare): New function.
2047 (expand_compare_loop): New function.
2048 (expand_block_compare): Call expand_compare_loop() when appropriate.
2049 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
2051 (-mblock-compare-inline-loop-limit): New option.
2053 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2056 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
2057 Reverse order of second and third operands in first alternative.
2058 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
2059 of first and second elements in UNSPEC_VPERMR vector.
2060 (altivec_expand_vec_perm_le): Likewise.
2062 2017-01-08 Jeff Law <law@redhat.com>
2064 PR rtl-optimizatin/81308
2065 * tree-switch-conversion.c (cfg_altered): New file scoped static.
2066 (process_switch): If group_case_labels makes a change, then set
2068 (pass_convert_switch::execute): If a switch is converted, then
2069 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
2071 PR rtl-optimization/81308
2072 * recog.c (split_all_insns): Conditionally cleanup the CFG after
2075 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
2077 PR target/83663 - Revert r255946
2078 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
2079 generation for cases where splatting a value is not useful.
2080 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
2081 across a vec_duplicate and a paradoxical subreg forming a vector
2082 mode to a vec_concat.
2084 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2086 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
2087 -march=armv8.3-a variants.
2088 * config/arm/t-multilib: Likewise.
2089 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
2091 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2093 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
2095 (cceq_ior_compare_complement): Give it a name so I can use it, and
2096 change boolean_or_operator predicate to boolean_operator so it can
2097 be used to generate a crand.
2098 (eqne): New code iterator.
2099 (bd/bd_neg): New code_attrs.
2100 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
2101 a single define_insn.
2102 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
2103 decrement (bdnzt/bdnzf/bdzt/bdzf).
2104 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
2105 with the new names of the branch decrement patterns, and added the
2106 names of the branch decrement conditional patterns.
2108 2018-01-08 Richard Biener <rguenther@suse.de>
2110 PR tree-optimization/83563
2111 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
2114 2018-01-08 Richard Biener <rguenther@suse.de>
2117 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
2119 2018-01-08 Richard Biener <rguenther@suse.de>
2121 PR tree-optimization/83685
2122 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
2123 references to abnormals.
2125 2018-01-08 Richard Biener <rguenther@suse.de>
2128 * dwarf2out.c (output_indirect_strings): Handle empty
2129 skeleton_debug_str_hash.
2130 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
2132 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2134 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
2135 (emit_store_direct): Likewise.
2136 (arc_trampoline_adjust_address): Likewise.
2137 (arc_asm_trampoline_template): New function.
2138 (arc_initialize_trampoline): Use asm_trampoline_template.
2139 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
2140 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
2141 * config/arc/arc.md (flush_icache): Delete pattern.
2143 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2145 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
2146 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
2149 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2152 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
2153 by not USED_FOR_TARGET.
2154 (make_pass_resolve_sw_modes): Likewise.
2156 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2158 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
2161 2018-01-08 Richard Biener <rguenther@suse.de>
2164 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
2166 2018-01-08 Richard Biener <rguenther@suse.de>
2169 * match.pd ((t * 2) / 2) -> t): Add missing :c.
2171 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
2174 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
2175 basic blocks with a small number of successors.
2176 (convert_control_dep_chain_into_preds): Improve handling of
2178 (dump_predicates): Split apart into...
2179 (dump_pred_chain): ...here...
2180 (dump_pred_info): ...and here.
2181 (can_one_predicate_be_invalidated_p): Add debugging printfs.
2182 (can_chain_union_be_invalidated_p): Improve check for invalidation
2184 (uninit_uses_cannot_happen): Avoid unnecessary if
2185 convert_control_dep_chain_into_preds yielded nothing.
2187 2018-01-06 Martin Sebor <msebor@redhat.com>
2189 PR tree-optimization/83640
2190 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
2191 subtracting negative offset from size.
2192 (builtin_access::overlap): Adjust offset bounds of the access to fall
2193 within the size of the object if possible.
2195 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
2197 PR rtl-optimization/83699
2198 * expmed.c (extract_bit_field_1): Restrict the vector usage of
2199 extract_bit_field_as_subreg to cases in which the extracted
2200 value is also a vector.
2202 * lra-constraints.c (process_alt_operands): Test for the equivalence
2203 substitutions when detecting a possible reload cycle.
2205 2018-01-06 Jakub Jelinek <jakub@redhat.com>
2208 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
2209 by default if flag_selective_schedling{,2}. Formatting fixes.
2211 PR rtl-optimization/83682
2212 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
2213 if it has non-VECTOR_MODE element mode.
2214 (vec_duplicate_p): Likewise.
2217 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
2218 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
2220 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2223 * config/i386/i386-builtin.def
2224 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
2225 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
2226 Require also OPTION_MASK_ISA_AVX512F in addition to
2227 OPTION_MASK_ISA_GFNI.
2228 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
2229 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
2230 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
2231 to OPTION_MASK_ISA_GFNI.
2232 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
2233 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
2234 OPTION_MASK_ISA_AVX512BW.
2235 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
2236 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
2237 addition to OPTION_MASK_ISA_GFNI.
2238 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
2239 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
2240 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
2241 to OPTION_MASK_ISA_GFNI.
2242 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
2243 a requirement for all ISAs rather than any of them with a few
2245 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
2247 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
2248 bitmasks to be enabled with 3 exceptions, instead of requiring any
2249 enabled ISA with lots of exceptions.
2250 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
2251 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
2252 Change avx512bw in isa attribute to avx512f.
2253 * config/i386/sgxintrin.h: Add license boilerplate.
2254 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
2255 to __AVX512F__ and __AVX512VL to __AVX512VL__.
2256 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
2257 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
2259 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
2260 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
2261 temporarily sse2 rather than sse if not enabled already.
2264 * config/i386/sse.md (VI248_VLBW): Rename to ...
2265 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
2266 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
2267 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
2268 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
2269 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
2270 mode iterator instead of VI248_VLBW.
2272 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
2274 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
2275 (record_modified): Skip clobbers; add debug output.
2276 (param_change_prob): Use sreal frequencies.
2278 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2280 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
2281 punt for user-aligned variables.
2283 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2285 * tree-chrec.c (chrec_contains_symbols): Return true for
2288 2018-01-05 Sudakshina Das <sudi.das@arm.com>
2291 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
2292 of (x|y) == x for BICS pattern.
2294 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2296 PR tree-optimization/83605
2297 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
2298 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
2301 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
2303 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
2304 * config/epiphany/rtems.h: New file.
2306 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2307 Uros Bizjak <ubizjak@gmail.com>
2310 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
2311 QIreg_operand instead of register_operand predicate.
2312 * config/i386/i386.c (ix86_rop_should_change_byte_p,
2313 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
2314 comments instead of -fmitigate[-_]rop.
2316 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2319 * cgraphunit.c (symbol_table::compile): Switch to text_section
2320 before calling assembly_start debug hook.
2321 * run-rtl-passes.c (run_rtl_passes): Likewise.
2324 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2326 * tree-vrp.c (extract_range_from_binary_expr_1): Check
2327 range_int_cst_p rather than !symbolic_range_p before calling
2328 extract_range_from_multiplicative_op_1.
2330 2017-01-04 Jeff Law <law@redhat.com>
2332 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
2333 redundant test in assertion.
2335 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2337 * doc/rtl.texi: Document machine_mode wrapper classes.
2339 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2341 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
2344 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2346 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
2347 the VEC_PERM_EXPR fold to fail.
2349 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2352 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
2353 to switched_sections.
2355 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2358 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
2361 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
2364 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
2365 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
2367 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2370 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
2371 is BLKmode and bitpos not zero or mode change is needed.
2373 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2376 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
2379 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
2382 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
2383 instead of MULT rtx. Update all corresponding splitters.
2385 (*ssub<modesuffix>): Ditto.
2387 (*cmp_sadd_di): Update split patterns.
2388 (*cmp_sadd_si): Ditto.
2389 (*cmp_sadd_sidi): Ditto.
2390 (*cmp_ssub_di): Ditto.
