(remove_cycle): Instead of physically removing call_info structures
to break call graph cycles, mark them using the broken_cycle flag.
(mark_overlay_section): Respect broken_cycle flag.
(unmark_overlay_section): Likewise.
(collect_lib_sections): Likewise.
(collect_overlays): Likewise.
(sum_stack): Likewise.
(mark_functions_via_relocs): Initialize callee->count to 1.
(pasted_function): Likewise.
(spu_elf_auto_overlay): Honor call counts when determining number
of stubs required in software i-cache mode.
* elf32-spu.c (spu_elf_find_overlays): Separate error return from
"no overlays" return. If there are overlays, create overlay
manager entry symbols here, so that..
(spu_elf_build_stubs): ..we don't need to set them up here.
Simplify entry symbol tests.
* elf32-spu.h (spu_elf_find_overlays): Update prototype.
ld/
* emultempl/spuelf.em (spu_before_allocation): Report errors from
spu_elf_find_overlays.
* elf32-spu.h (struct spu_elf_params): ovly_flavour now only 1 bit.
Add compact_stub.
(emum _ovly_flavour): Delete ovly_compact, ovly_none.
* elf32-spu.c (struct spu_link_hash_table): Replace ovly_load and
ovly_return fields with ovly_entry[2]. Adjust all users.
(spu_elf_find_overlays): Set ovly_entry[1] from __icache_call_handler
when soft-icache.
(spu_elf_build_stubs): Likewise.
(ovl_stub_size): Change arg to spu_elf_params pointer. Adjust for
ovly_flavour changes. Update all callers.
(ovl_stub_size_log2): New function.
(build_stub): Handle compact icache stubs. Use different manager
entry point for stubs in non-icache area.
(spu_elf_size_stubs): Don't allocate space for indirect branch
descriptors.
(spu_elf_build_stubs): And don't built them.
ld/
* emultempl/spu_icache.S: Add new entry to dummy handler.
* emultempl/spu_icache.o_c: Regenerate.
* emultempl/spuelf.em (params): Init new field.
(no_overlays): New static var.
(spu_before_allocation): Use it.
(OPTION_SPU_COMPACT_STUBS): Define.
(PARSE_AND_LIST_LONGOPTS, PARSE_AND_LIST_OPTIONS): Add compact-stubs.
(PARSE_AND_LIST_ARGS_CASES): Handle compact-stubs. Adjust no-overlays
handling.
ld/testsuite/
* ld-spu/icache1.d: Update for fixed set_id, new manager entry, and
reduced data.
parameter just because fixed sections fit into local store.
(spu_elf_auto_overlay): Do not declare as "noreturn". Skip
generating overlays if fixed sections plus reserved stack
and heap space fit into local store.
(find_function): Likewise.
(pasted_function): Don't error if no prior function found.
(discover_functions): Revert 2008-12-10 change. Extend first
function range to start of section.
* elf32-spu.c (struct spu_link_hash_table): Add init, line_size_log2,
num_lines_log2.
(struct got_entry): Add br_addr.
(struct call_info): Add priority.
(struct function_info): Add lr_store and sp_adjust.
(spu_elf_setup): Init line_size_log2 and num_lines_log2.
(spu_elf_find_overlays): For soft-icache, mark any section within cache
area as an overlay, and check that no other overlays exist. Look up
icache overlay manager entry sym.
(BRA_STUBS, BRA, BRASL): Define.
(enum _stub_type): Replace ovl_stub with call_ovl_stub and br*_ovl_stub.
(needs_ovl_stub): Adjust for soft-icache. Return priority encoded
in branch insn.
(count_stub, build_stub): Support soft-icache.
(build_spuear_stubs, process_stubs): Adjust build_stub call.
(spu_elf_size_stubs): Size soft-icache stubs.
(overlay_index): New function.
(spu_elf_build_stubs): Make static. Support soft-icache.
(spu_elf_check_vma): Don't turn off auto_overlay if soft-icache.
(find_function_stack_adjust): Save lr store and stack adjust insn
offsets.
(maybe_insert_function): Adjust find_function_stack_adjust call.
(mark_functions_via_relocs): Retrieve priority.
(remove_cycles): Only warn about pruned arcs when stack_analysis.
(sort_calls): Sort by priority first.
(mark_overlay_section): Ignore .ovl.init.
(sum_stack): Only print when stack_analysis.
(print_one_overlay_section): New function, extracted from..
(spu_elf_auto_overlay): ..here. Support soft-icache overlays.
(spu_elf_stack_analysis): Only print when htab->stack_analysis.
(spu_elf_final_link): Call spu_elf_stack_analysis for lrlive
analysis. Call spu_elf_build_stubs.
(spu_elf_relocate_section): For soft-icache encode overlay index
into addresses.
(spu_elf_output_symbol_hook): Support soft-icache.
