1 2005-01-14 Aldy Hernandez <aldyh@redhat.com>
3 * config/rs6000/rs6000.c (rs6000_generate_compare): Replace
4 flag_finite_math_only with flag_unsafe_math_optimizations.
6 2005-01-14 J. D. Johnston <jjohnst@us.ibm.com>
8 * config/s390/tpf-unwind.h (s390_fallback_frame_state): Correct
10 (__tpf_eh_return): Copy TPF private stack area from the module
11 boundary stack frame. Add check for when module addresses are
12 equal, but no stub address is found.
14 2005-01-14 Richard Earnshaw <rearnsha@arm.com>
17 * arm.h (struct machine_function): Add call_via field.
18 (thumb_call_via_label): Declare.
19 * arm.c (thumb_call_via_label): New variable.
20 (thumb_call_reg_needed): New variable.
21 (arm_output_function_epilogue): For Thumb code, output any per-function
22 call-indirect trampolines.
23 (thumb_call_via_reg): New function.
24 (arm_file_end): New function.
25 (TARGET_ASM_FILE_END): Call arm_file_end.
26 (aof_file_end): Likewise.
27 * arm-protos.h (thumb_call_via_reg): Declare.
28 * arm.md (call_reg_thumb, call_value_reg_thumb): Call
29 thumb_call_via_reg in normal case.
31 2005-01-14 Jakub Jelinek <jakub@redhat.com>
34 PR rtl-optimization/19348
35 * recog.c (peephole2_optimize): Do global life update if some peephole
36 decides it doesn't need at least one of its inputs and that change
37 influences liveness at the start of the basic block.
39 * basic-block.h (EXECUTE_IF_AND_COMPL_IN_REG_SET): Needs 2 REGSET
40 arguments instead of 1.
42 2005-01-14 Eric Botcazou <ebotcazou@libertysurf.fr>
45 * varasm.c (initializer_constant_valid_p) <ADDR_EXPR>: Return
46 zero for nested functions needing a static chain or functions
47 with a non-constant address.
49 2005-01-13 Roger Sayle <roger@eyesopen.com>
51 * simplify-rtx.c (simplify_binary_operation) <AND>: Optimize
52 (and (sign_extend X) C) into (zero_extend (and X C)).
54 2005-01-13 David O'Brien <obrien@FreeBSD.org>
56 * config/freebsd-spec.h: Make KSE pthread lib logic the default.
58 2005-01-13 Richard Henderson <rth@redhat.com>
63 * config/i386/i386.md (cmpdf, cmpsf, bunordered, bordered, buneq,
64 bunge, bungt, bunle, bunlt, bltgt): Enable for TARGET_SSE_MATH,
66 (cmpfp_i_387): Rename from cmpfp_i. Move after sse patterns.
67 (cmpfp_i_mixed): Rename from cmpfp_i_sse; use for TARGET_MIX_SSE_I387.
68 (cmpfp_i_sse): Rename from cmpfp_i_sse_only; use for TARGET_SSE_MATH.
69 (cmpfp_iu_mixed, cmpfp_iu_sse, cmpfp_iu_387): Similarly.
70 (fp_jcc_1_mixed, fp_jcc_1_sse, fp_jcc_1_387): Similarly.
71 (fp_jcc_2_mixed, fp_jcc_2_sse, fp_jcc_2_387): Similarly.
72 (fp_jcc_3_387, fp_jcc_4_387, fp_jcc_5_387, fp_jcc_6_387,
73 fp_jcc_7_387, fp_jcc_8_387): Rename from fp_jcc_N.
74 (movdicc_c_rex64): Rename with '*'.
75 (movsfcc, movdfcc): Add checks for 387 and sse math to condition.
76 (movsfcc_1_sse_min, movsfcc_1_sse_max, movsfcc_1_sse): New.
77 (movsfcc_1_387): Rename from movsfcc_1.
78 (movdfcc_1_sse_min, movdfcc_1_sse_max, movdfcc_1_sse): New.
79 (movdfcc_1, movdfcc_1_rex64): Add check for 387.
80 (sminsf3, smaxsf3, smindf3, smaxdf3): New.
81 (minsf3, minsf, minsf_nonieee, minsf_sse, mindf3, mindf,
82 mindf_nonieee, mindf_sse, maxsf3, maxsf, maxsf_nonieee, maxsf_sse,
83 maxdf3, maxdf, maxdf_nonieee, maxdf_sse, sse_movsfcc, sse_movsfcc_eq,
84 sse_movdfcc, sse_movdfcc_eq, sse_movsfcc_const0_1,
85 sse_movsfcc_const0_2, sse_movsfcc_const0_3, sse_movsfcc_const0_4,
86 sse_movdfcc_const0_1, sse_movdfcc_const0_2, sse_movdfcc_const0_3,
87 sse_movdfcc_const0_4): Remove.
88 * config/i386/i386.c (ix86_expand_fp_movcc): For TARGET_SSE_MATH,
89 recognize min/max early. Update for changed sse cmove patterns.
90 (ix86_split_sse_movcc): New.
91 * config/i386/i386-protos.h: Update.
93 2005-01-13 Steven Bosscher <stevenb@suse.de>
95 * tree-ssa-dse.c (fix_phi_uses): Use SSA operand iterators.
96 (fix_stmt_v_may_defs): Likewise.
98 2005-01-13 Richard Henderson <rth@redhat.com>
100 * config/i386/i386.c (ix86_expand_fp_absneg_operator): Use elt_mode
101 for converting the mask.
103 2005-01-13 David Edelsohn <edelsohn@gnu.org>
105 * config/rs6000/aix52.h (CPLUSPLUS_CPP_SPEC): Change _XOPEN_SOURCE
108 2005-01-13 Richard Henderson <rth@redhat.com>
110 * config/i386/i386.c (IX86_BUILTIN_MOVQ, IX86_BUILTIN_LOADD,
111 IX86_BUILTIN_STORED, IX86_BUILTIN_MOVQ2DQ,
112 IX86_BUILTIN_MOVDQ2Q): Remove.
113 (IX86_BUILTIN_VEC_EXT_V4SI): New.
114 (ix86_init_mmx_sse_builtins, ix86_expand_builtin): Update to match.
115 (ix86_expand_vector_extract): For V4S[FI], extract element 0 after
117 * config/i386/sse.md (sse_concatv2sf): Accept zero operand 2.
118 (sse2_pextrw): Fix immediate constraint.
119 (sse2_loadq, sse2_loadq_rex64): Remove.
120 * config/i386/emmintrin.h (_mm_cvtsi128_si32, _mm_cvtsi128_si64x):
121 Use __builtin_ia32_vec_ext_<size>.
122 (_mm_cvtsi32_si128, _mm_cvtsi64x_si128): Use _mm_set_epi<size>.
124 2005-01-13 Aldy Hernandez <aldyh@redhat.com>
126 * function.c (assign_parm_setup_block): Look inside original
127 entry_parm when inspecting PARALLEL.
129 2005-01-13 Ralf Corsepius <ralf.corsepius@rtems.org>
130 Joel Sherrill <joel@oarcorp.com>
133 * gthr-rtems.h (__gthread_recursive_mutex_t): New type.
134 (__GTHREAD_RECURSIVE_MUTEX_INIT_FUNCTION): Define to
135 rtems_gxx_recursive_mutex_init.
136 (__gthread_recursive_mutex_lock): New function.
137 (__gthread_recursive_mutex_trylock): Likewise.
138 (__gthread_recursive_mutex_unlock): Likewise.
140 2005-01-13 Ralf Corsepius <ralf.corsepius@rtems.org>
142 * config/i386/t-rtems-i386: Multilib on -mtune instead of -mcpu.
144 2005-01-13 Jan Beulich <jbeulich@novell.com>
146 * config/ia64/ia64.md (ashlti3, ashlti3_internal): New.
147 (ashrti3_internal): Indicate output is early clobber. Generate result
148 into output rather than first input. Use move for low word of output
149 if shift count is exactly 64.
150 (lshrti3_internal): Likewise.
152 2005-01-13 Hans-Peter Nilsson <hp@bitrange.com>
156 * reload1.c (delete_output_reload): Don't delete an output reload
157 if the pseudo lives longer than a single basic block. Adjust and
158 improve wording of comment.
160 2005-01-12 Andrew Pinski <pinskia@physics.uc.edu>
163 * config/darwin.c (machopic_select_section): Use TYPE_SIZE_UNIT instead
164 of TYPE_SIZE where we mean the number of bytes.
166 2005-01-12 Aldy Hernandez <aldyh@redhat.com>
168 * function.c (assign_parm_setup_block): Relax condition on
169 multi-register optimization.