2391 (*cmp_ssub_si): Ditto.
2392 (*cmp_ssub_sidi): Ditto.
2393 * config/alpha/predicates.md (const23_operand): New predicate.
2394 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
2395 Look for ASHIFT, not MULT inner operand.
2396 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
2398 2018-01-04 Martin Liska <mliska@suse.cz>
2400 PR gcov-profile/83669
2401 * gcov.c (output_intermediate_file): Add version to intermediate
2403 * doc/gcov.texi: Document new field 'version' in intermediate
2404 file format. Fix location of '-k' option of gcov command.
2406 2018-01-04 Martin Liska <mliska@suse.cz>
2409 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
2411 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2413 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
2415 2018-01-03 Martin Sebor <msebor@redhat.com>
2417 PR tree-optimization/83655
2418 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
2419 checking calls with invalid arguments.
2421 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2423 * tree-vect-stmts.c (vect_get_store_rhs): New function.
2424 (vectorizable_mask_load_store): Delete.
2425 (vectorizable_call): Return false for masked loads and stores.
2426 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
2427 instead of gimple_assign_rhs1.
2428 (vectorizable_load): Handle IFN_MASK_LOAD.
2429 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
2431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2433 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
2435 (vectorizable_mask_load_store): ...here.
2436 (vectorizable_load): ...and here.
2438 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2440 * tree-vect-stmts.c (vect_build_all_ones_mask)
2441 (vect_build_zero_merge_argument): New functions, split out from...
2442 (vectorizable_load): ...here.
2444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2446 * tree-vect-stmts.c (vect_check_store_rhs): New function,
2448 (vectorizable_mask_load_store): ...here.
2449 (vectorizable_store): ...and here.
2451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2453 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
2455 (vectorizable_mask_load_store): ...here.
2457 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2459 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
2460 (vect_model_store_cost): Take a vec_load_store_type instead of a
2462 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
2463 (vect_model_store_cost): Take a vec_load_store_type instead of a
2465 (vectorizable_mask_load_store): Update accordingly.
2466 (vectorizable_store): Likewise.
2467 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
2469 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2471 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
2472 IFN_MASK_LOAD calls here rather than...
2473 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
2475 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2476 Alan Hayward <alan.hayward@arm.com>
2477 David Sherwood <david.sherwood@arm.com>
2479 * expmed.c (extract_bit_field_1): For vector extracts,
2480 fall back to extract_bit_field_as_subreg if vec_extract
2483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2484 Alan Hayward <alan.hayward@arm.com>
2485 David Sherwood <david.sherwood@arm.com>
2487 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
2488 they are variable or constant sized.
2489 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
2490 slots for constant-sized data.
2492 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2493 Alan Hayward <alan.hayward@arm.com>
2494 David Sherwood <david.sherwood@arm.com>
2496 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
2497 handling COND_EXPRs with boolean comparisons, try to find a better
2498 basis for the mask type than the boolean itself.
2500 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2502 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
2503 is calculated and how it can be overridden.
2504 * genmodes.c (max_bitsize_mode_any_mode): New variable.
2505 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
2507 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
2510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2511 Alan Hayward <alan.hayward@arm.com>
2512 David Sherwood <david.sherwood@arm.com>
2514 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
2515 Remove the mode argument.
2516 (aarch64_simd_valid_immediate): Remove the mode and inverse
2518 * config/aarch64/iterators.md (bitsize): New iterator.
2519 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
2520 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
2521 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
2522 aarch64_simd_valid_immediate.
2523 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
2524 (aarch64_reg_or_bic_imm): Likewise.
2525 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
2526 with an insn_type enum and msl with a modifier_type enum.
2527 Replace element_width with a scalar_mode. Change the shift
2528 to unsigned int. Add constructors for scalar_float_mode and
2529 scalar_int_mode elements.
2530 (aarch64_vect_float_const_representable_p): Delete.
2531 (aarch64_can_const_movi_rtx_p)
2532 (aarch64_simd_scalar_immediate_valid_for_move)
2533 (aarch64_simd_make_constant): Update call to
2534 aarch64_simd_valid_immediate.
2535 (aarch64_advsimd_valid_immediate_hs): New function.
2536 (aarch64_advsimd_valid_immediate): Likewise.
2537 (aarch64_simd_valid_immediate): Remove mode and inverse
2538 arguments. Rewrite to use the above. Use const_vec_duplicate_p
2539 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
2540 and aarch64_float_const_representable_p on the result.
2541 (aarch64_output_simd_mov_immediate): Remove mode argument.
2542 Update call to aarch64_simd_valid_immediate and use of
2543 simd_immediate_info.
2544 (aarch64_output_scalar_simd_mov_immediate): Update call
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_precision): Prefix with CONST_MODE_PRECISION.
2552 (mode_nunits): Likewise CONST_MODE_NUNITS.
2553 * machmode.def (ADJUST_NUNITS): Document.
2554 * genmodes.c (mode_data::need_nunits_adj): New field.
2555 (blank_mode): Update accordingly.
2556 (adj_nunits): New variable.
2557 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
2559 (emit_mode_size_inline): Set need_bytesize_adj for all modes
2560 listed in adj_nunits.
2561 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
2562 listed in adj_nunits. Don't emit case statements for such modes.
2563 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
2564 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
2565 nothing if adj_nunits is nonnull.
2566 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
2567 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
2568 (emit_mode_fbit): Update use of print_maybe_const_decl.
2569 (emit_move_size): Likewise. Treat the array as non-const
2571 (emit_mode_adjustments): Handle adj_nunits.
2573 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2575 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
2576 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
2577 (VECTOR_MODES): Use it.
2578 (make_vector_modes): Take the prefix as an argument.
2580 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2581 Alan Hayward <alan.hayward@arm.com>
2582 David Sherwood <david.sherwood@arm.com>
2584 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
2585 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
2586 for MODE_VECTOR_BOOL.
2587 * machmode.def (VECTOR_BOOL_MODE): Document.
2588 * genmodes.c (VECTOR_BOOL_MODE): New macro.
2589 (make_vector_bool_mode): New function.
2590 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
2592 * lto-streamer-in.c (lto_input_mode_table): Likewise.
2593 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
2595 * stor-layout.c (int_mode_for_mode): Likewise.
2596 * tree.c (build_vector_type_for_mode): Likewise.
2597 * varasm.c (output_constant_pool_2): Likewise.
2598 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
2599 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
2600 for MODE_VECTOR_BOOL.
2601 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
2602 of mode class checks.
2603 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
2604 instead of a list of mode class checks.
2605 (expand_vector_scalar_condition): Likewise.
2606 (type_for_widest_vector_mode): Handle BImode as an inner mode.
2608 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2609 Alan Hayward <alan.hayward@arm.com>
2610 David Sherwood <david.sherwood@arm.com>
2612 * machmode.h (mode_size): Change from unsigned short to
2614 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
2615 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
2616 or if measurement_type is not polynomial.
2617 (fixed_size_mode::includes_p): Check for constant-sized modes.
2618 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
2619 return a poly_uint16 rather than an unsigned short.
2620 (emit_mode_size): Change the type of mode_size from unsigned short
2621 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
2622 (emit_mode_adjustments): Cope with polynomial vector sizes.
2623 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2625 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2627 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
2628 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
2629 * caller-save.c (setup_save_areas): Likewise.
2630 (replace_reg_with_saved_mem): Likewise.
2631 * calls.c (emit_library_call_value_1): Likewise.
2632 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
2633 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
2634 (gen_lowpart_for_combine): Likewise.
2635 * convert.c (convert_to_integer_1): Likewise.
2636 * cse.c (equiv_constant, cse_insn): Likewise.
2637 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
2638 (cselib_subst_to_values): Likewise.
2639 * dce.c (word_dce_process_block): Likewise.
2640 * df-problems.c (df_word_lr_mark_ref): Likewise.
2641 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
2642 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
2643 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
2644 (rtl_for_decl_location): Likewise.
