![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.info-20 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-20 CVS log | Up to [GCC 2.x] / gcc |
1.1.1.5 ! 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: 1.1.1.4 root 6: Published by the Free Software Foundation 59 Temple Place - Suite 330 7: Boston, MA 02111-1307 USA 1.1 root 8: 1.1.1.4 root 9: Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995 Free Software 10: Foundation, Inc. 1.1 root 11: 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. 15: 16: Permission is granted to copy and distribute modified versions of 17: this manual under the conditions for verbatim copying, provided also 1.1.1.3 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 26: versions, except that the sections entitled "GNU General Public 1.1.1.3 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 root 31: 32: 1.1.1.4 root 33: File: gcc.info, Node: Leaf Functions, Next: Stack Registers, Prev: Values in Registers, Up: Registers 1.1 root 34: 1.1.1.4 root 35: Handling Leaf Functions 36: ----------------------- 1.1.1.2 root 37: 1.1.1.4 root 38: On some machines, a leaf function (i.e., one which makes no calls) 39: can run more efficiently if it does not make its own register window. 40: Often this means it is required to receive its arguments in the 41: registers where they are passed by the caller, instead of the registers 42: where they would normally arrive. 43: 44: The special treatment for leaf functions generally applies only when 45: other conditions are met; for example, often they may use only those 46: registers for its own variables and temporaries. We use the term "leaf 47: function" to mean a function that is suitable for this special 48: handling, so that functions with no calls are not necessarily "leaf 49: functions". 50: 51: GNU CC assigns register numbers before it knows whether the function 52: is suitable for leaf function treatment. So it needs to renumber the 53: registers in order to output a leaf function. The following macros 54: accomplish this. 55: 56: `LEAF_REGISTERS' 57: A C initializer for a vector, indexed by hard register number, 58: which contains 1 for a register that is allowable in a candidate 59: for leaf function treatment. 60: 61: If leaf function treatment involves renumbering the registers, 62: then the registers marked here should be the ones before 63: renumbering--those that GNU CC would ordinarily allocate. The 64: registers which will actually be used in the assembler code, after 65: renumbering, should not be marked with 1 in this vector. 66: 67: Define this macro only if the target machine offers a way to 68: optimize the treatment of leaf functions. 69: 70: `LEAF_REG_REMAP (REGNO)' 71: A C expression whose value is the register number to which REGNO 72: should be renumbered, when a function is treated as a leaf 73: function. 74: 75: If REGNO is a register number which should not appear in a leaf 76: function before renumbering, then the expression should yield -1, 77: which will cause the compiler to abort. 78: 79: Define this macro only if the target machine offers a way to 80: optimize the treatment of leaf functions, and registers need to be 81: renumbered to do this. 82: 83: Normally, `FUNCTION_PROLOGUE' and `FUNCTION_EPILOGUE' must treat 84: leaf functions specially. It can test the C variable `leaf_function' 85: which is nonzero for leaf functions. (The variable `leaf_function' is 86: defined only if `LEAF_REGISTERS' is defined.) 1.1 root 87: 88: 1.1.1.4 root 89: File: gcc.info, Node: Stack Registers, Next: Obsolete Register Macros, Prev: Leaf Functions, Up: Registers 90: 91: Registers That Form a Stack 92: --------------------------- 93: 94: There are special features to handle computers where some of the 95: "registers" form a stack, as in the 80387 coprocessor for the 80386. 96: Stack registers are normally written by pushing onto the stack, and are 97: numbered relative to the top of the stack. 1.1 root 98: 1.1.1.4 root 99: Currently, GNU CC can only handle one group of stack-like registers, 100: and they must be consecutively numbered. 1.1 root 101: 1.1.1.4 root 102: `STACK_REGS' 103: Define this if the machine has any stack-like registers. 1.1 root 104: 1.1.1.4 root 105: `FIRST_STACK_REG' 106: The number of the first stack-like register. This one is the top 107: of the stack. 1.1.1.3 root 108: 1.1.1.4 root 109: `LAST_STACK_REG' 110: The number of the last stack-like register. This one is the 111: bottom of the stack. 1.1 root 112: 113: 1.1.1.4 root 114: File: gcc.info, Node: Obsolete Register Macros, Prev: Stack Registers, Up: Registers 1.1.1.3 root 115: 1.1.1.4 root 116: Obsolete Macros for Controlling Register Usage 117: ---------------------------------------------- 1.1 root 118: 1.1.1.4 root 119: These features do not work very well. They exist because they used 120: to be required to generate correct code for the 80387 coprocessor of the 121: 80386. They are no longer used by that machine description and may be 122: removed in a later version of the compiler. Don't use them! 123: 124: `OVERLAPPING_REGNO_P (REGNO)' 125: If defined, this is a C expression whose value is nonzero if hard 126: register number REGNO is an overlapping register. This means a 127: hard register which overlaps a hard register with a different 128: number. (Such overlap is undesirable, but occasionally it allows 129: a machine to be supported which otherwise could not be.) This 130: macro must return nonzero for *all* the registers which overlap 131: each other. GNU CC can use an overlapping register only in 132: certain limited ways. It can be used for allocation within a 133: basic block, and may be spilled for reloading; that is all. 134: 135: If this macro is not defined, it means that none of the hard 136: registers overlap each other. This is the usual situation. 