![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.info-13 CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 2.x] / gcc|
Return to gcc.info-13 CVS log | Up to [GCC 2.x] / gcc |
1.1.1.9 ! root 1: This is Info file gcc.info, produced by Makeinfo version 1.67 from the ! 2: input file gcc.texi. 1.1 root 3: 4: This file documents the use and the internals of the GNU compiler. 5: 1.1.1.8 root 6: Published by the Free Software Foundation 59 Temple Place - Suite 330 7: Boston, MA 02111-1307 USA 1.1.1.5 root 8: 1.1.1.8 root 9: Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995 Free Software 10: Foundation, Inc. 1.1 root 11: 1.1.1.3 root 12: Permission is granted to make and distribute verbatim copies of this 13: manual provided the copyright notice and this permission notice are 14: preserved on all copies. 1.1 root 15: 16: Permission is granted to copy and distribute modified versions of 17: this manual under the conditions for verbatim copying, provided also 1.1.1.7 root 18: that the sections entitled "GNU General Public License," "Funding for 19: Free Software," and "Protect Your Freedom--Fight `Look And Feel'" are 20: included exactly as in the original, and provided that the entire 21: resulting derived work is distributed under the terms of a permission 22: notice identical to this one. 1.1 root 23: 24: Permission is granted to copy and distribute translations of this 25: manual into another language, under the above conditions for modified 1.1.1.3 root 26: versions, except that the sections entitled "GNU General Public 1.1.1.7 root 27: License," "Funding for Free Software," and "Protect Your Freedom--Fight 28: `Look And Feel'", and this permission notice, may be included in 29: translations approved by the Free Software Foundation instead of in the 30: original English. 1.1.1.4 root 31: 32: 1.1.1.8 root 33: File: gcc.info, Node: VMS Misc, Prev: Global Declarations, Up: VMS 1.1.1.6 root 34: 1.1.1.8 root 35: Other VMS Issues 36: ================ 1.1.1.6 root 37: 1.1.1.8 root 38: GNU CC automatically arranges for `main' to return 1 by default if 39: you fail to specify an explicit return value. This will be interpreted 40: by VMS as a status code indicating a normal successful completion. 41: Version 1 of GNU CC did not provide this default. 42: 43: GNU CC on VMS works only with the GNU assembler, GAS. You need 44: version 1.37 or later of GAS in order to produce value debugging 45: information for the VMS debugger. Use the ordinary VMS linker with the 46: object files produced by GAS. 47: 48: Under previous versions of GNU CC, the generated code would 49: occasionally give strange results when linked to the sharable `VAXCRTL' 50: library. Now this should work. 51: 52: A caveat for use of `const' global variables: the `const' modifier 53: must be specified in every external declaration of the variable in all 54: of the source files that use that variable. Otherwise the linker will 55: issue warnings about conflicting attributes for the variable. Your 56: program will still work despite the warnings, but the variable will be 57: placed in writable storage. 58: 59: Although the VMS linker does distinguish between upper and lower case 60: letters in global symbols, most VMS compilers convert all such symbols 61: into upper case and most run-time library routines also have upper case 62: names. To be able to reliably call such routines, GNU CC (by means of 63: the assembler GAS) converts global symbols into upper case like other 64: VMS compilers. However, since the usual practice in C is to distinguish 65: case, GNU CC (via GAS) tries to preserve usual C behavior by augmenting 66: each name that is not all lower case. This means truncating the name 67: to at most 23 characters and then adding more characters at the end 68: which encode the case pattern of those 23. Names which contain at 69: least one dollar sign are an exception; they are converted directly into 70: upper case without augmentation. 71: 72: Name augmentation yields bad results for programs that use 73: precompiled libraries (such as Xlib) which were generated by another 74: compiler. You can use the compiler option `/NOCASE_HACK' to inhibit 75: augmentation; it makes external C functions and variables 76: case-independent as is usual on VMS. Alternatively, you could write 77: all references to the functions and variables in such libraries using 78: lower case; this will work on VMS, but is not portable to other 79: systems. The compiler option `/NAMES' also provides control over 80: global name handling. 81: 82: Function and variable names are handled somewhat differently with GNU 83: C++. The GNU C++ compiler performs "name mangling" on function names, 84: which means that it adds information to the function name to describe 85: the data types of the arguments that the function takes. One result of 86: this is that the name of a function can become very long. Since the 87: VMS linker only recognizes the first 31 characters in a name, special 88: action is taken to ensure that each function and variable has a unique 89: name that can be represented in 31 characters. 90: 91: If the name (plus a name augmentation, if required) is less than 32 92: characters in length, then no special action is performed. If the name 93: is longer than 31 characters, the assembler (GAS) will generate a hash 94: string based upon the function name, truncate the function name to 23 95: characters, and append the hash string to the truncated name. If the 96: `/VERBOSE' compiler option is used, the assembler will print both the 97: full and truncated names of each symbol that is truncated. 98: 99: The `/NOCASE_HACK' compiler option should not be used when you are 100: compiling programs that use libg++. libg++ has several instances of 101: objects (i.e. `Filebuf' and `filebuf') which become indistinguishable 102: in a case-insensitive environment. This leads to cases where you need 103: to inhibit augmentation selectively (if you were using libg++ and Xlib 104: in the same program, for example). There is no special feature for 105: doing this, but you can get the result by defining a macro for each 106: mixed case symbol for which you wish to inhibit augmentation. The 107: macro should expand into the lower case equivalent of itself. For 108: example: 1.1.1.7 root 109: 1.1.1.8 root 110: #define StuDlyCapS studlycaps 1.1.1.6 root 111: 1.1.1.8 root 112: These macro definitions can be placed in a header file to minimize 113: the number of changes to your source code. 1.1.1.6 root 114: 115: 1.1.1.8 root 116: File: gcc.info, Node: Portability, Next: Interface, Prev: VMS, Up: Top 1.1.1.6 root 117: 1.1.1.8 root 118: GNU CC and Portability 119: ********************** 1.1.1.6 root 120: 1.1.1.8 root 121: The main goal of GNU CC was to make a good, fast compiler for 122: machines in the class that the GNU system aims to run on: 32-bit 123: machines that address 8-bit bytes and have several general registers. 