1 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3 * config/aarch64/aarch64-simd.md
4 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
6 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
7 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
8 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
9 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
10 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
11 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
12 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
13 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
15 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
18 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
19 targ_options->x_arm_arch_string is non NULL.
21 2018-01-11 Tamar Christina <tamar.christina@arm.com>
23 * config/aarch64/aarch64.h
24 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
26 2018-01-11 Sudakshina Das <sudi.das@arm.com>
29 * expmed.c (emit_store_flag_force): Swap if const op0
30 and change VOIDmode to mode of op0.
32 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
34 PR rtl-optimization/83761
35 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
36 than bytes to mode_for_size.
38 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
41 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
42 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
45 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
48 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
50 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
51 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
54 2018-01-10 Michael Collison <michael.collison@arm.com>
56 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
57 * config/aarch64/aarch64-option-extension.def: Add
58 AARCH64_OPT_EXTENSION of 'fp16fml'.
59 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
60 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
61 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
62 * config/aarch64/constraints.md (Ui7): New constraint.
63 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
67 (VFMLA16_LOW): New int iterator.
68 (VFMLA16_HIGH): Ditto.
69 (UNSPEC_FMLAL): New unspec.
70 (UNSPEC_FMLSL): Ditto.
71 (UNSPEC_FMLAL2): Ditto.
72 (UNSPEC_FMLSL2): Ditto.
73 (f16mac): New code attribute.
74 * config/aarch64/aarch64-simd-builtins.def
75 (aarch64_fmlal_lowv2sf): Ditto.
76 (aarch64_fmlsl_lowv2sf): Ditto.
77 (aarch64_fmlalq_lowv4sf): Ditto.
78 (aarch64_fmlslq_lowv4sf): Ditto.
79 (aarch64_fmlal_highv2sf): Ditto.
80 (aarch64_fmlsl_highv2sf): Ditto.
81 (aarch64_fmlalq_highv4sf): Ditto.
82 (aarch64_fmlslq_highv4sf): Ditto.
83 (aarch64_fmlal_lane_lowv2sf): Ditto.
84 (aarch64_fmlsl_lane_lowv2sf): Ditto.
85 (aarch64_fmlal_laneq_lowv2sf): Ditto.
86 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
87 (aarch64_fmlalq_lane_lowv4sf): Ditto.
88 (aarch64_fmlsl_lane_lowv4sf): Ditto.
89 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
90 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
91 (aarch64_fmlal_lane_highv2sf): Ditto.
92 (aarch64_fmlsl_lane_highv2sf): Ditto.
93 (aarch64_fmlal_laneq_highv2sf): Ditto.
94 (aarch64_fmlsl_laneq_highv2sf): Ditto.
95 (aarch64_fmlalq_lane_highv4sf): Ditto.
96 (aarch64_fmlsl_lane_highv4sf): Ditto.
97 (aarch64_fmlalq_laneq_highv4sf): Ditto.
98 (aarch64_fmlsl_laneq_highv4sf): Ditto.
99 * config/aarch64/aarch64-simd.md:
100 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
101 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
102 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
103 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
104 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
105 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
106 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
107 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
108 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
109 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
110 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
111 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
112 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
113 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
114 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
115 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
116 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
117 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
118 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
119 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
120 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
121 (vfmlsl_low_u32): Ditto.
122 (vfmlalq_low_u32): Ditto.
123 (vfmlslq_low_u32): Ditto.
124 (vfmlal_high_u32): Ditto.
125 (vfmlsl_high_u32): Ditto.
126 (vfmlalq_high_u32): Ditto.
127 (vfmlslq_high_u32): Ditto.
128 (vfmlal_lane_low_u32): Ditto.
129 (vfmlsl_lane_low_u32): Ditto.
130 (vfmlal_laneq_low_u32): Ditto.
131 (vfmlsl_laneq_low_u32): Ditto.
132 (vfmlalq_lane_low_u32): Ditto.
133 (vfmlslq_lane_low_u32): Ditto.
134 (vfmlalq_laneq_low_u32): Ditto.
135 (vfmlslq_laneq_low_u32): Ditto.
136 (vfmlal_lane_high_u32): Ditto.
137 (vfmlsl_lane_high_u32): Ditto.
138 (vfmlal_laneq_high_u32): Ditto.
139 (vfmlsl_laneq_high_u32): Ditto.
140 (vfmlalq_lane_high_u32): Ditto.
141 (vfmlslq_lane_high_u32): Ditto.
142 (vfmlalq_laneq_high_u32): Ditto.
143 (vfmlslq_laneq_high_u32): Ditto.
144 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
145 (AARCH64_FL_FOR_ARCH8_4): New.
146 (AARCH64_ISA_F16FML): New ISA flag.
147 (TARGET_F16FML): New feature flag for fp16fml.
148 (doc/invoke.texi): Document new fp16fml option.
150 2018-01-10 Michael Collison <michael.collison@arm.com>
152 * config/aarch64/aarch64-builtins.c:
153 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
154 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
155 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
156 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
157 (AARCH64_ISA_SHA3): New ISA flag.
158 (TARGET_SHA3): New feature flag for sha3.
159 * config/aarch64/iterators.md (sha512_op): New int attribute.
160 (CRYPTO_SHA512): New int iterator.
161 (UNSPEC_SHA512H): New unspec.
162 (UNSPEC_SHA512H2): Ditto.
163 (UNSPEC_SHA512SU0): Ditto.
164 (UNSPEC_SHA512SU1): Ditto.
165 * config/aarch64/aarch64-simd-builtins.def
166 (aarch64_crypto_sha512hqv2di): New builtin.
167 (aarch64_crypto_sha512h2qv2di): Ditto.
168 (aarch64_crypto_sha512su0qv2di): Ditto.
169 (aarch64_crypto_sha512su1qv2di): Ditto.
170 (aarch64_eor3qv8hi): Ditto.
171 (aarch64_rax1qv2di): Ditto.
172 (aarch64_xarqv2di): Ditto.
173 (aarch64_bcaxqv8hi): Ditto.
174 * config/aarch64/aarch64-simd.md:
175 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
176 (aarch64_crypto_sha512su0qv2di): Ditto.
177 (aarch64_crypto_sha512su1qv2di): Ditto.
178 (aarch64_eor3qv8hi): Ditto.
179 (aarch64_rax1qv2di): Ditto.
180 (aarch64_xarqv2di): Ditto.
181 (aarch64_bcaxqv8hi): Ditto.
182 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
183 (vsha512h2q_u64): Ditto.
184 (vsha512su0q_u64): Ditto.
185 (vsha512su1q_u64): Ditto.
190 * config/arm/types.md (crypto_sha512): New type attribute.
191 (crypto_sha3): Ditto.
192 (doc/invoke.texi): Document new sha3 option.
194 2018-01-10 Michael Collison <michael.collison@arm.com>
196 * config/aarch64/aarch64-builtins.c:
197 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
198 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
199 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
200 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
201 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
202 (AARCH64_ISA_SM4): New ISA flag.
203 (TARGET_SM4): New feature flag for sm4.
204 * config/aarch64/aarch64-simd-builtins.def
205 (aarch64_sm3ss1qv4si): Ditto.
206 (aarch64_sm3tt1aq4si): Ditto.
207 (aarch64_sm3tt1bq4si): Ditto.
208 (aarch64_sm3tt2aq4si): Ditto.
209 (aarch64_sm3tt2bq4si): Ditto.
210 (aarch64_sm3partw1qv4si): Ditto.
211 (aarch64_sm3partw2qv4si): Ditto.
212 (aarch64_sm4eqv4si): Ditto.
213 (aarch64_sm4ekeyqv4si): Ditto.
214 * config/aarch64/aarch64-simd.md:
215 (aarch64_sm3ss1qv4si): Ditto.
216 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
217 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
218 (aarch64_sm4eqv4si): Ditto.
219 (aarch64_sm4ekeyqv4si): Ditto.
220 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
222 (CRYPTO_SM3TT): Ditto.
223 (CRYPTO_SM3PART): Ditto.
224 (UNSPEC_SM3SS1): New unspec.
225 (UNSPEC_SM3TT1A): Ditto.
226 (UNSPEC_SM3TT1B): Ditto.
227 (UNSPEC_SM3TT2A): Ditto.
228 (UNSPEC_SM3TT2B): Ditto.
229 (UNSPEC_SM3PARTW1): Ditto.
230 (UNSPEC_SM3PARTW2): Ditto.
231 (UNSPEC_SM4E): Ditto.
232 (UNSPEC_SM4EKEY): Ditto.
233 * config/aarch64/constraints.md (Ui2): New constraint.
234 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
235 * config/arm/types.md (crypto_sm3): New type attribute.
237 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
238 (vsm3tt1aq_u32): Ditto.
239 (vsm3tt1bq_u32): Ditto.
240 (vsm3tt2aq_u32): Ditto.
241 (vsm3tt2bq_u32): Ditto.
242 (vsm3partw1q_u32): Ditto.
243 (vsm3partw2q_u32): Ditto.
245 (vsm4ekeyq_u32): Ditto.
246 (doc/invoke.texi): Document new sm4 option.
248 2018-01-10 Michael Collison <michael.collison@arm.com>
250 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
251 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
252 (AARCH64_FL_FOR_ARCH8_4): New.
253 (AARCH64_FL_V8_4): New flag.
254 (doc/invoke.texi): Document new armv8.4-a option.
256 2018-01-10 Michael Collison <michael.collison@arm.com>
258 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
259 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
260 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
261 * config/aarch64/aarch64-option-extension.def: Add
262 AARCH64_OPT_EXTENSION of 'sha2'.
263 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
264 (crypto): Disable sha2 and aes if crypto disabled.
265 (crypto): Enable aes and sha2 if enabled.
266 (simd): Disable sha2 and aes if simd disabled.
267 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
269 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
270 (TARGET_SHA2): New feature flag for sha2.
271 (TARGET_AES): New feature flag for aes.
272 * config/aarch64/aarch64-simd.md:
273 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
274 conditional on TARGET_AES.
275 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
276 (aarch64_crypto_sha1hsi): Make pattern conditional
278 (aarch64_crypto_sha1hv4si): Ditto.