171 2005-01-12 Nick Clifton <nickc@redhat.com>
173 * config/sh/sh.md (udivsi3_sh2a, divsi3_sh2a): Give these patterns
174 an "in_delay_slot" attribute of "no" to prevent them being used in
175 delay slots. This is forbidden because they might generate
178 2005-01-12 Alan Modra <amodra@bigpond.net.au>
181 * config/rs6000/rs6000.md (movtf_internal): Replace r->o and m->r
184 2005-01-12 Nick Clifton <nickc@redhat.com>
186 * config/iq2000/iq2000.h (ASM_SPEC): Undefine (to stop -Qy being
187 passed on to GAS) but do not define, as GAS no longer supports or
188 needs the -m2000 option.
190 2005-01-11 Roger Sayle <roger@eyesopen.com>
192 * builtins.c (fold_builtin_fabs): Convert argument to the appropriate
193 type as args of unprototyped builtins aren't automatically promoted.
194 (fold_builtin_abs): Likewise.
196 2005-01-11 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
198 * Makefile.in: Set a `build-warn' variable.
200 2005-01-11 Zdenek Dvorak <dvorakz@suse.cz>
202 PR tree-optimization/17949
203 * tree-ssa-loop-ivopts.c (may_be_unaligned_p): New function.
204 (find_interesting_uses_address): Use it.
206 2005-01-11 Aldy Hernandez <aldyh@redhat.com>
208 * regrename.c (kill_value): Handle subreg's that won't simplify.
210 2005-01-11 Richard Henderson <rth@redhat.com>
213 * config/i386/i386.h (enum ix86_builtins): Move ...
214 * config/i386/i386.c: ... here.
215 (IX86_BUILTIN_MOVDDUP, IX86_BUILTIN_MMX_ZERO, IX86_BUILTIN_PEXTRW,
216 IX86_BUILTIN_PINSRW, IX86_BUILTIN_LOADAPS, IX86_BUILTIN_LOADSS,
217 IX86_BUILTIN_STORESS, IX86_BUILTIN_SSE_ZERO, IX86_BUILTIN_PEXTRW128,
218 IX86_BUILTIN_PINSRW128, IX86_BUILTIN_LOADAPD, IX86_BUILTIN_LOADSD,
219 IX86_BUILTIN_STOREAPD, IX86_BUILTIN_STORESD, IX86_BUILTIN_STOREHPD,
220 IX86_BUILTIN_STORELPD, IX86_BUILTIN_SETPD1, IX86_BUILTIN_SETPD,
221 IX86_BUILTIN_CLRPD, IX86_BUILTIN_LOADPD1, IX86_BUILTIN_LOADRPD,
222 IX86_BUILTIN_STOREPD1, IX86_BUILTIN_STORERPD, IX86_BUILTIN_LOADDQA,
223 IX86_BUILTIN_STOREDQA, IX86_BUILTIN_CLRTI,
224 IX86_BUILTIN_LOADDDUP): Remove.
225 (IX86_BUILTIN_VEC_INIT_V2SI, IX86_BUILTIN_VEC_INIT_V4HI,
226 IX86_BUILTIN_VEC_INIT_V8QI, IX86_BUILTIN_VEC_EXT_V2DF,
227 IX86_BUILTIN_VEC_EXT_V2DI, IX86_BUILTIN_VEC_EXT_V4SF,
228 IX86_BUILTIN_VEC_EXT_V8HI, IX86_BUILTIN_VEC_EXT_V4HI,
229 IX86_BUILTIN_VEC_SET_V8HI, IX86_BUILTIN_VEC_SET_V4HI): New.
230 (ix86_init_builtins): Make static.
231 (ix86_init_mmx_sse_builtins): Update for changed builtins.
232 (ix86_expand_binop_builtin): Only use ix86_fixup_binary_operands
233 if all the modes match. Otherwise, fake it.
234 (get_element_number, ix86_expand_vec_init_builtin,
235 ix86_expand_vec_ext_builtin, ix86_expand_vec_set_builtin): New.
236 (ix86_expand_builtin): Make static. Update for changed builtins.
237 (ix86_expand_vector_move_misalign): Use sse2_loadlpd with zero
238 operand instead of sse2_loadsd. Cast sse1 fallback to V4SFmode.
239 (ix86_expand_vector_init_duplicate): New.
240 (ix86_expand_vector_init_low_nonzero): New.
241 (ix86_expand_vector_init_one_var, ix86_expand_vector_init_general):
242 Split out from ix86_expand_vector_init; handle integer modes.
243 (ix86_expand_vector_init): Use them.
244 (ix86_expand_vector_set, ix86_expand_vector_extract): New.
245 * config/i386/i386-protos.h: Update.
246 * config/i386/predicates.md (reg_or_0_operand): New.
247 * config/i386/mmx.md (mov<MMXMODEI>_internal): Add 'r' variants.
248 (movv2sf_internal): Likewise. And a splitter to match them all.
249 (vec_dupv2sf, mmx_concatv2sf, vec_setv2sf, vec_extractv2sf,
250 vec_initv2sf, vec_dupv4hi, vec_dupv2si, mmx_concatv2si, vec_setv2si,
251 vec_extractv2si, vec_initv2si, vec_setv4hi, vec_extractv4hi,
252 vec_initv4hi, vec_setv8qi, vec_extractv8qi, vec_initv8qi): New.
253 (mmx_pinsrw): Fix operand ordering.
254 * config/i386/sse.md (movv4sf splitter): Use direct pattern,
255 rather than sse_loadss expander.
256 (movv2df splitter): Similarly.
257 (sse_loadss, sse_loadlss): Remove.
258 (vec_dupv4sf, sse_concatv2sf, sse_concatv4sf, vec_extractv4sf_0): New.
259 (vec_setv4sf, vec_setv2df): Use ix86_expand_vector_set.
260 (vec_extractv4sf, vec_extractv2df): Use ix86_expand_vector_extract.
261 (sse3_movddup): Rename with '*'.
262 (sse3_movddup splitter): Use gen_rtx_REG instead of gen_lowpart.
263 (sse2_loadsd): Remove.
264 (vec_dupv2df_sse3): Rename from sse3_loadddup.
265 (vec_dupv2df, vec_concatv2df_sse3, vec_concatv2df): New.
266 (sse2_pinsrw): Fix argument ordering.
267 (sse2_loadld, sse2_loadq): Add sse1 alternatives.
268 (sse2_stored): Remove 'r' destination.
269 (vec_dupv4si, vec_dupv2di, sse2_concatv2si, sse1_concatv2si,
270 vec_concatv4si_1, vec_concatv2di, vec_setv2di, vec_extractv2di,
271 vec_initv2di, vec_setv4si, vec_extractv4si, vec_initv4si,
272 vec_setv8hi, vec_extractv8hi, vec_initv8hi, vec_setv16qi,
273 vec_extractv16qi, vec_initv16qi): New.
275 * config/i386/emmintrin.h (__m128i, __m128d): Use typedef, not define.
276 (_mm_set_sd, _mm_set1_pd, _mm_setzero_pd, _mm_set_epi64x,
277 _mm_set_epi32, _mm_set_epi16, _mm_set_epi8, _mm_setzero_si128): Use
279 (_mm_load_pd, _mm_store_pd): Use plain dereference.
280 (_mm_load_si128, _mm_store_si128): Likewise.
281 (_mm_load1_pd): Use _mm_set1_pd.
282 (_mm_load_sd): Use _mm_set_sd.
283 (_mm_store_sd, _mm_storeh_pd): Use __builtin_ia32_vec_ext_v2df.
284 (_mm_store1_pd, _mm_storer_pd): Use _mm_store_pd.
285 (_mm_set_epi64): Use _mm_set_epi64x.
286 (_mm_set1_epi64x, _mm_set1_epi64, _mm_set1_epi32, _mm_set_epi16,
287 _mm_set1_epi8, _mm_setr_epi64, _mm_setr_epi32, _mm_setr_epi16,
288 _mm_setr_epi8): Use _mm_set_foo form.
289 (_mm_loadl_epi64, _mm_movpi64_epi64, _mm_move_epi64): Use _mm_set_epi64.
290 (_mm_storel_epi64, _mm_movepi64_pi64): Use __builtin_ia32_vec_ext_v2di.
291 (_mm_extract_epi16): Use __builtin_ia32_vec_ext_v8hi.
292 (_mm_insert_epi16): Use __builtin_ia32_vec_set_v8hi.
293 * config/i386/mmintrin.h (_mm_setzero_si64): Use plain cast.
294 (_mm_set_pi32): Use __builtin_ia32_vec_init_v2si.
295 (_mm_set_pi16): Use __builtin_ia32_vec_init_v4hi.
296 (_mm_set_pi8): Use __builtin_ia32_vec_init_v8qi.