2645 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
2646 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
2647 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
2648 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
2649 (expand_expr_real_1): Likewise.
2650 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
2651 (pad_below): Likewise.
2652 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
2653 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
2654 * ira.c (get_subreg_tracking_sizes): Likewise.
2655 * ira-build.c (ira_create_allocno_objects): Likewise.
2656 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
2657 (ira_sort_regnos_for_alter_reg): Likewise.
2658 * ira-costs.c (record_operand_costs): Likewise.
2659 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
2660 (resolve_simple_move): Likewise.
2661 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
2662 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
2663 (lra_constraints): Likewise.
2664 (CONST_POOL_OK_P): Reject variable-sized modes.
2665 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
2666 (add_pseudo_to_slot, lra_spill): Likewise.
2667 * omp-low.c (omp_clause_aligned_alignment): Likewise.
2668 * optabs-query.c (get_best_extraction_insn): Likewise.
2669 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
2670 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
2671 (expand_mult_highpart, valid_multiword_target_p): Likewise.
2672 * recog.c (offsettable_address_addr_space_p): Likewise.
2673 * regcprop.c (maybe_mode_change): Likewise.
2674 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
2675 * regrename.c (build_def_use): Likewise.
2676 * regstat.c (dump_reg_info): Likewise.
2677 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
2678 (find_reloads, find_reloads_subreg_address): Likewise.
2679 * reload1.c (eliminate_regs_1): Likewise.
2680 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
2681 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
2682 (simplify_binary_operation_1, simplify_subreg): Likewise.
2683 * targhooks.c (default_function_arg_padding): Likewise.
2684 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
2685 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
2686 (verify_gimple_assign_ternary): Likewise.
2687 * tree-inline.c (estimate_move_cost): Likewise.
2688 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2689 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
2690 (get_address_cost_ainc): Likewise.
2691 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
2692 (vect_supportable_dr_alignment): Likewise.
2693 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
2694 (vectorizable_reduction): Likewise.
2695 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
2696 (vectorizable_operation, vectorizable_load): Likewise.
2697 * tree.c (build_same_sized_truth_vector_type): Likewise.
2698 * valtrack.c (cleanup_auto_inc_dec): Likewise.
2699 * var-tracking.c (emit_note_insn_var_location): Likewise.
2700 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
2701 (ADDR_VEC_ALIGN): Likewise.
2703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2704 Alan Hayward <alan.hayward@arm.com>
2705 David Sherwood <david.sherwood@arm.com>
2707 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
2709 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
2710 or if measurement_type is polynomial.
2711 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
2712 * combine.c (make_extraction): Likewise.
2713 * dse.c (find_shift_sequence): Likewise.
2714 * dwarf2out.c (mem_loc_descriptor): Likewise.
2715 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
2716 (extract_bit_field, extract_low_bits): Likewise.
2717 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
2718 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
2719 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
2720 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
2721 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
2722 * reload.c (find_reloads): Likewise.
2723 * reload1.c (alter_reg): Likewise.
2724 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
2725 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
2726 * tree-if-conv.c (predicate_mem_writes): Likewise.
2727 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
2728 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
2729 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
2730 * valtrack.c (dead_debug_insert_temp): Likewise.
2731 * varasm.c (mergeable_constant_section): Likewise.
2732 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
2734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2735 Alan Hayward <alan.hayward@arm.com>
2736 David Sherwood <david.sherwood@arm.com>
2738 * expr.c (expand_assignment): Cope with polynomial mode sizes
2739 when assigning to a CONCAT.
2741 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2742 Alan Hayward <alan.hayward@arm.com>
2743 David Sherwood <david.sherwood@arm.com>
2745 * machmode.h (mode_precision): Change from unsigned short to
2747 (mode_to_precision): Return a poly_uint16 rather than an unsigned
2749 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
2750 or if measurement_type is not polynomial.
2751 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
2752 in which the mode is already known to be a scalar_int_mode.
2753 * genmodes.c (emit_mode_precision): Change the type of mode_precision
2754 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
2756 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2757 for GET_MODE_PRECISION.
2758 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2759 for GET_MODE_PRECISION.
2760 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
2762 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
2763 (expand_field_assignment, make_extraction): Likewise.
2764 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
2765 (get_last_value): Likewise.
2766 * convert.c (convert_to_integer_1): Likewise.
2767 * cse.c (cse_insn): Likewise.
2768 * expr.c (expand_expr_real_1): Likewise.
2769 * lra-constraints.c (simplify_operand_subreg): Likewise.
2770 * optabs-query.c (can_atomic_load_p): Likewise.
2771 * optabs.c (expand_atomic_load): Likewise.
2772 (expand_atomic_store): Likewise.
2773 * ree.c (combine_reaching_defs): Likewise.
2774 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
2775 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
2776 * tree.h (type_has_mode_precision_p): Likewise.
2777 * ubsan.c (instrument_si_overflow): Likewise.
2779 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2780 Alan Hayward <alan.hayward@arm.com>
2781 David Sherwood <david.sherwood@arm.com>
2783 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
2784 polynomial numbers of units.
2785 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
2786 (valid_vector_subparts_p): New function.
2787 (build_vector_type): Remove temporary shim and take the number
2788 of units as a poly_uint64 rather than an int.
2789 (build_opaque_vector_type): Take the number of units as a
2790 poly_uint64 rather than an int.
2791 * tree.c (build_vector_from_ctor): Handle polynomial
2792 TYPE_VECTOR_SUBPARTS.
2793 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
2794 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
2795 (build_vector_from_val): If the number of units is variable,
2796 use build_vec_duplicate_cst for constant operands and
2797 VEC_DUPLICATE_EXPR otherwise.
2798 (make_vector_type): Remove temporary is_constant ().
2799 (build_vector_type, build_opaque_vector_type): Take the number of
2800 units as a poly_uint64 rather than an int.
2801 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
2803 * cfgexpand.c (expand_debug_expr): Likewise.
2804 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
2805 (store_constructor, expand_expr_real_1): Likewise.
2806 (const_scalar_mask_from_tree): Likewise.
2807 * fold-const-call.c (fold_const_reduction): Likewise.
2808 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
2809 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
2810 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
2811 (fold_relational_const): Likewise.
2812 (native_interpret_vector): Likewise. Change the size from an
2813 int to an unsigned int.
2814 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
2815 TYPE_VECTOR_SUBPARTS.
2816 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
2817 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
2818 duplicating a non-constant operand into a variable-length vector.
2819 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
2820 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
2821 * ipa-icf.c (sem_variable::equals): Likewise.
2822 * match.pd: Likewise.
2823 * omp-simd-clone.c (simd_clone_subparts): Likewise.
2824 * print-tree.c (print_node): Likewise.
2825 * stor-layout.c (layout_type): Likewise.
2826 * targhooks.c (default_builtin_vectorization_cost): Likewise.
2827 * tree-cfg.c (verify_gimple_comparison): Likewise.
2828 (verify_gimple_assign_binary): Likewise.
2829 (verify_gimple_assign_ternary): Likewise.
2830 (verify_gimple_assign_single): Likewise.
2831 * tree-pretty-print.c (dump_generic_node): Likewise.
2832 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2833 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
2834 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
2835 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
2836 (vect_shift_permute_load_chain): Likewise.
2837 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
2838 (expand_vector_condition, optimize_vector_constructor): Likewise.
2839 (lower_vec_perm, get_compute_type): Likewise.
2840 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
2841 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
2842 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
2843 (vect_recog_mask_conversion_pattern): Likewise.
2844 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
2845 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
2846 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2847 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
2848 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
2849 (vectorizable_shift, vectorizable_operation, vectorizable_store)
2850 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
2851 (supportable_widening_operation): Likewise.
2852 (supportable_narrowing_operation): Likewise.
2853 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
2855 * varasm.c (output_constant): Likewise.
2857 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2858 Alan Hayward <alan.hayward@arm.com>
2859 David Sherwood <david.sherwood@arm.com>
2861 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
2862 so that both the length == 3 and length != 3 cases set up their
2863 own permute vectors. Add comments explaining why we know the
2864 number of elements is constant.
