- Generated by
1.8.1.1
|
GCC Middle and Back End API Reference
|
Data Structures | |
| struct | label_alignment |
| struct | debug_prefix_map |
Typedefs | |
| typedef struct debug_prefix_map | debug_prefix_map |
| typedef struct debug_prefix_map debug_prefix_map |
??? This is probably the wrong place for these.
Structure recording the mapping from source file and directory names at compile time to those to be embedded in debug information.
| void add_debug_prefix_map | ( | ) |
Record a debug file prefix mapping. ARG is the argument to -fdebug-prefix-map and must be of the form OLD=NEW.
References emit_note_before().
|
static |
The differences in addresses between a branch and its target might grow or shrink depending on the alignment the start insn of the range (the branch for a forward branch or the label for a backward branch) starts out on; if these differences are used naively, they can even oscillate infinitely. We therefore want to compute a 'worst case' address difference that is independent of the alignment the start insn of the range end up on, and that is at least as large as the actual difference. The function align_fuzz calculates the amount we have to add to the naively computed difference, by traversing the part of the alignment chain of the start insn of the range that is in front of the end insn of the range, and considering for each alignment the maximum amount that it might contribute to a size increase. For casesi tables, we also want to know worst case minimum amounts of address difference, in case a machine description wants to introduce some common offset that is added to all offsets in a table. For this purpose, align_fuzz with a growth argument of 0 computes the appropriate adjustment.
Compute the maximum delta by which the difference of the addresses of START and END might grow / shrink due to a different address for start which changes the size of alignment insns between START and END. KNOWN_ALIGN_LOG is the alignment known for START. GROWTH should be ~0 if the objective is to compute potential code size increase, and 0 if the objective is to compute potential shrink. The return value is undefined for any other value of GROWTH.
|
static |
|
static |
Given BODY, the body of a jump instruction, alter the jump condition as required by the bits that are set in cc_status.flags. Not all of the bits there can be handled at this level in all cases. The value is normally 0. 1 means that the condition has become always true. -1 means that the condition has become always false. 2 means that COND has been altered.
Jump becomes unconditional.
Jump becomes no-op.
Jump becomes unconditional.
Jump becomes no-op.
Jump becomes unconditional.
Jump becomes no-op.
The flags are valid if signed condition operators are converted
to unsigned.
| rtx alter_subreg | ( | ) |
If X is a SUBREG, try to replace it with a REG or a MEM, based on the thing it is a subreg of. Do it anyway if FINAL_P.
simplify_subreg does not remove subreg from volatile references.
We are required to. For paradoxical subregs on big-endian machines, SUBREG_BYTE
contains 0 instead of the proper offset. See simplify_subreg. Simplify_subreg can't handle some REG cases, but we have to.
Referenced by output_asm_insn().
| void app_disable | ( | void | ) |
Disable APP processing of subsequent output. Called from varasm.c before most kinds of output.
| void app_enable | ( | void | ) |
Enable APP processing of subsequent output. Used before the output from an `asm' statement.
| void asm_fprintf | ( | ) |
A poor man's fprintf, with the added features of %I, %R, %L, and %U. %R prints the value of REGISTER_PREFIX. %L prints the value of LOCAL_LABEL_PREFIX. %U prints the value of USER_LABEL_PREFIX. %I prints the value of IMMEDIATE_PREFIX. %O runs ASM_OUTPUT_OPCODE to transform what follows in the string. Also supported are %d, %i, %u, %x, %X, %o, %c, %s and %%. We handle alternate assembler dialects here, just like output_asm_insn.
This is a prefix to the 'd', 'i', 'u', 'x', 'X', and
'o' cases, but we do not check for those cases. It
means that the value is a HOST_WIDE_INT, which may be
either `long' or `long long'. Uppercase letters are reserved for general use by asm_fprintf
and so are not available to target specific code. In order to
prevent the ASM_FPRINTF_EXTENSIONS macro from using them then,
they are defined here. As they get turned into real extensions
to asm_fprintf they should be removed from this list.
|
static |
|
static |
Given the body of an INSN known to be generated by an ASM statement, return the number of machine instructions likely to be generated for this insn. This is used to compute its length.
| int asm_str_count | ( | ) |
Return the number of machine instructions likely to be generated for the inline-asm template.
|
static |
Given a CALL_INSN, find and return the nested CALL.
|
static |
Emit lexical block notes needed to change scope from S1 to S2.