124: Elegance, theoretical power and simplicity are only secondary. 125: 126: GNU CC gets most of the information about the target machine from a 127: machine description which gives an algebraic formula for each of the 128: machine's instructions. This is a very clean way to describe the 129: target. But when the compiler needs information that is difficult to 130: express in this fashion, I have not hesitated to define an ad-hoc 131: parameter to the machine description. The purpose of portability is to 132: reduce the total work needed on the compiler; it was not of interest 133: for its own sake. 134: 135: GNU CC does not contain machine dependent code, but it does contain 136: code that depends on machine parameters such as endianness (whether the 137: most significant byte has the highest or lowest address of the bytes in 138: a word) and the availability of autoincrement addressing. In the 139: RTL-generation pass, it is often necessary to have multiple strategies 140: for generating code for a particular kind of syntax tree, strategies 141: that are usable for different combinations of parameters. Often I have 142: not tried to address all possible cases, but only the common ones or 143: only the ones that I have encountered. As a result, a new target may 144: require additional strategies. You will know if this happens because 145: the compiler will call `abort'. Fortunately, the new strategies can be 146: added in a machine-independent fashion, and will affect only the target 147: machines that need them. 1.1.1.6 root 148: 1.1.1.7 root 149: 1.1.1.8 root 150: File: gcc.info, Node: Interface, Next: Passes, Prev: Portability, Up: Top 1.1.1.6 root 151: 1.1.1.8 root 152: Interfacing to GNU CC Output 153: **************************** 1.1.1.6 root 154: 1.1.1.8 root 155: GNU CC is normally configured to use the same function calling 156: convention normally in use on the target system. This is done with the 157: machine-description macros described (*note Target Macros::.). 158: 159: However, returning of structure and union values is done differently 160: on some target machines. As a result, functions compiled with PCC 161: returning such types cannot be called from code compiled with GNU CC, 162: and vice versa. This does not cause trouble often because few Unix 163: library routines return structures or unions. 164: 165: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes 166: long in the same registers used for `int' or `double' return values. 167: (GNU CC typically allocates variables of such types in registers also.) 168: Structures and unions of other sizes are returned by storing them into 169: an address passed by the caller (usually in a register). The 170: machine-description macros `STRUCT_VALUE' and `STRUCT_INCOMING_VALUE' 171: tell GNU CC where to pass this address. 172: 173: By contrast, PCC on most target machines returns structures and 174: unions of any size by copying the data into an area of static storage, 175: and then returning the address of that storage as if it were a pointer 176: value. The caller must copy the data from that memory area to the 177: place where the value is wanted. This is slower than the method used 178: by GNU CC, and fails to be reentrant. 179: 180: On some target machines, such as RISC machines and the 80386, the 181: standard system convention is to pass to the subroutine the address of 182: where to return the value. On these machines, GNU CC has been 183: configured to be compatible with the standard compiler, when this method 184: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes. 185: 186: GNU CC uses the system's standard convention for passing arguments. 187: On some machines, the first few arguments are passed in registers; in 188: others, all are passed on the stack. It would be possible to use 189: registers for argument passing on any machine, and this would probably 190: result in a significant speedup. But the result would be complete 191: incompatibility with code that follows the standard convention. So this 192: change is practical only if you are switching to GNU CC as the sole C 193: compiler for the system. We may implement register argument passing on 194: certain machines once we have a complete GNU system so that we can 195: compile the libraries with GNU CC. 196: 197: On some machines (particularly the Sparc), certain types of arguments 198: are passed "by invisible reference". This means that the value is 199: stored in memory, and the address of the memory location is passed to 200: the subroutine. 201: 202: If you use `longjmp', beware of automatic variables. ANSI C says 203: that automatic variables that are not declared `volatile' have undefined 204: values after a `longjmp'. And this is all GNU CC promises to do, 205: because it is very difficult to restore register variables correctly, 206: and one of GNU CC's features is that it can put variables in registers 207: without your asking it to. 208: 209: If you want a variable to be unaltered by `longjmp', and you don't 210: want to write `volatile' because old C compilers don't accept it, just 211: take the address of the variable. If a variable's address is ever 212: taken, even if just to compute it and ignore it, then the variable 213: cannot go in a register: 214: 215: { 216: int careful; 217: &careful; 218: ... 219: } 220: 221: Code compiled with GNU CC may call certain library routines. Most of 222: them handle arithmetic for which there are no instructions. This 223: includes multiply and divide on some machines, and floating point 224: operations on any machine for which floating point support is disabled 225: with `-msoft-float'. Some standard parts of the C library, such as 226: `bcopy' or `memcpy', are also called automatically. The usual function 227: call interface is used for calling the library routines. 