2865 (vect_permute_load_chain): Likewise.
2867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2868 Alan Hayward <alan.hayward@arm.com>
2869 David Sherwood <david.sherwood@arm.com>
2871 * machmode.h (mode_nunits): Change from unsigned char to
2873 (ONLY_FIXED_SIZE_MODES): New macro.
2874 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
2875 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
2876 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
2878 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
2879 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
2880 or if measurement_type is not polynomial.
2881 * genmodes.c (ZERO_COEFFS): New macro.
2882 (emit_mode_nunits_inline): Make mode_nunits_inline return a
2884 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
2885 Use ZERO_COEFFS when emitting initializers.
2886 * data-streamer.h (bp_pack_poly_value): New function.
2887 (bp_unpack_poly_value): Likewise.
2888 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
2889 for GET_MODE_NUNITS.
2890 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
2891 for GET_MODE_NUNITS.
2892 * tree.c (make_vector_type): Remove temporary shim and make
2893 the real function take the number of units as a poly_uint64
2895 (build_vector_type_for_mode): Handle polynomial nunits.
2896 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
2897 * emit-rtl.c (const_vec_series_p_1): Likewise.
2898 (gen_rtx_CONST_VECTOR): Likewise.
2899 * fold-const.c (test_vec_duplicate_folding): Likewise.
2900 * genrecog.c (validate_pattern): Likewise.
2901 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
2902 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
2903 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
2904 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
2905 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
2906 * rtlanal.c (subreg_get_info): Likewise.
2907 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2908 (vect_grouped_load_supported): Likewise.
2909 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
2910 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
2911 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
2912 (simplify_const_unary_operation, simplify_binary_operation_1)
2913 (simplify_const_binary_operation, simplify_ternary_operation)
2914 (test_vector_ops_duplicate, test_vector_ops): Likewise.
2915 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
2916 instead of CONST_VECTOR_NUNITS.
2917 * varasm.c (output_constant_pool_2): Likewise.
2918 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
2919 explicit-encoded elements in the XVEC for variable-length vectors.
2921 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2923 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
2925 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2926 Alan Hayward <alan.hayward@arm.com>
2927 David Sherwood <david.sherwood@arm.com>
2929 * coretypes.h (fixed_size_mode): Declare.
2930 (fixed_size_mode_pod): New typedef.
2931 * builtins.h (target_builtins::x_apply_args_mode)
2932 (target_builtins::x_apply_result_mode): Change type to
2933 fixed_size_mode_pod.
2934 * builtins.c (apply_args_size, apply_result_size, result_vector)
2935 (expand_builtin_apply_args_1, expand_builtin_apply)
2936 (expand_builtin_return): Update accordingly.
2938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2940 * cse.c (hash_rtx_cb): Hash only the encoded elements.
2941 * cselib.c (cselib_hash_rtx): Likewise.
2942 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
2943 CONST_VECTOR encoding.
2945 2017-01-03 Jakub Jelinek <jakub@redhat.com>
2946 Jeff Law <law@redhat.com>
2949 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
2950 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
2951 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
2952 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
2955 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
2956 explicitly probe *sp in a noreturn function if there were any callee
2957 register saves or frame pointer is needed.
2959 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2962 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
2963 BLKmode for ternary, binary or unary expressions.
2966 * var-tracking.c (delete_vta_debug_insn): New inline function.
2967 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
2968 insns from get_insns () to NULL instead of each bb separately.
2969 Use delete_vta_debug_insn. No longer static.
2970 (vt_debug_insns_local, variable_tracking_main_1): Adjust
2971 delete_vta_debug_insns callers.
2972 * rtl.h (delete_vta_debug_insns): Declare.
2973 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
2974 instead of variable_tracking_main.
2976 2018-01-03 Martin Sebor <msebor@redhat.com>
2978 PR tree-optimization/83603
2979 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
2980 arguments past the endof the argument list in functions declared
2981 without a prototype.
2982 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
2983 Avoid checking when arguments are null.
2985 2018-01-03 Martin Sebor <msebor@redhat.com>
2988 * doc/extend.texi (attribute const): Fix a typo.
2989 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
2990 issuing -Wsuggest-attribute for void functions.
2992 2018-01-03 Martin Sebor <msebor@redhat.com>
2994 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
2995 offset_int::from instead of wide_int::to_shwi.
2996 (maybe_diag_overlap): Remove assertion.
2997 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
2998 * gimple-ssa-sprintf.c (format_directive): Same.
2999 (parse_directive): Same.
3000 (sprintf_dom_walker::compute_format_length): Same.
3001 (try_substitute_return_value): Same.
3003 2017-01-03 Jeff Law <law@redhat.com>
3006 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
3007 non-constant residual for zero at runtime and avoid probing in
3008 that case. Reorganize code for trailing problem to mirror handling
3011 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3013 PR tree-optimization/83501
3014 * tree-ssa-strlen.c (get_string_cst): New.
3015 (handle_char_store): Call get_string_cst.
3017 2018-01-03 Martin Liska <mliska@suse.cz>
3019 PR tree-optimization/83593
3020 * tree-ssa-strlen.c: Include tree-cfg.h.
3021 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
3022 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
3023 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
3025 (strlen_dom_walker::before_dom_children): Call
3026 gimple_purge_dead_eh_edges. Dump tranformation with details
3028 (strlen_dom_walker::before_dom_children): Update call by adding
3029 new argument cleanup_eh.
3030 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
3032 2018-01-03 Martin Liska <mliska@suse.cz>
3035 * cif-code.def (VARIADIC_THUNK): New enum value.
3036 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
3039 2018-01-03 Jan Beulich <jbeulich@suse.com>
3041 * sse.md (mov<mode>_internal): Tighten condition for when to use
3042 vmovdqu<ssescalarsize> for TI and OI modes.
3044 2018-01-03 Jakub Jelinek <jakub@redhat.com>
3046 Update copyright years.
3048 2018-01-03 Martin Liska <mliska@suse.cz>
3051 * ipa-visibility.c (function_and_variable_visibility): Skip
3052 functions with noipa attribure.
3054 2018-01-03 Jakub Jelinek <jakub@redhat.com>
3056 * gcc.c (process_command): Update copyright notice dates.
3057 * gcov-dump.c (print_version): Ditto.
3058 * gcov.c (print_version): Ditto.
3059 * gcov-tool.c (print_version): Ditto.
3060 * gengtype.c (create_file): Ditto.
3061 * doc/cpp.texi: Bump @copying's copyright year.
3062 * doc/cppinternals.texi: Ditto.
3063 * doc/gcc.texi: Ditto.
3064 * doc/gccint.texi: Ditto.
3065 * doc/gcov.texi: Ditto.
3066 * doc/install.texi: Ditto.
3067 * doc/invoke.texi: Ditto.
3069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3071 * vector-builder.h (vector_builder::m_full_nelts): Change from
3072 unsigned int to poly_uint64.
3073 (vector_builder::full_nelts): Update prototype accordingly.
3074 (vector_builder::new_vector): Likewise.
3075 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
3076 (vector_builder::operator ==): Likewise.
3077 (vector_builder::finalize): Likewise.
3078 * int-vector-builder.h (int_vector_builder::int_vector_builder):
3079 Take the number of elements as a poly_uint64 rather than an
3081 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
3082 from unsigned int to poly_uint64.
3083 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
3084 (vec_perm_indices::new_vector): Likewise.
3085 (vec_perm_indices::length): Likewise.
3086 (vec_perm_indices::nelts_per_input): Likewise.
3087 (vec_perm_indices::input_nelts): Likewise.
3088 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
3089 number of elements per input as a poly_uint64 rather than an
3090 unsigned int. Use the original encoding for variable-length
3091 vectors, rather than clamping each individual element.
3092 For the second and subsequent elements in each pattern,
3093 clamp the step and base before clamping their sum.
3094 (vec_perm_indices::series_p): Handle polynomial element counts.