Close scopes.
Open scopes.
|
static |
Return scope resulting from combination of S1 and S2.
References cfun, and function::decl.
| void cleanup_subreg_operands | ( | ) |
For each operand in INSN, simplify (subreg (reg)) so that it refers directly to the desired hard register.
The following test cannot use recog_data.operand when testing
for a SUBREG: the underlying object might have been changed
already if we are inside a match_operator expression that
matches the else clause. Instead we test the underlying
expression directly.
| unsigned int compute_alignments | ( | void | ) |
Compute branch alignments based on frequency information in the CFG.
If not optimizing or optimizing for size, don't assign any alignments.
There are two purposes to align block with no fallthru incoming edge:
1) to avoid fetch stalls when branch destination is near cache boundary
2) to improve cache efficiency in case the previous block is not executed
(so it does not need to be in the cache).
We to catch first case, we align frequently executed blocks.
To catch the second, we align blocks that are executed more frequently
than the predecessor and the predecessor is likely to not be executed
when function is called. In case block is frequent and reached mostly by non-fallthru edge,
align it. It is most likely a first block of loop.
References edge_def::flags.
| int dbr_sequence_length | ( | void | ) |
Return the number of slots filled in the current delayed branch sequence (we don't count the insn needing the delay slot). Zero if not in a delayed branch sequence.
References label_alignment::alignment, and label_alignment::max_skip.
| void default_function_pro_epilogue | ( | FILE * | file, |
| HOST_WIDE_INT | size | ||
| ) |
Default target function prologue and epilogue assembler output. If not overridden for epilogue code, then the function body itself contains return instructions wherever needed.
Referenced by reemit_insn_block_notes().
| void default_function_switched_text_sections | ( | FILE * | , |
| tree | , | ||
| bool | |||
| ) |
Default target function switched text sections.
References app_on, and asm_out_file.
| int default_jump_align_max_skip | ( | ) |
| int default_label_align_after_barrier_max_skip | ( | ) |
| int default_label_align_max_skip | ( | ) |
| int default_loop_align_max_skip | ( | ) |
|
static |
Helper function to parse assembler dialects in the asm string. This is called from output_asm_insn and asm_fprintf.
If we want the first dialect, do nothing. Otherwise, skip
DIALECT_NUMBER of strings ending with '|'. Skip over any character after a percent sign.
Skip to close brace.
Skip over any character after a percent sign.
|
static |
Dumper helper for basic block information. FILE is the assembly output file, and INSN is the instruction being emitted.
|
static |
Return true if DWARF2 debug info can be emitted for DECL.
| void final | ( | ) |
Output assembler code for some insns: all or part of a function. For description of args, see `final_start_function', above.
Used for -dA dump.
If CC tracking across branches is enabled, record the insn which
jumps to each branch only reached from one place. There is no cfg for a thunk.
Output the insns.
This can be triggered by bugs elsewhere in the compiler if
new insns are created after init_insn_lengths is called. Remove CFI notes, to avoid compare-debug failures.
References targetm.
|
static |
|
static |
| void final_end_function | ( | void | ) |
Output assembler code for the end of a function. For clarity, args are same as those of `final_start_function' even though not all of them are needed.
Finally, output the function epilogue:
code to restore the stack frame and return to the caller. And debug output.
| int final_forward_branch_p | ( | ) |
Return 1 if branch is a forward branch. Uses insn_shuid array, so it works only in the final pass. May be used by output templates to customary add branch prediction hints.
We've hit some insns that does not have id information available.
References RTL_PASS.
The final scan for one insn, INSN. Args are same as in `final', except that INSN is the insn being scanned. Value returned is the next insn to be scanned. NOPEEPHOLES is the flag to disallow peephole processing (currently used for within delayed branch sequence output). SEEN is used to track the end of the prologue, for emitting debug information. We force the emission of a line note after both NOTE_INSN_PROLOGUE_END and NOTE_INSN_FUNCTION_BEG, or at the beginning of the second basic block, whichever comes first.