228: 229: These library routines should be defined in the library `libgcc.a', 230: which GNU CC automatically searches whenever it links a program. On 231: machines that have multiply and divide instructions, if hardware 232: floating point is in use, normally `libgcc.a' is not needed, but it is 233: searched just in case. 234: 235: Each arithmetic function is defined in `libgcc1.c' to use the 236: corresponding C arithmetic operator. As long as the file is compiled 237: with another C compiler, which supports all the C arithmetic operators, 238: this file will work portably. However, `libgcc1.c' does not work if 239: compiled with GNU CC, because each arithmetic function would compile 240: into a call to itself! 1.1.1.6 root 241: 242: 1.1.1.8 root 243: File: gcc.info, Node: Passes, Next: RTL, Prev: Interface, Up: Top 1.1.1.6 root 244: 1.1.1.8 root 245: Passes and Files of the Compiler 246: ******************************** 1.1.1.6 root 247: 1.1.1.8 root 248: The overall control structure of the compiler is in `toplev.c'. This 249: file is responsible for initialization, decoding arguments, opening and 250: closing files, and sequencing the passes. 251: 252: The parsing pass is invoked only once, to parse the entire input. 253: The RTL intermediate code for a function is generated as the function 254: is parsed, a statement at a time. Each statement is read in as a 255: syntax tree and then converted to RTL; then the storage for the tree 256: for the statement is reclaimed. Storage for types (and the expressions 257: for their sizes), declarations, and a representation of the binding 258: contours and how they nest, remain until the function is finished being 259: compiled; these are all needed to output the debugging information. 260: 261: Each time the parsing pass reads a complete function definition or 262: top-level declaration, it calls either the function 263: `rest_of_compilation', or the function `rest_of_decl_compilation' in 264: `toplev.c', which are responsible for all further processing necessary, 265: ending with output of the assembler language. All other compiler 266: passes run, in sequence, within `rest_of_compilation'. When that 267: function returns from compiling a function definition, the storage used 268: for that function definition's compilation is entirely freed, unless it 269: is an inline function (*note An Inline Function is As Fast As a Macro: 270: Inline.). 271: 272: Here is a list of all the passes of the compiler and their source 273: files. Also included is a description of where debugging dumps can be 274: requested with `-d' options. 275: 276: * Parsing. This pass reads the entire text of a function definition, 277: constructing partial syntax trees. This and RTL generation are no 278: longer truly separate passes (formerly they were), but it is 279: easier to think of them as separate. 280: 281: The tree representation does not entirely follow C syntax, because 282: it is intended to support other languages as well. 283: 284: Language-specific data type analysis is also done in this pass, 285: and every tree node that represents an expression has a data type 286: attached. Variables are represented as declaration nodes. 287: 288: Constant folding and some arithmetic simplifications are also done 289: during this pass. 290: 291: The language-independent source files for parsing are 292: `stor-layout.c', `fold-const.c', and `tree.c'. There are also 293: header files `tree.h' and `tree.def' which define the format of 294: the tree representation. 295: 296: The source files to parse C are `c-parse.in', `c-decl.c', 297: `c-typeck.c', `c-aux-info.c', `c-convert.c', and `c-lang.c' along 298: with header files `c-lex.h', and `c-tree.h'. 299: 300: The source files for parsing C++ are `cp-parse.y', `cp-class.c', 301: `cp-cvt.c', `cp-decl.c', `cp-decl2.c', `cp-dem.c', `cp-except.c', 302: `cp-expr.c', `cp-init.c', `cp-lex.c', `cp-method.c', `cp-ptree.c', 303: `cp-search.c', `cp-tree.c', `cp-type2.c', and `cp-typeck.c', along 304: with header files `cp-tree.def', `cp-tree.h', and `cp-decl.h'. 305: 306: The special source files for parsing Objective C are 307: `objc-parse.y', `objc-actions.c', `objc-tree.def', and 308: `objc-actions.h'. Certain C-specific files are used for this as 309: well. 310: 311: The file `c-common.c' is also used for all of the above languages. 312: 313: * RTL generation. This is the conversion of syntax tree into RTL 314: code. It is actually done statement-by-statement during parsing, 315: but for most purposes it can be thought of as a separate pass. 316: 317: This is where the bulk of target-parameter-dependent code is found, 318: since often it is necessary for strategies to apply only when 319: certain standard kinds of instructions are available. The purpose 320: of named instruction patterns is to provide this information to 321: the RTL generation pass. 322: 323: Optimization is done in this pass for `if'-conditions that are 324: comparisons, boolean operations or conditional expressions. Tail 325: recursion is detected at this time also. Decisions are made about 326: how best to arrange loops and how to output `switch' statements. 327: 328: The source files for RTL generation include `stmt.c', `calls.c', 329: `expr.c', `explow.c', `expmed.c', `function.c', `optabs.c' and 330: `emit-rtl.c'. Also, the file `insn-emit.c', generated from the 331: machine description by the program `genemit', is used in this 332: pass. The header file `expr.h' is used for communication within 333: this pass. 334: 335: The header files `insn-flags.h' and `insn-codes.h', generated from 336: the machine description by the programs `genflags' and `gencodes', 337: tell this pass which standard names are available for use and 338: which patterns correspond to them. 339: 340: Aside from debugging information output, none of the following 341: passes refers to the tree structure representation of the function 342: (only part of which is saved). 343: 344: The decision of whether the function can and should be expanded 345: inline in its subsequent callers is made at the end of rtl 346: generation. The function must meet certain criteria, currently 347: related to the size of the function and the types and number of 348: parameters it has. Note that this function may contain loops, 349: recursive calls to itself (tail-recursive functions can be 350: inlined!), gotos, in short, all constructs supported by GNU CC. 351: The file `integrate.c' contains the code to save a function's rtl 352: for later inlining and to inline that rtl when the function is 353: called. The header file `integrate.h' is also used for this 354: purpose. 