297 (_mm_set1_pi16, _mm_set1_pi8): Use _mm_set_piN variant.
298 * config/i386/pmmintrin.h (_mm_loaddup_pd): Use _mm_load1_pd.
299 (_mm_movedup_pd): Use _mm_shuffle_pd.
300 * config/i386/xmmintrin.h (_mm_setzero_ps, _mm_set_ss,
301 _mm_set1_ps, _mm_set_ps, _mm_setr_ps): Use constructor form.
302 (_mm_cvtpi16_ps, _mm_cvtpu16_ps, _mm_cvtpi8_ps, _mm_cvtpu8_ps,
303 _mm_cvtps_pi8, _mm_cvtpi32x2_ps): Avoid __builtin_ia32_mmx_zero;
305 (_mm_load_ss, _mm_load1_ps): Use _mm_set* form.
306 (_mm_load_ps, _mm_loadr_ps): Use raw dereference.
307 (_mm_store_ss): Use __builtin_ia32_vec_ext_v4sf.
308 (_mm_store_ps): Use raw dereference.
309 (_mm_store1_ps): Use _mm_storeu_ps.
310 (_mm_storer_ps): Use _mm_store_ps.
311 (_mm_extract_pi16): Use __builtin_ia32_vec_ext_v4hi.
312 (_mm_insert_pi16): Use __builtin_ia32_vec_set_v4hi.
314 2005-01-11 Stan Shebs <shebs@apple.com>
316 * config/rs6000/rs6000.c (machopic_output_stub): Issue
317 ldu instead of lwzu if 64-bit -mdynamic-no-pic.
319 2005-01-11 Andrew Pinski <pinskia@physics.uc.edu>
322 * config/rs6000/rs6000.c (rs6000_special_round_type_align):
323 Skip all DECLs except for FIELD_DECLs.
325 2005-01-11 Andreas Krebbel <krebbel1@de.ibm.com>
327 * config/s390/s390.c (override_options): Return error if
328 -mbackchain, -mpacked-stack and -mhard-float are used together.
329 (s390_va_start): Remove the backchain && packed-stack special case.
330 (s390_gimplify_va_arg): Likewise.
331 * doc/invoke.texi: Remove the ABI incompatibility note.
333 2005-01-11 Andreas Krebbel <krebbel1@de.ibm.com>
335 * config/s390/s390.c (struct s390_frame_layout): Remove
337 (s390_frame_info, s390_emit_prologue): Replace occurrences of
338 save_backchain_p with TARGET_BACKCHAIN.
340 2005-01-11 Alan Modra <amodra@bigpond.net.au>
343 * builtins.c (std_gimplify_va_arg_expr): Adjust alignment of *ap.
344 * expr.h (struct locate_and_pad_arg_data): Add "boundary".
345 * function.c (locate_and_pad_parm): Set new field.
346 (assign_parm_find_stack_rtl): Use it instead of FUNCTION_ARG_BOUNDARY.
347 Tweak where_pad test to include "none". Always set mem align for
349 (assign_parm_adjust_stack_rtl): Discard stack_parm if alignment
350 not sufficient for type.
351 (assign_parm_setup_block): If stack_parm is zero on entry, always
352 make a new stack local. Block move old stack parm if necessary
353 to new aligned stack local.
354 (assign_parm_setup_stack): Use a block move to handle
355 potentially misaligned entry_parm.
356 (assign_parms_unsplit_complex): Specify required alignment when
357 creating stack local.
358 * calls.c (compute_argument_addresses): Override alignment of stack
359 arg calculated from its type with the alignment given by
360 FUNCTION_ARG_BOUNDARY.
361 (store_one_arg): Likewise.
363 2005-01-11 Jan Beulich <jbeulich@novell.com>
365 * config/ia64/ia64.md (zero_extendsidi2): Replace zxt4 by addp4.
366 Change respective itanium_class attribute to ialu.
367 (shladdp4_internal): New.
368 * config/ia64/predicates.md (shladd_log2_operand): New.
370 2005-01-11 Richard Henderson <rth@redhat.com>
372 * expr.c (store_constructor): Use rtvec_alloc instead of
373 alloca+gen_rtvec_v, and an incorrect number passed to alloca.
375 2005-01-11 Kazu Hirata <kazu@cs.umass.edu>
377 * config/alpha/alpha.c, config/i386/mmx.md: Fix comment typos.
379 2005-01-11 Alan Modra <amodra@bigpond.net.au>
381 * varasm.c (default_section_type_flags_1): Don't set SECTION_SMALL.
382 * config/ia64/ia64.c (TARGET_SECTION_TYPE_FLAGS): Define.
383 (TARGET_RWRELOC): Define.
384 (ia64_rwreloc_section_type_flags): Delete.
385 (ia64_section_type_flags): New function.
386 * config/ia64/hpux.h (TARGET_SECTION_TYPE_FLAGS): Don't define.
387 (TARGET_RWRELOC): Define.
389 2005-01-10 David Mosberger <davidm@hpl.hp.com>
392 * config/ia64/ia64.c (process_set): For alloc insn, only call
393 process_epilogue is !frame_pointer_needed.
395 2005-01-10 Roger Sayle <roger@eyesopen.com>
398 * c-common.c (c_common_truthvalue_conversion): TRUTH_NOT_EXPR is a
399 unary operator and can't be treated as a binary/comparison operator.
401 2005-01-10 Richard Henderson <rth@redhat.com>
403 * config/i386/i386.c (ix86_function_value): Use type_natural_mode.
404 (ix86_return_in_memory): Likewise.
405 (function_arg_advance): Likewise. Mirror structure in function_arg
406 for choosing register to advance.
408 2005-01-10 Kazu Hirata <kazu@cs.umass.edu>
410 * tree-vectorizer.c, tree.def: Fix comment typos.
412 2005-01-10 Mark Dettinger <dettinge@de.ibm.com>
414 * config/s390/s390.c (struct processor_costs): 4 new fields:
415 ddbr, ddr, debr, der.
416 (s390_rtx_costs): More precise handling of divide instructions.
418 2005-01-10 David Edelsohn <edelsohn@gnu.org>
420 * config/rs6000/t-aix43 (BOOT_LDFLAGS): Define.
422 2005-01-10 Jan Beulich <jbeulich@novell.com>
424 * config/ia64/ia64.c (ia64_in_small_data_p): Also handle the section
425 names resulting from -ffunction-sections/-fdata-sections and linkonce
427 * varasm.c (default_section_type_flags_1): Also set SECTION_SMALL
428 based on the section name. Rearrange the section name comparison logic
429 slightly so that each section name is compared against at most once.
431 2005-01-10 Ben Elliston <bje@au.ibm.com>
433 * doc/invoke.texi (Code Gen Options): Add PowerPC to the list of
434 targets for which -fPIC reduces limitations on the GOT size.
436 2005-01-09 Falk Hueffner <falk@debian.org>
438 * fold-const.c (fold): Also handle EXACT_DIV_EXPR when folding
441 2005-01-09 David Edelsohn <edelsohn@gnu.org>
444 * collect2.c (main): Set aixrtl_flag for -brtl option.
445 (resolve_lib_name): Search for .so file extension before .a
448 2005-01-09 Dorit Naishlos <dorit@il.ibm.com>
450 * tree-vectorizer.c (vect_enhance_data_refs_alignment): Add dump prints.
451 (vect_analyze_data_refs_alignment): Add dump prints.
453 2005-01-09 Ira Rosen <irar@il.ibm.com>
455 * tree-vectorizer.c (vect_analyze_offset_expr): Use
456 expr_invariant_in_loop_p.
457 Initialize outputs first thing in the function.
458 (vect_update_ivs_after_vectorizer): Call initial_condition_in_loop_num.
459 (vect_is_simple_iv_evolution): Call initial_condition_in_loop_num.
460 (vect_analyze_pointer_ref_access): Check that the initial condition of
461 the access function is loop invariant.
463 2005-01-09 Richard Henderson <rth@redhat.com>
465 * config/i386/i386.c (bdesc_2arg): Update names for mmx_ prefixes.
466 (ix86_expand_builtin): Likewise. Frob MASKMOVQ wrt the input mem
467 just like MASKMOVDQU. Return plain zero for MMX_ZERO.