Ignore deleted insns. These can occur when we split insns (due to a
template of "#") while not optimizing. Output debugging info about the symbol-block beginning.
Mark this block as output.
End of a symbol-block.
Emit the label. We may have deleted the CODE_LABEL because
the label could be proved to be unreachable, though still
referenced (in the form of having its address taken. Similarly, but need to use different namespace for it.
The target port might emit labels in the output function for
some insn, e.g. sh.c output_branchy_insn. If this label is followed by a jump-table, make sure we put
the label in the read-only section. Also possibly write the
label and jump table together. In this case, the case vector is being moved by the
target, so don't output the label at all. Leave that
to the back end macros. Reset this early so it is correct for ASM statements.
An INSN, JUMP_INSN or CALL_INSN.
First check for special kinds that recog doesn't recognize. If there is a REG_CC_SETTER note on this insn, it means that
the setting of the condition code was done in the delay slot
of the insn that branched here. So recover the cc status
from the insn that set it. Detect insns that are really jump-tables
and output them as such. Output this line note if it is the first or the last line
note in a row. There's no telling what that did to the condition codes.
Detect `asm' construct with operands.
There's no telling what that did to the condition codes.
Get out the operand values.
Inhibit dying on what would otherwise be compiler bugs.
Output the insn using them.
A delayed-branch sequence
The first insn in this SEQUENCE might be a JUMP_INSN that will
force the restoration of a comparison that was previously
thought unnecessary. If that happens, cancel this sequence
and cause that insn to be restored. We loop in case any instruction in a delay slot gets
split. If the insn requiring the delay slot was a CALL_INSN, the
insns in the delay slot are actually executed before the
called function. Hence we don't preserve any CC-setting
actions in these insns and the CC must be marked as being
clobbered by the function. We have a real machine instruction as rtl.
Check for redundant test and compare instructions
(when the condition codes are already set up as desired).
This is done only when optimizing; if not optimizing,
it should be possible for the user to alter a variable
with the debugger in between statements
and the next statement should reexamine the variable
to compute the condition codes. Don't delete insn if it has an addressing side-effect.
or if anything in it is volatile.
We don't really delete the insn; just ignore it.
If this is a conditional branch, maybe modify it
if the cc's are in a nonstandard state
so that it accomplishes the same thing that it would
do straightforwardly if the cc's were set up normally. This function may alter the contents of its argument
and clear some of the cc_status.flags bits.
It may also return 1 meaning condition now always true
or -1 meaning condition now always false
or 2 meaning condition nontrivial but altered. If condition now has fixed value, replace the IF_THEN_ELSE
with its then-operand or its else-operand. The jump is now either unconditional or a no-op.
If it has become a no-op, don't try to output it.
(It would not be recognized.) Replace (set (pc) (return)) with (return).
Rerecognize the instruction if it has changed.
If this is a conditional trap, maybe modify it if the cc's
are in a nonstandard state so that it accomplishes the same
thing that it would do straightforwardly if the cc's were
set up normally. This function may alter the contents of its argument
and clear some of the cc_status.flags bits.
It may also return 1 meaning condition now always true
or -1 meaning condition now always false
or 2 meaning condition nontrivial but altered. If TRAP_CONDITION has become always false, delete the
instruction. If TRAP_CONDITION has become always true, replace
TRAP_CONDITION with const_true_rtx. Rerecognize the instruction if it has changed.
Make same adjustments to instructions that examine the
condition codes without jumping and instructions that
handle conditional moves (if this machine has either one). Do machine-specific peephole optimizations if desired.
When peepholing, if there were notes within the peephole,
emit them before the peephole. Put the notes in the proper position for a later
rescan. For example, the SH target can do this
when generating a far jump in a delayed branch
sequence. PEEPHOLE might have changed this.
Try to recognize the instruction.
If successful, verify that the operands satisfy the
constraints for the instruction. Crash if they don't,
since `reload' should have changed them so that they do. Dump the insn in the assembly for debugging (-dAP).