355: 356: The option `-dr' causes a debugging dump of the RTL code after 357: this pass. This dump file's name is made by appending `.rtl' to 358: the input file name. 359: 360: * Jump optimization. This pass simplifies jumps to the following 361: instruction, jumps across jumps, and jumps to jumps. It deletes 362: unreferenced labels and unreachable code, except that unreachable 363: code that contains a loop is not recognized as unreachable in this 364: pass. (Such loops are deleted later in the basic block analysis.) 365: It also converts some code originally written with jumps into 366: sequences of instructions that directly set values from the 367: results of comparisons, if the machine has such instructions. 368: 369: Jump optimization is performed two or three times. The first time 370: is immediately following RTL generation. The second time is after 371: CSE, but only if CSE says repeated jump optimization is needed. 372: The last time is right before the final pass. That time, 373: cross-jumping and deletion of no-op move instructions are done 374: together with the optimizations described above. 375: 376: The source file of this pass is `jump.c'. 377: 378: The option `-dj' causes a debugging dump of the RTL code after 379: this pass is run for the first time. This dump file's name is 380: made by appending `.jump' to the input file name. 381: 382: * Register scan. This pass finds the first and last use of each 383: register, as a guide for common subexpression elimination. Its 384: source is in `regclass.c'. 385: 386: * Jump threading. This pass detects a condition jump that branches 387: to an identical or inverse test. Such jumps can be `threaded' 388: through the second conditional test. The source code for this 389: pass is in `jump.c'. This optimization is only performed if 390: `-fthread-jumps' is enabled. 391: 392: * Common subexpression elimination. This pass also does constant 393: propagation. Its source file is `cse.c'. If constant propagation 394: causes conditional jumps to become unconditional or to become 395: no-ops, jump optimization is run again when CSE is finished. 396: 397: The option `-ds' causes a debugging dump of the RTL code after 398: this pass. This dump file's name is made by appending `.cse' to 399: the input file name. 400: 401: * Loop optimization. This pass moves constant expressions out of 402: loops, and optionally does strength-reduction and loop unrolling 403: as well. Its source files are `loop.c' and `unroll.c', plus the 404: header `loop.h' used for communication between them. Loop 405: unrolling uses some functions in `integrate.c' and the header 406: `integrate.h'. 407: 408: The option `-dL' causes a debugging dump of the RTL code after 409: this pass. This dump file's name is made by appending `.loop' to 410: the input file name. 411: 412: * If `-frerun-cse-after-loop' was enabled, a second common 413: subexpression elimination pass is performed after the loop 414: optimization pass. Jump threading is also done again at this time 415: if it was specified. 416: 417: The option `-dt' causes a debugging dump of the RTL code after 418: this pass. This dump file's name is made by appending `.cse2' to 419: the input file name. 420: 421: * Stupid register allocation is performed at this point in a 422: nonoptimizing compilation. It does a little data flow analysis as 423: well. When stupid register allocation is in use, the next pass 424: executed is the reloading pass; the others in between are skipped. 425: The source file is `stupid.c'. 426: 427: * Data flow analysis (`flow.c'). This pass divides the program into 428: basic blocks (and in the process deletes unreachable loops); then 429: it computes which pseudo-registers are live at each point in the 430: program, and makes the first instruction that uses a value point at 431: the instruction that computed the value. 432: 433: This pass also deletes computations whose results are never used, 434: and combines memory references with add or subtract instructions 435: to make autoincrement or autodecrement addressing. 436: 437: The option `-df' causes a debugging dump of the RTL code after 438: this pass. This dump file's name is made by appending `.flow' to 439: the input file name. If stupid register allocation is in use, this 440: dump file reflects the full results of such allocation. 441: 442: * Instruction combination (`combine.c'). This pass attempts to 443: combine groups of two or three instructions that are related by 444: data flow into single instructions. It combines the RTL 445: expressions for the instructions by substitution, simplifies the 446: result using algebra, and then attempts to match the result 447: against the machine description. 448: 449: The option `-dc' causes a debugging dump of the RTL code after 450: this pass. This dump file's name is made by appending `.combine' 451: to the input file name. 452: 453: * Instruction scheduling (`sched.c'). This pass looks for 454: instructions whose output will not be available by the time that 455: it is used in subsequent instructions. (Memory loads and floating 456: point instructions often have this behavior on RISC machines). It 457: re-orders instructions within a basic block to try to separate the 458: definition and use of items that otherwise would cause pipeline 459: stalls. 460: 461: Instruction scheduling is performed twice. The first time is 462: immediately after instruction combination and the second is 463: immediately after reload. 464: 465: The option `-dS' causes a debugging dump of the RTL code after this 466: pass is run for the first time. The dump file's name is made by 467: appending `.sched' to the input file name. 468: 469: * Register class preferencing. The RTL code is scanned to find out 470: which register class is best for each pseudo register. The source 471: file is `regclass.c'. 472: 473: * Local register allocation (`local-alloc.c'). This pass allocates 474: hard registers to pseudo registers that are used only within one 475: basic block. Because the basic block is linear, it can use fast 476: and powerful techniques to do a very good job. 477: 478: The option `-dl' causes a debugging dump of the RTL code after 479: this pass. This dump file's name is made by appending `.lreg' to 480: the input file name. 481: 482: * Global register allocation (`global.c'). This pass allocates hard 483: registers for the remaining pseudo registers (those whose life 484: spans are not contained in one basic block). 