468 * config/i386/i386.md (MMXMODEI, mov<MMXMODEI>,
469 mov<MMXMODEI>_internal_rex64, mov<MMXMODEI>_internal, movv2sf,
470 movv2sf_internal_rex64, movv2sf_internal, MMXMODE,
471 movmisalign<MMXMODE>, mmx_pmovmskb, mmx_maskmovq, mmx_maskmovq_rex,
472 sse_movntdi, addv8qi3, addv4hi3, addv2si3, mmx_adddi3, ssaddv8qi3,
473 ssaddv4hi3, usaddv8qi3, usaddv4hi3, subv8qi3, subv4hi3, subv2si3,
474 mmx_subdi3, sssubv8qi3, sssubv4hi3, ussubv8qi3, ussubv4hi3,
475 mulv4hi3, smulv4hi3_highpart, umulv4hi3_highpart, mmx_pmaddwd,
476 sse2_umulsidi3, mmx_iordi3, mmx_xordi3, mmx_anddi3, mmx_nanddi3,
477 mmx_uavgv8qi3, mmx_uavgv4hi3, mmx_psadbw, mmx_pinsrw, mmx_pinsrw,
478 mmx_pextrw, mmx_pshufw, eqv8qi3, eqv4hi3, eqv2si3, gtv8qi3, gtv4hi3,
479 gtv2si3, umaxv8qi3, smaxv4hi3, uminv8qi3, sminv4hi3, ashrv4hi3,
480 ashrv2si3, lshrv4hi3, lshrv2si3, mmx_lshrdi3, ashlv4hi3, ashlv2si3,
481 mmx_ashldi3, mmx_packsswb, mmx_packssdw, mmx_packuswb, mmx_punpckhbw,
482 mmx_punpckhwd, mmx_punpckhdq, mmx_punpcklbw, mmx_punpcklwd,
483 mmx_punpckldq, emms, addv2sf3, subv2sf3, subrv2sf3, gtv2sf3, gev2sf3,
484 eqv2sf3, pfmaxv2sf3, pfminv2sf3, mulv2sf3, femms, pf2id, pf2iw,
485 pfacc, pfnacc, pfpnacc, pi2fw, floatv2si2, pfrcpv2sf2, pfrcpit1v2sf3,
486 pfrcpit2v2sf3, pfrsqrtv2sf2, pfrsqit1v2sf3, pmulhrwv4hi3, pswapdv2si2,
487 pswapdv2sf2): Move to mmx.md; rename as necessary with leading
489 (mmx_clrdi, pavgusb): Remove.
490 (ldmxcsr, stmxcsr, sfence, sfence_insn): Move to sse.md; rename
491 with leading sse_ prefix.
492 * config/i386/sse.md: Receive them.
493 * config/i386/mmx.md: New file.
494 (MMXMODE12, MMXMODE24, mmxvecsize): New.
495 (subrv2sf3): Turn into expander for normal subtraction.
496 (mmx_addv2sf3, mmx_mulv2sf3, mmx_smaxv2sf3, mmx_sminv2sf3,
497 mmx_eqv2sf3, mmx_mulv4hi3, mmx_smulv4hi3_highpart,
498 mmx_umulv4hi3_highpart, mmx_pmaddwd, mmx_pmulhrwv4hi3, sse2_umulsidi3,
499 mmx_umaxv8qi3, mmx_smaxv4hi3, mmx_uminv8qi3, mmx_sminv4hi3): Mark
500 commutative; use ix86_binary_operator_ok.
501 (mmx_add<MMXMODEI>3, mmx_ssadd<MMXMODE12>3, mmx_usadd<MMXMODE12>3,
502 mmx_sub<MMXMODEI>3, mmx_sssub<MMXMODE12>3, mmx_ussub<MMXMODE12>3
503 mmx_ashr<MMXMODE24>3, mmx_lshr<MMXMODE23>3, mmx_ashl<MMXMODE24>3
504 mmx_eq<MMXMODEI>3, mmx_gt<MMXMODEI>3, mmx_and<MMXMODEI>3,
505 mmx_nand<MMXMODEI>3, mmx_ior<MMXMODEI>3, mmx_xor<MMXMODEI>3):
506 Macroize from existing patterns; use ix86_binary_operator_ok.
507 (mmx_packsswb, mmx_packssdw, mmx_packuswb): Add memory alternative.
508 (mmx_punpckhbw, mmx_punpcklbw, mmx_punpckhwd, mmx_punpcklwd,
509 mmx_punpckhdq, mmx_punpckhdq, mmx_punpckldq): Likewise. Model
510 with vec_select+vec_concat.
511 (mmx_pshufw, mmx_pshufw_1): Likewise.
512 (mmx_uavgv8qi3): Merge pavgusb. Model correcty.
513 (mmx_uavgv4hi3): Model correctly.
514 * config/i386/mmintrin.h (_mm_and_si64, _mm_andnot_si64, _mm_or_si64,
515 _mm_xor_si64): Remove casts.
517 2005-01-09 Zdenek Dvorak <dvorakz@suse.cz>
519 PR tree-optimization/19224
520 * tree-scalar-evolution.c (get_instantiated_value,
521 set_instantiated_value): New functions.
522 (instantiate_parameters_1): Cache the results.
523 (instantiate_parameters, resolve_mixers): Initialize and free
526 2005-01-08 David Edelsohn <edelsohn@gnu.org>
528 * config/i386/i386.md (addhi_4): Correct reference in comment.
531 2005-01-08 Richard Henderson <rth@redhat.com>
533 * config/i386/emmintrin.h (_mm_cvtsi128_si32): Move earlier.
534 (_mm_cvtsi128_si64x): Likewise.
535 (_mm_srl_epi64, _mm_srl_epi32, _mm_srl_epi16, _mm_sra_epi32,
536 _mm_sra_epi16, _mm_sll_epi64, _mm_sll_epi32, _mm_sll_epi16): Use
537 the _mm_{srl,sll}i_foo counterpart, and _mm_cvtsi128_si32.
538 * config/i386/i386-modes.def: Add V16HI, V32QI, V4DF, V8SF.
539 * config/i386/i386-protos.h: Update.
540 * config/i386/i386.c (print_operand): Add 'H'.
541 (ix86_fixup_binary_operands): Split out from ...
542 (ix86_expand_binary_operator): ... here.
543 (ix86_fixup_binary_operands_no_copy): New.
544 (ix86_expand_fp_absneg_operator): Handle vector mode results.
545 (bdesc_2arg): Update names for sse{,2,3}_ prefixes.
546 (ix86_init_mmx_sse_builtins): Remove *maskncmp* special cases.
547 (safe_vector_operand): Use CONST0_RTX.
548 (ix86_expand_binop_builtin): Use ix86_fixup_binary_operands.
549 (ix86_expand_builtin): Merge CODE_FOR_sse2_maskmovdqu_rex64 and
550 CODE_FOR_sse2_maskmovdqu. Special case SSE version of MASKMOVDQU
551 expansion. Update names for sse{,2,3}_ prefixes. Remove *maskncmp*
553 * config/i386/i386.h (IX86_BUILTIN_CMPNGTSS): New.
554 (IX86_BUILTIN_CMPNGESS): New.
555 * config/i386/i386.md (UNSPEC_FIX_NOTRUNC): New.
556 (attr type): Add sselog1.
557 (attr unit, attr memory): Handle it.
558 (movti, movti_internal, movti_rex64): Move near other integer moves.
559 (movtf, movtf_internal): Move near other fp moves.