If the final dump is requested as slim RTL, dump slim
RTL to the assembly file also. Some target machines need to prescan each insn before
it is output. Update `cc_status' for this instruction.
The instruction's output routine may change it further.
If the output routine for a jump insn needs to depend
on the cc status, it should look at cc_prev_status. Find the proper template for this insn.
If the C code returns 0, it means that it is a jump insn
which follows a deleted test insn, and that test insn
needs to be reinserted. We have already processed the notes between the setter and
the user. Make sure we don't process them again, this is
particularly important if one of the notes is a block
scope note or an EH note. If the template is the string "#", it means that this insn must
be split. If we didn't split the insn, go away.
If we have a length attribute, this instruction should have
been split in shorten_branches, to ensure that we would have
valid length info for the splitees. ??? This will put the directives in the wrong place if
get_insn_template outputs assembly directly. However calling it
before get_insn_template breaks if the insns is split. Output assembler code from the template.
Some target machines need to postscan each insn after
it is output.
References debug_hooks.
| void final_start_function | ( | rtx | first, |
| FILE * | file, | ||
| int | optimize_p | ||
| ) |
Output assembler code for the start of a function,
and initialize some of the variables in this file
for the new function. The label for the function and associated
assembler pseudo-ops have already been output in `assemble_start_function'.
FIRST is the first insn of the rtl for the function being compiled.
FILE is the file to write assembler code to.
OPTIMIZE_P is nonzero if we should eliminate redundant
test and compare insns. The Sun386i and perhaps other machines don't work right
if the profiling code comes after the prologue. If debugging, assign block numbers to all of the blocks in this
function. We never actually put out begin/end notes for the top-level
block in the function. But, conceptually, that block is
always needed. Issue a warning
First output the function prologue: code to set up the stack frame.
If the machine represents the prologue as RTL, the profiling code must
be emitted when NOTE_INSN_PROLOGUE_END is scanned.
References assemble_integer(), data_section, floor_log2(), switch_to_section(), and targetm.
| void fprint_ul | ( | ) |
Write an unsigned long as decimal to a file, fast.
python says: len(str(2**64)) == 20
It's probably too small to bother with string reversal and fputs.
References df_regs_ever_live_p(), and global_regs.
| void fprint_whex | ( | ) |
Write a HOST_WIDE_INT number in hex form 0x1234, fast.
References get_insns().
Referenced by dw2_asm_output_data_uleb128_raw().
| int get_attr_length | ( | ) |
Obtain the current length of an insn. If branch shortening has been done, get its actual length. Otherwise, get its maximum length.
Obtain the current length of an insn. If branch shortening has been done, get its actual length. Otherwise, use FALLBACK_FN to calculate the length.
| int get_attr_min_length | ( | ) |
Obtain the current length of an insn. If branch shortening has been done, get its actual length. Otherwise, get its minimum length.
Referenced by decls_for_scope(), and reorder_basic_blocks().
| const char* get_insn_template | ( | ) |
References asm_out_file, need_profile_function, and profile_function().
|
static |
If OP is a REG or MEM and we can find a MEM_EXPR corresponding to it or its address, return that expr . Set *PADDRESSP to 1 if the expr corresponds to the address of the object and 0 if to the object.
Otherwise we have an address, so indicate it and look at the address.
First check if we have a decl for the address, then look at the right side
if it is a PLUS. Otherwise, strip off arithmetic and keep looking.
But don't allow the address to itself be indirect.
|
static |
Grow the LABEL_ALIGN array after new labels are created.
Range of labels grows monotonically in the function. Failing here
means that the initialization of array got lost.
Referenced by update_alignments().
| void init_final | ( | ) |
Initialize data in final at the beginning of a compilation.
| void init_insn_lengths | ( | void | ) |
Indicate that branch shortening hasn't yet been done.
| int insn_current_reference_address | ( | ) |
Compute a worst-case reference address of a branch so that it can be safely used in the presence of aligned labels. Since the size of the branch itself is unknown, the size of the branch is not included in the range. I.e. for a forward branch, the reference address is the end address of the branch as known from the previous branch shortening pass, minus a value to account for possible size increase due to alignment. For a backward branch, it is the start address of the branch as known from the current pass, plus a value to account for possible size increase due to alignment. NB.: Therefore, the maximum offset allowed for backward branches needs to exclude the branch size.