(spu_elf_modify_program_headers: Likewise.
* elf32-spu.h (struct spu_elf_params): Add lrlive_analysis. Rename
num_regions to num_lines. Add line_size and max_branch.
(enum _ovly_flavour): Add ovly_soft_icache.
(spu_elf_build_stubs): Delete.
gas/
* config/tc-spu.c (md_pseudo_table): Add "brinfo".
(brinfo): New var.
(md_assemble): Poke brinfo into branch instructions.
(spu_brinfo): New function.
(md_apply_fix): Don't assume insn fields start off at zero, mask
them to remove possible brinfo.
ld/
* emultempl/spuelf.em (params): Init new fields.
(num_lines_set, line_size_set, icache_mgr, icache_mgr_stream): New vars.
(spu_place_special_section): Adjust placement for soft-icache. Pad
soft-icache section to a fixed size. Clear addr_tree.
(spu_elf_load_ovl_mgr): Support soft-icache. Map overlay manager
sections a little more intelligently.
(gld${EMULATION_NAME}_finish): Don't call spu_elf_build_stubs.
(OPTION_SPU_NUM_LINES): Rename from OPTION_SPU_NUM_REGIONS.
(OPTION_SPU_SOFT_ICACHE, OPTION_SPU_LINE_SIZE): Define.
(OPTION_SPU_LRLIVE): Define.
(PARSE_AND_LIST_LONGOPTS): Add new soft-icache options.
(PARSE_AND_LIST_OPTIONS): Likewise.
(PARSE_AND_LIST_ARGS_CASES): Handle them.
* emultempl/spu_icache.S: Dummy file.
* emultempl/spu_icache.o_c: Regenerate.
* Makefile.am (eelf32_spu.c): Depend on spu_icache.o_c.
(spu_icache.o_c): Add rule to build.
(CLEANFILES): Zap temp files.
(EXTRA_DIST): Add spu_icache.o_c.
* Makefile.in: Regenerate.
ld/testsuite/
* ld-spu/ovl.d: Allow for absolute branches in stubs.
* ld-spu/ovl2.d: Likewise.
* elf32-spu.g (struct spu_elf_params, enum _ovly_flavour): New.
(spu_elf_setup): Declare.
(spu_elf_create_sections, spu_elf_size_stubs): Update prototype.
(spu_elf_build_stubs, spu_elf_check_vma): Likewise.
* elf32-spu.c (struct spu_link_hash_table): Add "params". Remove
various other fields now in "params". Adjust code throughout.
(struct call_info, struct function_info): Move earlier in file.
(struct spu_elf_stack_info): Likewise.
(spu_elf_setup): New function.
(spu_elf_create_sections): Remove args other than "info".
(spu_elf_size_stubs, spu_elf_build_stubs, spu_elf_check_vma): Likewise.
(maybe_needs_stubs): Remove "output_bfd" arg. Adjust all calls.
(interesting_section): Similarly with "obfd" arg.
(needs_ovl_stub): Adjust output_section test.
(allocate_spuear_stubs): Likewise.
(OVL_STUB_SIZE): Don't define.
(ovl_stub_size): New function, use in place of OVL_STUB_SIZE.
(build_stub): Test params->ovly_flavour rather than OVL_STUB_SIZE.
(spu_elf_auto_overlay): Remove args other than "info". Make use
of size returned from spu_elf_load_ovl_mgr.
(spu_elf_stack_analysis): Remove args other than "info".
(spu_elf_relocate_section): Tidy setting of "ea".
ld/
* emultempl/spuelf.em (params): New var, used instead of various others.
Adjust use throughout file.
(spu_after_open): Call spu_elf_setup.
(spu_place_special_section): Tidy.
(spu_elf_load_ovl_mgr): Return total size of sections loaded. Move
code setting overlay section alignment to..
(spu_before_allocation): ..here.
defined in linker scripts.
(discover_functions): Consider STT_SECTION symbols too.
(collect_lib_sections): Don't cut short call tree traversal
when function size is too large.
of insns scanned. Correct sp tests. Handle "fsmbi" and "andbi".
(mark_detached_root): New function.
(build_call_tree): Call it.
(sort_calls): Don't do void* arithmetic.
(allocate_spuear_stubs, build_spuear_stubs): Return value from
count_stub/build_stub.
(spu_elf_build_stubs): Correct location of stub reloc error message.
(process_stubs): Likewise.
(discover_functions): Don't used cached symbols.
(maybe_insert_function): Correct condition under which function
array is realloc'd.
(mark_functions_via_relocs): Delete unused variable.
* elf32-spu.c (spu_elf_special_sections): Add "._ea".
(spu_elf_relocate_section): Handle relocations against symbols
defined in ._ea specially.
binutils/
* embedspu.sh: Take note of R_SPU_PPU32/64 relocs without a symbol,
and if present, put image in ".data.speelf". Put program handle
in ".data.spehandle".
ld/emulparams/
* elf32_spu.sh (OTHER_SECTIONS): Add "._ea".
