changelog.00

这个是LINUX下的GDB调度工具的源码

	putting the values into local variables.
	(first_undef, last_undef): New structures to remember names to
	#undef if -R.
	(lf_print_c_extraction): If -R and this is semantics, emit names
	as #defines pointing to the cache structure, rather than loading
	the values into local variables.
	(lf_print_c_semantic_function): If -R, #undef all of the names
	defined in lf_print_c_extraction.
	(main): Recognize -R.

	* idecode_fields.h (SPR_*): Redefine spr_* macros as SPR_* to
	avoid a name confict if -R passed to igen.

	* ppc-instructions (mfspr, mtspr): Rename spr field to SPR.
	(model_data): Add field to count the various # of CRs that the
	mtcrf instruction used.
	(model_mon_info): Return structures counting the # of CRs that the
	mtcrf instruction used.
	(branches, sync instructions): Do not call model functions if
	WITH_MODEL_ISSUE is 0.

	* mon.c (stdlib.h): Include if the system supplies one.
 	(mon_sort_instruction_names): New function to sort instruction
	names alphabetically.
	(mon_print_info): Call qsort with mon_sort_instruction_names to
	sort instruction names.  Don't abort if WITH_MODEL_ISSUE is 0.
	
	* debug.h (ITRACE): Make printf_filtered arguments type correct.
	* idecode_expression.h (CR0_COMPARE): Ditto.
	* psim.c (psim_read_register): Ditto.

	* igen.c (lf_print_my_prefix): Use __attribute__((__unused__)) to
	silence compiler warnings about unused automatically generated
	variables.
	(lf_print_c_extraction): Ditto.
	* idecode_expression.h (FPSCR_BEGIN): Ditto.

	* ppc-cache-rules: Define rules for making a bitmask for all
	registers.

	* ppc-instructions: Rewrite model specific functions to use the
	bitmask of the register number, instead of using the register
	pointer to get the register number, and then making the bitmask.

Wed Nov 22 15:24:27 1995  Michael Meissner  <[e-mail]>

	* ppc-instructions (model_branches): Add conditional argument to
	count the number of times each type of conditional branch is used.
	(conditional branches): Pass B0 or -1 to model_branches.
	(model_mon_info): Print out conditional branch counts.
	(model-data): Add support for printing out conditional branch
	types.

Tue Nov 21 16:31:25 1995  Michael Meissner  <[e-mail]>

	* igen.c (insn_table_load_insns): Add support for model-static for
	internal functions that should not be inlined.
	(lf_print_c_semantic): Remove model_cleanup.
	(gen_model_{c,h}): Ditto.

	* ppc-instructions: Redo model specific support once again.  Add
	floating point support to the model specific information.  Flesh
	out all of the floating mutiply add/subtract instructions.  Add
	better tracing support to the model specific information.

Sun Nov 19 23:00:52 1995  Michael Meissner  <[e-mail]>

	* ppc-instructions (model data, model_busy): Rather than using a
	bit mask for the busy units, just use a char array.  Also, only
	support 2 function units an insn can use, rather than a loop.

Fri Nov 17 14:08:08 1995  Michael Meissner  <[e-mail]>

	* table.c (table_entry_read): Move setting entry->line_nr to after
	the model specific fields so the line numbers for the annex are
	correct.

	* cpu.c (cpu_{create,init,halt}): Check for WITH_MODEL_ISSUE
	before calling the model functions.

	* debug.c (trace_descriptor): Add trace_model support.
	* debug.h (trace_options): Ditto.

	* igen.c (gen_icache_h): Create type idecode_cache as void if not
	caching instructions.
	(gen_model_{c,h}): Drop model_issue support.  Add support for
	model_cleanup.
	(lf_print_my_prefix): Initialize a const itable_index with the
	current index.
	(lf_print_c_semantic): Call model_cleanup at the end of the
	function to check for instructions that aren't supported yet by
	the scheduling code.
	
	* mon.h (count_type): New type for counters.
	* mon.c: Use count_type instead of unsigned.

	* ppc-instructions: Redo scheduling code once again.  Make it all
	inline friendly.  Instead of having general code emitted by igen,
	go the route of having each semantic routine call the appropriate
	module.

Thu Nov 16 09:52:26 1995  Michael Meissner  <[e-mail]>

	* table.c (table_entry_read): Allow the annex to have blank lines.

	* ppc-instructions: Change lines in model stuff that just have a
	tab to just newline.  Add 601 support.  Document most instructions
	in terms of model specific timing information.  Drop 'FUNCTION'
	from PPC_FUNCTION_UNIT_xxx enums.  Change PPC_UNIT_UNKNOWN ->
	PPC_UNIT_BAD.  Add TRACE(trace_tbd) for all data cache
	instruction.s.  Signal illegal instruciton if data cache block
	invalidate is issued from problem state.