This can happen for example on the PA; the objective is to know the
offset to address something in front of the start of the function.
Thus, we can treat it like a backward branch.
We assume here that FUNCTION_BOUNDARY / BITS_PER_UNIT is larger than
any alignment we'd encounter, so we skip the call to align_fuzz. BRANCH has no proper alignment chain set, so use SEQ.
BRANCH also has no INSN_SHUID. Forward branch.
Backward branch.
| int label_to_alignment | ( | ) |
For the benefit of port specific code do this also as a function.
| int label_to_max_skip | ( | ) |
| int leaf_function_p | ( | void | ) |
Return nonzero if this function has no function calls.
Referenced by split_live_ranges_for_shrink_wrap().
|
static |
|
static |
Scan all instructions and renumber all registers into those available in leaf functions.
Renumber only the actual patterns.
The reg-notes can contain frame pointer refs,
and renumbering them could crash, and should not be needed.
References current_function_decl, dump_flags, dump_function_header(), final_insns_dump_p, flag_dump_unnumbered, and get_insns().
| void leaf_renumber_regs_insn | ( | ) |
Scan IN_RTX and its subexpressions, and renumber all regs into those available in leaf functions.
Renumber all input-registers into output-registers.
renumbered_regs would be 1 for an output-register;
they Don't renumber the same reg twice.
Don't try to renumber pseudo regs. It is possible for a pseudo reg
to reach here as part of a REG_NOTE. Inside a SEQUENCE, we find insns.
Renumber just the patterns of these insns,
just as we do for the top-level insns.
Referenced by sprint_ul().
| rtl_opt_pass* make_pass_clean_state | ( | ) |
| rtl_opt_pass* make_pass_compute_alignments | ( | ) |
References log(), and next_nonnote_insn().
| rtl_opt_pass* make_pass_final | ( | ) |
| rtl_opt_pass* make_pass_shorten_branches | ( | ) |
|
static |
Helper rtx-iteration-function for mark_symbol_refs_as_used and output_operand. Marks SYMBOL_REFs as referenced through use of assemble_external.
If we have a used symbol, we may have to emit assembly
annotations corresponding to whether the symbol is external, weak
or has non-default visibility.
Referenced by output_asm_insn().
| void mark_symbol_refs_as_used | ( | ) |
Marks SYMBOL_REFs in x as referenced through use of assemble_external.
| void no_asm_to_stream | ( | ) |
Default target hook that outputs nothing to a stream.
|
static |
|
static |
Return whether a source line note needs to be emitted before INSN. Sets IS_STMT to TRUE if the line should be marked as a possible breakpoint location.
If the discriminator changed, but the line number did not,
output the line table entry with is_stmt false so the
debugger does not treat this as a breakpoint location.
References CC_STATUS::flags, reverse_condition(), and swap_condition().
| int only_leaf_regs_used | ( | void | ) |
On some machines, a function with no call insns can run faster if it doesn't create its own register window. When output, the leaf function should use only the "output" registers. Ordinarily, the function would be compiled to use the "input" registers to find its arguments; it is a candidate for leaf treatment if it uses only the "input" registers. Leaf function treatment means renumbering so the function uses the "output" registers instead.
Return 1 if this function uses only the registers that can be safely renumbered.
| void output_addr_const | ( | ) |
Print an integer constant expression in assembler syntax. Addition and subtraction are the only arithmetic that may appear in these expressions.
Fall through.
This used to output parentheses around the expression,
but that does not work on the 386 (either ATT or BSD assembler). We can use %d if the number is one word and positive.
We can't handle floating point constants;
PRINT_OPERAND must handle them. Some assemblers need integer constants to appear last (eg masm).
Avoid outputting things like x-x or x+5-x,
since some assemblers can't handle that.