3095 (vec_perm_indices::all_in_range_p): Likewise.
3096 (vec_perm_indices_to_tree): Likewise.
3097 (vec_perm_indices_to_rtx): Likewise.
3098 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
3099 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
3100 (tree_vector_builder::new_binary_operation): Handle polynomial
3101 element counts. Return false if we need to know the number
3102 of elements at compile time.
3103 * fold-const.c (fold_vec_perm): Punt if the number of elements
3104 isn't known at compile time.
3106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3108 * vec-perm-indices.h (vec_perm_builder): Change element type
3109 from HOST_WIDE_INT to poly_int64.
3110 (vec_perm_indices::element_type): Update accordingly.
3111 (vec_perm_indices::clamp): Handle polynomial element_types.
3112 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
3113 (vec_perm_indices::all_in_range_p): Likewise.
3114 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
3116 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
3117 polynomial vec_perm_indices element types.
3118 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
3119 * fold-const.c (fold_vec_perm): Likewise.
3120 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
3121 * tree-vect-generic.c (lower_vec_perm): Likewise.
3122 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
3123 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
3124 element type to HOST_WIDE_INT.
3126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3127 Alan Hayward <alan.hayward@arm.com>
3128 David Sherwood <david.sherwood@arm.com>
3130 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
3131 rather than an int. Use plus_constant.
3132 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
3133 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
3135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3136 Alan Hayward <alan.hayward@arm.com>
3137 David Sherwood <david.sherwood@arm.com>
3139 * calls.c (emit_call_1, expand_call): Change struct_value_size from
3140 a HOST_WIDE_INT to a poly_int64.
3142 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3143 Alan Hayward <alan.hayward@arm.com>
3144 David Sherwood <david.sherwood@arm.com>
3146 * calls.c (load_register_parameters): Cope with polynomial
3147 mode sizes. Require a constant size for BLKmode parameters
3148 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
3149 forces a parameter to be padded at the lsb end in order to
3150 fill a complete number of words, require the parameter size
3151 to be ordered wrt UNITS_PER_WORD.
3153 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3154 Alan Hayward <alan.hayward@arm.com>
3155 David Sherwood <david.sherwood@arm.com>
3157 * reload1.c (spill_stack_slot_width): Change element type
3158 from unsigned int to poly_uint64_pod.
3159 (alter_reg): Treat mode sizes as polynomial.
3161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3162 Alan Hayward <alan.hayward@arm.com>
3163 David Sherwood <david.sherwood@arm.com>
3165 * reload.c (complex_word_subreg_p): New function.
3166 (reload_inner_reg_of_subreg, push_reload): Use it.
3168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3169 Alan Hayward <alan.hayward@arm.com>
3170 David Sherwood <david.sherwood@arm.com>
3172 * lra-constraints.c (process_alt_operands): Reject matched
3173 operands whose sizes aren't ordered.
3174 (match_reload): Refer to this check here.
3176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3177 Alan Hayward <alan.hayward@arm.com>
3178 David Sherwood <david.sherwood@arm.com>
3180 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
3181 that the mode size is in the set {1, 2, 4, 8, 16}.
3183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3184 Alan Hayward <alan.hayward@arm.com>
3185 David Sherwood <david.sherwood@arm.com>
3187 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
3188 Use plus_constant instead of gen_rtx_PLUS.
3190 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3191 Alan Hayward <alan.hayward@arm.com>
3192 David Sherwood <david.sherwood@arm.com>
3194 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
3195 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
3196 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
3197 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
3198 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
3199 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
3200 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
3201 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
3202 * config/i386/i386.c (ix86_push_rounding): ...this new function.
3203 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
3205 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
3206 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
3207 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
3208 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
3209 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
3210 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
3211 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
3212 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
3213 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
3214 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
3216 * expr.c (emit_move_resolve_push): Treat the input and result
3217 of PUSH_ROUNDING as a poly_int64.
3218 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
3219 (emit_push_insn): Likewise.
3220 * lra-eliminations.c (mark_not_eliminable): Likewise.
3221 * recog.c (push_operand): Likewise.
3222 * reload1.c (elimination_effects): Likewise.
3223 * rtlanal.c (nonzero_bits1): Likewise.
3224 * calls.c (store_one_arg): Likewise. Require the padding to be
3225 known at compile time.
3227 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3228 Alan Hayward <alan.hayward@arm.com>
3229 David Sherwood <david.sherwood@arm.com>
3231 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
3232 Use plus_constant instead of gen_rtx_PLUS.
3234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3235 Alan Hayward <alan.hayward@arm.com>
3236 David Sherwood <david.sherwood@arm.com>
3238 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
3241 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3242 Alan Hayward <alan.hayward@arm.com>
3243 David Sherwood <david.sherwood@arm.com>
3245 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
3246 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
3247 via stack temporaries. Treat the mode size as polynomial too.
3249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3250 Alan Hayward <alan.hayward@arm.com>
3251 David Sherwood <david.sherwood@arm.com>
3253 * expr.c (expand_expr_real_2): When handling conversions involving
3254 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
3255 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
3256 as a poly_uint64 too.
3258 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3259 Alan Hayward <alan.hayward@arm.com>
3260 David Sherwood <david.sherwood@arm.com>
3262 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
3264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3265 Alan Hayward <alan.hayward@arm.com>
3266 David Sherwood <david.sherwood@arm.com>
3268 * combine.c (can_change_dest_mode): Handle polynomial
3269 REGMODE_NATURAL_SIZE.
3270 * expmed.c (store_bit_field_1): Likewise.
3271 * expr.c (store_constructor): Likewise.
3272 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
3273 and polynomial REGMODE_NATURAL_SIZE.
3274 (gen_lowpart_common): Likewise.
3275 * reginfo.c (record_subregs_of_mode): Likewise.
3276 * rtlanal.c (read_modify_subreg_p): Likewise.
3278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3279 Alan Hayward <alan.hayward@arm.com>
3280 David Sherwood <david.sherwood@arm.com>
3282 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
3283 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
3291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3292 Alan Hayward <alan.hayward@arm.com>
3293 David Sherwood <david.sherwood@arm.com>
3295 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
3296 in a POINTER_PLUS_EXPR.
3298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3299 Alan Hayward <alan.hayward@arm.com>
3300 David Sherwood <david.sherwood@arm.com>
3302 * omp-simd-clone.c (simd_clone_subparts): New function.
3303 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
3304 (ipa_simd_modify_function_body): Likewise.
3306 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3307 Alan Hayward <alan.hayward@arm.com>
3308 David Sherwood <david.sherwood@arm.com>
3310 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
3311 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
3312 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
3313 (expand_vector_condition, vector_element): Likewise.
3314 (subparts_gt): New function.
3315 (get_compute_type): Use subparts_gt.
3316 (count_type_subparts): Delete.
3317 (expand_vector_operations_1): Use subparts_gt instead of
3318 count_type_subparts.
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-data-refs.c (vect_no_alias_p): Replace with...
3325 (vect_compile_time_alias): ...this new function. Do the calculation
3326 on poly_ints rather than trees.
3327 (vect_prune_runtime_alias_test_list): Update call accordingly.
3329 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3330 Alan Hayward <alan.hayward@arm.com>
3331 David Sherwood <david.sherwood@arm.com>
3333 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
3335 (vect_schedule_slp_instance): Likewise.
3337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3338 Alan Hayward <alan.hayward@arm.com>
3339 David Sherwood <david.sherwood@arm.com>
3341 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
3342 constant and extern definitions for variable-length vectors.
3343 (vect_get_constant_vectors): Note that the number of units
3344 is known to be constant.
3346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3347 Alan Hayward <alan.hayward@arm.com>
3348 David Sherwood <david.sherwood@arm.com>
3350 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
3351 of units as polynomial. Choose between WIDE and NARROW based
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-stmts.c (simd_clone_subparts): New function.