560 (SSEMODE, SSEMODEI, vec_setv2df, vec_extractv2df, vec_initv2df,
561 vec_setv4sf, vec_extractv4sf, vec_initv4sf, movv4sf, movv4sf_internal,
562 movv2df, movv2df_internal, mov<SSEMODEI>, mov<SSEMODEI>_internal,
563 movmisalign<SSEMODE>, sse_movups_1, sse_movmskps, sse_movntv4sf,
564 sse_movhlps, sse_movlhps, sse_storehps, sse_loadhps, sse_storelps,
565 sse_loadlps, sse_loadss, sse_loadss_1, sse_movss, sse_storess,
566 sse_shufps, addv4sf3, vmaddv4sf3, subv4sf3, vmsubv4sf3, negv4sf2,
567 mulv4sf3, vmmulv4sf3, divv4sf3, vmdivv4sf3, rcpv4sf2, vmrcpv4sf2,
568 rsqrtv4sf2, vmrsqrtv4sf2, sqrtv4sf2, vmsqrtv4sf2, sse_andv4sf3,
569 sse_nandv4sf3, sse_iorv4sf3, sse_xorv4sf3, sse2_andv2df3,
570 sse2_nandv2df3, sse2_iorv2df3, sse2_xorv2df3, sse2_andv2di3,
571 sse2_nandv2di3, sse2_iorv2di3, sse2_xorv2di3, maskcmpv4sf3,
572 vmmaskcmpv4sf3, sse_comi, sse_ucomi, sse_unpckhps, sse_unpcklps,
573 smaxv4sf3, vmsmaxv4sf3, sminv4sf3, vmsminv4sf3, cvtpi2ps, cvtps2pi,
574 cvttps2pi, cvtsi2ss, cvtsi2ssq, cvtss2si, cvtss2siq, cvttss2si,
575 cvttss2siq, addv2df3, vmaddv2df3, subv2df3, vmsubv2df3, mulv2df3,
576 vmmulv2df3, divv2df3, vmdivv2df3, smaxv2df3, vmsmaxv2df3, sminv2df3,
577 vmsminv2df3, sqrtv2df2, vmsqrtv2df2, maskcmpv2df3, vmmaskcmpv2df3,
578 sse2_comi, sse2_ucomi, sse2_movmskpd, sse2_pmovmskb, sse2_maskmovdqu,
579 sse2_maskmovdqu_rex64, sse2_movntv2df, sse2_movntv2di, sse2_movntsi,
580 cvtdq2ps, cvtps2dq, cvttps2dq, cvtdq2pd, cvtpd2dq, cvttpd2dq,
581 cvtpd2pi, cvttpd2pi, cvtpi2pd, cvtsd2si, cvtsd2siq, cvttsd2si,
582 cvttsd2siq, cvtsi2sd, cvtsi2sdq, cvtsd2ss, cvtss2sd, cvtpd2ps,
583 cvtps2pd, addv16qi3, addv8hi3, addv4si3, addv2di3, ssaddv16qi3,
584 ssaddv8hi3, usaddv16qi3, usaddv8hi3, subv16qi3, subv8hi3, subv4si3,
585 subv2di3, sssubv16qi3, sssubv8hi3, ussubv16qi3, ussubv8hi3, mulv8hi3,
586 smulv8hi3_highpart, umulv8hi3_highpart, sse2_umulsidi3,
587 sse2_umulv2siv2di3, sse2_pmaddwd, sse2_uavgv16qi3, sse2_uavgv8hi3,
588 sse2_psadbw, sse2_pinsrw, sse2_pextrw, sse2_pshufd, sse2_pshuflw,
589 sse2_pshufhw, eqv16qi3, eqv8hi3, eqv4si3, gtv16qi3, gtv8hi3,
590 gtv4si3, umaxv16qi3, smaxv8hi3, uminv16qi3, sminv8hi3, ashrv8hi3,
591 ashrv4si3, lshrv8hi3, lshrv4si3, lshrv2di3, ashlv8hi3, ashlv4si3,
592 ashlv2di3, sse2_ashlti3, sse2_lshrti3, sse2_unpckhpd, sse2_unpcklpd,
593 sse2_packsswb, sse2_packssdw, sse2_packuswb, sse2_punpckhbw,
594 sse2_punpckhwd, sse2_punpckhdq, sse2_punpcklbw, sse2_punpcklwd,
595 sse2_punpckldq, sse2_punpcklqdq, sse2_punpckhqdq, sse2_movupd,
596 sse2_movdqu, sse2_movdq2q, sse2_movdq2q_rex64, sse2_movq2dq,
597 sse2_movq2dq_rex64, sse2_loadd, sse2_stored, sse2_storehpd,
598 sse2_loadhpd, sse2_storelpd, sse2_loadlpd, sse2_movsd, sse2_loadsd,
599 sse2_loadsd_1, sse2_storesd, sse2_shufpd, sse2_clflush, sse2_mfence,
600 mfence_insn, sse2_lfence, lfence_insn, mwait, monitor, addsubv4sf3,
601 addsubv2df3, haddv4sf3, haddv2df3, hsubv4sf3, hsubv2df3, movshdup,
602 movsldup, lddqu, loadddup, movddup): Move to sse.md. Any with
603 non-optabs meanings renamed with an "sse{,2,3}_" prefix at the
605 (SSEPUSH, push<SSEPUSH>): Remove.
606 (MMXPUSH, push<MMXPUSH>): Remove.
607 (sse_movaps, sse_movaps_1, sse_movups): Remove.
608 (sse2_movapd, sse2_movdqa, sse2_movq): Remove.
609 (sse2_andti3, sse2_nandti3, sse2_iorti3, sse2_xorti3): Remove.
610 (sse_clrv4sf, sse_clrv2df, sse2_clrti): Remove.
611 (maskncmpv4sf3, vmmaskncmpv4sf3): Remove.
612 (maskncmpv2df3, vmmaskncmpv2df3): Remove.
613 (ashrv8hi3_ti, ashrv4si3_ti, lshrv8hi3_ti, lshrv4si3_ti): Remove.
614 (lshrv2di3_ti, ashlv8hi3_ti, ashlv4si3_ti, ashlv2di3_ti): Remove.
615 * config/i386/athlon.md (athlon_sselog_load): Handle sselog1.
616 (athlon_sselog_load_k8, athlon_sselog, athlon_sselog_k8): Likewise.
617 * config/i386/ppro.md (ppro_sse_div_V4SF_load): Fix memory attr.
618 (ppro_sse_log_V4SF_load): Similarly. Handle sselog1.
619 (ppro_sse_log_V4SF): Handle sselog1.
620 * config/i386/predicates.md (const_0_to_1_operand): New.
621 (const_0_to_255_mul_8_operand): New.
622 (const_1_to_31_operand): Rename from const_int_1_31_operand.
623 (const_2_to_3_operand, const_4_to_7_operand): New.
624 * config/i386/sse.md: New file.
625 (SSEMODE12, SSEMODE24, SSEMODE124, SSEMODE248, ssevecsize): New.
626 (sse_movups): Rename from sse_movups_1.
627 (sse_loadlss): Rename from sse_loadss_1.
628 (andv4sf3, iorv4sf3, xorv4sf3, andv2df3): Remove the sse prefix
630 (negv4sf2): Use ix86_expand_fp_absneg_operator.
631 (absv4sf2, negv2df, absv2df): New.
632 (addv4sf3): Add expander to call ix86_fixup_binary_operands_no_copy.
633 (subv4sf3, mulv4sf3, divv4sf3, smaxv4sf3, sminv4sf3, andv4sf3,
634 iorv4sf3, xorv4sf3, addv2df3, subv2df3, mulv2df3, divv2df3,
635 smaxv2df3, sminv2df3, andv2df3, iorv2df3, xorv2df3, mulv8hi3,
636 umaxv16qi3, smaxv8hi3, uminv16qi3, sminv8hi3): Likewise.
637 (sse3_addsubv4sf3): Model correctly.
638 sse3_haddv4sf3, sse3_hsubv4sf3, sse3_addsubv2df3, sse3_haddv2df3,
639 sse3_hsubv2df3, sse2_ashlti3, sse2_lshrti3): Likewise.
640 (sse_movhlps): Model with vec_select+vec_concat.
641 (sse_movlhps, sse_unpckhps, sse_unpcklps, sse3_movshdup,
642 sse3_movsldup, sse_shufps, sse_shufps_1, sse2_unpckhpd, sse3_movddup,
643 sse2_unpcklpd, sse2_shufpd, sse2_shufpd_1, sse2_punpckhbw,
644 sse2_punpcklbw, sse2_punpckhwd, sse2_punpcklwd, sse2_punpckhdq,
645 sse2_punpckldq, sse2_punpckhqdq, sse2_punpcklqdq, sse2_pshufd,
646 sse2_pshufd_1, sse2_pshuflw, sse2_pshuflw_1, sse2_pshufhw,
647 sse2_pshufhw_1): Likewise.
648 (neg<SSEMODEI>2, one_cmpl<SSEMODEI>2): New.
649 (add<SSEMODEI>3, sse2_ssadd<SSEMODE12>3, sse2_usadd<SSEMODE12>3,
650 sub<SSEMODEI>3, sse2_sssub<SSEMODE12>3, sse2_ussub<SSEMODE12>3,
651 ashr<SSEMODE24>3, lshr<SSEMODE248>3, sse2_eq<SSEMODE124>3,
652 sse2_gt<SSEMODDE124>3, and<SSEMODEI>3, sse_nand<SSEMODEI>3,
653 ior<SSEMODEI>3, xor<SSEMODEI>3): Macroize from existing patterns.
654 (addv4sf3, sse_vmaddv4sf3, mulv4sf3, sse_vmmulv4sf3, smaxv4sf3,
655 sse_vmsmaxv4sf3, sminv4sf3, sse_vmsminv4sf3, addv2df3, sse2_vmaddv2df3,
656 mulv2df3, sse2_vmmulv2df3, smaxv2df3, sse2_vmsmaxv2df3, sminv2df3,
657 sse2_vmsminv2df3, umaxv16qi3, smaxv8hi3, uminv16qi3
658 sminv8hi3): Mark commutative
659 operands. Use ix86_binary_operator_ok.