Referenced by output_asm_operand_names().
| void output_address | ( | ) |
Print a memory reference operand for address X using machine-dependent assembler syntax.
Referenced by output_asm_operand_names().
|
static |
|
static |
Emit the appropriate declaration for an alternate-entry-point symbol represented by INSN, to FILE. INSN is a CODE_LABEL with LABEL_KIND != LABEL_NORMAL. The case fall-through in this function is intentional.
References targetm.
| void output_asm_insn | ( | ) |
Output text from TEMPLATE to the assembler output file,
obeying %-directions to substitute operands taken from
the vector OPERANDS.
%N (for N a digit) means print operand N in usual manner.
%lN means require operand N to be a CODE_LABEL or LABEL_REF
and print the label name with no punctuation.
%cN means require operand N to be a constant
and print the constant expression with no punctuation.
%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference
to that address.
%nN means expect operand N to be a constant
and print a constant expression for minus the value
of the operand, with no other punctuation. An insn may return a null string template
in a case where no assembler code is needed. %% outputs a single %. %{, %} and %| print {, } and | respectively
if ASSEMBLER_DIALECT defined and these characters have a special
meaning as dialect delimiters. %= outputs a number which is unique to each insn in the entire
compilation. This is useful for making local labels that are
referred to more than once in a given insn. % followed by a letter and some digits
outputs an operand in a special way depending on the letter.
Letters `acln' are implemented directly.
Other letters are passed to `output_operand' so that
the TARGET_PRINT_OPERAND hook can define them. % followed by a digit outputs an operand the default way.
% followed by punctuation: output something for that
punctuation character alone, with no operand. The
TARGET_PRINT_OPERAND hook decides what is actually done. Write out the variable names for operands, if we know them.
References alter_subreg(), asm_out_file, for_each_rtx(), mark_symbol_ref_as_used(), and targetm.
| void output_asm_label | ( | ) |
Output a LABEL_REF, or a bare CODE_LABEL, as an assembler symbol.
Referenced by output_asm_operand_names().
|
static |
Output of assembler code from a template, and its subroutines.
Annotate the assembly with a comment describing the pattern and alternative used.
Clear this so only the first assembler insn
of any rtl insn will get the special comment for -dp.
|
static |
|
static |
Output operand names for assembler instructions. OPERANDS is the operand vector, OPORDER is the order to write the operands, and NOPS is the number of operands to write.
References asm_out_file, HOST_WIDE_INT_PRINT_DEC, insn_noperands, output_addr_const(), output_address(), output_asm_label(), output_operand(), and output_operand_lossage().
| void output_operand | ( | ) |
Print operand X using machine-dependent assembler syntax. CODE is a non-digit that preceded the operand-number in the % spec, such as 'z' if the spec was `%z3'. CODE is 0 if there was no char between the % and the digits. When CODE is a non-letter, X is 0. The meanings of the letters are machine-dependent and controlled by TARGET_PRINT_OPERAND.
X must not be a pseudo reg.
Referenced by output_asm_operand_names().
| void output_operand_lossage | ( | ) |
Report inconsistency between the assembler template and the operands. In an `asm', it's the user's fault; otherwise, the compiler's fault.
Referenced by get_random_seed(), and output_asm_operand_names().
| void output_quoted_string | ( | ) |
Output a quoted string.
Referenced by dbxout_begin_stabn(), and dbxout_begin_stabn_sline().
|
static |
|
static |
|
static |
Referenced by get_insn_template(), and reemit_insn_block_notes().
|
static |
References cfun, get_max_uid(), init_recog(), function::is_thunk, and debug_prefix_map::next.
|
static |
Rebuild all the NOTE_INSN_BLOCK_BEG and NOTE_INSN_BLOCK_END notes based on the scope tree and the newly reordered instructions.
Prevent lexical blocks from straddling section boundaries.
Avoid putting scope notes between jump table and its label.
For sequences compute scope resulting from merging all scopes
of instructions nested inside. change_scope emits before the insn, not after.