279 (aarch64_be_crypto_sha1hv4si): Ditto.
280 (aarch64_crypto_sha1su1v4si): Ditto.
281 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
282 (aarch64_crypto_sha1su0v4si): Ditto.
283 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
284 (aarch64_crypto_sha256su0v4si): Ditto.
285 (aarch64_crypto_sha256su1v4si): Ditto.
286 (doc/invoke.texi): Document new aes and sha2 options.
288 2018-01-10 Martin Sebor <msebor@redhat.com>
290 PR tree-optimization/83781
291 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
294 2018-01-11 Martin Sebor <msebor@gmail.com>
295 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
297 PR tree-optimization/83501
298 PR tree-optimization/81703
300 * tree-ssa-strlen.c (get_string_cst): Rename...
301 (get_string_len): ...to this. Handle global constants.
302 (handle_char_store): Adjust.
304 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
305 Jim Wilson <jimw@sifive.com>
307 * config/riscv/riscv-protos.h (riscv_output_return): New.
308 * config/riscv/riscv.c (struct machine_function): New naked_p field.
309 (riscv_attribute_table, riscv_output_return),
310 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
311 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
312 (riscv_compute_frame_info): Only compute frame->mask if not a naked
314 (riscv_expand_prologue): Add early return for naked function.
315 (riscv_expand_epilogue): Likewise.
316 (riscv_function_ok_for_sibcall): Return false for naked function.
317 (riscv_set_current_function): New.
318 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
319 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
320 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
321 * doc/extend.texi (RISC-V Function Attributes): New.
323 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
325 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
326 check for 128-bit long double before checking TCmode.
327 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
328 128-bit long doubles before checking TFmode or TCmode.
329 (FLOAT128_IBM_P): Likewise.
331 2018-01-10 Martin Sebor <msebor@redhat.com>
333 PR tree-optimization/83671
334 * builtins.c (c_strlen): Unconditionally return zero for the empty
336 Use -Warray-bounds for warnings.
337 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
338 for non-constant array indices with COMPONENT_REF, arrays of
339 arrays, and pointers to arrays.
340 (gimple_fold_builtin_strlen): Determine and set length range for
341 non-constant character arrays.
343 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
346 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
349 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
351 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
353 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
356 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
357 VECTOR_MEM_ALTIVEC_OR_VSX_P.
358 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
359 indexed_or_indirect_operand predicate.
360 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
361 (*vsx_le_perm_load_v8hi): Likewise.
362 (*vsx_le_perm_load_v16qi): Likewise.
363 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
364 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
365 (*vsx_le_perm_store_v8hi): Likewise.
366 (*vsx_le_perm_store_v16qi): Likewise.
367 (eight unnamed splitters): Likewise.
369 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
371 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
372 * config/rs6000/emmintrin.h: Likewise.
373 * config/rs6000/mmintrin.h: Likewise.
374 * config/rs6000/xmmintrin.h: Likewise.
376 2018-01-10 David Malcolm <dmalcolm@redhat.com>
379 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
381 * tree.c (tree_nop_conversion): Return true for location wrapper
383 (maybe_wrap_with_location): New function.
384 (selftest::check_strip_nops): New function.
385 (selftest::test_location_wrappers): New function.
386 (selftest::tree_c_tests): Call it.
387 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
388 (maybe_wrap_with_location): New decl.
389 (EXPR_LOCATION_WRAPPER_P): New macro.
390 (location_wrapper_p): New inline function.
391 (tree_strip_any_location_wrapper): New inline function.
393 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
396 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
397 stack_realign_offset for the largest alignment of stack slot
399 (ix86_find_max_used_stack_alignment): New function.
400 (ix86_finalize_stack_frame_flags): Use it. Set
401 max_used_stack_alignment if we don't realign stack.
402 * config/i386/i386.h (machine_function): Add
403 max_used_stack_alignment.
405 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
407 * config/arm/arm.opt (-mbranch-cost): New option.
408 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
411 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
414 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
415 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
417 2018-01-10 Richard Biener <rguenther@suse.de>
420 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
421 early out so it also covers the case where we have a non-NULL
424 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
426 PR tree-optimization/83753
427 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
428 for non-strided grouped accesses if the number of elements is 1.
430 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
433 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
434 * i386.h (TARGET_USE_GATHER): Define.
435 * x86-tune.def (X86_TUNE_USE_GATHER): New.
437 2018-01-10 Martin Liska <mliska@suse.cz>
440 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
441 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
443 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
444 CLEANUP_NO_PARTITIONING is not set.
446 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
448 * doc/rtl.texi: Remove documentation of (const ...) wrappers
449 for vectors, as a partial revert of r254296.
450 * rtl.h (const_vec_p): Delete.
451 (const_vec_duplicate_p): Don't test for vector CONSTs.
452 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
453 * expmed.c (make_tree): Likewise.
456 * common.md (E, F): Use CONSTANT_P instead of checking for
458 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
459 checking for CONST_VECTOR.
461 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
464 * predict.c (force_edge_cold): Handle in more sane way edges
467 2018-01-09 Carl Love <cel@us.ibm.com>
469 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
471 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
472 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
473 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
474 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
475 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
476 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
477 * config/rs6000/rs6000-protos.h: Add extern defition for
478 rs6000_generate_float2_double_code.
479 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
481 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
482 (float2_v2df): Add define_expand.
484 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
487 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
488 op_mode in the force_to_mode call.
490 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
492 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
493 instead of checking each element individually.
494 (aarch64_evpc_uzp): Likewise.
495 (aarch64_evpc_zip): Likewise.
496 (aarch64_evpc_ext): Likewise.
497 (aarch64_evpc_rev): Likewise.
498 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
499 instead of checking each element individually. Return true without
501 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
502 whether all selected elements come from the same input, instead of
503 checking each element individually. Remove calls to gen_rtx_REG,
504 start_sequence and end_sequence and instead assert that no rtl is
507 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
509 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
510 order of HIGH and CONST checks.
512 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
514 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
515 if the destination isn't an SSA_NAME.
517 2018-01-09 Richard Biener <rguenther@suse.de>
519 PR tree-optimization/83668
520 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
522 (canonicalize_loop_form): ... here, renamed from ...
523 (canonicalize_loop_closed_ssa_form): ... this and amended to
524 swap successor edges for loop exit blocks to make us use
525 the RPO order we need for initial schedule generation.
527 2018-01-09 Joseph Myers <joseph@codesourcery.com>
529 PR tree-optimization/64811
530 * match.pd: When optimizing comparisons with Inf, avoid
531 introducing or losing exceptions from comparisons with NaN.
533 2018-01-09 Martin Liska <mliska@suse.cz>
536 * asan.c (shadow_mem_size): Add gcc_assert.
538 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
540 Don't save registers in main().
543 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
544 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
545 * config/avr/avr.c (avr_set_current_function): Don't error if
546 naked, OS_task or OS_main are specified at the same time.
547 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
549 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
551 * common/config/avr/avr-common.c (avr_option_optimization_table):
552 Switch on -mmain-is-OS_task for optimizing compilations.
554 2018-01-09 Richard Biener <rguenther@suse.de>
556 PR tree-optimization/83572
557 * graphite.c: Include cfganal.h.
558 (graphite_transform_loops): Connect infinite loops to exit
559 and remove fake edges at the end.
561 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
563 * ipa-inline.c (edge_badness): Revert accidental checkin.
565 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
568 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
569 symbols; not inline clones.
571 2018-01-09 Jakub Jelinek <jakub@redhat.com>
574 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
575 hard registers. Formatting fixes.
577 PR preprocessor/83722
578 * gcc.c (try_generate_repro): Pass
579 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
580 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
583 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
584 Kito Cheng <kito.cheng@gmail.com>
586 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
587 (riscv_leaf_function_p): Delete.
588 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
590 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
592 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
594 (do_ifelse): New function.
595 (do_isel): New function.
596 (do_sub3): New function.
597 (do_add3): New function.
598 (do_load_mask_compare): New function.
599 (do_overlap_load_compare): New function.
600 (expand_compare_loop): New function.
601 (expand_block_compare): Call expand_compare_loop() when appropriate.
602 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
604 (-mblock-compare-inline-loop-limit): New option.
606 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
609 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
610 Reverse order of second and third operands in first alternative.
611 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
612 of first and second elements in UNSPEC_VPERMR vector.
613 (altivec_expand_vec_perm_le): Likewise.
615 2017-01-08 Jeff Law <law@redhat.com>
617 PR rtl-optimizatin/81308
618 * tree-switch-conversion.c (cfg_altered): New file scoped static.
619 (process_switch): If group_case_labels makes a change, then set
621 (pass_convert_switch::execute): If a switch is converted, then
622 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
624 PR rtl-optimization/81308
625 * recog.c (split_all_insns): Conditionally cleanup the CFG after
628 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
630 PR target/83663 - Revert r255946
631 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
632 generation for cases where splatting a value is not useful.
633 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
634 across a vec_duplicate and a paradoxical subreg forming a vector
635 mode to a vec_concat.
637 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
639 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
640 -march=armv8.3-a variants.
641 * config/arm/t-multilib: Likewise.
642 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
644 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
646 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
648 (cceq_ior_compare_complement): Give it a name so I can use it, and
649 change boolean_or_operator predicate to boolean_operator so it can
650 be used to generate a crand.
651 (eqne): New code iterator.
652 (bd/bd_neg): New code_attrs.
653 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
654 a single define_insn.
655 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
656 decrement (bdnzt/bdnzf/bdzt/bdzf).
657 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
658 with the new names of the branch decrement patterns, and added the
659 names of the branch decrement conditional patterns.
661 2018-01-08 Richard Biener <rguenther@suse.de>
663 PR tree-optimization/83563
664 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
667 2018-01-08 Richard Biener <rguenther@suse.de>
670 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
672 2018-01-08 Richard Biener <rguenther@suse.de>
674 PR tree-optimization/83685
675 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
676 references to abnormals.
678 2018-01-08 Richard Biener <rguenther@suse.de>
681 * dwarf2out.c (output_indirect_strings): Handle empty
682 skeleton_debug_str_hash.
683 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
685 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
687 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
688 (emit_store_direct): Likewise.