660 (sse_unpckhps, sse_unpcklps, sse2_packsswb, sse2_packssdw,
661 sse2_packuswb, sse2_punpckhbw, sse2_punpcklbw, sse2_punpckhwd,
662 sse2_punpcklwd, sse2_punpckhdq, sse2_punpckldq, sse2_punpckhqdq,
663 sse2_punpcklqdq): Allow operand2 in memory.
664 (sse_movhlps, sse_movlhps, sse2_unpckhpd, sse2_unpcklpd
665 sse2_movsd): Add memory alternatives.
666 (sse_storelps): Turn expander into an insn; split after reload.
667 (sse_storess, sse2_loadhpd, sse2_loadlpd): Add non-xmm inputs.
668 (sse2_storehpd, sse2_storelpd): Add non-xmm outputs.
670 2005-01-08 Eric Botcazou <ebotcazou@libertysurf.fr>
672 * configure.ac (DWARF-2 debug_line): Use objdump.
673 * configure: Regenerate.
675 2005-01-08 Jeff Law <law@redhat.com>
676 Diego Novillo <dnovillo@redhat.com>
678 PR tree-optimization/18241
679 * tree-nrv.c (tree_nrv): Ignore volatile return values.
680 * tree-ssa-dse.c (dse_optimize_stmt): Do not optimize
681 statements with volatile operands.
682 * tree-ssa-operands.c (add_stmt_operand): Do add volatile
683 operands after marking a statement with has_volatile_ops.
685 2005-01-08 Roger Sayle <roger@eyesopen.com>
687 * tree.c (int_fits_type_p): Always honor integer constant
688 TYPE_MIN_VALUE and TYPE_MAX_VALUE if they exist.
690 2005-01-08 Roger Sayle <roger@eyesopen.com>
692 * ifcvt.c (find_if_case_1): Reinstate 2005-01-04 change, now that
693 the latent bug in rtl_delete_block has been resolved.
695 2005-01-08 Richard Sandiford <rsandifo@redhat.com>
697 * config/mips/t-iris6 (MULTILIB_DIRNAMES): Use -mabi argument values.
698 (MULTILIB_OSDIRNAMES): Use the standard lib, lib32 and lib64.
700 2005-01-08 Richard Sandiford <rsandifo@redhat.com>
702 * config/mips/t-slibgcc-irix (SHLIB_LINK): Install a copy of the
703 library as @multilib_dir@/$(SHLIB_SONAME).
705 2005-01-07 Eric Botcazou <ebotcazou@libertysurf.fr>
707 * configure.ac (HAVE_AS_OFFSETABLE_LO10): Fix typo.
708 * configure: Regenerate.
710 2005-01-07 Jakub Jelinek <jakub@redhat.com>
712 * c-common.c (handle_mode_attribute): For ENUMERAL_TYPE, also copy
715 2005-01-07 David Edelsohn <edelsohn@gnu.org>
718 * config/rs6000/rs6000.c (rs6000_legitimize_reload_address):
719 Convert non-word aligned offset address using ld/std into
722 2005-01-07 Richard Henderson <rth@redhat.com>
724 * config/i386/i386.md (sse_loadhps splitter): Fix operand number typo.
726 2005-01-07 Richard SAndiford <rsandifo@redhat.com>
728 * tree.h (IS_EXPR_CODE_CLASS): Use a straight-forward range check.
730 2005-01-07 Jakub Jelinek <jakub@redhat.com>
732 PR tree-optimization/19283
733 * fold-const.c (fold_widened_comparison): Return NULL if shorter_type
734 is not shorter than the original type.
736 PR rtl-optimization/19012
737 * config/i386/i386.md (addqi_1_slp): Set memory attribute.
739 PR rtl-optimization/18861
740 * cfgbuild.c (BLOCK_USED_BY_TABLEJUMP): Define.
741 (FULL_STATE): Define.
742 (mark_tablejump_edge): New function.
743 (purge_dead_tablejump_edges): New function.
744 (find_bb_boundaries): Use it.
746 PR tree-optimization/18828
747 * builtins.c (expand_builtin_next_arg): Remove argument and all
748 the argument checking.
749 (expand_builtin): Adjust caller.
750 (expand_builtin_va_start): Likewise. Remove error for too many
752 (fold_builtin_next_arg): Issue error for too many arguments.
753 After checking arguments, replace them with magic arguments that
754 prevent further checking of the args.
756 2005-01-06 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
758 * pa64-hpux.h (STARTFILE_SPEC): Fix typo in spec.
760 2005-01-06 Roger Sayle <roger@eyesopen.com>
763 * gcc.c (input_suffix_matches): Tweak the semantics of %{.s:...}
764 and %{.S:...} (and their negative variants) to test whether the
765 input file is assembler or pre-processed-assembler independent of
766 the actual filename extension.
768 2005-01-06 Roger Sayle <roger@eyesopen.com>
770 * simplify-rtx.c (simplify_subreg): Simplify truncations of shifts
771 of sign or zero extended values.
773 2005-01-06 Geoffrey Keating <geoffk@apple.com>
775 * c-cppbuiltin.c (builtin_define_float_constants): Set __*_EPSILON__
776 for IBM long double format correctly.
778 2005-01-06 Daniel Berlin <dberlin@dberlin.org>
780 Fix PR tree-optimization/18792
782 * tree-data-ref.c (build_classic_dist_vector): Change first_loop
783 to first_loop_depth, and use loop depth instead of loop number.
784 (build_classic_dir_vector): Ditto.
785 (compute_data_dependences_for_loop): Use depth, not loop number.
786 * tree-loop-linear.c (try_interchange_loops): Use loop depth, not loop
787 number. Pass in loops, instead of loop numbers.
788 (gather_interchange_stats): Ditto.
789 (linear_transform_loops): Ditto.
791 2005-01-06 Richard Sandiford <rsandifo@redhat.com>
794 * loop.c (get_monotonic_increment, biased_biv_fits_mode_p,
795 biv_fits_mode_p, extension_within_bounds_p): New functions.
796 (check_ext_dependent_givs): Use them.
798 2005-01-06 Roger Sayle <roger@eyesopen.com>
800 * cfgrtl.c (rtl_delete_block): A basic block may be followed by
801 more than one barrier, in which case we should delete them all.
803 2005-01-06 Tobias Schlueter <tobias.schlueter@physik.uni-muenchen.de>
805 * gcc.c (process_command): Change year in 'gcc --version' to 2005.
807 2005-01-05 Daniel Berlin <dberlin@dberlin.org>
809 Fix PR middle-end/19286
811 * dwarf2out.c (gen_subprogram_die): If we've already tried to
812 output this subprogram, simply ignore this attempt to do it again.
813 (add_abstract_origin_attribute): Don't abort trying to add the abstract
814 origin attribute if it's not possible.
815 (gen_block_die): Don't ignore subblocks of "unused" blocks.
816 (decls_for_scope): Ditto.
817 * gimple-low.c (mark_blocks_with_used_subblocks): Remove.
818 (mark_used_blocks): Don't call mark_blocks_with_used_subblocks.
820 2005-01-05 Richard Henderson <rth@redhat.com>
823 * config/i386/i386.c (BUILTIN_DESC_SWAP_OPERANDS): New.
824 (bdesc_2arg): Use it.
825 (ix86_expand_binop_builtin): Force operands into registers
827 (ix86_expand_unop_builtin, ix86_expand_unop1_builtin,
828 ix86_expand_sse_compare, ix86_expand_sse_comi,
829 ix86_expand_builtin): Likewise.
831 2005-01-05 Richard Henderson <rth@redhat.com>
833 * config/ia64/ia64.c (rtx_needs_barrier): Handle CONST_VECTOR
835 * config/ia64/vect.md (mulv8qi3): Re-implement with mix_[rl].
836 (mulv4hi3): Set itanium_class mmmul.
837 (fpack_sfxf, fpack_xfsf, fpack_xfxf): Remove.
838 (fpack): Rename from fpack_sfsf.
840 2005-01-05 Richard Henderson <rth@redhat.com>
843 * reload1.c (do_input_reload): Restrict the optimization deleteing
844 a previous output reload to RELOAD_FOR_INPUT.
846 2005-01-05 Steven Bosscher <stevenb@suse.de>
848 * combine.c (expand_compound_operation) <ZERO_EXTRACT>: Add
849 comment that we fall through after case.
850 (mark_used_regs_combine): Don't expect a SIGN_EXTRACT in a SET_DEST.
851 (distribute_links): Likewise.
852 * cse.c (cse_insn): Likewise.
853 * cselib.c (cselib_invalidate_mem): Likewise.
854 * df.c: Update comments at the top of the file.
855 (read_modify_subreg_p): Update comments here too.
856 (df_def_record_1): Don't expect a SIGN_EXTRACT in a SET_DEST.
857 * flow.c (mark_set_1): Likewise.
858 (mark_used_regs): Likewise.