137: 138: `INSN_CLOBBERS_REGNO_P (INSN, REGNO)' 139: If defined, this is a C expression whose value should be nonzero if 140: the insn INSN has the effect of mysteriously clobbering the 141: contents of hard register number REGNO. By "mysterious" we mean 142: that the insn's RTL expression doesn't describe such an effect. 143: 144: If this macro is not defined, it means that no insn clobbers 145: registers mysteriously. This is the usual situation; all else 146: being equal, it is best for the RTL expression to show all the 147: activity. 148: 149: `PRESERVE_DEATH_INFO_REGNO_P (REGNO)' 150: If defined, this is a C expression whose value is nonzero if 151: accurate `REG_DEAD' notes are needed for hard register number REGNO 152: at the time of outputting the assembler code. When this is so, a 153: few optimizations that take place after register allocation and 154: could invalidate the death notes are not done when this register is 155: involved. 156: 157: You would arrange to preserve death info for a register when some 158: of the code in the machine description which is executed to write 159: the assembler code looks at the death notes. This is necessary 160: only when the actual hardware feature which GNU CC thinks of as a 161: register is not actually a register of the usual sort. (It might, 162: for example, be a hardware stack.) 1.1 root 163: 1.1.1.4 root 164: If this macro is not defined, it means that no death notes need to 165: be preserved. This is the usual situation. 1.1 root 166: 1.1.1.2 root 167: 1.1.1.4 root 168: File: gcc.info, Node: Register Classes, Next: Stack and Calling, Prev: Registers, Up: Target Macros 1.1 root 169: 1.1.1.4 root 170: Register Classes 171: ================ 1.1 root 172: 1.1.1.4 root 173: On many machines, the numbered registers are not all equivalent. 174: For example, certain registers may not be allowed for indexed 175: addressing; certain registers may not be allowed in some instructions. 176: These machine restrictions are described to the compiler using 177: "register classes". 178: 179: You define a number of register classes, giving each one a name and 180: saying which of the registers belong to it. Then you can specify 181: register classes that are allowed as operands to particular instruction 182: patterns. 183: 184: In general, each register will belong to several classes. In fact, 185: one class must be named `ALL_REGS' and contain all the registers. 186: Another class must be named `NO_REGS' and contain no registers. Often 187: the union of two classes will be another class; however, this is not 188: required. 189: 190: One of the classes must be named `GENERAL_REGS'. There is nothing 191: terribly special about the name, but the operand constraint letters `r' 192: and `g' specify this class. If `GENERAL_REGS' is the same as 193: `ALL_REGS', just define it as a macro which expands to `ALL_REGS'. 194: 195: Order the classes so that if class X is contained in class Y then X 196: has a lower class number than Y. 197: 198: The way classes other than `GENERAL_REGS' are specified in operand 199: constraints is through machine-dependent operand constraint letters. 200: You can define such letters to correspond to various classes, then use 201: them in operand constraints. 202: 203: You should define a class for the union of two classes whenever some 204: instruction allows both classes. For example, if an instruction allows 205: either a floating point (coprocessor) register or a general register 206: for a certain operand, you should define a class `FLOAT_OR_GENERAL_REGS' 207: which includes both of them. Otherwise you will get suboptimal code. 208: 209: You must also specify certain redundant information about the 210: register classes: for each class, which classes contain it and which 211: ones are contained in it; for each pair of classes, the largest class 212: contained in their union. 213: 214: When a value occupying several consecutive registers is expected in a 215: certain class, all the registers used must belong to that class. 216: Therefore, register classes cannot be used to enforce a requirement for 217: a register pair to start with an even-numbered register. The way to 218: specify this requirement is with `HARD_REGNO_MODE_OK'. 