689 (arc_trampoline_adjust_address): Likewise.
690 (arc_asm_trampoline_template): New function.
691 (arc_initialize_trampoline): Use asm_trampoline_template.
692 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
693 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
694 * config/arc/arc.md (flush_icache): Delete pattern.
696 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
698 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
699 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
702 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
705 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
706 by not USED_FOR_TARGET.
707 (make_pass_resolve_sw_modes): Likewise.
709 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
711 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
714 2018-01-08 Richard Biener <rguenther@suse.de>
717 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
719 2018-01-08 Richard Biener <rguenther@suse.de>
722 * match.pd ((t * 2) / 2) -> t): Add missing :c.
724 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
727 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
728 basic blocks with a small number of successors.
729 (convert_control_dep_chain_into_preds): Improve handling of
731 (dump_predicates): Split apart into...
732 (dump_pred_chain): ...here...
733 (dump_pred_info): ...and here.
734 (can_one_predicate_be_invalidated_p): Add debugging printfs.
735 (can_chain_union_be_invalidated_p): Improve check for invalidation
737 (uninit_uses_cannot_happen): Avoid unnecessary if
738 convert_control_dep_chain_into_preds yielded nothing.
740 2018-01-06 Martin Sebor <msebor@redhat.com>
742 PR tree-optimization/83640
743 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
744 subtracting negative offset from size.
745 (builtin_access::overlap): Adjust offset bounds of the access to fall
746 within the size of the object if possible.
748 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
750 PR rtl-optimization/83699
751 * expmed.c (extract_bit_field_1): Restrict the vector usage of
752 extract_bit_field_as_subreg to cases in which the extracted
753 value is also a vector.
755 * lra-constraints.c (process_alt_operands): Test for the equivalence
756 substitutions when detecting a possible reload cycle.
758 2018-01-06 Jakub Jelinek <jakub@redhat.com>
761 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
762 by default if flag_selective_schedling{,2}. Formatting fixes.
764 PR rtl-optimization/83682
765 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
766 if it has non-VECTOR_MODE element mode.
767 (vec_duplicate_p): Likewise.
770 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
771 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
773 2018-01-05 Jakub Jelinek <jakub@redhat.com>
776 * config/i386/i386-builtin.def
777 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
778 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
779 Require also OPTION_MASK_ISA_AVX512F in addition to
780 OPTION_MASK_ISA_GFNI.
781 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
782 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
783 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
784 to OPTION_MASK_ISA_GFNI.
785 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
786 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
787 OPTION_MASK_ISA_AVX512BW.
788 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
789 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
790 addition to OPTION_MASK_ISA_GFNI.
791 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
792 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
793 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
794 to OPTION_MASK_ISA_GFNI.
795 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
796 a requirement for all ISAs rather than any of them with a few
798 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
800 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
801 bitmasks to be enabled with 3 exceptions, instead of requiring any
802 enabled ISA with lots of exceptions.
803 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
804 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
805 Change avx512bw in isa attribute to avx512f.
806 * config/i386/sgxintrin.h: Add license boilerplate.
807 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
808 to __AVX512F__ and __AVX512VL to __AVX512VL__.
809 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
810 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
812 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
813 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
814 temporarily sse2 rather than sse if not enabled already.
817 * config/i386/sse.md (VI248_VLBW): Rename to ...
818 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
819 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
820 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
821 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
822 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
823 mode iterator instead of VI248_VLBW.
825 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
827 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
828 (record_modified): Skip clobbers; add debug output.
829 (param_change_prob): Use sreal frequencies.
831 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
833 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
834 punt for user-aligned variables.
836 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
838 * tree-chrec.c (chrec_contains_symbols): Return true for
841 2018-01-05 Sudakshina Das <sudi.das@arm.com>
844 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
845 of (x|y) == x for BICS pattern.
847 2018-01-05 Jakub Jelinek <jakub@redhat.com>
849 PR tree-optimization/83605
850 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
851 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
854 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
856 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
857 * config/epiphany/rtems.h: New file.
859 2018-01-04 Jakub Jelinek <jakub@redhat.com>
860 Uros Bizjak <ubizjak@gmail.com>
863 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
864 QIreg_operand instead of register_operand predicate.
865 * config/i386/i386.c (ix86_rop_should_change_byte_p,
866 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
867 comments instead of -fmitigate[-_]rop.
869 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
872 * cgraphunit.c (symbol_table::compile): Switch to text_section
873 before calling assembly_start debug hook.
874 * run-rtl-passes.c (run_rtl_passes): Likewise.
877 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
879 * tree-vrp.c (extract_range_from_binary_expr_1): Check
880 range_int_cst_p rather than !symbolic_range_p before calling
881 extract_range_from_multiplicative_op_1.
883 2017-01-04 Jeff Law <law@redhat.com>
885 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
886 redundant test in assertion.
888 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
890 * doc/rtl.texi: Document machine_mode wrapper classes.
892 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
894 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
897 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
899 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
900 the VEC_PERM_EXPR fold to fail.
902 2018-01-04 Jakub Jelinek <jakub@redhat.com>
905 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
906 to switched_sections.
908 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
911 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
914 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
917 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
918 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
920 2018-01-04 Jakub Jelinek <jakub@redhat.com>
923 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
924 is BLKmode and bitpos not zero or mode change is needed.
926 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
929 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
932 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
935 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
936 instead of MULT rtx. Update all corresponding splitters.
938 (*ssub<modesuffix>): Ditto.
940 (*cmp_sadd_di): Update split patterns.
941 (*cmp_sadd_si): Ditto.
942 (*cmp_sadd_sidi): Ditto.
943 (*cmp_ssub_di): Ditto.
944 (*cmp_ssub_si): Ditto.
945 (*cmp_ssub_sidi): Ditto.
946 * config/alpha/predicates.md (const23_operand): New predicate.
947 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
948 Look for ASHIFT, not MULT inner operand.
949 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
951 2018-01-04 Martin Liska <mliska@suse.cz>
953 PR gcov-profile/83669
954 * gcov.c (output_intermediate_file): Add version to intermediate
956 * doc/gcov.texi: Document new field 'version' in intermediate
957 file format. Fix location of '-k' option of gcov command.
959 2018-01-04 Martin Liska <mliska@suse.cz>
962 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
964 2018-01-04 Jakub Jelinek <jakub@redhat.com>
966 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
968 2018-01-03 Martin Sebor <msebor@redhat.com>
970 PR tree-optimization/83655
971 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
972 checking calls with invalid arguments.
974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
976 * tree-vect-stmts.c (vect_get_store_rhs): New function.
977 (vectorizable_mask_load_store): Delete.
978 (vectorizable_call): Return false for masked loads and stores.
979 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
980 instead of gimple_assign_rhs1.
981 (vectorizable_load): Handle IFN_MASK_LOAD.
982 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
986 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
988 (vectorizable_mask_load_store): ...here.
989 (vectorizable_load): ...and here.
991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
993 * tree-vect-stmts.c (vect_build_all_ones_mask)
994 (vect_build_zero_merge_argument): New functions, split out from...
995 (vectorizable_load): ...here.
997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
999 * tree-vect-stmts.c (vect_check_store_rhs): New function,
1001 (vectorizable_mask_load_store): ...here.
1002 (vectorizable_store): ...and here.
1004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1006 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
1008 (vectorizable_mask_load_store): ...here.
1010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1012 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
1013 (vect_model_store_cost): Take a vec_load_store_type instead of a
1015 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
1016 (vect_model_store_cost): Take a vec_load_store_type instead of a
1018 (vectorizable_mask_load_store): Update accordingly.
1019 (vectorizable_store): Likewise.
1020 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
1022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1024 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
1025 IFN_MASK_LOAD calls here rather than...
1026 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
1028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1029 Alan Hayward <alan.hayward@arm.com>
1030 David Sherwood <david.sherwood@arm.com>
1032 * expmed.c (extract_bit_field_1): For vector extracts,
1033 fall back to extract_bit_field_as_subreg if vec_extract
1036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1037 Alan Hayward <alan.hayward@arm.com>
1038 David Sherwood <david.sherwood@arm.com>
1040 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
1041 they are variable or constant sized.
1042 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
1043 slots for constant-sized data.
1045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1046 Alan Hayward <alan.hayward@arm.com>
1047 David Sherwood <david.sherwood@arm.com>
1049 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
1050 handling COND_EXPRs with boolean comparisons, try to find a better
1051 basis for the mask type than the boolean itself.
1053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1055 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
1056 is calculated and how it can be overridden.
1057 * genmodes.c (max_bitsize_mode_any_mode): New variable.
1058 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
1060 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
1063 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1064 Alan Hayward <alan.hayward@arm.com>
1065 David Sherwood <david.sherwood@arm.com>
1067 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
1068 Remove the mode argument.
1069 (aarch64_simd_valid_immediate): Remove the mode and inverse
1071 * config/aarch64/iterators.md (bitsize): New iterator.
1072 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
1073 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
1074 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
1075 aarch64_simd_valid_immediate.
1076 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
1077 (aarch64_reg_or_bic_imm): Likewise.
1078 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
1079 with an insn_type enum and msl with a modifier_type enum.
1080 Replace element_width with a scalar_mode. Change the shift
1081 to unsigned int. Add constructors for scalar_float_mode and
1082 scalar_int_mode elements.
1083 (aarch64_vect_float_const_representable_p): Delete.
1084 (aarch64_can_const_movi_rtx_p)
1085 (aarch64_simd_scalar_immediate_valid_for_move)
1086 (aarch64_simd_make_constant): Update call to
1087 aarch64_simd_valid_immediate.
1088 (aarch64_advsimd_valid_immediate_hs): New function.
1089 (aarch64_advsimd_valid_immediate): Likewise.
1090 (aarch64_simd_valid_immediate): Remove mode and inverse
1091 arguments. Rewrite to use the above. Use const_vec_duplicate_p
1092 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
1093 and aarch64_float_const_representable_p on the result.
1094 (aarch64_output_simd_mov_immediate): Remove mode argument.
1095 Update call to aarch64_simd_valid_immediate and use of
1096 simd_immediate_info.
