|
| 1 | +2025-03-07 Jakub Jelinek < [email protected]> |
| 2 | + |
| 3 | + PR c/117178 |
| 4 | + * tree.cc (get_attr_nonstring_decl): Look through all ARRAY_REFs, not |
| 5 | + just one and handle COMPONENT_REF and MEM_REF after skipping those |
| 6 | + rather than only when there wasn't ARRAY_REF. Formatting fix. |
| 7 | + |
| 8 | +2025-03-07 Kees Cook < [email protected]> |
| 9 | + |
| 10 | + |
| 11 | + PR c/117178 |
| 12 | + * doc/invoke.texi (Wunterminated-string-initialization): Document |
| 13 | + the new interaction between this warning and -Wc++-compat and that |
| 14 | + initialization of decls with nonstring attribute aren't warned about. |
| 15 | + |
| 16 | +2025-03-07 Sandra Loosemore < [email protected]> |
| 17 | + |
| 18 | + PR sanitizer/56682 |
| 19 | + * doc/invoke.texi (Instrumentation Options): Document that -g |
| 20 | + is useful with -fsanitize=thread and -fsanitize=address. |
| 21 | + Also mention -fno-omit-frame-pointer per the asan wiki. |
| 22 | + |
| 23 | +2025-03-07 Richard Sandiford < [email protected]> |
| 24 | + |
| 25 | + |
| 26 | + |
| 27 | + PR rtl-optimization/117477 |
| 28 | + * config/aarch64/aarch64.cc (aarch64_count_saves): New function. |
| 29 | + (aarch64_count_above_hard_fp_saves, aarch64_callee_save_cost) |
| 30 | + (aarch64_frame_allocation_cost): Likewise. |
| 31 | + (TARGET_CALLEE_SAVE_COST): Define. |
| 32 | + (TARGET_FRAME_ALLOCATION_COST): Likewise. |
| 33 | + * config/i386/i386.cc (ix86_ira_callee_saved_register_cost_scale): |
| 34 | + Replace with... |
| 35 | + (ix86_callee_save_cost): ...this new hook. |
| 36 | + (TARGET_IRA_CALLEE_SAVED_REGISTER_COST_SCALE): Delete. |
| 37 | + (TARGET_CALLEE_SAVE_COST): Define. |
| 38 | + * target.h (spill_cost_type, frame_cost_type): New enums. |
| 39 | + * target.def (callee_save_cost, frame_allocation_cost): New hooks. |
| 40 | + (ira_callee_saved_register_cost_scale): Delete. |
| 41 | + * doc/tm.texi.in (TARGET_IRA_CALLEE_SAVED_REGISTER_COST_SCALE): Delete. |
| 42 | + (TARGET_CALLEE_SAVE_COST, TARGET_FRAME_ALLOCATION_COST): New hooks. |
| 43 | + * doc/tm.texi: Regenerate. |
| 44 | + * hard-reg-set.h (hard_reg_set_popcount): New function. |
| 45 | + * ira-color.cc (allocated_memory_p): New variable. |
| 46 | + (allocated_callee_save_regs): Likewise. |
| 47 | + (record_allocation): New function. |
| 48 | + (assign_hard_reg): Use targetm.frame_allocation_cost to model |
| 49 | + the cost of the first spill or first caller save. Use |
| 50 | + targetm.callee_save_cost to model the cost of using new callee-saved |
| 51 | + registers. Apply the exit rather than entry frequency to the cost |
| 52 | + of restoring a register or deallocating the frame. Update the |
| 53 | + new variables above. |
| 54 | + (improve_allocation): Use record_allocation. |
| 55 | + (color): Initialize allocated_callee_save_regs. |
| 56 | + (ira_color): Initialize allocated_memory_p. |
| 57 | + * targhooks.h (default_callee_save_cost): Declare. |
| 58 | + (default_frame_allocation_cost): Likewise. |
| 59 | + * targhooks.cc (default_callee_save_cost): New function. |
| 60 | + (default_frame_allocation_cost): Likewise. |
| 61 | + |
| 62 | +2025-03-07 Sandra Loosemore < [email protected]> |
| 63 | + |
| 64 | + PR target/116708 |
| 65 | + * doc/invoke.texi (x86 Options): Clarify how -msse4 and -mno-sse4 |
| 66 | + interact with other SSE options. |
| 67 | + |
| 68 | +2025-03-07 Martin Jambor < [email protected]> |
| 69 | + |
| 70 | + PR ipa/118318 |
| 71 | + * ipa-cp.cc (adjust_clone_incoming_counts): Add a compatible_p check. |
| 72 | + |
| 73 | +2025-03-07 Richard Earnshaw < [email protected]> |
| 74 | + |
| 75 | + * config/arm/arm_neon.h: Try harder to detect if we have |
| 76 | + the softfp ABI enabled. |
| 77 | + |
| 78 | +2025-03-07 Jakub Jelinek < [email protected]> |
| 79 | + |
| 80 | + PR c/112960 |
| 81 | + PR c/117029 |
| 82 | + * doc/extend.texi (Complex): Add I and J suffixes to the list of |
| 83 | + complex suffixes, adjust for all of those being part of ISO C2Y, |
| 84 | + clarify that for -fno-ext-numeric-literals none of those are |
| 85 | + recognized as GNU extensions and for C++14 i is considered UDL |