859 * gcse.c (mems_conflict_for_gcse_p): Likewise.
860 (canon_list_insert): Likewise.
861 (mark_set): Likewise.
862 (try_replace_reg): Likewise.
863 (store_killed_in_insn): Likewise.
864 * loop.c (count_one_set): Likewise.
865 (basic_induction_var): Likewise.
866 * postreload-gcse.c (find_mem_conflicts): Likewise.
867 * postreload.c (reload_combine_note_store): Likewise.
868 (move2add_note_store): Likewise.
869 * reload.c (find_equiv_reg): Likewise.
870 (mark_referenced_resources): Likewise.
871 * rtlanal.c (set_noop_p): Likewise.
872 (note_stores): Likewise.
873 (note_uses): Likewise.
874 * sched-deps.c (sched_analyze_1): Likewise.
875 * sched-rgn.c (check_live_1): Likewise.
876 (update_live_1): Likewise.
877 * config/i860/i860.c: Likewise.
879 * rtl.dec (SIGN_EXTRACT): Document that this cannot appear as
881 (ZERO_EXTRACT): Mention that this one can be an lvalue.
883 * doc/rtl.texi: Update documentation for bit-fields and SET.
885 2005-01-05 Roger Sayle <roger@eyesopen.com>
887 * ifcvt.c (find_if_case_1): Revert 2005-01-04 change.
889 2005-01-05 Jan Hubicka <jh@suse.cz>
890 Richard Henderson <rth@redhat.com>
893 * config/i386/i386.c (ix86_expand_move): Handle tls symbols
896 2005-01-05 Richard Henderson <rth@redhat.com>
899 * config/i386/i386.md (sse_movhps, sse_movlps): Remove.
900 (sse_shufps): Change operand 3 to const_int_operand.
901 (sse2_storelps): Fix typo in template.
902 (sse_storehps, sse_loadhps, sse_storelps, sse_loadlps): New.
903 * config/i386/i386.c (ix86_expand_vector_move_misalign): Use them.
904 (ix86_expand_builtin): Likewise.
906 2005-01-05 Julian Brown <julian@codesourcery.com>
908 * config/arm/arm.c (arm_return_in_memory): Treat complex types
909 as aggregates for AAPCS ABIs.
911 2005-01-05 Stan Shebs <shebs@apple.com>
913 * unwind-dw2-fde-darwin.c (examine_objects): Use 64-bit
914 Mach-O getters if ppc64.
915 * config/darwin.c (darwin_asm_output_dwarf_delta): Obey
917 * config/darwin.h (ASM_PREFERRED_EH_DATA_FORMAT): Use
918 signed four-byte field for global code case.
919 (STARTFILE_SPEC): Avoid crt2.o for 64-bit compilation.
921 2005-01-05 Roger Sayle <roger@eyesopen.com>
924 * c-common.c: Include real.h.
925 (c_common_truthvalue_conversion): Avoid destructively modifying expr.
926 Correctly handle TREE_CONSTANT_OVERFLOW for INTEGER_CST.
927 Correctly handle TREE_CONSTANT_OVERFLOW and NaNs for REAL_CST.
928 * Makefile.in (c-common.o): Update dependencies.
930 2005-01-05 Joseph S. Myers <joseph@codesourcery.com>
932 * c-parse.in (asm_string): Add trailing semicolon.
934 2005-01-05 Joseph S. Myers <joseph@codesourcery.com>
936 * c-parse.in (asm_string): New. Don't allow wide strings in
938 (simple_asm_expr, asm_argument, asm_operand, asm_clobbers): Use
939 asm_string instead of STRING.
941 2005-01-05 Joseph S. Myers <joseph@codesourcery.com>
943 * c-typeck.c (constructor_no_implicit): Remove.
944 (set_designator, process_init_element): Don't check
945 constructor_no_implicit.
947 2005-01-05 J"orn Rennecke <joern.rennecke@st.com>
948 Kaz Kojima <kkojima@gcc.gnu.org>
951 * lcm.c (create_pre_exit): New.
952 (optimize_mode_switching): In MODE_ENTRY / MODE_EXIT case, set
953 ENTRY_EXIT_EXTRA to 3. Use create_pre_exit.
955 2004-01-05 Richard Earnshaw <rearnsha@arm.com>
957 * arm.h (TARGET_OPTIONS): Correctly record -mhard-float and
958 -msoft-float in target_float_switch.
959 * arm.c (arm_override_options): Fix processing of target_float_switch.
961 2004-01-05 Richard Earnshaw <rearnsha@arm.com>
963 * arm/vfp.md (arm_movsi_vfp): Hide VFP register classes from register
966 2004-01-05 Uros Bizjak <uros@kss-loka.si>
968 * doc/invoke.texi (Intel 386 and AMD x86-64 Options):
969 Replace i387 with 'i386 compiler' in -mfpmath=sse option.
971 2005-01-04 Roger Sayle <roger@eyesopen.com>
973 * ifcvt.c (find_if_case_1): Avoid creating an empty forwarder block,
974 if deleting the then-block allows the test-block to fallthru to the
977 2005-01-04 Andrew Pinski <pinskia@physics.uc.edu>
980 * c-decl.c (diagnose_mismatched_decls): Accept "extern inline" declared
981 after the full declaration if the are in two different TUs.
983 2005-01-04 Richard Henderson <rth@redhat.com>
986 * tree-sra.c (generate_one_element_init): Just
987 call gimplify_and_add.
988 (generate_element_init): Record the
989 new referenced variables and mark them for renaming
991 (generate_element_init_1): This.
992 (scalarize_init): Don't call push_gimplify_context/
993 pop_gimplify_context.
995 2005-01-04 Geoffrey Keating <geoffk@apple.com>
997 * toplev.c (get_src_pwd): Handle failure of getpwd().
999 2005-01-04 Roger Sayle <roger@eyesopen.com>
1001 * fold-const.c (fold_single_bit_test): Delete unreachable handling
1003 (fold): Don't call fold_single_bit_test with a TRUTH_NOT_EXPR, as
1004 all the cases handled by it are inverted by invert_truthvalue.
1006 2005-01-04 Eric Botcazou <ebotcazou@libertysurf.fr>
1008 * config/sparc/sparc.md (save_register_windowdi): Add missing mode.
1009 (save_register_windowsi): Likewise.
1011 2005-01-04 Richard Henderson <rth@redhat.com>
1013 * tree-mudflap.c (mf_varname_tree): Fix thinko setting declname.
1015 2005-01-04 Uros Bizjak <uros@kss-loka.si>
1018 * cse.c (fold_rtx) [RTX_COMPARE, RTX_COMM_COMPARE]: Don't attempt
1019 any simplifications of vector mode comparison operators.
1020 * simplify-rtx.c (simplify_relational_operation): Fix variable name.
1022 2005-01-04 Paolo Bonzini <bonzini@gnu.org>
1023 Devang Patel <dpatel@apple.com>
1025 PR tree-optimization/18308
1026 * tree-if-conv.c (add_to_dst_predicate_list): Gimplify
1027 the operands before creating a new expression.
1028 * dojump.c (do_jump): Make drop_through_label available
1029 for all cases. Add expansion of COND_EXPR.
1031 2005-01-04 Ira Rosen <irar@il.ibm.com>
1033 * tree-vectorizer.c (vect_analyze_offset_expr): Test for
1034 INTEGER_CST instead of TREE_CONSTANT.
1035 (vect_gen_niters_for_prolog_loop): Test for INTEGER_CST
1036 instead of TREE_CONSTANT.
1037 (vect_analyze_pointer_ref_access): Test for INTEGER_CST
1038 instead of TREE_CONSTANT.
1040 2005-01-04 Eric Botcazou <ebotcazou@libertysurf.fr>
1042 * tree-eh.c (replace_goto_queue): Return early if the queue is empty.
1044 2005-01-04 Uros Bizjak <uros@kss-loka.si>
1047 * config/i386/i386.md (*fop_df_1_i387): Disable for TARGET_SSE_MATH.
1048 (*fop_df_1_i387): Disable for (TARGET_SSE2 && TARGET_SSE_MATH).
1050 2005-01-03 Richard Henderson <rth@redhat.com>
1052 * fold-const.c (force_fit_type): Cope with types larger than 2 HWI.
1053 (fold_convert_const_int_from_int, fold_convert_const_int_from_real,
1054 fold_convert_const_real_from_real): Split out from ...
1055 (fold_convert_const): ... here.
1057 2005-01-03 Richard Henderson <rth@redhat.com>
1060 * config/i386/i386.md (movdi_2): Separate SSE1 and SSE2 alternatives.
1061 (mov<MMXMODEI>_internal): Likewise.
1062 (movdf_nointeger): Prefer Y while not preferring, but allowing, x.