1097 (aarch64_output_scalar_simd_mov_immediate): Update call
1100 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1101 Alan Hayward <alan.hayward@arm.com>
1102 David Sherwood <david.sherwood@arm.com>
1104 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
1105 (mode_nunits): Likewise CONST_MODE_NUNITS.
1106 * machmode.def (ADJUST_NUNITS): Document.
1107 * genmodes.c (mode_data::need_nunits_adj): New field.
1108 (blank_mode): Update accordingly.
1109 (adj_nunits): New variable.
1110 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
1112 (emit_mode_size_inline): Set need_bytesize_adj for all modes
1113 listed in adj_nunits.
1114 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
1115 listed in adj_nunits. Don't emit case statements for such modes.
1116 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
1117 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
1118 nothing if adj_nunits is nonnull.
1119 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
1120 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
1121 (emit_mode_fbit): Update use of print_maybe_const_decl.
1122 (emit_move_size): Likewise. Treat the array as non-const
1124 (emit_mode_adjustments): Handle adj_nunits.
1126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1128 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
1129 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
1130 (VECTOR_MODES): Use it.
1131 (make_vector_modes): Take the prefix as an argument.
1133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1134 Alan Hayward <alan.hayward@arm.com>
1135 David Sherwood <david.sherwood@arm.com>
1137 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
1138 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
1139 for MODE_VECTOR_BOOL.
1140 * machmode.def (VECTOR_BOOL_MODE): Document.
1141 * genmodes.c (VECTOR_BOOL_MODE): New macro.
1142 (make_vector_bool_mode): New function.
1143 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
1145 * lto-streamer-in.c (lto_input_mode_table): Likewise.
1146 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
1148 * stor-layout.c (int_mode_for_mode): Likewise.
1149 * tree.c (build_vector_type_for_mode): Likewise.
1150 * varasm.c (output_constant_pool_2): Likewise.
1151 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
1152 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
1153 for MODE_VECTOR_BOOL.
1154 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
1155 of mode class checks.
1156 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
1157 instead of a list of mode class checks.
1158 (expand_vector_scalar_condition): Likewise.
1159 (type_for_widest_vector_mode): Handle BImode as an inner mode.
1161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1162 Alan Hayward <alan.hayward@arm.com>
1163 David Sherwood <david.sherwood@arm.com>
1165 * machmode.h (mode_size): Change from unsigned short to
1167 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
1168 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1169 or if measurement_type is not polynomial.
1170 (fixed_size_mode::includes_p): Check for constant-sized modes.
1171 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
1172 return a poly_uint16 rather than an unsigned short.
1173 (emit_mode_size): Change the type of mode_size from unsigned short
1174 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
1175 (emit_mode_adjustments): Cope with polynomial vector sizes.
1176 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1178 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1180 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
1181 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
1182 * caller-save.c (setup_save_areas): Likewise.
1183 (replace_reg_with_saved_mem): Likewise.
1184 * calls.c (emit_library_call_value_1): Likewise.
1185 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
1186 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
1187 (gen_lowpart_for_combine): Likewise.
1188 * convert.c (convert_to_integer_1): Likewise.
1189 * cse.c (equiv_constant, cse_insn): Likewise.
1190 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
1191 (cselib_subst_to_values): Likewise.
1192 * dce.c (word_dce_process_block): Likewise.
1193 * df-problems.c (df_word_lr_mark_ref): Likewise.
1194 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
1195 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
1196 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
1197 (rtl_for_decl_location): Likewise.
1198 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
1199 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
1200 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
1201 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
1202 (expand_expr_real_1): Likewise.
1203 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
1204 (pad_below): Likewise.
1205 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1206 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
1207 * ira.c (get_subreg_tracking_sizes): Likewise.
1208 * ira-build.c (ira_create_allocno_objects): Likewise.
1209 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
1210 (ira_sort_regnos_for_alter_reg): Likewise.
1211 * ira-costs.c (record_operand_costs): Likewise.
1212 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
1213 (resolve_simple_move): Likewise.
1214 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
1215 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
1216 (lra_constraints): Likewise.
1217 (CONST_POOL_OK_P): Reject variable-sized modes.
1218 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
1219 (add_pseudo_to_slot, lra_spill): Likewise.
1220 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1221 * optabs-query.c (get_best_extraction_insn): Likewise.
1222 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1223 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
1224 (expand_mult_highpart, valid_multiword_target_p): Likewise.
1225 * recog.c (offsettable_address_addr_space_p): Likewise.
1226 * regcprop.c (maybe_mode_change): Likewise.
1227 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
1228 * regrename.c (build_def_use): Likewise.
1229 * regstat.c (dump_reg_info): Likewise.
1230 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
1231 (find_reloads, find_reloads_subreg_address): Likewise.
1232 * reload1.c (eliminate_regs_1): Likewise.
1233 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
1234 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
1235 (simplify_binary_operation_1, simplify_subreg): Likewise.
1236 * targhooks.c (default_function_arg_padding): Likewise.
1237 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
1238 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
1239 (verify_gimple_assign_ternary): Likewise.
1240 * tree-inline.c (estimate_move_cost): Likewise.
1241 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1242 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
1243 (get_address_cost_ainc): Likewise.
1244 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
1245 (vect_supportable_dr_alignment): Likewise.
1246 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1247 (vectorizable_reduction): Likewise.
1248 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
1249 (vectorizable_operation, vectorizable_load): Likewise.
1250 * tree.c (build_same_sized_truth_vector_type): Likewise.
1251 * valtrack.c (cleanup_auto_inc_dec): Likewise.
1252 * var-tracking.c (emit_note_insn_var_location): Likewise.
1253 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
1254 (ADDR_VEC_ALIGN): Likewise.
1256 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1257 Alan Hayward <alan.hayward@arm.com>
1258 David Sherwood <david.sherwood@arm.com>
1260 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
1262 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1263 or if measurement_type is polynomial.
1264 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
1265 * combine.c (make_extraction): Likewise.
1266 * dse.c (find_shift_sequence): Likewise.
1267 * dwarf2out.c (mem_loc_descriptor): Likewise.
1268 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
1269 (extract_bit_field, extract_low_bits): Likewise.
1270 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
1271 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
1272 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
1273 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
1274 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1275 * reload.c (find_reloads): Likewise.
1276 * reload1.c (alter_reg): Likewise.
1277 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
1278 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
1279 * tree-if-conv.c (predicate_mem_writes): Likewise.
1280 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
1281 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
1282 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
1283 * valtrack.c (dead_debug_insert_temp): Likewise.
1284 * varasm.c (mergeable_constant_section): Likewise.
1285 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
1287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1288 Alan Hayward <alan.hayward@arm.com>
1289 David Sherwood <david.sherwood@arm.com>
1291 * expr.c (expand_assignment): Cope with polynomial mode sizes
1292 when assigning to a CONCAT.
1294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1295 Alan Hayward <alan.hayward@arm.com>
1296 David Sherwood <david.sherwood@arm.com>
1298 * machmode.h (mode_precision): Change from unsigned short to
1300 (mode_to_precision): Return a poly_uint16 rather than an unsigned
1302 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
1303 or if measurement_type is not polynomial.
1304 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
1305 in which the mode is already known to be a scalar_int_mode.
1306 * genmodes.c (emit_mode_precision): Change the type of mode_precision
1307 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
1309 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1310 for GET_MODE_PRECISION.
1311 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1312 for GET_MODE_PRECISION.
1313 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
1315 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
1316 (expand_field_assignment, make_extraction): Likewise.
1317 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
1318 (get_last_value): Likewise.
1319 * convert.c (convert_to_integer_1): Likewise.
1320 * cse.c (cse_insn): Likewise.
1321 * expr.c (expand_expr_real_1): Likewise.
1322 * lra-constraints.c (simplify_operand_subreg): Likewise.
1323 * optabs-query.c (can_atomic_load_p): Likewise.
1324 * optabs.c (expand_atomic_load): Likewise.
1325 (expand_atomic_store): Likewise.
1326 * ree.c (combine_reaching_defs): Likewise.
1327 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
1328 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
1329 * tree.h (type_has_mode_precision_p): Likewise.
1330 * ubsan.c (instrument_si_overflow): Likewise.
1332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1333 Alan Hayward <alan.hayward@arm.com>
1334 David Sherwood <david.sherwood@arm.com>
1336 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
1337 polynomial numbers of units.
1338 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
1339 (valid_vector_subparts_p): New function.
1340 (build_vector_type): Remove temporary shim and take the number
1341 of units as a poly_uint64 rather than an int.
1342 (build_opaque_vector_type): Take the number of units as a
1343 poly_uint64 rather than an int.
1344 * tree.c (build_vector_from_ctor): Handle polynomial
1345 TYPE_VECTOR_SUBPARTS.
1346 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
1347 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
1348 (build_vector_from_val): If the number of units is variable,
1349 use build_vec_duplicate_cst for constant operands and
1350 VEC_DUPLICATE_EXPR otherwise.
1351 (make_vector_type): Remove temporary is_constant ().
1352 (build_vector_type, build_opaque_vector_type): Take the number of
1353 units as a poly_uint64 rather than an int.
1354 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
1356 * cfgexpand.c (expand_debug_expr): Likewise.
1357 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
1358 (store_constructor, expand_expr_real_1): Likewise.
1359 (const_scalar_mask_from_tree): Likewise.
1360 * fold-const-call.c (fold_const_reduction): Likewise.
1361 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
1362 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
1363 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
1364 (fold_relational_const): Likewise.
1365 (native_interpret_vector): Likewise. Change the size from an
1366 int to an unsigned int.
1367 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
1368 TYPE_VECTOR_SUBPARTS.
1369 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
1370 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
1371 duplicating a non-constant operand into a variable-length vector.
1372 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
1373 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
1374 * ipa-icf.c (sem_variable::equals): Likewise.
1375 * match.pd: Likewise.
1376 * omp-simd-clone.c (simd_clone_subparts): Likewise.
1377 * print-tree.c (print_node): Likewise.
1378 * stor-layout.c (layout_type): Likewise.
1379 * targhooks.c (default_builtin_vectorization_cost): Likewise.
1380 * tree-cfg.c (verify_gimple_comparison): Likewise.
1381 (verify_gimple_assign_binary): Likewise.
1382 (verify_gimple_assign_ternary): Likewise.
