OSDN Git Service

2004-11-22 David Edelsohn <edelsohn@gnu.org>
[pf3gnuchains/gcc-fork.git] / gcc / ChangeLog
index 6b8a217..f46c34b 100644 (file)
@@ -1,3 +1,129 @@
+2004-11-22  David Edelsohn  <edelsohn@gnu.org>
+           Fariborz Jahanian  <fjahanian@apple.com>
+
+       * config/rs6000/rs6000.c (rs6000_return_in_memory): Test
+       TARGET_ALTIVEC_ABI.
+       (function_arg_boundary): Align 8 to 16 byte vectors like SPE.
+       Align 16 byte and larger vectors like Altivec.
+       (function_arg_advance): Pass any 16 byte vector like Altivec vector.
+       (function_arg): Same.
+       (rs6000_pass_by_references): Test TARGET_ALTIVEC_ABI.
+
+2004-11-22  Andrew Pinski  <pinskia@physics.uc.edu>
+
+       PR tree-opt/18572
+       * tree-sra.c (scalarize_init): Unshare the rhs before gimplifying
+       it.
+
+2004-11-22  Richard Henderson  <rth@redhat.com>
+
+       * function.c (assign_parm_setup_block): Revert 11-13 change to
+       forcably re-mode a (possibly BLKmode) hard register.
+
+2004-11-22  Eric Botcazou  <ebotcazou@libertysurf.fr>
+
+       * config/sparc/sparc.c (mem_min_alignment): Check MEM_ALIGN.
+
+2004-11-22  David Edelsohn  <edelsohn@gnu.org>
+
+       * config/rs6000/power4.md: Remove delay between dispatch and issue
+       associated with dispatch group slots 3 and 4.
+       * config/rs6000/power5.md: Same.
+
+2004-11-21  Jeff Law  <law@redhat.com>
+
+       * cfg.c (cached_make_edge): Use find_edge rather than an inlined
+       variant.
+       * cfgbuild.c (make_edges): Likewise.
+       * cfghooks.c (can_duplicate_block_p): Likewise.
+       * cfgloop.c (loop_latch_edge): Likewise.
+       * cfgloopmanip.c (force_single_succ_latches): Likewise.
+       * cfgrtl.c (rtl_flow_call_edges_add): Likewise.
+       * predict.c (predict_loops, propagate_freq): Likewise.
+       * tracer.c (tail_duplicate): Likewise.
+       * tree-cfg.c (disband_implicit_edges): Likewise.
+       (tree_forwarder_block_p, tree_flow_call_edges_add): Likewise.
+
+2004-11-22  Nick Clifton  <nickc@redhat.com>
+
+       * sbitmap.c (sbitmap_union_of_preds): Remove redundant
+       initialisation of 'e'.
+
+2004-11-22  Kazu Hirata  <kazu@cs.umass.edu>
+
+       PR rtl-optimization/18599
+       * regrename.c (copyprop_hardreg_forward): Speed up by putting
+       BB_VISITED flags on basic blocks as we process them.
+
+2004-11-22  Nathan Sidwell  <nathan@codesourcery.com>
+
+       * config/rs6000/altivec.md (VI_char): New mode attribute.
+       (addv16qi3, addv8hi3, addv4ai3): Replace with ...
+       (add<mode>3): ... this.
+       (subv16qi3, subv8hi3, subv4ai3): Replace with ...
+       (sub<mode>3): ... this.
+       (smaxv16qi3, smaxv8hi3, smaxv4ai3): Replace with ...
+       (smax<mode>3): ... this.
+       (sminv16qi3, sminv8hi3, sminv4ai3): Replace with ...
+       (smin<mode>3): ... this.
+       (umaxv16qi3, umaxv8hi3, umaxv4ai3): Replace with ...
+       (umax<mode>3): ... this.
+       (uminv16qi3, uminv8hi3, uminv4ai3): Replace with ...
+       (umin<mode>3): ... this.
+       (andv16qi3, andv8hi3, andv4ai3): Replace with ...
+       (and<mode>3): ... this.
+       (iorv16qi3, iorv8hi3, iorv4ai3): Replace with ...
+       (ior<mode>3): ... this.
+       (xorv16qi3, xorv8hi3, xorv4ai3): Replace with ...
+       (xor<mode>3): ... this.
+       (andv16qi3, andv8hi3, andv4ai3): Replace with ...
+       (and<mode>3): ... this.
+       (iorv16qi3, iorv8hi3, iorv4ai3): Replace with ...
+       (ior<mode>3): ... this.
+       (altivec_vnor): Replace with ...
+       (altivec_nor<mode>): ... this.
+       (one_cmplv16qi2, one_cmplv8hi2, one_complv4ai2): Replace with ...
+       (one_cmpl<mode>2): ... this.
+       (altivec_vandc): New expander.
+       (*andc<mode>3): New insn.
+       * config/rs6000/rs6000.c (bdesc_2arg): Adjust for new insn names.
+
+2004-11-22  Dorit Naishlos  <dorit@il.ibm.com>
+
+       PR tree-opt/18536
+       * tree-vectorizer.c (make_loop_iterate_ntimes): Use type of niters when
+       creating iv instead of integer_type.
+       (vect_build_loop_niters): Remove redundant code.
+       (vect_transform_loop_bound): Removed (duplicates functionality of
+       make_loop_iterate_ntimes.
+       (vect_gen_niters_for_prolog_loop): Add documentation. Call
+       lang_hooks.types.type_for_size to create a type of the required size.
+       Use that type instead of integer_type. Remove redundant code.
+       (vect_update_niters_after_peeling): Use type of niters instead of
+       integer_type.
+       (vect_transform_loop): Unify handling of known and unknown loop bound
+       cases. Call make_loop_iterate_ntimes instead of 
+       vect_transform_loop_bound.
+
+2004-11-22  Andrew Pinski  <pinskia@physics.uc.edu>
+
+       PR target/18217
+       * config/rs6000/t-darwin (LIB2FUNCS_EXTRA): Move darwin-fallback.c to
+       (LIB2ADDEH): Here.
+
+2004-11-22  Dorit Naishlos  <dorit@il.ibm.com>
+
+       PR tree-opt/18544
+       * tree-vectorizer.c (vect_analyze_data_refs): Fail if memtag not found.
+
+2004-11-22 Kazuhiro Inaoka <inaoka.kazuhiro@renesas.com>
+
+       * config.gcc: Use t-slibgcc-elf to build libgcc_s.so on
+        m32r*linux.
+       * config/m32r/t-linux (SHLIB_MAPFILES): Override to use
+       m32r/libgcc-glibc.ver.
+       * config/m32r/libgcc-glibc.ver: Add New file.
+
 2004-11-22  Nathan Sidwell  <nathan@codesourcery.com>
 
        * bitmap.h (struct bitmap_obstack): New obstack type.