* elf32ppc.sh: If building with spu support, put ".data.spehandle"
sections at the start of ".data" and provide a symbol to locate
the directory of embedded spe programs.
ld/testsuite/
* ld-spu/ear.s: Align various sections.
* ld-spu/embed.rd: Update.
* elf32-spu.c: Include libiberty.h.
(struct spu_link_hash_table): Add local_stire, overlay_fixed, reserved,
non_ovly_stub, spu_elf_load_ovl_mgr, spu_elf_open_overlay_script,
spu_elf_relink, auto_overlay fields.
(AUTO_OVERLAY, AUTO_RELINK, OVERLAY_RODATA): Define.
(needs_ovl_stub): Flip test so that call to non-function warning
is emitted during relocate_section rather than earlier.
(spu_elf_check_vma): Stash --auto-overlay parameters, and clear
auto_overlay if no section exceeds local store.
(struct call_info): Add count, max_depth, is_pasted fields.
(struct function_info): Add rodata, last_caller, call_count,
depth, new visit flags.
(insert_callee): Increment call count.
(copy_callee): New function.
(mark_functions_via_relocs): Investigate all reloc types to count
possible function pointer stubs for --auto-overlay. Track
last_caller and increment function call_count.
(pasted_function): Insert a "call" into call info for pasted section.
(remove_cycles): Track max depth of calls. Don't emit call graph
pruning warning for --auto-overlay.
(build_call_tree): Don't transfer_calls for --auto-overlay.
Adjust remove_cycles call.
(sort_calls, sort_lib, sort_bfds): New functions.
(struct _mos_param, struct _uos_param, struct _cl_param): New.
(mark_overlay_section, unmark_overlay_section): New functions.
(collect_lib_sectios, auto_ovl_lib_functions): New functions.
(collect_overlays, find_pasted_call): New functions.
(sum_stack): Deal with is_pasted "calls". Exit before printing
when --auto-overlay.
(spu_elf_auto_overlay): New function.
(spu_elf_final_link): Call spu_elf_auto_overlay.
* elf32-spu.h (spu_elf_check_vma): Update prototype.
ld/
* emultempl/spuelf.em (auto_overlay, auto_overlay_file,
auto_overlay_fixed, auto_overlay_reserved, my_argc, my_argv): New vars.
(spu_before_allocation): Warn on --auto-overlay and existing overlays.
(struct tflist, clean_tmp): Move.
(new_tmp_file): New function, extracted from..
(embedded_spu_file): ..here.
(spu_elf_open_overlay_script, spu_elf_relink): New function.
(gld${EMULATION_NAME}_finish): Pass a bunch of --auto-overlay params.
Warn on --auto-overlay and zero local store.
(gld${EMULATION_NAME}_choose_target): New function to stash argv.
(OPTION_SPU_AUTO_OVERLAY, OPTION_SPU_AUTO_RELINK,
OPTION_SPU_OVERLAY_RODATA, OPTION_SPU_FIXED_SPACE,
OPTION_SPU_RESERVED_SPACE, OPTION_SPU_NO_AUTO_OVERLAY): Define.
(PARSE_AND_LIST_LONGOPTS): Add entries for new options.
(PARSE_AND_LIST_OPTIONS): Likewise.
(PARSE_AND_LIST_ARGS_CASES): Likewise.
(LDEMUL_CHOOSE_TARGET): Define.
call is always first.
(interesting_section): Move.
(mark_functions_via_relocs): Fold interesting_section and
reloc_count tests in callers to here. Simplify output section
owner test.
(discover_functions): Set "gaps" when no symbols and some
"interesting_section". Run pasted_function loop for no symbol
bfds.
(for_each_node, transfer_calls): New functions.
(mark_non_root): Adjust to suit for_each_node.
(call_graph_traverse): Likewise. Fix memory leak. Rename to..
(remove_cycles): ..this.
(build_call_tree): Use for_each_node and transfer_calls.
(struct _sum_stack_param): New.
(sum_stack): Adjust to suit for_each_node. Return error on
malloc failure. Move code to print root node cumulative stack from..
(spu_elf_stack_analysis): ..here. Use for_each_node.
common code to..
(maybe_needs_stub): ..here, a new function that also omits stubs
for .eh_frame, and..
(needs_ovl_stub): ..here. Create stubs for labels in code section
referenced by switch jump table.
(spu_elf_find_overlays): Set htab->ovly_load and htab->ovly_return.
(enum _insn_type): Delete.
(enum _stub_type): New.
(count_stub, build_stub): Adjust.
(allocate_spuear_stubs, build_spuear_stubs): Adjust.
Ulrich Weigand
Alan Modra