485: 486: * Reloading. This pass renumbers pseudo registers with the hardware 487: registers numbers they were allocated. Pseudo registers that did 488: not get hard registers are replaced with stack slots. Then it 489: finds instructions that are invalid because a value has failed to 490: end up in a register, or has ended up in a register of the wrong 491: kind. It fixes up these instructions by reloading the 492: problematical values temporarily into registers. Additional 493: instructions are generated to do the copying. 494: 495: The reload pass also optionally eliminates the frame pointer and 496: inserts instructions to save and restore call-clobbered registers 497: around calls. 498: 499: Source files are `reload.c' and `reload1.c', plus the header 500: `reload.h' used for communication between them. 501: 502: The option `-dg' causes a debugging dump of the RTL code after 503: this pass. This dump file's name is made by appending `.greg' to 504: the input file name. 505: 506: * Instruction scheduling is repeated here to try to avoid pipeline 507: stalls due to memory loads generated for spilled pseudo registers. 508: 509: The option `-dR' causes a debugging dump of the RTL code after 510: this pass. This dump file's name is made by appending `.sched2' 511: to the input file name. 512: 513: * Jump optimization is repeated, this time including cross-jumping 514: and deletion of no-op move instructions. 515: 516: The option `-dJ' causes a debugging dump of the RTL code after 517: this pass. This dump file's name is made by appending `.jump2' to 518: the input file name. 519: 520: * Delayed branch scheduling. This optional pass attempts to find 521: instructions that can go into the delay slots of other 522: instructions, usually jumps and calls. The source file name is 523: `reorg.c'. 524: 525: The option `-dd' causes a debugging dump of the RTL code after 526: this pass. This dump file's name is made by appending `.dbr' to 527: the input file name. 528: 529: * Conversion from usage of some hard registers to usage of a register 530: stack may be done at this point. Currently, this is supported only 531: for the floating-point registers of the Intel 80387 coprocessor. 532: The source file name is `reg-stack.c'. 533: 534: The options `-dk' causes a debugging dump of the RTL code after 535: this pass. This dump file's name is made by appending `.stack' to 536: the input file name. 537: 538: * Final. This pass outputs the assembler code for the function. It 539: is also responsible for identifying spurious test and compare 540: instructions. Machine-specific peephole optimizations are 541: performed at the same time. The function entry and exit sequences 542: are generated directly as assembler code in this pass; they never 543: exist as RTL. 544: 545: The source files are `final.c' plus `insn-output.c'; the latter is 546: generated automatically from the machine description by the tool 547: `genoutput'. The header file `conditions.h' is used for 548: communication between these files. 549: 550: * Debugging information output. This is run after final because it 551: must output the stack slot offsets for pseudo registers that did 552: not get hard registers. Source files are `dbxout.c' for DBX 553: symbol table format, `sdbout.c' for SDB symbol table format, and 554: `dwarfout.c' for DWARF symbol table format. 555: 556: Some additional files are used by all or many passes: 557: 558: * Every pass uses `machmode.def' and `machmode.h' which define the 559: machine modes. 560: 561: * Several passes use `real.h', which defines the default 562: representation of floating point constants and how to operate on 563: them. 564: 565: * All the passes that work with RTL use the header files `rtl.h' and 566: `rtl.def', and subroutines in file `rtl.c'. The tools `gen*' also 567: use these files to read and work with the machine description RTL. 568: 569: * Several passes refer to the header file `insn-config.h' which 570: contains a few parameters (C macro definitions) generated 571: automatically from the machine description RTL by the tool 572: `genconfig'. 573: 574: * Several passes use the instruction recognizer, which consists of 575: `recog.c' and `recog.h', plus the files `insn-recog.c' and 576: `insn-extract.c' that are generated automatically from the machine 577: description by the tools `genrecog' and `genextract'. 578: 579: * Several passes use the header files `regs.h' which defines the 580: information recorded about pseudo register usage, and 581: `basic-block.h' which defines the information recorded about basic 582: blocks. 583: 584: * `hard-reg-set.h' defines the type `HARD_REG_SET', a bit-vector 585: with a bit for each hard register, and some macros to manipulate 586: it. This type is just `int' if the machine has few enough hard 587: registers; otherwise it is an array of `int' and some of the 588: macros expand into loops. 589: 590: * Several passes use instruction attributes. A definition of the 591: attributes defined for a particular machine is in file 592: `insn-attr.h', which is generated from the machine description by 593: the program `genattr'. The file `insn-attrtab.c' contains 594: subroutines to obtain the attribute values for insns. It is 595: generated from the machine description by the program `genattrtab'. 1.1.1.6 root 596: 597: 1.1.1.8 root 598: File: gcc.info, Node: RTL, Next: Machine Desc, Prev: Passes, Up: Top 1.1.1.6 root 599: 1.1.1.8 root 600: RTL Representation 601: ****************** 1.1.1.6 root 602: 1.1.1.8 root 603: Most of the work of the compiler is done on an intermediate 604: representation called register transfer language. In this language, 605: the instructions to be output are described, pretty much one by one, in 606: an algebraic form that describes what the instruction does. 607: 608: RTL is inspired by Lisp lists. It has both an internal form, made 609: up of structures that point at other structures, and a textual form 610: that is used in the machine description and in printed debugging dumps. 611: The textual form uses nested parentheses to indicate the pointers in 612: the internal form. 613: 614: * Menu: 615: 616: * RTL Objects:: Expressions vs vectors vs strings vs integers. 617: * Accessors:: Macros to access expression operands or vector elts. 618: * Flags:: Other flags in an RTL expression. 619: * Machine Modes:: Describing the size and format of a datum. 