	* igen.c (max_model_fields_len): New static to keep track of the
	max size for the model specific fields.
	(model_c_insn): Use max_model_fields_len to size fields.
	(insn_table_insert_insn): Set max_model_fields_len.
	(model_table_insert): Ditto.
	(gen_model_{c,h}): Model_issue is now called with a processor
	argument.

	* debug.c (trace_description): Add support for trace_tbd.

	* mon.c (mon_issue): Pass processor argument to model_issue.

	* Makefile.in: Delete all function unit support, since the newer
	table driven model support replaces it.
	* cpu.{c,h}: Ditto.
	* mon.c: Ditto.
	* inline.{c,h}: Ditto.
	* std-config.h: Ditto.
	* options.c: Ditto.
	* configure{,.in}: Ditto.
	* Makefile.in: Ditto.
	* psim.c: Ditto.
	* function_unit.{c,h}: Delete these now usused files.

	* std-config.h (WITH_MODEL_ISSUE): Add new macro on whether to
	trace instructions in a model specific manor.
	* options.c (print_options): Print it out.
	* configure{,.in}: Add --enable-sim-model-issue option.
	* Makefile.in: Add --enable-sim-model-issue flags.
	* igen.c (lf_print_c_semantic): Add call to mon_issue here.  Check
	for WITH_MODEL_ISSUE.
	* mon.c (mon_issue): Remove call to mon_issue_here.

	* ppc-instructions: Move branch tracing to the actual branch
	instructions, rather than testing it in model_issue.  Add code to
	code successful/unsuccessful branch predictions, and the number of
	conditional branches that fell through.

Wed Nov 15 17:32:13 1995  Michael Meissner  <[e-mail]>

	* cpu.h (cpu_model): Add declaration.

	* cpu.c (struct _cpu): Add model_ptr to hold model specific
	information.
	(cpu_model): Return the model internal pointer.
	(cpu_{create,init,halt}): Call the appropriate model function.

	* inline.c (mon.c): Move include of mon.c after model.c.

	* mon.c (_cpu_mon): Add fields to count unaligned memory
	references.
	(mon_issue): Call model_issue, not function_unit_issue.
	(mon_{read,write}): Count # of unaligned memory accesses.
	(mon_print_info): Switch to calling model_mon_info and
	model_mon_info_free instead of function_unit version.  Print out
	number of unaligned reads/writes.

	* {ppc-instructions,igen.c}: More global changes to add model
	specific features.

	* inline.{c,h}: Provide for inlining options.c.
	* options.{c,h}: Ditto.
	* std-config.h: Add OPTIONS_INLINE.

Tue Nov 14 04:47:25 1995  Michael Meissner  <[e-mail]>

	* Makefile.in (devices.o, main.o): Update dependency.

	* igen.c (gen_model_h): Use correct variable in loop.
	(gen_model_c): Use strcmp, strcasecmp.
	(gen_model_c): Use EXTERN_MODEL for arrays.
	(gen_model_h): Use STATIC_MODEL for arrays.
	(lf_print_c_semantic_function_header): Delete unused function.

	* main.c (cpu.h): Include cpu.h to get model.h.

	* inline.h ({EXTERN,STATIC}_MODEL): Define.

Mon Nov 13 09:14:13 1995  Michael Meissner  <[e-mail]>

	* igen.c ({insn,model}_table_fields): Spell mnemonic correctly.
	(gen_itable_h,itable_c_insn): Ditto.
	(model support): Move model support around, add support for
	model-data, model-internal.  Use annex field for model-macros
	now.

	* configure.in (--enable-sim-inline): If --enable-sim-inline=no,
	also define INLINE as nothing.
	* configure: Regenerate.

	* std-config.h (INLINE): Rather than nuking INLINE, only define it
	as __inline__ if any of the INLINE flags are non-zero.

	* options.c (print_options): Print out WITH_XOR_ENDAIN.

Mon Nov 13 23:03:45 1995  Andrew Cagney   <cagneyhighland.com.au>

	* ppc-instructions (rfi): Add missing code.

	* cpu.c (cpu_get_time_base): Fix calculation of current value of
        time base register.

	* ppc-spr-table (TBL, TBU): Fix TBL/TBU entries - was confusing
        m[tf]tb with m[tf]spr.

	* ppc-instructions (mtspr, mfspr): Fix mttbl - wasn't storing
        lower word.

Mon Nov 13 21:35:37 1995  Andrew Cagney   <cagneyhighland.com.au>

	* std-config.h (INLINE, STATIC_INLINE): Was being set to static
        inline..  Only problem being that with ppc-opcode-simple this gave
        it the chance to inline all the idecode functions with potentially
        disasterous results on a 16mb PC.  For moment hobble INLINE.

	* configure.in, std-config.h (WITH_SMP): Make that 5 processors by
        default ...

	* configure.in: Tweek flags passed to gcc for --with-sim-warnings.
        Firstly make them errors and secondly remove the options gcc-245
        doesn't reconize.