References default_function_pro_epilogue(), HAVE_prologue, need_profile_function, profile_function(), and targetm.
| const char* remap_debug_filename | ( | ) |
Perform user-specified mapping of debug filename prefixes. Return the new name corresponding to FILENAME.
|
static |
It is very important to decompose the RTL instruction chain here:
debug information keeps pointing into CODE_LABEL insns inside the function
body. If these remain pointing to the other insns, we end up preserving
whole RTL chain and attached detailed debug info in memory. In case the function was not output,
don't leave any temporary anonymous types
queued up for sdb output. Clear out the insn_length contents now that they are no
longer valid. Show no temporary slots allocated.
We can reduce stack alignment on call site only when we are sure that
the function body just produced will be actually used in the final
executable. Make sure volatile mem refs aren't considered valid operands for
arithmetic insns. We must call this here if this is a nested inline
function, since the above code leaves us in the init_recog state,
and the function context push/pop code does not save/restore volatile_ok.
??? Maybe it isn't necessary for expand_start_function to call this
anymore if we do it here? We're done with this function. Free up memory if we can.
|
static |
Turn the RTL into assembly.
Get the function's name, as described by its RTL. This may be
different from the DECL_NAME name used in the source file. The IA-64 ".handlerdata" directive must be issued before the ".endp"
directive that closes the procedure descriptor. Similarly, for x64 SEH.
Otherwise it's not strictly necessary, but it doesn't hurt either. Free up reg info memory.
Write DBX symbols if requested.
Note that for those inline functions where we don't initially
know for certain that we will be generating an out-of-line copy,
the first invocation of this routine (rest_of_compilation) will
skip over this code by doing a `goto exit_rest_of_compilation;'.
Later on, wrapup_global_declarations will (indirectly) call
rest_of_compilation again for those inline functions that need
to have out-of-line copies generated. During that call, we
*will* be routed past here. Release the blocks that are linked to DECL_INITIAL() to free the memory.
|
static |
Shorten branches.
| void shorten_branches | ( | ) |
Make a pass over all insns and compute their actual lengths by shortening any branches of variable length if possible.
shorten_branches might be called multiple times: for example, the SH port splits out-of-range conditional branches in MACHINE_DEPENDENT_REORG. In order to do this, it needs proper length information, which it obtains by calling shorten_branches. This cannot be collapsed with shorten_branches itself into a single pass unless we also want to integrate reorg.c, since the branch splitting exposes new instructions with delay slots.
Compute maximum UID and allocate label_align / uid_shuid.
Free uid_shuid before reallocating it.
Initialize label_align and set up uid_shuid to be strictly
monotonically rising with insn order. We use max_log here to keep track of the maximum alignment we want to
impose on the next CODE_LABEL (or the current one if we are processing
the CODE_LABEL itself). Merge in alignments computed by compute_alignments.
ADDR_VECs only take room if read-only data goes into the text
section. Allocate the rest of the arrays.
Syntax errors can lead to labels being outside of the main insn stream.
Initialize insn_addresses, so that we get reproducible results. Initialize uid_align. We scan instructions
from end to start, and keep in align_tab[n] the last seen insn
that does an alignment of at least n+1, i.e. the successor
in the alignment chain for an insn that does / has a known
alignment of n. Found an alignment label.
When optimizing, we start assuming minimum length, and keep increasing
lengths as we find the need for this, till nothing changes.
When not optimizing, we start assuming maximum lengths, and
do a single pass to update the lengths. Look for ADDR_DIFF_VECs, and initialize their minimum and maximum
label fields. Compute initial lengths, addresses, and varying flags for each insn.
This only takes room if read-only data goes into the text
section. Alignment is handled by ADDR_VEC_ALIGN.
Inside a delay slot sequence, we do not do any branch shortening
if the shortening could change the number of delay slots
of the branch. If needed, do any adjustment.
Now loop over all the insns finding varying length insns. For each,
get the current insn length. If it has changed, reflect the change.
When nothing changes for a full pass, we are done. If the mode of a following jump table was changed, we
may need to update the alignment of this label. Avoid automatic aggregate initialization.
Try to find a known alignment for rel_lab.