| 86 | + even for -fext-numeric-literals when <complex> is included. |
| 87 | + |
| 88 | +2025-03-07 Simon Martin < [email protected]> |
| 89 | + |
| 90 | + * tree-vect-data-refs.cc: Define INCLUDE_ALGORITHM. |
| 91 | + |
| 92 | +2025-03-07 Tamar Christina < [email protected]> |
| 93 | + |
| 94 | + PR tree-optimization/118464 |
| 95 | + PR tree-optimization/116855 |
| 96 | + * doc/invoke.texi (min-pagesize): Update docs with vectorizer use. |
| 97 | + * tree-vect-data-refs.cc (vect_analyze_early_break_dependences): Delay |
| 98 | + checks. |
| 99 | + (vect_compute_data_ref_alignment): Remove alignment checks and move to |
| 100 | + get_load_store_type, increase group access alignment. |
| 101 | + (vect_enhance_data_refs_alignment): Add note to comment needing |
| 102 | + investigating. |
| 103 | + (vect_analyze_data_refs_alignment): Likewise. |
| 104 | + (vect_supportable_dr_alignment): For group loads look at first DR. |
| 105 | + * tree-vect-stmts.cc (get_load_store_type): |
| 106 | + Perform safety checks for early break pfa. |
| 107 | + * tree-vectorizer.h (dr_set_safe_speculative_read_required, |
| 108 | + dr_safe_speculative_read_required, DR_SCALAR_KNOWN_BOUNDS): New. |
| 109 | + (need_peeling_for_alignment): Renamed to... |
| 110 | + (safe_speculative_read_required): .. This |
| 111 | + (class dr_vec_info): Add scalar_access_known_in_bounds. |
| 112 | + |
| 113 | +2025-03-07 Tamar Christina < [email protected]> |
| 114 | + |
| 115 | + PR tree-optimization/118464 |
| 116 | + PR tree-optimization/116855 |
| 117 | + * config/aarch64/aarch64-sve.md (@extract_<last_op>_<mode>, |
| 118 | + @fold_extract_<last_op>_<mode>, |
| 119 | + @aarch64_fold_extract_vector_<last_op>_<mode>): Change SVE_FULL to |
| 120 | + SVE_ALL. |
| 121 | + * config/aarch64/iterators.md (vccore): Add more partial types. |
| 122 | + |
| 123 | +2025-03-07 Richard Biener < [email protected]> |
| 124 | + |
| 125 | + PR tree-optimization/119145 |
| 126 | + * tree-vectorizer.cc (try_vectorize_loop_1): Avoid BB |
| 127 | + vectorizing an if-converted loop body when there's a .MASK_CALL |
| 128 | + in the loop body. |
| 129 | + |
| 130 | +2025-03-07 Christophe Lyon < [email protected]> |
| 131 | + |
| 132 | + PR target/115485 |
| 133 | + * config/arm/arm.cc (require_pic_register): Fix typos in |
| 134 | + comment. Handle fixed arm_pic_register. |
| 135 | + |
| 136 | +2025-03-07 Richard Sandiford < [email protected]> |
| 137 | + |
| 138 | + PR tree-optimization/116125 |
| 139 | + * tree-vect-data-refs.cc (vect_prune_runtime_alias_test_list): Make |
| 140 | + the dr_with_seg_len alignment fields describe tha access sizes as |
| 141 | + well as the pointer alignment. |
| 142 | + * tree-data-ref.cc (create_intersect_range_checks): Don't compensate |
| 143 | + for invalid alignment fields here. |
| 144 | + |
| 145 | +2025-03-07 Richard Sandiford < [email protected]> |
| 146 | + |
| 147 | + PR target/119133 |
| 148 | + * config/aarch64/aarch64.md |
| 149 | + (*aarch64_bfi<GPI:mode><ALLX:mode>_<SUBDI_BITS>): Use |
| 150 | + force_lowpart_subreg. |
| 151 | + |
| 152 | +2025-03-07 Richard Biener < [email protected]> |
| 153 | + |
| 154 | + PR middle-end/118801 |
| 155 | + * tree-ssa-dce.cc (eliminate_unnecessary_stmts): Prune |
| 156 | + sequences of uninterrupted DEBUG BEGIN_STMTs, keeping only |
| 157 | + the last of a set with unique location. |
| 158 | + |
| 159 | +2025-03-07 Sandra Loosemore < [email protected]> |
| 160 | + |
| 161 | + PR c/113515 |
| 162 | + * doc/invoke.texi (Warning Options): Improve -Wstringop-overflow |
| 163 | + documentation. |
| 164 | + |
| 165 | +2025-03-07 Haochen Jiang < [email protected]> |
| 166 | + |
| 167 | + * config/i386/avx10_2-512convertintrin.h |
| 168 | + (_mm512_mask_cvtbf8_ph): Correct mask width. |
| 169 | + (_mm512_maskz_cvtbf8_ph): Ditto. |
| 170 | + * config/i386/avx10_2convertintrin.h |
| 171 | + (_mm256_mask_cvtbf8_ph): Ditto. |
| 172 | + (_mm256_maskz_cvtbf8_ph): Ditto. |
| 173 | + |
1 | 174 | 2025-03-06 Alexey Merzlyakov < [email protected]>
|
2 | 175 |
|
3 | 176 | PR rtl-optimization/119099
|
|
0 commit comments