1383 (verify_gimple_assign_single): Likewise.
1384 * tree-pretty-print.c (dump_generic_node): Likewise.
1385 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1386 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
1387 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
1388 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
1389 (vect_shift_permute_load_chain): Likewise.
1390 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
1391 (expand_vector_condition, optimize_vector_constructor): Likewise.
1392 (lower_vec_perm, get_compute_type): Likewise.
1393 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1394 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
1395 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
1396 (vect_recog_mask_conversion_pattern): Likewise.
1397 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
1398 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
1399 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1400 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
1401 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
1402 (vectorizable_shift, vectorizable_operation, vectorizable_store)
1403 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
1404 (supportable_widening_operation): Likewise.
1405 (supportable_narrowing_operation): Likewise.
1406 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
1408 * varasm.c (output_constant): Likewise.
1410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1411 Alan Hayward <alan.hayward@arm.com>
1412 David Sherwood <david.sherwood@arm.com>
1414 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
1415 so that both the length == 3 and length != 3 cases set up their
1416 own permute vectors. Add comments explaining why we know the
1417 number of elements is constant.
1418 (vect_permute_load_chain): Likewise.
1420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1421 Alan Hayward <alan.hayward@arm.com>
1422 David Sherwood <david.sherwood@arm.com>
1424 * machmode.h (mode_nunits): Change from unsigned char to
1426 (ONLY_FIXED_SIZE_MODES): New macro.
1427 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
1428 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
1429 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
1431 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
1432 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
1433 or if measurement_type is not polynomial.
1434 * genmodes.c (ZERO_COEFFS): New macro.
1435 (emit_mode_nunits_inline): Make mode_nunits_inline return a
1437 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
1438 Use ZERO_COEFFS when emitting initializers.
1439 * data-streamer.h (bp_pack_poly_value): New function.
1440 (bp_unpack_poly_value): Likewise.
1441 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1442 for GET_MODE_NUNITS.
1443 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1444 for GET_MODE_NUNITS.
1445 * tree.c (make_vector_type): Remove temporary shim and make
1446 the real function take the number of units as a poly_uint64
1448 (build_vector_type_for_mode): Handle polynomial nunits.
1449 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
1450 * emit-rtl.c (const_vec_series_p_1): Likewise.
1451 (gen_rtx_CONST_VECTOR): Likewise.
1452 * fold-const.c (test_vec_duplicate_folding): Likewise.
1453 * genrecog.c (validate_pattern): Likewise.
1454 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
1455 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1456 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
1457 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
1458 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
1459 * rtlanal.c (subreg_get_info): Likewise.
1460 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1461 (vect_grouped_load_supported): Likewise.
1462 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
1463 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
1464 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
1465 (simplify_const_unary_operation, simplify_binary_operation_1)
1466 (simplify_const_binary_operation, simplify_ternary_operation)
1467 (test_vector_ops_duplicate, test_vector_ops): Likewise.
1468 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
1469 instead of CONST_VECTOR_NUNITS.
1470 * varasm.c (output_constant_pool_2): Likewise.
1471 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
1472 explicit-encoded elements in the XVEC for variable-length vectors.
1474 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1476 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
1478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1479 Alan Hayward <alan.hayward@arm.com>
1480 David Sherwood <david.sherwood@arm.com>
1482 * coretypes.h (fixed_size_mode): Declare.
1483 (fixed_size_mode_pod): New typedef.
1484 * builtins.h (target_builtins::x_apply_args_mode)
1485 (target_builtins::x_apply_result_mode): Change type to
1486 fixed_size_mode_pod.
1487 * builtins.c (apply_args_size, apply_result_size, result_vector)
1488 (expand_builtin_apply_args_1, expand_builtin_apply)
1489 (expand_builtin_return): Update accordingly.
1491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1493 * cse.c (hash_rtx_cb): Hash only the encoded elements.
1494 * cselib.c (cselib_hash_rtx): Likewise.
1495 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
1496 CONST_VECTOR encoding.
1498 2017-01-03 Jakub Jelinek <jakub@redhat.com>
1499 Jeff Law <law@redhat.com>
1502 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
1503 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
1504 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
1505 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
1508 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
1509 explicitly probe *sp in a noreturn function if there were any callee
1510 register saves or frame pointer is needed.
1512 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1515 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
1516 BLKmode for ternary, binary or unary expressions.
1519 * var-tracking.c (delete_vta_debug_insn): New inline function.
1520 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
1521 insns from get_insns () to NULL instead of each bb separately.
1522 Use delete_vta_debug_insn. No longer static.
1523 (vt_debug_insns_local, variable_tracking_main_1): Adjust
1524 delete_vta_debug_insns callers.
1525 * rtl.h (delete_vta_debug_insns): Declare.
1526 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
1527 instead of variable_tracking_main.
1529 2018-01-03 Martin Sebor <msebor@redhat.com>
1531 PR tree-optimization/83603
1532 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
1533 arguments past the endof the argument list in functions declared
1534 without a prototype.
1535 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
1536 Avoid checking when arguments are null.
1538 2018-01-03 Martin Sebor <msebor@redhat.com>
1541 * doc/extend.texi (attribute const): Fix a typo.
1542 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
1543 issuing -Wsuggest-attribute for void functions.
1545 2018-01-03 Martin Sebor <msebor@redhat.com>
1547 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
1548 offset_int::from instead of wide_int::to_shwi.
1549 (maybe_diag_overlap): Remove assertion.
1550 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
1551 * gimple-ssa-sprintf.c (format_directive): Same.
1552 (parse_directive): Same.
1553 (sprintf_dom_walker::compute_format_length): Same.
1554 (try_substitute_return_value): Same.
1556 2017-01-03 Jeff Law <law@redhat.com>
1559 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
1560 non-constant residual for zero at runtime and avoid probing in
1561 that case. Reorganize code for trailing problem to mirror handling
1564 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1566 PR tree-optimization/83501
1567 * tree-ssa-strlen.c (get_string_cst): New.
1568 (handle_char_store): Call get_string_cst.
1570 2018-01-03 Martin Liska <mliska@suse.cz>
1572 PR tree-optimization/83593
1573 * tree-ssa-strlen.c: Include tree-cfg.h.
1574 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
1575 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
1576 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
1578 (strlen_dom_walker::before_dom_children): Call
1579 gimple_purge_dead_eh_edges. Dump tranformation with details
1581 (strlen_dom_walker::before_dom_children): Update call by adding
1582 new argument cleanup_eh.
1583 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
1585 2018-01-03 Martin Liska <mliska@suse.cz>
1588 * cif-code.def (VARIADIC_THUNK): New enum value.
1589 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
1592 2018-01-03 Jan Beulich <jbeulich@suse.com>
1594 * sse.md (mov<mode>_internal): Tighten condition for when to use
1595 vmovdqu<ssescalarsize> for TI and OI modes.
1597 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1599 Update copyright years.
1601 2018-01-03 Martin Liska <mliska@suse.cz>
1604 * ipa-visibility.c (function_and_variable_visibility): Skip
1605 functions with noipa attribure.
1607 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1609 * gcc.c (process_command): Update copyright notice dates.
1610 * gcov-dump.c (print_version): Ditto.
1611 * gcov.c (print_version): Ditto.
1612 * gcov-tool.c (print_version): Ditto.
1613 * gengtype.c (create_file): Ditto.
1614 * doc/cpp.texi: Bump @copying's copyright year.
1615 * doc/cppinternals.texi: Ditto.
1616 * doc/gcc.texi: Ditto.
1617 * doc/gccint.texi: Ditto.
1618 * doc/gcov.texi: Ditto.
1619 * doc/install.texi: Ditto.
1620 * doc/invoke.texi: Ditto.
1622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1624 * vector-builder.h (vector_builder::m_full_nelts): Change from
1625 unsigned int to poly_uint64.
1626 (vector_builder::full_nelts): Update prototype accordingly.
1627 (vector_builder::new_vector): Likewise.
1628 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
1629 (vector_builder::operator ==): Likewise.
1630 (vector_builder::finalize): Likewise.
1631 * int-vector-builder.h (int_vector_builder::int_vector_builder):
1632 Take the number of elements as a poly_uint64 rather than an
1634 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
1635 from unsigned int to poly_uint64.
1636 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
1637 (vec_perm_indices::new_vector): Likewise.
1638 (vec_perm_indices::length): Likewise.
1639 (vec_perm_indices::nelts_per_input): Likewise.
1640 (vec_perm_indices::input_nelts): Likewise.
1641 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
1642 number of elements per input as a poly_uint64 rather than an
1643 unsigned int. Use the original encoding for variable-length
1644 vectors, rather than clamping each individual element.
1645 For the second and subsequent elements in each pattern,
1646 clamp the step and base before clamping their sum.
1647 (vec_perm_indices::series_p): Handle polynomial element counts.
1648 (vec_perm_indices::all_in_range_p): Likewise.
1649 (vec_perm_indices_to_tree): Likewise.
1650 (vec_perm_indices_to_rtx): Likewise.
1651 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
1652 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
1653 (tree_vector_builder::new_binary_operation): Handle polynomial
1654 element counts. Return false if we need to know the number
1655 of elements at compile time.
1656 * fold-const.c (fold_vec_perm): Punt if the number of elements
1657 isn't known at compile time.
1659 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1661 * vec-perm-indices.h (vec_perm_builder): Change element type
1662 from HOST_WIDE_INT to poly_int64.
1663 (vec_perm_indices::element_type): Update accordingly.
1664 (vec_perm_indices::clamp): Handle polynomial element_types.
1665 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1666 (vec_perm_indices::all_in_range_p): Likewise.
1667 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
1669 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
1670 polynomial vec_perm_indices element types.
1671 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
1672 * fold-const.c (fold_vec_perm): Likewise.
1673 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
1674 * tree-vect-generic.c (lower_vec_perm): Likewise.
1675 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1676 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
1677 element type to HOST_WIDE_INT.
1679 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1680 Alan Hayward <alan.hayward@arm.com>
1681 David Sherwood <david.sherwood@arm.com>
1683 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
1684 rather than an int. Use plus_constant.