3359 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 * tree-vect-stmts.c (vectorizable_call): Treat the number of
3366 vectors as polynomial. Use build_index_vector for
3369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3370 Alan Hayward <alan.hayward@arm.com>
3371 David Sherwood <david.sherwood@arm.com>
3373 * tree-vect-stmts.c (get_load_store_type): Treat the number of
3374 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
3375 for variable-length vectors.
3376 (vectorizable_mask_load_store): Treat the number of units as
3377 polynomial, asserting that it is constant if the condition has
3378 already been enforced.
3379 (vectorizable_store, vectorizable_load): Likewise.
3381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3382 Alan Hayward <alan.hayward@arm.com>
3383 David Sherwood <david.sherwood@arm.com>
3385 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
3386 of units as polynomial. Punt if we can't tell at compile time
3387 which vector contains the final result.
3389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3390 Alan Hayward <alan.hayward@arm.com>
3391 David Sherwood <david.sherwood@arm.com>
3393 * tree-vect-loop.c (vectorizable_induction): Treat the number
3394 of units as polynomial. Punt on SLP inductions. Use an integer
3395 VEC_SERIES_EXPR for variable-length integer reductions. Use a
3396 cast of such a series for variable-length floating-point
3399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3400 Alan Hayward <alan.hayward@arm.com>
3401 David Sherwood <david.sherwood@arm.com>
3403 * tree.h (build_index_vector): Declare.
3404 * tree.c (build_index_vector): New function.
3405 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
3406 of units as polynomial, forcibly converting it to a constant if
3407 vectorizable_reduction has already enforced the condition.
3408 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
3409 to create a {1,2,3,...} vector.
3410 (vectorizable_reduction): Treat the number of units as polynomial.
3411 Choose vectype_in based on the largest scalar element size rather
3412 than the smallest number of units. Enforce the restrictions
3415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3416 Alan Hayward <alan.hayward@arm.com>
3417 David Sherwood <david.sherwood@arm.com>
3419 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
3420 number of units as polynomial.
3422 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3423 Alan Hayward <alan.hayward@arm.com>
3424 David Sherwood <david.sherwood@arm.com>
3426 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
3427 * target.def (autovectorize_vector_sizes): Return the vector sizes
3428 by pointer, using vector_sizes rather than a bitmask.
3429 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
3430 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
3431 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
3433 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
3434 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
3435 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
3436 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
3437 * omp-general.c (omp_max_vf): Likewise.
3438 * omp-low.c (omp_clause_aligned_alignment): Likewise.
3439 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
3440 * tree-vect-loop.c (vect_analyze_loop): Likewise.
3441 * tree-vect-slp.c (vect_slp_bb): Likewise.
3442 * doc/tm.texi: Regenerate.
3443 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
3445 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
3446 the vector size as a poly_uint64 rather than an unsigned int.
3447 (current_vector_size): Change from an unsigned int to a poly_uint64.
3448 (get_vectype_for_scalar_type): Update accordingly.
3449 * tree.h (build_truth_vector_type): Take the size and number of
3450 units as a poly_uint64 rather than an unsigned int.
3451 (build_vector_type): Add a temporary overload that takes
3452 the number of units as a poly_uint64 rather than an unsigned int.
3453 * tree.c (make_vector_type): Likewise.
3454 (build_truth_vector_type): Take the number of units as a poly_uint64
3455 rather than an unsigned int.
3457 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3458 Alan Hayward <alan.hayward@arm.com>
3459 David Sherwood <david.sherwood@arm.com>
3461 * target.def (get_mask_mode): Take the number of units and length
3462 as poly_uint64s rather than unsigned ints.
3463 * targhooks.h (default_get_mask_mode): Update accordingly.
3464 * targhooks.c (default_get_mask_mode): Likewise.
3465 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
3466 * doc/tm.texi: Regenerate.
3468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3469 Alan Hayward <alan.hayward@arm.com>
3470 David Sherwood <david.sherwood@arm.com>
3472 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
3473 * omp-general.c (omp_max_vf): Likewise.
3474 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
3475 (expand_omp_simd): Handle polynomial safelen.
3476 * omp-low.c (omplow_simd_context): Add a default constructor.
3477 (omplow_simd_context::max_vf): Change from int to poly_uint64.
3478 (lower_rec_simd_input_clauses): Update accordingly.
3479 (lower_rec_input_clauses): Likewise.
3481 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3482 Alan Hayward <alan.hayward@arm.com>
3483 David Sherwood <david.sherwood@arm.com>
3485 * tree-vectorizer.h (vect_nunits_for_cost): New function.
3486 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
3487 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
3488 (vect_analyze_slp_cost): Likewise.
3489 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
3490 (vect_model_load_cost): Likewise.
3492 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3493 Alan Hayward <alan.hayward@arm.com>
3494 David Sherwood <david.sherwood@arm.com>
3496 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
3497 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
3498 from an unsigned int * to a poly_uint64_pod *.
3499 (calculate_unrolling_factor): New function.
3500 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
3502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3503 Alan Hayward <alan.hayward@arm.com>
3504 David Sherwood <david.sherwood@arm.com>
3506 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
3507 from an unsigned int to a poly_uint64.
3508 (_loop_vec_info::slp_unrolling_factor): Likewise.
3509 (_loop_vec_info::vectorization_factor): Change from an int
3511 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
3512 (vect_get_num_vectors): New function.
3513 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
3514 (vect_get_num_copies): Use vect_get_num_vectors.
3515 (vect_analyze_data_ref_dependences): Change max_vf from an int *
3516 to an unsigned int *.
3517 (vect_analyze_data_refs): Change min_vf from an int * to a
3519 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
3520 than an unsigned HOST_WIDE_INT.
3521 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
3522 (vect_analyze_data_ref_dependence): Change max_vf from an int *
3523 to an unsigned int *.
3524 (vect_analyze_data_ref_dependences): Likewise.
3525 (vect_compute_data_ref_alignment): Handle polynomial vf.
3526 (vect_enhance_data_refs_alignment): Likewise.
3527 (vect_prune_runtime_alias_test_list): Likewise.
3528 (vect_shift_permute_load_chain): Likewise.
3529 (vect_supportable_dr_alignment): Likewise.
3530 (dependence_distance_ge_vf): Take the vectorization factor as a
3531 poly_uint64 rather than an unsigned HOST_WIDE_INT.
3532 (vect_analyze_data_refs): Change min_vf from an int * to a
3534 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
3535 vfm1 as a poly_uint64 rather than an int. Make the same change
3536 for the returned bound_scalar.
3537 (vect_gen_vector_loop_niters): Handle polynomial vf.
3538 (vect_do_peeling): Likewise. Update call to
3539 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
3540 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
3542 * tree-vect-loop.c (vect_determine_vectorization_factor)
3543 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
3544 (vect_get_known_peeling_cost): Likewise.
3545 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
3546 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
3547 (vect_transform_loop): Likewise. Use the lowest possible VF when
3548 updating the upper bounds of the loop.
3549 (vect_min_worthwhile_factor): Make static. Return an unsigned int
3551 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
3552 polynomial unroll factors.
3553 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
3554 (vect_make_slp_decision): Likewise.
3555 (vect_supported_load_permutation_p): Likewise, and polynomial
3557 (vect_analyze_slp_cost): Handle polynomial vf.
3558 (vect_slp_analyze_node_operations): Likewise.
3559 (vect_slp_analyze_bb_1): Likewise.
3560 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
3561 than an unsigned HOST_WIDE_INT.
3562 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
3563 (vectorizable_load): Handle polynomial vf.
3564 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
3566 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
3568 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3569 Alan Hayward <alan.hayward@arm.com>
3570 David Sherwood <david.sherwood@arm.com>
3572 * match.pd: Handle bit operations involving three constants
3573 and try to fold one pair.
3575 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3577 * tree-vect-loop-manip.c: Include gimple-fold.h.
3578 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
3579 niters_maybe_zero parameters. Handle other cases besides a step of 1.
3580 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
3581 Add a path that uses a step of VF instead of 1, but disable it
3583 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
3584 and niters_no_overflow parameters. Update calls to
3585 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
3586 Create a new SSA name if the latter choses to use a ste other
3587 than zero, and return it via niters_vector_mult_vf_var.