620: * Constants:: Expressions with constant values. 621: * Regs and Memory:: Expressions representing register contents or memory. 622: * Arithmetic:: Expressions representing arithmetic on other expressions. 623: * Comparisons:: Expressions representing comparison of expressions. 624: * Bit Fields:: Expressions representing bitfields in memory or reg. 625: * Conversions:: Extending, truncating, floating or fixing. 626: * RTL Declarations:: Declaring volatility, constancy, etc. 627: * Side Effects:: Expressions for storing in registers, etc. 628: * Incdec:: Embedded side-effects for autoincrement addressing. 629: * Assembler:: Representing `asm' with operands. 630: * Insns:: Expression types for entire insns. 631: * Calls:: RTL representation of function call insns. 632: * Sharing:: Some expressions are unique; others *must* be copied. 633: * Reading RTL:: Reading textual RTL from a file. 1.1.1.6 root 634: 635: 1.1.1.8 root 636: File: gcc.info, Node: RTL Objects, Next: Accessors, Prev: RTL, Up: RTL 1.1.1.6 root 637: 1.1.1.8 root 638: RTL Object Types 639: ================ 1.1.1.6 root 640: 1.1.1.8 root 641: RTL uses five kinds of objects: expressions, integers, wide integers, 642: strings and vectors. Expressions are the most important ones. An RTL 643: expression ("RTX", for short) is a C structure, but it is usually 644: referred to with a pointer; a type that is given the typedef name `rtx'. 645: 646: An integer is simply an `int'; their written form uses decimal 647: digits. A wide integer is an integral object whose type is 648: `HOST_WIDE_INT' (*note Config::.); their written form uses decimal 649: digits. 650: 651: A string is a sequence of characters. In core it is represented as a 652: `char *' in usual C fashion, and it is written in C syntax as well. 653: However, strings in RTL may never be null. If you write an empty 654: string in a machine description, it is represented in core as a null 655: pointer rather than as a pointer to a null character. In certain 656: contexts, these null pointers instead of strings are valid. Within RTL 657: code, strings are most commonly found inside `symbol_ref' expressions, 658: but they appear in other contexts in the RTL expressions that make up 659: machine descriptions. 660: 661: A vector contains an arbitrary number of pointers to expressions. 662: The number of elements in the vector is explicitly present in the 663: vector. The written form of a vector consists of square brackets 664: (`[...]') surrounding the elements, in sequence and with whitespace 665: separating them. Vectors of length zero are not created; null pointers 666: are used instead. 667: 668: Expressions are classified by "expression codes" (also called RTX 669: codes). The expression code is a name defined in `rtl.def', which is 670: also (in upper case) a C enumeration constant. The possible expression 671: codes and their meanings are machine-independent. The code of an RTX 672: can be extracted with the macro `GET_CODE (X)' and altered with 673: `PUT_CODE (X, NEWCODE)'. 674: 675: The expression code determines how many operands the expression 676: contains, and what kinds of objects they are. In RTL, unlike Lisp, you 677: cannot tell by looking at an operand what kind of object it is. 678: Instead, you must know from its context--from the expression code of 679: the containing expression. For example, in an expression of code 680: `subreg', the first operand is to be regarded as an expression and the 681: second operand as an integer. In an expression of code `plus', there 682: are two operands, both of which are to be regarded as expressions. In 683: a `symbol_ref' expression, there is one operand, which is to be 684: regarded as a string. 685: 686: Expressions are written as parentheses containing the name of the 687: expression type, its flags and machine mode if any, and then the 688: operands of the expression (separated by spaces). 689: 690: Expression code names in the `md' file are written in lower case, 691: but when they appear in C code they are written in upper case. In this 692: manual, they are shown as follows: `const_int'. 1.1.1.6 root 693: 1.1.1.8 root 694: In a few contexts a null pointer is valid where an expression is 695: normally wanted. The written form of this is `(nil)'. 1.1.1.5 root 696: 1.1.1.8 root 697: 698: File: gcc.info, Node: Accessors, Next: Flags, Prev: RTL Objects, Up: RTL 1.1.1.5 root 699: 1.1.1.8 root 700: Access to Operands 701: ================== 1.1.1.5 root 702: 1.1.1.8 root 703: For each expression type `rtl.def' specifies the number of contained 704: objects and their kinds, with four possibilities: `e' for expression 705: (actually a pointer to an expression), `i' for integer, `w' for wide 706: integer, `s' for string, and `E' for vector of expressions. The 707: sequence of letters for an expression code is called its "format". 708: Thus, the format of `subreg' is `ei'. 709: 710: A few other format characters are used occasionally: 711: 712: `u' 713: `u' is equivalent to `e' except that it is printed differently in 714: debugging dumps. It is used for pointers to insns. 715: 716: `n' 717: `n' is equivalent to `i' except that it is printed differently in 718: debugging dumps. It is used for the line number or code number of 719: a `note' insn. 720: 721: `S' 722: `S' indicates a string which is optional. In the RTL objects in 723: core, `S' is equivalent to `s', but when the object is read, from 724: an `md' file, the string value of this operand may be omitted. An 725: omitted string is taken to be the null string. 726: 727: `V' 728: `V' indicates a vector which is optional. In the RTL objects in 729: core, `V' is equivalent to `E', but when the object is read from 730: an `md' file, the vector value of this operand may be omitted. An 731: omitted vector is effectively the same as a vector of no elements. 732: 733: `0' 734: `0' means a slot whose contents do not fit any normal category. 735: `0' slots are not printed at all in dumps, and are often used in 736: special ways by small parts of the compiler. 737: 738: There are macros to get the number of operands, the format, and the 739: class of an expression code: 740: 741: `GET_RTX_LENGTH (CODE)' 742: Number of operands of an RTX of code CODE. 743: 744: `GET_RTX_FORMAT (CODE)' 745: The format of an RTX of code CODE, as a C string. 746: 747: `GET_RTX_CLASS (CODE)' 748: A single character representing the type of RTX operation that code 749: CODE performs. 