1685 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
1686 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
1688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1689 Alan Hayward <alan.hayward@arm.com>
1690 David Sherwood <david.sherwood@arm.com>
1692 * calls.c (emit_call_1, expand_call): Change struct_value_size from
1693 a HOST_WIDE_INT to a poly_int64.
1695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1696 Alan Hayward <alan.hayward@arm.com>
1697 David Sherwood <david.sherwood@arm.com>
1699 * calls.c (load_register_parameters): Cope with polynomial
1700 mode sizes. Require a constant size for BLKmode parameters
1701 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
1702 forces a parameter to be padded at the lsb end in order to
1703 fill a complete number of words, require the parameter size
1704 to be ordered wrt UNITS_PER_WORD.
1706 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1707 Alan Hayward <alan.hayward@arm.com>
1708 David Sherwood <david.sherwood@arm.com>
1710 * reload1.c (spill_stack_slot_width): Change element type
1711 from unsigned int to poly_uint64_pod.
1712 (alter_reg): Treat mode sizes as polynomial.
1714 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1715 Alan Hayward <alan.hayward@arm.com>
1716 David Sherwood <david.sherwood@arm.com>
1718 * reload.c (complex_word_subreg_p): New function.
1719 (reload_inner_reg_of_subreg, push_reload): Use it.
1721 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1722 Alan Hayward <alan.hayward@arm.com>
1723 David Sherwood <david.sherwood@arm.com>
1725 * lra-constraints.c (process_alt_operands): Reject matched
1726 operands whose sizes aren't ordered.
1727 (match_reload): Refer to this check here.
1729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1730 Alan Hayward <alan.hayward@arm.com>
1731 David Sherwood <david.sherwood@arm.com>
1733 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
1734 that the mode size is in the set {1, 2, 4, 8, 16}.
1736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1737 Alan Hayward <alan.hayward@arm.com>
1738 David Sherwood <david.sherwood@arm.com>
1740 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
1741 Use plus_constant instead of gen_rtx_PLUS.
1743 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1744 Alan Hayward <alan.hayward@arm.com>
1745 David Sherwood <david.sherwood@arm.com>
1747 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
1748 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
1749 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
1750 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
1751 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
1752 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
1753 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
1754 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
1755 * config/i386/i386.c (ix86_push_rounding): ...this new function.
1756 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
1758 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
1759 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
1760 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
1761 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
1762 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
1763 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
1764 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
1765 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
1766 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
1767 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
1769 * expr.c (emit_move_resolve_push): Treat the input and result
1770 of PUSH_ROUNDING as a poly_int64.
1771 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
1772 (emit_push_insn): Likewise.
1773 * lra-eliminations.c (mark_not_eliminable): Likewise.
1774 * recog.c (push_operand): Likewise.
1775 * reload1.c (elimination_effects): Likewise.
1776 * rtlanal.c (nonzero_bits1): Likewise.
1777 * calls.c (store_one_arg): Likewise. Require the padding to be
1778 known at compile time.
1780 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1781 Alan Hayward <alan.hayward@arm.com>
1782 David Sherwood <david.sherwood@arm.com>
1784 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
1785 Use plus_constant instead of gen_rtx_PLUS.
1787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1788 Alan Hayward <alan.hayward@arm.com>
1789 David Sherwood <david.sherwood@arm.com>
1791 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
1794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1795 Alan Hayward <alan.hayward@arm.com>
1796 David Sherwood <david.sherwood@arm.com>
1798 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
1799 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
1800 via stack temporaries. Treat the mode size as polynomial too.
1802 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1803 Alan Hayward <alan.hayward@arm.com>
1804 David Sherwood <david.sherwood@arm.com>
1806 * expr.c (expand_expr_real_2): When handling conversions involving
1807 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
1808 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
1809 as a poly_uint64 too.
1811 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1812 Alan Hayward <alan.hayward@arm.com>
1813 David Sherwood <david.sherwood@arm.com>
1815 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
1817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1818 Alan Hayward <alan.hayward@arm.com>
1819 David Sherwood <david.sherwood@arm.com>
1821 * combine.c (can_change_dest_mode): Handle polynomial
1822 REGMODE_NATURAL_SIZE.
1823 * expmed.c (store_bit_field_1): Likewise.
1824 * expr.c (store_constructor): Likewise.
1825 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
1826 and polynomial REGMODE_NATURAL_SIZE.
1827 (gen_lowpart_common): Likewise.
1828 * reginfo.c (record_subregs_of_mode): Likewise.
1829 * rtlanal.c (read_modify_subreg_p): Likewise.
1831 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1832 Alan Hayward <alan.hayward@arm.com>
1833 David Sherwood <david.sherwood@arm.com>
1835 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
1836 numbers of elements.
1838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1839 Alan Hayward <alan.hayward@arm.com>
1840 David Sherwood <david.sherwood@arm.com>
1842 * match.pd: Cope with polynomial numbers of vector elements.
1844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1845 Alan Hayward <alan.hayward@arm.com>
1846 David Sherwood <david.sherwood@arm.com>
1848 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1849 in a POINTER_PLUS_EXPR.
1851 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1852 Alan Hayward <alan.hayward@arm.com>
1853 David Sherwood <david.sherwood@arm.com>
1855 * omp-simd-clone.c (simd_clone_subparts): New function.
1856 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1857 (ipa_simd_modify_function_body): Likewise.
1859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1860 Alan Hayward <alan.hayward@arm.com>
1861 David Sherwood <david.sherwood@arm.com>
1863 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1864 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1865 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1866 (expand_vector_condition, vector_element): Likewise.
1867 (subparts_gt): New function.
1868 (get_compute_type): Use subparts_gt.
1869 (count_type_subparts): Delete.
1870 (expand_vector_operations_1): Use subparts_gt instead of
1871 count_type_subparts.
1873 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1874 Alan Hayward <alan.hayward@arm.com>
1875 David Sherwood <david.sherwood@arm.com>
1877 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1878 (vect_compile_time_alias): ...this new function. Do the calculation
1879 on poly_ints rather than trees.
1880 (vect_prune_runtime_alias_test_list): Update call accordingly.
1882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1883 Alan Hayward <alan.hayward@arm.com>
1884 David Sherwood <david.sherwood@arm.com>
1886 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1888 (vect_schedule_slp_instance): Likewise.
1890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1891 Alan Hayward <alan.hayward@arm.com>
1892 David Sherwood <david.sherwood@arm.com>
1894 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1895 constant and extern definitions for variable-length vectors.
1896 (vect_get_constant_vectors): Note that the number of units
1897 is known to be constant.
1899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1900 Alan Hayward <alan.hayward@arm.com>
1901 David Sherwood <david.sherwood@arm.com>
1903 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1904 of units as polynomial. Choose between WIDE and NARROW based
1907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1908 Alan Hayward <alan.hayward@arm.com>
1909 David Sherwood <david.sherwood@arm.com>
1911 * tree-vect-stmts.c (simd_clone_subparts): New function.
1912 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1914 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1915 Alan Hayward <alan.hayward@arm.com>
1916 David Sherwood <david.sherwood@arm.com>
1918 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1919 vectors as polynomial. Use build_index_vector for
1922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1923 Alan Hayward <alan.hayward@arm.com>
1924 David Sherwood <david.sherwood@arm.com>
1926 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1927 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1928 for variable-length vectors.
1929 (vectorizable_mask_load_store): Treat the number of units as
1930 polynomial, asserting that it is constant if the condition has
1931 already been enforced.
1932 (vectorizable_store, vectorizable_load): Likewise.
1934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1935 Alan Hayward <alan.hayward@arm.com>
1936 David Sherwood <david.sherwood@arm.com>
1938 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1939 of units as polynomial. Punt if we can't tell at compile time
1940 which vector contains the final result.
1942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1943 Alan Hayward <alan.hayward@arm.com>
1944 David Sherwood <david.sherwood@arm.com>
1946 * tree-vect-loop.c (vectorizable_induction): Treat the number
1947 of units as polynomial. Punt on SLP inductions. Use an integer
1948 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1949 cast of such a series for variable-length floating-point
1952 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1953 Alan Hayward <alan.hayward@arm.com>
1954 David Sherwood <david.sherwood@arm.com>
1956 * tree.h (build_index_vector): Declare.
1957 * tree.c (build_index_vector): New function.
1958 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1959 of units as polynomial, forcibly converting it to a constant if
1960 vectorizable_reduction has already enforced the condition.
1961 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1962 to create a {1,2,3,...} vector.
1963 (vectorizable_reduction): Treat the number of units as polynomial.
1964 Choose vectype_in based on the largest scalar element size rather
1965 than the smallest number of units. Enforce the restrictions
1968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1969 Alan Hayward <alan.hayward@arm.com>
1970 David Sherwood <david.sherwood@arm.com>
1972 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1973 number of units as polynomial.
1975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1976 Alan Hayward <alan.hayward@arm.com>
1977 David Sherwood <david.sherwood@arm.com>
1979 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1980 * target.def (autovectorize_vector_sizes): Return the vector sizes
1981 by pointer, using vector_sizes rather than a bitmask.
1982 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1983 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1984 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1986 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1987 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1988 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1989 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1990 * omp-general.c (omp_max_vf): Likewise.
1991 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1992 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1993 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1994 * tree-vect-slp.c (vect_slp_bb): Likewise.
1995 * doc/tm.texi: Regenerate.
1996 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1998 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1999 the vector size as a poly_uint64 rather than an unsigned int.
2000 (current_vector_size): Change from an unsigned int to a poly_uint64.
2001 (get_vectype_for_scalar_type): Update accordingly.
2002 * tree.h (build_truth_vector_type): Take the size and number of
2003 units as a poly_uint64 rather than an unsigned int.
2004 (build_vector_type): Add a temporary overload that takes
2005 the number of units as a poly_uint64 rather than an unsigned int.
2006 * tree.c (make_vector_type): Likewise.
2007 (build_truth_vector_type): Take the number of units as a poly_uint64
2008 rather than an unsigned int.
2010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2011 Alan Hayward <alan.hayward@arm.com>
2012 David Sherwood <david.sherwood@arm.com>
2014 * target.def (get_mask_mode): Take the number of units and length
2015 as poly_uint64s rather than unsigned ints.
2016 * targhooks.h (default_get_mask_mode): Update accordingly.