3588 * tree-vect-loop.c (vect_transform_loop): Update calls to
3589 vect_do_peeling, vect_gen_vector_loop_niters and
3590 slpeel_make_loop_iterate_ntimes.
3591 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
3592 (vect_gen_vector_loop_niters): Update declarations after above changes.
3594 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
3596 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
3597 128-bit round to integer instructions.
3598 (ceil<mode>2): Likewise.
3599 (btrunc<mode>2): Likewise.
3600 (round<mode>2): Likewise.
3602 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3604 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
3605 unaligned VSX load/store on P8/P9.
3606 (expand_block_clear): Allow the use of unaligned VSX
3607 load/store on P8/P9.
3609 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3611 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
3613 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
3614 swap associated with both a load and a store.
3616 2018-01-02 Andrew Waterman <andrew@sifive.com>
3618 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
3619 * config/riscv/riscv.md (clear_cache): Use it.
3621 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
3623 * web.c: Remove out-of-date comment.
3625 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3627 * expr.c (fixup_args_size_notes): Check that any existing
3628 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
3629 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
3630 (emit_single_push_insn): ...here.
3632 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3634 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
3635 (const_vector_encoded_nelts): New function.
3636 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
3637 (const_vector_int_elt, const_vector_elt): Declare.
3638 * emit-rtl.c (const_vector_int_elt_1): New function.
3639 (const_vector_elt): Likewise.
3640 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
3641 of CONST_VECTOR_ELT.
3643 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3645 * expr.c: Include rtx-vector-builder.h.
3646 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
3647 directly on the tree encoding.
3648 (const_vector_from_tree): Likewise.
3649 * optabs.c: Include rtx-vector-builder.h.
3650 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
3651 sequence of "u" values.
3652 * vec-perm-indices.c: Include rtx-vector-builder.h.
3653 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
3654 directly on the vec_perm_indices encoding.
3656 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3658 * doc/rtl.texi (const_vector): Describe new encoding scheme.
3659 * Makefile.in (OBJS): Add rtx-vector-builder.o.
3660 * rtx-vector-builder.h: New file.
3661 * rtx-vector-builder.c: Likewise.
3662 * rtl.h (rtx_def::u2): Add a const_vector field.
3663 (CONST_VECTOR_NPATTERNS): New macro.
3664 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
3665 (CONST_VECTOR_DUPLICATE_P): Likewise.
3666 (CONST_VECTOR_STEPPED_P): Likewise.
3667 (CONST_VECTOR_ENCODED_ELT): Likewise.
3668 (const_vec_duplicate_p): Check for a duplicated vector encoding.
3669 (unwrap_const_vec_duplicate): Likewise.
3670 (const_vec_series_p): Check for a non-duplicated vector encoding.
3671 Say that the function only returns true for integer vectors.
3672 * emit-rtl.c: Include rtx-vector-builder.h.
3673 (gen_const_vec_duplicate_1): Delete.
3674 (gen_const_vector): Call gen_const_vec_duplicate instead of
3675 gen_const_vec_duplicate_1.
3676 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
3677 (gen_const_vec_duplicate): Use rtx_vector_builder.
3678 (gen_const_vec_series): Likewise.
3679 (gen_rtx_CONST_VECTOR): Likewise.
3680 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
3681 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
3682 Build a new vector rather than modifying a CONST_VECTOR in-place.
3683 (handle_special_swappables): Update call accordingly.
3684 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
3685 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
3686 Build a new vector rather than modifying a CONST_VECTOR in-place.
3687 (handle_special_swappables): Update call accordingly.
3689 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3691 * simplify-rtx.c (simplify_const_binary_operation): Use
3692 CONST_VECTOR_ELT instead of XVECEXP.
3694 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3696 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
3697 the selector elements to be different from the data elements
3698 if the selector is a VECTOR_CST.
3699 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
3700 ssizetype for the selector.
3702 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3704 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
3705 before testing each element individually.
3706 * tree-vect-generic.c (lower_vec_perm): Likewise.
3708 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3710 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
3711 * selftest-run-tests.c (selftest::run_tests): Call it.
3712 * vector-builder.h (vector_builder::operator ==): New function.
3713 (vector_builder::operator !=): Likewise.
3714 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
3715 (vec_perm_indices::all_from_input_p): New function.
3716 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
3717 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
3718 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
3719 instead of reading the VECTOR_CST directly. Detect whether both
3720 vector inputs are the same before constructing the vec_perm_indices,
3721 and update the number of inputs argument accordingly. Use the
3722 utility functions added above. Only construct sel2 if we need to.
3724 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3726 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
3727 the broadcast of the low byte.
3728 (expand_mult_highpart): Use an explicit encoding for the permutes.
3729 * optabs-query.c (can_mult_highpart_p): Likewise.
3730 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
3731 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3732 (vectorizable_bswap): Likewise.
3733 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
3734 explicit encoding for the power-of-2 permutes.
3735 (vect_permute_store_chain): Likewise.
3736 (vect_grouped_load_supported): Likewise.
3737 (vect_permute_load_chain): Likewise.
3739 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3741 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
3742 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
3743 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
3744 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
3745 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
3746 (vect_gen_perm_mask_any): Likewise.
3748 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3750 * int-vector-builder.h: New file.
3751 * vec-perm-indices.h: Include int-vector-builder.h.
3752 (vec_perm_indices): Redefine as an int_vector_builder.
3753 (auto_vec_perm_indices): Delete.
3754 (vec_perm_builder): Redefine as a stand-alone class.
3755 (vec_perm_indices::vec_perm_indices): New function.
3756 (vec_perm_indices::clamp): Likewise.
3757 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
3758 (vec_perm_indices::new_vector): New function.
3759 (vec_perm_indices::new_expanded_vector): Update for new
3760 vec_perm_indices class.
3761 (vec_perm_indices::rotate_inputs): New function.
3762 (vec_perm_indices::all_in_range_p): Operate directly on the
3763 encoded form, without computing elided elements.
3764 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
3765 encoding. Update for new vec_perm_indices class.
3766 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
3767 the given vec_perm_builder.
3768 (expand_vec_perm_var): Update vec_perm_builder constructor.
3769 (expand_mult_highpart): Use vec_perm_builder instead of
3770 auto_vec_perm_indices.
3771 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
3772 vec_perm_indices instead of auto_vec_perm_indices. Use a single
3773 or double series encoding as appropriate.
3774 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
3775 vec_perm_indices instead of auto_vec_perm_indices.
3776 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3777 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3778 (vect_permute_store_chain): Likewise.
3779 (vect_grouped_load_supported): Likewise.
3780 (vect_permute_load_chain): Likewise.
3781 (vect_shift_permute_load_chain): Likewise.
3782 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
3783 (vect_transform_slp_perm_load): Likewise.
3784 (vect_schedule_slp_instance): Likewise.
3785 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3786 (vectorizable_mask_load_store): Likewise.
3787 (vectorizable_bswap): Likewise.
3788 (vectorizable_store): Likewise.
3789 (vectorizable_load): Likewise.
3790 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
3791 vec_perm_indices instead of auto_vec_perm_indices. Use
3792 tree_to_vec_perm_builder to read the vector from a tree.
3793 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
3794 vec_perm_builder instead of a vec_perm_indices.
3795 (have_whole_vector_shift): Use vec_perm_builder and
3796 vec_perm_indices instead of auto_vec_perm_indices. Leave the
3797 truncation to calc_vec_perm_mask_for_shift.
3798 (vect_create_epilog_for_reduction): Likewise.
3799 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
3800 from auto_vec_perm_indices to vec_perm_indices.
3801 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
3802 instead of changing individual elements.
3803 (aarch64_vectorize_vec_perm_const): Use new_vector to install
3804 the vector in d.perm.
3805 * config/arm/arm.c (expand_vec_perm_d::perm): Change
3806 from auto_vec_perm_indices to vec_perm_indices.