750: 751: The following classes are defined: 752: 753: `o' 754: An RTX code that represents an actual object, such as `reg' or 755: `mem'. `subreg' is not in this class. 756: 757: `<' 758: An RTX code for a comparison. The codes in this class are 759: `NE', `EQ', `LE', `LT', `GE', `GT', `LEU', `LTU', `GEU', 760: `GTU'. 761: 762: `1' 763: An RTX code for a unary arithmetic operation, such as `neg'. 764: 765: `c' 766: An RTX code for a commutative binary operation, other than 767: `NE' and `EQ' (which have class `<'). 768: 769: `2' 770: An RTX code for a noncommutative binary operation, such as 771: `MINUS'. 772: 773: `b' 774: An RTX code for a bitfield operation, either `ZERO_EXTRACT' or 775: `SIGN_EXTRACT'. 776: 777: `3' 778: An RTX code for other three input operations, such as 779: `IF_THEN_ELSE'. 780: 781: `i' 782: An RTX code for a machine insn (`INSN', `JUMP_INSN', and 783: `CALL_INSN'). 784: 785: `m' 786: An RTX code for something that matches in insns, such as 787: `MATCH_DUP'. 788: 789: `x' 790: All other RTX codes. 791: 792: Operands of expressions are accessed using the macros `XEXP', 793: `XINT', `XWINT' and `XSTR'. Each of these macros takes two arguments: 794: an expression-pointer (RTX) and an operand number (counting from zero). 795: Thus, 796: 797: XEXP (X, 2) 798: 799: accesses operand 2 of expression X, as an expression. 800: 801: XINT (X, 2) 802: 803: accesses the same operand as an integer. `XSTR', used in the same 804: fashion, would access it as a string. 805: 806: Any operand can be accessed as an integer, as an expression or as a 807: string. You must choose the correct method of access for the kind of 808: value actually stored in the operand. You would do this based on the 809: expression code of the containing expression. That is also how you 810: would know how many operands there are. 811: 812: For example, if X is a `subreg' expression, you know that it has two 813: operands which can be correctly accessed as `XEXP (X, 0)' and `XINT (X, 814: 1)'. If you did `XINT (X, 0)', you would get the address of the 815: expression operand but cast as an integer; that might occasionally be 816: useful, but it would be cleaner to write `(int) XEXP (X, 0)'. `XEXP 817: (X, 1)' would also compile without error, and would return the second, 818: integer operand cast as an expression pointer, which would probably 819: result in a crash when accessed. Nothing stops you from writing `XEXP 820: (X, 28)' either, but this will access memory past the end of the 821: expression with unpredictable results. 822: 823: Access to operands which are vectors is more complicated. You can 824: use the macro `XVEC' to get the vector-pointer itself, or the macros 825: `XVECEXP' and `XVECLEN' to access the elements and length of a vector. 826: 827: `XVEC (EXP, IDX)' 828: Access the vector-pointer which is operand number IDX in EXP. 829: 830: `XVECLEN (EXP, IDX)' 831: Access the length (number of elements) in the vector which is in 832: operand number IDX in EXP. This value is an `int'. 833: 834: `XVECEXP (EXP, IDX, ELTNUM)' 835: Access element number ELTNUM in the vector which is in operand 836: number IDX in EXP. This value is an RTX. 837: 838: It is up to you to make sure that ELTNUM is not negative and is 839: less than `XVECLEN (EXP, IDX)'. 840: 841: All the macros defined in this section expand into lvalues and 842: therefore can be used to assign the operands, lengths and vector 843: elements as well as to access them. 1.1.1.5 root 844: 845: 1.1.1.8 root 846: File: gcc.info, Node: Flags, Next: Machine Modes, Prev: Accessors, Up: RTL 1.1.1.5 root 847: 1.1.1.8 root 848: Flags in an RTL Expression 849: ========================== 1.1.1.5 root 850: 1.1.1.8 root 851: RTL expressions contain several flags (one-bit bitfields) that are 852: used in certain types of expression. Most often they are accessed with 853: the following macros: 854: 855: `MEM_VOLATILE_P (X)' 856: In `mem' expressions, nonzero for volatile memory references. 857: Stored in the `volatil' field and printed as `/v'. 858: 859: `MEM_IN_STRUCT_P (X)' 860: In `mem' expressions, nonzero for reference to an entire 861: structure, union or array, or to a component of one. Zero for 862: references to a scalar variable or through a pointer to a scalar. 863: Stored in the `in_struct' field and printed as `/s'. 864: 865: `REG_LOOP_TEST_P' 866: In `reg' expressions, nonzero if this register's entire life is 867: contained in the exit test code for some loop. Stored in the 868: `in_struct' field and printed as `/s'. 869: 870: `REG_USERVAR_P (X)' 871: In a `reg', nonzero if it corresponds to a variable present in the 872: user's source code. Zero for temporaries generated internally by 873: the compiler. Stored in the `volatil' field and printed as `/v'. 874: 875: `REG_FUNCTION_VALUE_P (X)' 876: Nonzero in a `reg' if it is the place in which this function's 877: value is going to be returned. (This happens only in a hard 878: register.) Stored in the `integrated' field and printed as `/i'. 879: 880: The same hard register may be used also for collecting the values 881: of functions called by this one, but `REG_FUNCTION_VALUE_P' is zero 882: in this kind of use. 883: 884: `SUBREG_PROMOTED_VAR_P' 885: Nonzero in a `subreg' if it was made when accessing an object that 886: was promoted to a wider mode in accord with the `PROMOTED_MODE' 887: machine description macro (*note Storage Layout::.). In this 888: case, the mode of the `subreg' is the declared mode of the object 889: and the mode of `SUBREG_REG' is the mode of the register that 890: holds the object. Promoted variables are always either sign- or 891: zero-extended to the wider mode on every assignment. Stored in 892: the `in_struct' field and printed as `/s'. 893: 894: `SUBREG_PROMOTED_UNSIGNED_P' 895: Nonzero in a `subreg' that has `SUBREG_PROMOTED_VAR_P' nonzero if 896: the object being referenced is kept zero-extended and zero if it 897: is kept sign-extended. Stored in the `unchanging' field and 898: printed as `/u'. 899: 900: `RTX_UNCHANGING_P (X)' 901: Nonzero in a `reg' or `mem' if the value is not changed. (This 902: flag is not set for memory references via pointers to constants. 903: Such pointers only guarantee that the object will not be changed 904: explicitly by the current function. The object might be changed by 905: other functions or by aliasing.) Stored in the `unchanging' field 906: and printed as `/u'. 