2017 * targhooks.c (default_get_mask_mode): Likewise.
2018 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
2019 * doc/tm.texi: Regenerate.
2021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2022 Alan Hayward <alan.hayward@arm.com>
2023 David Sherwood <david.sherwood@arm.com>
2025 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
2026 * omp-general.c (omp_max_vf): Likewise.
2027 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
2028 (expand_omp_simd): Handle polynomial safelen.
2029 * omp-low.c (omplow_simd_context): Add a default constructor.
2030 (omplow_simd_context::max_vf): Change from int to poly_uint64.
2031 (lower_rec_simd_input_clauses): Update accordingly.
2032 (lower_rec_input_clauses): Likewise.
2034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2035 Alan Hayward <alan.hayward@arm.com>
2036 David Sherwood <david.sherwood@arm.com>
2038 * tree-vectorizer.h (vect_nunits_for_cost): New function.
2039 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
2040 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
2041 (vect_analyze_slp_cost): Likewise.
2042 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
2043 (vect_model_load_cost): Likewise.
2045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2046 Alan Hayward <alan.hayward@arm.com>
2047 David Sherwood <david.sherwood@arm.com>
2049 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
2050 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
2051 from an unsigned int * to a poly_uint64_pod *.
2052 (calculate_unrolling_factor): New function.
2053 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
2055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2056 Alan Hayward <alan.hayward@arm.com>
2057 David Sherwood <david.sherwood@arm.com>
2059 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
2060 from an unsigned int to a poly_uint64.
2061 (_loop_vec_info::slp_unrolling_factor): Likewise.
2062 (_loop_vec_info::vectorization_factor): Change from an int
2064 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
2065 (vect_get_num_vectors): New function.
2066 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
2067 (vect_get_num_copies): Use vect_get_num_vectors.
2068 (vect_analyze_data_ref_dependences): Change max_vf from an int *
2069 to an unsigned int *.
2070 (vect_analyze_data_refs): Change min_vf from an int * to a
2072 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2073 than an unsigned HOST_WIDE_INT.
2074 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
2075 (vect_analyze_data_ref_dependence): Change max_vf from an int *
2076 to an unsigned int *.
2077 (vect_analyze_data_ref_dependences): Likewise.
2078 (vect_compute_data_ref_alignment): Handle polynomial vf.
2079 (vect_enhance_data_refs_alignment): Likewise.
2080 (vect_prune_runtime_alias_test_list): Likewise.
2081 (vect_shift_permute_load_chain): Likewise.
2082 (vect_supportable_dr_alignment): Likewise.
2083 (dependence_distance_ge_vf): Take the vectorization factor as a
2084 poly_uint64 rather than an unsigned HOST_WIDE_INT.
2085 (vect_analyze_data_refs): Change min_vf from an int * to a
2087 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
2088 vfm1 as a poly_uint64 rather than an int. Make the same change
2089 for the returned bound_scalar.
2090 (vect_gen_vector_loop_niters): Handle polynomial vf.
2091 (vect_do_peeling): Likewise. Update call to
2092 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
2093 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
2095 * tree-vect-loop.c (vect_determine_vectorization_factor)
2096 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
2097 (vect_get_known_peeling_cost): Likewise.
2098 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
2099 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
2100 (vect_transform_loop): Likewise. Use the lowest possible VF when
2101 updating the upper bounds of the loop.
2102 (vect_min_worthwhile_factor): Make static. Return an unsigned int
2104 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
2105 polynomial unroll factors.
2106 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
2107 (vect_make_slp_decision): Likewise.
2108 (vect_supported_load_permutation_p): Likewise, and polynomial
2110 (vect_analyze_slp_cost): Handle polynomial vf.
2111 (vect_slp_analyze_node_operations): Likewise.
2112 (vect_slp_analyze_bb_1): Likewise.
2113 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2114 than an unsigned HOST_WIDE_INT.
2115 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
2116 (vectorizable_load): Handle polynomial vf.
2117 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
2119 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
2121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2122 Alan Hayward <alan.hayward@arm.com>
2123 David Sherwood <david.sherwood@arm.com>
2125 * match.pd: Handle bit operations involving three constants
2126 and try to fold one pair.
2128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2130 * tree-vect-loop-manip.c: Include gimple-fold.h.
2131 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
2132 niters_maybe_zero parameters. Handle other cases besides a step of 1.
2133 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
2134 Add a path that uses a step of VF instead of 1, but disable it
2136 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
2137 and niters_no_overflow parameters. Update calls to
2138 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
2139 Create a new SSA name if the latter choses to use a ste other
2140 than zero, and return it via niters_vector_mult_vf_var.
2141 * tree-vect-loop.c (vect_transform_loop): Update calls to
2142 vect_do_peeling, vect_gen_vector_loop_niters and
2143 slpeel_make_loop_iterate_ntimes.
2144 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
2145 (vect_gen_vector_loop_niters): Update declarations after above changes.
2147 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
2149 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
2150 128-bit round to integer instructions.
2151 (ceil<mode>2): Likewise.
2152 (btrunc<mode>2): Likewise.
2153 (round<mode>2): Likewise.
2155 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2157 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
2158 unaligned VSX load/store on P8/P9.
2159 (expand_block_clear): Allow the use of unaligned VSX
2160 load/store on P8/P9.
2162 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2164 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
2166 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
2167 swap associated with both a load and a store.
2169 2018-01-02 Andrew Waterman <andrew@sifive.com>
2171 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
2172 * config/riscv/riscv.md (clear_cache): Use it.
2174 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
2176 * web.c: Remove out-of-date comment.
2178 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2180 * expr.c (fixup_args_size_notes): Check that any existing
2181 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
2182 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
2183 (emit_single_push_insn): ...here.
2185 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2187 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
2188 (const_vector_encoded_nelts): New function.
2189 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
2190 (const_vector_int_elt, const_vector_elt): Declare.
2191 * emit-rtl.c (const_vector_int_elt_1): New function.
2192 (const_vector_elt): Likewise.
2193 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
2194 of CONST_VECTOR_ELT.
2196 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2198 * expr.c: Include rtx-vector-builder.h.
2199 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
2200 directly on the tree encoding.
2201 (const_vector_from_tree): Likewise.
2202 * optabs.c: Include rtx-vector-builder.h.
2203 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
2204 sequence of "u" values.
2205 * vec-perm-indices.c: Include rtx-vector-builder.h.
2206 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
2207 directly on the vec_perm_indices encoding.
2209 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2211 * doc/rtl.texi (const_vector): Describe new encoding scheme.
2212 * Makefile.in (OBJS): Add rtx-vector-builder.o.
2213 * rtx-vector-builder.h: New file.
2214 * rtx-vector-builder.c: Likewise.
2215 * rtl.h (rtx_def::u2): Add a const_vector field.
2216 (CONST_VECTOR_NPATTERNS): New macro.
2217 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
2218 (CONST_VECTOR_DUPLICATE_P): Likewise.
2219 (CONST_VECTOR_STEPPED_P): Likewise.
2220 (CONST_VECTOR_ENCODED_ELT): Likewise.
2221 (const_vec_duplicate_p): Check for a duplicated vector encoding.
2222 (unwrap_const_vec_duplicate): Likewise.
2223 (const_vec_series_p): Check for a non-duplicated vector encoding.
2224 Say that the function only returns true for integer vectors.
2225 * emit-rtl.c: Include rtx-vector-builder.h.
2226 (gen_const_vec_duplicate_1): Delete.
2227 (gen_const_vector): Call gen_const_vec_duplicate instead of
2228 gen_const_vec_duplicate_1.
2229 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
2230 (gen_const_vec_duplicate): Use rtx_vector_builder.
2231 (gen_const_vec_series): Likewise.
2232 (gen_rtx_CONST_VECTOR): Likewise.
2233 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
2234 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2235 Build a new vector rather than modifying a CONST_VECTOR in-place.
2236 (handle_special_swappables): Update call accordingly.
2237 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
2238 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2239 Build a new vector rather than modifying a CONST_VECTOR in-place.
2240 (handle_special_swappables): Update call accordingly.
2242 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2244 * simplify-rtx.c (simplify_const_binary_operation): Use
2245 CONST_VECTOR_ELT instead of XVECEXP.
2247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2249 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
2250 the selector elements to be different from the data elements
2251 if the selector is a VECTOR_CST.
2252 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
2253 ssizetype for the selector.
2255 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2257 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
2258 before testing each element individually.
2259 * tree-vect-generic.c (lower_vec_perm): Likewise.
2261 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2263 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
2264 * selftest-run-tests.c (selftest::run_tests): Call it.
2265 * vector-builder.h (vector_builder::operator ==): New function.
2266 (vector_builder::operator !=): Likewise.
2267 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
2268 (vec_perm_indices::all_from_input_p): New function.
2269 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
2270 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
2271 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
2272 instead of reading the VECTOR_CST directly. Detect whether both
2273 vector inputs are the same before constructing the vec_perm_indices,
2274 and update the number of inputs argument accordingly. Use the
2275 utility functions added above. Only construct sel2 if we need to.
2277 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2279 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
2280 the broadcast of the low byte.
2281 (expand_mult_highpart): Use an explicit encoding for the permutes.
2282 * optabs-query.c (can_mult_highpart_p): Likewise.
2283 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
2284 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2285 (vectorizable_bswap): Likewise.
2286 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
2287 explicit encoding for the power-of-2 permutes.
2288 (vect_permute_store_chain): Likewise.
2289 (vect_grouped_load_supported): Likewise.
2290 (vect_permute_load_chain): Likewise.
2292 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2294 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
2295 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
2296 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
2297 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
2298 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
2299 (vect_gen_perm_mask_any): Likewise.
2301 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2303 * int-vector-builder.h: New file.
2304 * vec-perm-indices.h: Include int-vector-builder.h.
2305 (vec_perm_indices): Redefine as an int_vector_builder.
2306 (auto_vec_perm_indices): Delete.
2307 (vec_perm_builder): Redefine as a stand-alone class.
2308 (vec_perm_indices::vec_perm_indices): New function.
2309 (vec_perm_indices::clamp): Likewise.
2310 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
2311 (vec_perm_indices::new_vector): New function.