3807 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
3808 instead of changing individual elements.
3809 (arm_vectorize_vec_perm_const): Use new_vector to install
3810 the vector in d.perm.
3811 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
3812 Update vec_perm_builder constructor.
3813 (rs6000_expand_interleave): Likewise.
3814 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
3815 (rs6000_expand_interleave): Likewise.
3817 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3819 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
3820 to qimode could truncate the indices.
3821 * optabs.c (expand_vec_perm_var): Likewise.
3823 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3825 * Makefile.in (OBJS): Add vec-perm-indices.o.
3826 * vec-perm-indices.h: New file.
3827 * vec-perm-indices.c: Likewise.
3828 * target.h (vec_perm_indices): Replace with a forward class
3830 (auto_vec_perm_indices): Move to vec-perm-indices.h.
3831 * optabs.h: Include vec-perm-indices.h.
3832 (expand_vec_perm): Delete.
3833 (selector_fits_mode_p, expand_vec_perm_var): Declare.
3834 (expand_vec_perm_const): Declare.
3835 * target.def (vec_perm_const_ok): Replace with...
3836 (vec_perm_const): ...this new hook.
3837 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
3838 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
3839 * doc/tm.texi: Regenerate.
3840 * optabs.def (vec_perm_const): Delete.
3841 * doc/md.texi (vec_perm_const): Likewise.
3842 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
3843 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
3844 expand_vec_perm for constant permutation vectors. Assert that
3845 the mode of variable permutation vectors is the integer equivalent
3846 of the mode that is being permuted.
3847 * optabs-query.h (selector_fits_mode_p): Declare.
3848 * optabs-query.c: Include vec-perm-indices.h.
3849 (selector_fits_mode_p): New function.
3850 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
3851 is defined, instead of checking whether the vec_perm_const_optab
3852 exists. Use targetm.vectorize.vec_perm_const instead of
3853 targetm.vectorize.vec_perm_const_ok. Check whether the indices
3854 fit in the vector mode before using a variable permute.
3855 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
3856 vec_perm_indices instead of an rtx.
3857 (expand_vec_perm): Replace with...
3858 (expand_vec_perm_const): ...this new function. Take the selector
3859 as a vec_perm_indices rather than an rtx. Also take the mode of
3860 the selector. Update call to shift_amt_for_vec_perm_mask.
3861 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
3862 Use vec_perm_indices::new_expanded_vector to expand the original
3863 selector into bytes. Check whether the indices fit in the vector
3864 mode before using a variable permute.
3865 (expand_vec_perm_var): Make global.
3866 (expand_mult_highpart): Use expand_vec_perm_const.
3867 * fold-const.c: Includes vec-perm-indices.h.
3868 * tree-ssa-forwprop.c: Likewise.
3869 * tree-vect-data-refs.c: Likewise.
3870 * tree-vect-generic.c: Likewise.
3871 * tree-vect-loop.c: Likewise.
3872 * tree-vect-slp.c: Likewise.
3873 * tree-vect-stmts.c: Likewise.
3874 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
3876 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
3877 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
3878 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
3879 (aarch64_vectorize_vec_perm_const): ...this new function.
3880 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3881 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3882 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
3883 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
3884 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3885 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3886 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
3888 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
3889 check for NEON modes.
3890 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
3891 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
3892 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
3893 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
3895 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
3896 the old VEC_PERM_CONST conditions.
3897 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
3898 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
3899 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
3900 (ia64_vectorize_vec_perm_const_ok): Merge into...
3901 (ia64_vectorize_vec_perm_const): ...this new function.
3902 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
3903 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
3904 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
3905 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
3906 * config/mips/mips.c (mips_expand_vec_perm_const)
3907 (mips_vectorize_vec_perm_const_ok): Merge into...
3908 (mips_vectorize_vec_perm_const): ...this new function.
3909 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
3910 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
3911 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
3912 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
3913 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
3914 (rs6000_expand_vec_perm_const): Delete.
3915 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
3917 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3918 (altivec_expand_vec_perm_const_le): Take each operand individually.
3919 Operate on constant selectors rather than rtxes.
3920 (altivec_expand_vec_perm_const): Likewise. Update call to
3921 altivec_expand_vec_perm_const_le.
3922 (rs6000_expand_vec_perm_const): Delete.
3923 (rs6000_vectorize_vec_perm_const_ok): Delete.
3924 (rs6000_vectorize_vec_perm_const): New function.
3925 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3926 an element count and rtx array.
3927 (rs6000_expand_extract_even): Update call accordingly.
3928 (rs6000_expand_interleave): Likewise.
3929 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
3930 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
3931 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
3932 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
3933 (rs6000_expand_vec_perm_const): Delete.
3934 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3935 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3936 (altivec_expand_vec_perm_const_le): Take each operand individually.
3937 Operate on constant selectors rather than rtxes.
3938 (altivec_expand_vec_perm_const): Likewise. Update call to
3939 altivec_expand_vec_perm_const_le.
3940 (rs6000_expand_vec_perm_const): Delete.
3941 (rs6000_vectorize_vec_perm_const_ok): Delete.
3942 (rs6000_vectorize_vec_perm_const): New function. Remove stray
3943 reference to the SPE evmerge intructions.
3944 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3945 an element count and rtx array.
3946 (rs6000_expand_extract_even): Update call accordingly.
3947 (rs6000_expand_interleave): Likewise.
3948 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
3949 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
3951 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3953 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3955 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
3956 vector mode and that that mode matches the mode of the data
3958 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
3959 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
3960 directly using expand_vec_perm_1 when forcing selectors into
3962 (expand_vec_perm_var): New function, split out from expand_vec_perm.
3964 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3966 * optabs-query.h (can_vec_perm_p): Delete.
3967 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
3968 * optabs-query.c (can_vec_perm_p): Split into...
3969 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
3970 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
3971 particular selector is valid.
3972 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3973 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3974 (vect_grouped_load_supported): Likewise.
3975 (vect_shift_permute_load_chain): Likewise.
3976 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
3977 (vect_transform_slp_perm_load): Likewise.
3978 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3979 (vectorizable_bswap): Likewise.
3980 (vect_gen_perm_mask_checked): Likewise.
3981 * fold-const.c (fold_ternary_loc): Likewise. Don't take
3982 implementations of variable permutation vectors into account
3983 when deciding which selector to use.
3984 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
3985 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
3986 with a false third argument.
3987 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
3988 to test whether the constant selector is valid and can_vec_perm_var_p
3989 to test whether a variable selector is valid.
3991 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3993 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
3994 * optabs-query.c (can_vec_perm_p): Likewise.
3995 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
3996 instead of vec_perm_indices.
3997 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
3998 (vect_gen_perm_mask_checked): Likewise,
3999 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
4000 (vect_gen_perm_mask_checked): Likewise,
4002 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4004 * optabs-query.h (qimode_for_vec_perm): Declare.
4005 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
4006 (qimode_for_vec_perm): ...this new function.
4007 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
4009 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4011 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
4012 does not have a conditional at the top.
4014 2018-01-02 Richard Biener <rguenther@suse.de>
4016 * ipa-inline.c (big_speedup_p): Fix expression.
4018 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
4021 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
4024 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
4028 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
4029 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
4030 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
4031 cond_taken_branch_cost 3->4.
4033 2018-01-01 Jakub Jelinek <jakub@redhat.com>
4035 PR tree-optimization/83581
4036 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
4037 TODO_cleanup_cfg if any changes have been made.
4040 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
4041 convert_modes if target mode has the right side, but different mode
4045 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
4046 last argument when extracting from CONCAT. If either from_real or
4047 from_imag is NULL, use expansion through memory. If result is not
4048 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
4049 the parts directly to inner mode, if even that fails, use expansion
4053 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
4054 check for bswap in mode rather than HImode and use that in expand_unop
4057 Copyright (C) 2018 Free Software Foundation, Inc.
4059 Copying and distribution of this file, with or without modification,
4060 are permitted in any medium without royalty provided the copyright
4061 notice and this notice are preserved.