Mon Nov 13 17:57:24 1995  Andrew Cagney  <[e-mail]>

	* misc.c (zalloc), cpu.c (cpu_init), devices
 	(console_io_read_buffer_callback, icu_io_read_buffer_callback,
 	vm_io_read_buffer_callback), main.c (zalloc), mon.c (memset),
 	sim_calls.c (zalloc) : replace bzero() with memset().

	* emul_netbsd.c (write_direntries), psim.c (psim_read_register,
 	psim_write_register): replace bcopy() with memcpy().
	
Sun Nov 12 20:55:41 1995  Andrew Cagney   <cagneyhighland.com.au>

	* configure.in: for --disable-sim-inline (--enable-sim-inline=no),
        force DEFAULT_INLINE to 0 rather then trusting the std
        configuration.

Sun Nov 12 20:55:41 1995  Andrew Cagney   <cagneyhighland.com.au>

	* igen.c (lf_print_idecode_table, idecode_table_leaf): Fix
        generation of switch entries in tables - treat the same as
        cracking/semantic functions.

	* igen.c (idecode_switch_end, idecode_switch_leaf): Fix generation
        of a boolean switch statement (field zero or non-zero).

	* ppc-opcode-test-1, ppc-opcode-test-2: New files.  These test the
        switch/table generation ability of igen.

	* igen.c (idecode_switch_leaf): Fix code output when a switch
        statement needs to look up a table.

	* igen.c (idecode_declare_if_switch): New function called from
        gen_idecode_c - need to declare any idecode switch functions
        before they are used in idecode tables.

	* igen.c (lf_print_c_cracker_function, idecode_crack_leaf,
        idecode_crack_insn): Add is_inline_function argument to code
        printing cracker functions which indicates if STATIC_IDECODE or
        STATIC_INLINE_IDECODE should be used for definition.  For
        idecode_crack_insn (which implies not duplicating/expanding) don't
        declare function as inline - we assume that the only time this is
        code is generated is when things are being tested.  For
        idecode_crack_leaf, make static (instead of INLINE) if the
        instructions parent is a table as function will always be called
        via a table.

	* igen.c (idecode_expand_if_switch): Declare as STATIC_IDECODE not
        STATIC_INLINE_IDECODE. Only the outermost idecode switch will be
        called directly, all others are called via a table.

	* igen.c (lf_print_semantic_function_header, semantics_h_leaf,
        semantics_h_insn, semantics_h_function,
        lf_print_c_semantic_function, semantics_c_function): Add
        is_inline_function argument to lf_print_semantic_function_header
        to indicate if an inline or static function declaration/definition
        should be output.  Depending on situtation call accordingly:
        functions (not instruction semantic routines) are always inline;
        Semantic routines are made inline when there is no icache (cache
        will contain the function address) and are duplicating (see above)
        and the parent of the instruction is a switch statement.

	* igen.c (opcode_field_new): Delete.  Code changed to use ZALLOC
        and moved to insn_table_find_opcode_field.

	* table.c (table_open): Fix typo (nr_model_fields vs nr_fields).

	* igen.c (model_c_insn): Suggestion - document the name of the
        instruction on each line of the instruction model table.

Fri Nov 10 00:44:38 1995  Andrew Cagney   <cagneyhighland.com.au>

	* emul_netbsd.c (do_ioctl): Cleanup compilation.

	* sim_callbacks.h (__attribute__): Only define if not defined (was
        already defined on NetBSD host).

Wed Nov  8 21:49:52 1995  Andrew Cagney   <cagneyhighland.com.au>

	* std-config.h (WITH_XOR_ENDIAN), configure.in, Makefile.in: New
        macro, indicates if the PowerPC's horrible XOR endian mode should
        be suported.  Add to configure and make.

	* vm_n.h (vm_data_map_read_N, vm_data_map_write_N), vm.c
        (vm_instruction_map_read): If XOR endian, xor the address
	with a value from an xor table (indexed by size of access).

	* vm.c (vm_synchronize_context), cpu.c (cpu_synchronize_context):
        set up xor table to xor if there is a conflict between the
        CURRENT_TARGET_ENDIAN and the endian indicated in the MSR.  Move
        check of suported change of endian mode from cpu.c to vm.c.

	* vm.c (vm_data_map_write_buffer, vm_data_map_read_buffer):
        Hopefully added correct hack to handle XOR endian mode.

	FIXME: If NONSTRICT alignment and XOR ENDIAN and MSR indicates
        little endian mode, the model accepts miss aligned transfers.

	FIXME: Need to create an `init' device that, during
        initializatioin for XOR mode, it mushes (XOR address) all the dma
        data before passing it on to the core for storage. Just like the
        real thing really.

Wed Nov  8 21:49:52 1995  Andrew Cagney   <cagneyhighland.com.au>

	* devices.c (halt_io_write_buffer_callback): Use value written to
        halt device to determine exit status.  Thus allowing
        success/failure of OEA tests.

Wed Nov  8 00:10:38 1995  Andrew Cagney   <cagneyhighland.com.au>

	* ppc-instructions (icbi): If icache present flush it.