907: 908: `RTX_INTEGRATED_P (INSN)' 909: Nonzero in an insn if it resulted from an in-line function call. 910: Stored in the `integrated' field and printed as `/i'. This may be 911: deleted; nothing currently depends on it. 912: 913: `SYMBOL_REF_USED (X)' 914: In a `symbol_ref', indicates that X has been used. This is 915: normally only used to ensure that X is only declared external 916: once. Stored in the `used' field. 917: 918: `SYMBOL_REF_FLAG (X)' 919: In a `symbol_ref', this is used as a flag for machine-specific 920: purposes. Stored in the `volatil' field and printed as `/v'. 921: 922: `LABEL_OUTSIDE_LOOP_P' 923: In `label_ref' expressions, nonzero if this is a reference to a 924: label that is outside the innermost loop containing the reference 925: to the label. Stored in the `in_struct' field and printed as `/s'. 926: 927: `INSN_DELETED_P (INSN)' 928: In an insn, nonzero if the insn has been deleted. Stored in the 929: `volatil' field and printed as `/v'. 930: 931: `INSN_ANNULLED_BRANCH_P (INSN)' 932: In an `insn' in the delay slot of a branch insn, indicates that an 933: annulling branch should be used. See the discussion under 934: `sequence' below. Stored in the `unchanging' field and printed as 935: `/u'. 936: 937: `INSN_FROM_TARGET_P (INSN)' 938: In an `insn' in a delay slot of a branch, indicates that the insn 939: is from the target of the branch. If the branch insn has 940: `INSN_ANNULLED_BRANCH_P' set, this insn should only be executed if 941: the branch is taken. For annulled branches with this bit clear, 942: the insn should be executed only if the branch is not taken. 943: Stored in the `in_struct' field and printed as `/s'. 944: 945: `CONSTANT_POOL_ADDRESS_P (X)' 946: Nonzero in a `symbol_ref' if it refers to part of the current 947: function's "constants pool". These are addresses close to the 948: beginning of the function, and GNU CC assumes they can be addressed 949: directly (perhaps with the help of base registers). Stored in the 950: `unchanging' field and printed as `/u'. 951: 952: `CONST_CALL_P (X)' 953: In a `call_insn', indicates that the insn represents a call to a 954: const function. Stored in the `unchanging' field and printed as 955: `/u'. 956: 957: `LABEL_PRESERVE_P (X)' 958: In a `code_label', indicates that the label can never be deleted. 959: Labels referenced by a non-local goto will have this bit set. 960: Stored in the `in_struct' field and printed as `/s'. 961: 962: `SCHED_GROUP_P (INSN)' 963: During instruction scheduling, in an insn, indicates that the 964: previous insn must be scheduled together with this insn. This is 965: used to ensure that certain groups of instructions will not be 966: split up by the instruction scheduling pass, for example, `use' 967: insns before a `call_insn' may not be separated from the 968: `call_insn'. Stored in the `in_struct' field and printed as `/s'. 969: 970: These are the fields which the above macros refer to: 971: 972: `used' 973: Normally, this flag is used only momentarily, at the end of RTL 974: generation for a function, to count the number of times an 975: expression appears in insns. Expressions that appear more than 976: once are copied, according to the rules for shared structure 977: (*note Sharing::.). 978: 979: In a `symbol_ref', it indicates that an external declaration for 980: the symbol has already been written. 981: 982: In a `reg', it is used by the leaf register renumbering code to 983: ensure that each register is only renumbered once. 984: 985: `volatil' 986: This flag is used in `mem', `symbol_ref' and `reg' expressions and 987: in insns. In RTL dump files, it is printed as `/v'. 988: 989: In a `mem' expression, it is 1 if the memory reference is volatile. 990: Volatile memory references may not be deleted, reordered or 991: combined. 992: 993: In a `symbol_ref' expression, it is used for machine-specific 994: purposes. 995: 996: In a `reg' expression, it is 1 if the value is a user-level 997: variable. 0 indicates an internal compiler temporary. 998: 999: In an insn, 1 means the insn has been deleted. 1000: 1001: `in_struct' 1002: In `mem' expressions, it is 1 if the memory datum referred to is 1003: all or part of a structure or array; 0 if it is (or might be) a 1004: scalar variable. A reference through a C pointer has 0 because 1005: the pointer might point to a scalar variable. This information 1006: allows the compiler to determine something about possible cases of 1007: aliasing. 1008: 1009: In an insn in the delay slot of a branch, 1 means that this insn 1010: is from the target of the branch. 1011: 1012: During instruction scheduling, in an insn, 1 means that this insn 1013: must be scheduled as part of a group together with the previous 1014: insn. 1015: 1016: In `reg' expressions, it is 1 if the register has its entire life 1017: contained within the test expression of some loop. 1018: 1019: In `subreg' expressions, 1 means that the `subreg' is accessing an 1020: object that has had its mode promoted from a wider mode. 1021: 1022: In `label_ref' expressions, 1 means that the referenced label is 1023: outside the innermost loop containing the insn in which the 1024: `label_ref' was found. 1025: 1026: In `code_label' expressions, it is 1 if the label may never be 1027: deleted. This is used for labels which are the target of 1028: non-local gotos. 1029: 1030: In an RTL dump, this flag is represented as `/s'. 1031: 1032: `unchanging' 1033: In `reg' and `mem' expressions, 1 means that the value of the 1034: expression never changes. 1035: 1036: In `subreg' expressions, it is 1 if the `subreg' references an 1037: unsigned object whose mode has been promoted to a wider mode. 1038: 1039: In an insn, 1 means that this is an annulling branch. 1040: 1041: In a `symbol_ref' expression, 1 means that this symbol addresses 1042: something in the per-function constants pool. 1043: 1044: In a `call_insn', 1 means that this instruction is a call to a 1045: const function. 1046: 1047: In an RTL dump, this flag is represented as `/u'. 1048: 1049: `integrated' 1050: In some kinds of expressions, including insns, this flag means the 1051: rtl was produced by procedure integration. 1052: 1053: In a `reg' expression, this flag indicates the register containing 1054: the value to be returned by the current function. On machines 1055: that pass parameters in registers, the same register number may be 1056: used for parameters as well, but this flag is not set on such uses. 1.1 root 1057:
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.