2312 (vec_perm_indices::new_expanded_vector): Update for new
2313 vec_perm_indices class.
2314 (vec_perm_indices::rotate_inputs): New function.
2315 (vec_perm_indices::all_in_range_p): Operate directly on the
2316 encoded form, without computing elided elements.
2317 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
2318 encoding. Update for new vec_perm_indices class.
2319 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
2320 the given vec_perm_builder.
2321 (expand_vec_perm_var): Update vec_perm_builder constructor.
2322 (expand_mult_highpart): Use vec_perm_builder instead of
2323 auto_vec_perm_indices.
2324 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
2325 vec_perm_indices instead of auto_vec_perm_indices. Use a single
2326 or double series encoding as appropriate.
2327 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
2328 vec_perm_indices instead of auto_vec_perm_indices.
2329 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2330 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2331 (vect_permute_store_chain): Likewise.
2332 (vect_grouped_load_supported): Likewise.
2333 (vect_permute_load_chain): Likewise.
2334 (vect_shift_permute_load_chain): Likewise.
2335 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2336 (vect_transform_slp_perm_load): Likewise.
2337 (vect_schedule_slp_instance): Likewise.
2338 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2339 (vectorizable_mask_load_store): Likewise.
2340 (vectorizable_bswap): Likewise.
2341 (vectorizable_store): Likewise.
2342 (vectorizable_load): Likewise.
2343 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
2344 vec_perm_indices instead of auto_vec_perm_indices. Use
2345 tree_to_vec_perm_builder to read the vector from a tree.
2346 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
2347 vec_perm_builder instead of a vec_perm_indices.
2348 (have_whole_vector_shift): Use vec_perm_builder and
2349 vec_perm_indices instead of auto_vec_perm_indices. Leave the
2350 truncation to calc_vec_perm_mask_for_shift.
2351 (vect_create_epilog_for_reduction): Likewise.
2352 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
2353 from auto_vec_perm_indices to vec_perm_indices.
2354 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2355 instead of changing individual elements.
2356 (aarch64_vectorize_vec_perm_const): Use new_vector to install
2357 the vector in d.perm.
2358 * config/arm/arm.c (expand_vec_perm_d::perm): Change
2359 from auto_vec_perm_indices to vec_perm_indices.
2360 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2361 instead of changing individual elements.
2362 (arm_vectorize_vec_perm_const): Use new_vector to install
2363 the vector in d.perm.
2364 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
2365 Update vec_perm_builder constructor.
2366 (rs6000_expand_interleave): Likewise.
2367 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
2368 (rs6000_expand_interleave): Likewise.
2370 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2372 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
2373 to qimode could truncate the indices.
2374 * optabs.c (expand_vec_perm_var): Likewise.
2376 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2378 * Makefile.in (OBJS): Add vec-perm-indices.o.
2379 * vec-perm-indices.h: New file.
2380 * vec-perm-indices.c: Likewise.
2381 * target.h (vec_perm_indices): Replace with a forward class
2383 (auto_vec_perm_indices): Move to vec-perm-indices.h.
2384 * optabs.h: Include vec-perm-indices.h.
2385 (expand_vec_perm): Delete.
2386 (selector_fits_mode_p, expand_vec_perm_var): Declare.
2387 (expand_vec_perm_const): Declare.
2388 * target.def (vec_perm_const_ok): Replace with...
2389 (vec_perm_const): ...this new hook.
2390 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
2391 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
2392 * doc/tm.texi: Regenerate.
2393 * optabs.def (vec_perm_const): Delete.
2394 * doc/md.texi (vec_perm_const): Likewise.
2395 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
2396 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
2397 expand_vec_perm for constant permutation vectors. Assert that
2398 the mode of variable permutation vectors is the integer equivalent
2399 of the mode that is being permuted.
2400 * optabs-query.h (selector_fits_mode_p): Declare.
2401 * optabs-query.c: Include vec-perm-indices.h.
2402 (selector_fits_mode_p): New function.
2403 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
2404 is defined, instead of checking whether the vec_perm_const_optab
2405 exists. Use targetm.vectorize.vec_perm_const instead of
2406 targetm.vectorize.vec_perm_const_ok. Check whether the indices
2407 fit in the vector mode before using a variable permute.
2408 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
2409 vec_perm_indices instead of an rtx.
2410 (expand_vec_perm): Replace with...
2411 (expand_vec_perm_const): ...this new function. Take the selector
2412 as a vec_perm_indices rather than an rtx. Also take the mode of
2413 the selector. Update call to shift_amt_for_vec_perm_mask.
2414 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
2415 Use vec_perm_indices::new_expanded_vector to expand the original
2416 selector into bytes. Check whether the indices fit in the vector
2417 mode before using a variable permute.
2418 (expand_vec_perm_var): Make global.
2419 (expand_mult_highpart): Use expand_vec_perm_const.
2420 * fold-const.c: Includes vec-perm-indices.h.
2421 * tree-ssa-forwprop.c: Likewise.
2422 * tree-vect-data-refs.c: Likewise.
2423 * tree-vect-generic.c: Likewise.
2424 * tree-vect-loop.c: Likewise.
2425 * tree-vect-slp.c: Likewise.
2426 * tree-vect-stmts.c: Likewise.
2427 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
2429 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
2430 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
2431 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
2432 (aarch64_vectorize_vec_perm_const): ...this new function.
2433 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2434 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2435 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
2436 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
2437 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2438 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2439 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
2441 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
2442 check for NEON modes.
2443 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
2444 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
2445 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
2446 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
2448 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
2449 the old VEC_PERM_CONST conditions.
2450 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
2451 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
2452 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
2453 (ia64_vectorize_vec_perm_const_ok): Merge into...
2454 (ia64_vectorize_vec_perm_const): ...this new function.
2455 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
2456 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
2457 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
2458 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
2459 * config/mips/mips.c (mips_expand_vec_perm_const)
2460 (mips_vectorize_vec_perm_const_ok): Merge into...
2461 (mips_vectorize_vec_perm_const): ...this new function.
2462 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
2463 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
2464 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
2465 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
2466 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
2467 (rs6000_expand_vec_perm_const): Delete.
2468 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
2470 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2471 (altivec_expand_vec_perm_const_le): Take each operand individually.
2472 Operate on constant selectors rather than rtxes.
2473 (altivec_expand_vec_perm_const): Likewise. Update call to
2474 altivec_expand_vec_perm_const_le.
2475 (rs6000_expand_vec_perm_const): Delete.
2476 (rs6000_vectorize_vec_perm_const_ok): Delete.
2477 (rs6000_vectorize_vec_perm_const): New function.
2478 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2479 an element count and rtx array.
2480 (rs6000_expand_extract_even): Update call accordingly.
2481 (rs6000_expand_interleave): Likewise.
2482 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
2483 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
2484 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
2485 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
2486 (rs6000_expand_vec_perm_const): Delete.
2487 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2488 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2489 (altivec_expand_vec_perm_const_le): Take each operand individually.
2490 Operate on constant selectors rather than rtxes.
2491 (altivec_expand_vec_perm_const): Likewise. Update call to
2492 altivec_expand_vec_perm_const_le.
2493 (rs6000_expand_vec_perm_const): Delete.
2494 (rs6000_vectorize_vec_perm_const_ok): Delete.
2495 (rs6000_vectorize_vec_perm_const): New function. Remove stray
2496 reference to the SPE evmerge intructions.
2497 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2498 an element count and rtx array.
2499 (rs6000_expand_extract_even): Update call accordingly.
2500 (rs6000_expand_interleave): Likewise.
2501 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
2502 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
2504 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2506 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2508 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
2509 vector mode and that that mode matches the mode of the data
2511 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
2512 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
2513 directly using expand_vec_perm_1 when forcing selectors into
2515 (expand_vec_perm_var): New function, split out from expand_vec_perm.
2517 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2519 * optabs-query.h (can_vec_perm_p): Delete.
2520 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
2521 * optabs-query.c (can_vec_perm_p): Split into...
2522 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
2523 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
2524 particular selector is valid.
2525 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2526 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2527 (vect_grouped_load_supported): Likewise.
2528 (vect_shift_permute_load_chain): Likewise.
2529 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2530 (vect_transform_slp_perm_load): Likewise.
2531 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2532 (vectorizable_bswap): Likewise.
2533 (vect_gen_perm_mask_checked): Likewise.
2534 * fold-const.c (fold_ternary_loc): Likewise. Don't take
2535 implementations of variable permutation vectors into account
2536 when deciding which selector to use.
2537 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
2538 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
2539 with a false third argument.
2540 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
2541 to test whether the constant selector is valid and can_vec_perm_var_p
2542 to test whether a variable selector is valid.
2544 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2546 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
2547 * optabs-query.c (can_vec_perm_p): Likewise.
2548 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
2549 instead of vec_perm_indices.
2550 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
2551 (vect_gen_perm_mask_checked): Likewise,
2552 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
2553 (vect_gen_perm_mask_checked): Likewise,
2555 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2557 * optabs-query.h (qimode_for_vec_perm): Declare.
2558 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
2559 (qimode_for_vec_perm): ...this new function.
2560 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
2562 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2564 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
2565 does not have a conditional at the top.
2567 2018-01-02 Richard Biener <rguenther@suse.de>
2569 * ipa-inline.c (big_speedup_p): Fix expression.
2571 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2574 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
2577 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2581 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
2582 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
2583 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
2584 cond_taken_branch_cost 3->4.
2586 2018-01-01 Jakub Jelinek <jakub@redhat.com>
2588 PR tree-optimization/83581
2589 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
2590 TODO_cleanup_cfg if any changes have been made.
2593 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
2594 convert_modes if target mode has the right side, but different mode
2598 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
2599 last argument when extracting from CONCAT. If either from_real or
2600 from_imag is NULL, use expansion through memory. If result is not
2601 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
2602 the parts directly to inner mode, if even that fails, use expansion
2606 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
2607 check for bswap in mode rather than HImode and use that in expand_unop
2610 Copyright (C) 2018 Free Software Foundation, Inc.
2612 Copying and distribution of this file, with or without modification,
2613 are permitted in any medium without royalty provided the copyright
2614 notice and this notice are preserved.