See the comment on addr_diff_vec_flags in rtl.h for the
meaning of the flags values. base: REL_LAB vec: INSN Anything after INSN has still addresses from the last
pass; adjust these so that they reflect our current
estimate for this pass. We want to know the worst case, i.e. lowest possible value
for the offset of MIN_LAB. If MIN_LAB is after REL_LAB,
its offset is positive, and we have to be wary of code shrink;
otherwise, it is negative, and we have to be vary of code
size increase. If INSN is between REL_LAB and MIN_LAB, the size
changes we are about to make can change the alignment
within the observed offset, therefore we have to break
it up into two parts that are independent. Likewise, determine the highest lowest possible value
for the offset of MAX_LAB. insn_current_length returns 0 for insns with a
non-varying length. If needed, do any adjustment.
For a non-optimizing compile, do only a single pass.
References targetm.
| int sprint_ul | ( | ) |
Write an unsigned long as decimal to a string, fast. s must be wide enough to not overflow, at least 21 chars. Returns the length of the string (without terminating '\0').
Reverse the string.
References leaf_renumber_regs_insn().
|
static |
Internal function that prints an unsigned long in decimal in reverse. The output string IS NOT null-terminated.
alternate version, without modulo
oldval = value;
value /= 10;
s[i] = "0123456789" [oldval - 10*value];
i++
| void update_alignments | ( | ) |
Update the already computed alignment information. LABEL_PAIRS is a vector made up of pairs of labels for which the alignment information of the first element will be copied from that of the second element.
References free(), get_insns(), get_max_uid(), grow_label_align(), log(), max_label_num(), and max_uid.
|
static |
Do alter_subreg on all the SUBREGs contained in X.
|
static |
Nonzero if have enabled APP processing of our assembler output.
Referenced by default_function_switched_text_sections().
|
static |
Number of unmatched NOTE_INSN_BLOCK_BEG notes we have seen.
| CC_STATUS cc_prev_status |
During output of an insn, this contains a copy of cc_status from before the insn.
| CC_STATUS cc_status |
This variable contains machine-dependent flags (defined in tm.h) set and examined by output routines that describe how to interpret the condition codes properly.
| rtx current_insn_predicate |
Nonnull if the insn currently being emitted was a COND_EXEC pattern.
| rtx current_output_insn |
Last insn processed by final_scan_insn.
|
static |
Last insn processed by final_scan_insn.
|
static |
Linked list of such structures.
|
static |
Number of the assembler dialect to use, starting at 0.
|
static |
Discriminator of current block.
| bool final_insns_dump_p |
True if printing into -fdump-final-insns= dump.
Referenced by leaf_renumber_regs().
| rtx final_sequence |
If we are outputting an insn sequence, this contains the sequence rtx. Zero otherwise.
|
static |
Whether to force emission of a line note before the next insn.
|
static |
Highest line number in current block.
|
static |
Likewise for function.
| vec<int> insn_addresses_ |
|
static |
Assign a unique number to each insn that is output. This can be used to generate unique local labels.
| int insn_current_address |
Address of insn being processed. Used by `insn_current_length'.
| int insn_current_align |
known invariant alignment of insn being processed.
| int insn_last_address |
Address of insn being processed in previous iteration.
|
static |
The next two pages contain routines used to compute the length of an insn and to shorten branches.
Arrays for insn lengths, and addresses. The latter is referenced by `insn_current_length'.
|
static |
Max uid for which the above arrays are valid.
|
static |
Number of operands of this insn, for an `asm' with operands.
Referenced by output_asm_operand_names().
|
static |
|
static |
Last discriminator written to assembly.
|
static |
Filename of last NOTE.
|
static |
Compare optimization flag.
|
static |
Line number of last NOTE.
| const int length_unit_log |
|
static |
|
static |
|
static |
True if profile_function should be called, but hasn't been called yet.
Referenced by get_insn_template(), and reemit_insn_block_notes().
|
static |
Override filename and line number.
|
static |
| rtx this_is_asm_operands |
Nonzero while outputting an `asm' with operands. This means that inconsistencies are the user's fault, so don't die. The precise value is the insn being output, to pass to error_for_asm.
|
static |
|
static |
1.8.1.1