1063 Add V2SF case; use it for SSE1; don't use TI.
1064 (movdf_integer): Likewise.
1065 (mov<SSEMODEI>_internal, movti_internal): Force V4SF for SSE1.
1067 2005-01-03 Ira Rosen <irar@il.ibm.com>
1069 * tree-vectorizer.c (vect_strip_conversions): New function.
1070 (vect_analyze_offset_expr): Call vect_strip_conversions. Add
1071 check for binary class.
1073 2005-01-03 Daniel Berlin <dberlin@dberlin.org>
1077 * dwarf2out.c (block_ultimate_origin): Follow decl origin if origin
1079 * gimple-low.c (mark_blocks_with_used_vars): New function.
1080 (mark_blocks_with_used_subblocks): Ditto.
1081 (mark_used_blocks): Ditto.
1082 (pass_mark_used_blocks): New pass.
1083 * tree-inline.c: Include debug.h.
1084 (expand_call_inline): Call outlining_inline_function here.
1085 * tree-optimize.c (init_tree_optimization_passes): Add
1086 pass_mark_used_blocks.
1087 * tree-pass.h (pass_mark_used_blocks): New.
1088 * Makefile.in (tree-inline.o): Add debug.h dependency.
1090 2005-01-03 Geoffrey Keating <geoffk@apple.com>
1092 * config/darwin.c (darwin_handle_weak_import_attribute): Permit
1093 VAR_DECLs to have weak_import attribute.
1095 * config/rs6000/darwin-fallback.c: Use 'ucontext_t' rather than
1098 2004-01-03 Steven Bosscher <stevenb@suse.de>
1100 * passes.c (rest_of_compilation): Don't run regmove if only
1101 flag_expensive_optimizations, require flag_regmove instead.
1103 2005-01-03 Eric Botcazou <ebotcazou@libertysurf.fr>
1105 * config/sparc/litecoff.h (TARGET_OS_CPP_BUILTINS): Do not
1107 * config/sparc/sysv4-only.h (TARGET_OS_CPP_BUILTINS): Likewise.
1108 * config/sparc/rtemself.h (TARGET_SUB_OS_CPP_BUILTINS): Likewise.
1109 * config/sparc/sol2-64.h (TARGET_SUB_OS_CPP_BUILTINS): Delete.
1110 * config/sparc/sp64-elf.h (TARGET_SUB_OS_CPP_BUILTINS): Likewise.
1112 2005-01-03 Richard Henderson <rth@redhat.com>
1114 * config/ia64/ia64.c (TARGET_VECTOR_MODE_SUPPORTED_P): New.
1115 (ia64_const_ok_for_letter_p): New.
1116 (ia64_const_double_ok_for_letter_p): New.
1117 (ia64_extra_constraint): New.
1118 (ia64_expand_vecint_compare): New.
1119 (ia64_expand_vcondu_v2si): New.
1120 (ia64_expand_vecint_cmov): New.
1121 (ia64_expand_vecint_minmax): New.
1122 (ia64_print_operand): Add 'v'.
1123 (ia64_preferred_reload_class): New.
1124 (ia64_vector_mode_supported_p): New.
1125 * config/ia64/ia64.h (UNITS_PER_SIMD_WORD): New.
1126 (PREFERRED_RELOAD_CLASS): Move to function.
1127 (CONST_OK_FOR_LETTER_P): Move to function.
1128 (CONST_DOUBLE_OK_FOR_LETTER_P): Move to function.
1129 (CONSTRAINT_OK_FOR_Q, CONSTRAINT_OK_FOR_R): Remove.
1130 (CONSTRAINT_OK_FOR_S, CONSTRAINT_OK_FOR_T): Remove.
1131 (EXTRA_CONSTRAINT): Move to function.
1132 * config/ia64/ia64.md: Include vect.md.
1133 (itanium_class): Add mmalua.
1135 * config/ia64/itanium1.md (1_mmalua): New. Add it to bypasses.
1137 * config/ia64/itanium2.md (2_mmalua, 2b_mmalua): Similarly.
1138 * config/ia64/predicates.md (gr_reg_or_0_operand): Accept any
1140 (const_int_2bit_operand): New.
1141 (fr_reg_or_0_operand): New.
1142 * config/ia64/ia64-modes.def: Add vector modes.
1143 * config/ia64/ia64-protos.h: Update.
1144 * config/ia64/vect.md: New file.
1146 2005-01-03 Richard Henderson <rth@redhat.com>
1148 * simplify-rtx.c (simplify_binary_operation): Handle VEC_CONCAT.
1150 2005-01-03 Uros Bizjak <uros@kss-loka.si>
1153 * config/i386/i386.c (log1psf2): Change mode of operands[0,1]
1155 (log1pdf2): Change mode of operands[0,1] to DFmode.
1157 2005-01-03 Eric Botcazou <ebotcazou@libertysurf.fr>
1159 * config/sparc/sparc.h (SPARC_RELAXED_ORDERING): Define to false.
1160 * config/sparc/linux.h (SPARC_RELAXED_ORDERING): Define to true.
1161 * config/sparc/linux64.h (SPARC_RELAXED_ORDERING): Likewise.
1162 * config/sparc/sparc.c (TARGET_RELAXED_ORDERING): Define to
1163 SPARC_RELAXED_ORDERING.
1165 2005-01-03 Richard Henderson <rth@redhat.com>
1166 Uros Bizjak <uros@kss-loka.si>
1169 * config/i386/i386.c (ix86_expand_builtin): [IX86_BUILTIN_PINSRW,
1170 IX86_BUILTIN_PINSRW128]: Fix wrong selector range in error message.
1171 * config/i386/i386.md (mmx_pinsrw, sse2_pinsrw): Fix selector
1173 (*mmx_pinsrw, *sse2_pinsrw): New patterns.
1174 * config/i386/i386/predicates.md (const_pow2_1_to_8_operand,
1175 const_pow2_1_to_128_operand): New predicates.
1177 2005-01-02 Greg McGary <greg@mcgary.org>
1179 * tree-mudflap.c (mf_varname_tree): decl_printable_name handles
1181 (mf_xform_derefs_1) [RESULT_DECL, STRING_CST]: Handle as innermost
1182 object. [ptr_type]: Remove unused variable.
1183 Remember nearest addressable array-element or record-component
1184 when checking bitfield components. Tolerate empty BIND_EXPR.
1186 2005-01-01 Richard Henderson <rth@redhat.com>
1188 * tree-vectorizer.c (vect_analyze_offset_expr): Strip conversions
1189 that don't narrow the value. Fail for other conversions.
1191 2005-01-01 Richard Henderson <rth@redhat.com>
1194 * c-decl.c (pop_file_scope): Call maybe_apply_pending_pragma_weaks.
1195 * c-lang.c (finish_file): Don't do it here.
1196 * objc/objc-act.c (objc_finish_file): Likewise.
1198 * cgraph.c (decl_assembler_name_equal): New.
1199 (cgraph_node_for_asm, cgraph_varpool_node_for_asm): New.
1200 (cgraph_varpool_node): Actually link up cgraph_varpool_nodes.
1201 * cgraph.h (struct cgraph_varpool_node): Add next.
1202 (cgraph_node_for_asm, cgraph_varpool_node_for_asm): Declare.
1203 * varasm.c (assemble_alias): Mark the target as needed.
1205 2005-01-01 Andrew Pinski <pinskia@physics.uc.edu>
1208 * function.c (get_arg_pointer_save_area): Use entry_of_function
1209 instead of get_insns.
1211 2005-01-01 Roger Sayle <roger@eyesopen.com>
1212 Andrew Pinski <pinskia@physics.uc.edu>
1213 James E. Wilson <wilson@specifixinc.com>
1215 PR rtl-optimization/12092
1216 * loop.c (emit_prefetch_instructions): Do nothing if PREFETCH_BLOCK
1219 2005-01-01 Roger Sayle <roger@eyesopen.com>
1220 Olivier Hainque <hainque@act-europe.fr>
1222 * tree.c (int_fits_type_p): A narrower type always fits in a
1223 wider one, except for negative values into unsigned types.
1225 2005-01-01 Roger Sayle <roger@eyesopen.com>
1227 * tree.c (int_fits_type_p): Compare the result of force_fit_type
1228 with the original constant rather than require TREE_OVERFLOW.
1230 2005-01-01 Steven Bosscher <stevenb@suse.de>
1233 * c-decl.c (finish_function): If compiling C99, annotate the
1234 compiler generated return with the current file name and line 0.
1235 * tree-cfg.c (remove_useless_stmts_warn_notreached): Only warn if
1236 the source line is greater than 0.
1237 (remove_bb): Likewise.
1239 See ChangeLog.12 for earlier changes.