219: 220: Register classes used for input-operands of bitwise-and or shift 221: instructions have a special requirement: each such class must have, for 222: each fixed-point machine mode, a subclass whose registers can transfer 223: that mode to or from memory. For example, on some machines, the 224: operations for single-byte values (`QImode') are limited to certain 225: registers. When this is so, each register class that is used in a 226: bitwise-and or shift instruction must have a subclass consisting of 227: registers from which single-byte values can be loaded or stored. This 228: is so that `PREFERRED_RELOAD_CLASS' can always have a possible value to 229: return. 230: 231: `enum reg_class' 232: An enumeral type that must be defined with all the register class 233: names as enumeral values. `NO_REGS' must be first. `ALL_REGS' 234: must be the last register class, followed by one more enumeral 235: value, `LIM_REG_CLASSES', which is not a register class but rather 236: tells how many classes there are. 237: 238: Each register class has a number, which is the value of casting 239: the class name to type `int'. The number serves as an index in 240: many of the tables described below. 241: 242: `N_REG_CLASSES' 243: The number of distinct register classes, defined as follows: 244: 245: #define N_REG_CLASSES (int) LIM_REG_CLASSES 246: 247: `REG_CLASS_NAMES' 248: An initializer containing the names of the register classes as C 249: string constants. These names are used in writing some of the 250: debugging dumps. 251: 252: `REG_CLASS_CONTENTS' 253: An initializer containing the contents of the register classes, as 254: integers which are bit masks. The Nth integer specifies the 255: contents of class N. The way the integer MASK is interpreted is 256: that register R is in the class if `MASK & (1 << R)' is 1. 257: 258: When the machine has more than 32 registers, an integer does not 259: suffice. Then the integers are replaced by sub-initializers, 260: braced groupings containing several integers. Each 261: sub-initializer must be suitable as an initializer for the type 262: `HARD_REG_SET' which is defined in `hard-reg-set.h'. 263: 264: `REGNO_REG_CLASS (REGNO)' 265: A C expression whose value is a register class containing hard 266: register REGNO. In general there is more than one such class; 267: choose a class which is "minimal", meaning that no smaller class 268: also contains the register. 269: 270: `BASE_REG_CLASS' 271: A macro whose definition is the name of the class to which a valid 272: base register must belong. A base register is one used in an 273: address which is the register value plus a displacement. 274: 275: `INDEX_REG_CLASS' 276: A macro whose definition is the name of the class to which a valid 277: index register must belong. An index register is one used in an 278: address where its value is either multiplied by a scale factor or 279: added to another register (as well as added to a displacement). 280: 281: `REG_CLASS_FROM_LETTER (CHAR)' 282: A C expression which defines the machine-dependent operand 283: constraint letters for register classes. If CHAR is such a 284: letter, the value should be the register class corresponding to 285: it. Otherwise, the value should be `NO_REGS'. The register 286: letter `r', corresponding to class `GENERAL_REGS', will not be 287: passed to this macro; you do not need to handle it. 288: 289: `REGNO_OK_FOR_BASE_P (NUM)' 290: A C expression which is nonzero if register number NUM is suitable 291: for use as a base register in operand addresses. It may be either 292: a suitable hard register or a pseudo register that has been 293: allocated such a hard register. 294: 295: `REGNO_OK_FOR_INDEX_P (NUM)' 296: A C expression which is nonzero if register number NUM is suitable 297: for use as an index register in operand addresses. It may be 298: either a suitable hard register or a pseudo register that has been 299: allocated such a hard register. 300: 301: The difference between an index register and a base register is 302: that the index register may be scaled. If an address involves the 303: sum of two registers, neither one of them scaled, then either one 304: may be labeled the "base" and the other the "index"; but whichever 305: labeling is used must fit the machine's constraints of which 306: registers may serve in each capacity. The compiler will try both 307: labelings, looking for one that is valid, and will reload one or 308: both registers only if neither labeling works. 1.1 root 309: 1.1.1.4 root 310: `PREFERRED_RELOAD_CLASS (X, CLASS)' 311: A C expression that places additional restrictions on the register 312: class to use when it is necessary to copy value X into a register 313: in class CLASS. The value is a register class; perhaps CLASS, or 314: perhaps another, smaller class. On many machines, the following 315: definition is safe: 316: 317: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 318: 319: Sometimes returning a more restrictive class makes better code. 320: For example, on the 68000, when X is an integer constant that is 321: in range for a `moveq' instruction, the value of this macro is 322: always `DATA_REGS' as long as CLASS includes the data registers. 323: Requiring a data register guarantees that a `moveq' will be used. 324: 325: If X is a `const_double', by returning `NO_REGS' you can force X 326: into a memory constant. This is useful on certain machines where 327: immediate floating values cannot be loaded into certain kinds of 328: registers. 1.1 root 329: 1.1.1.4 root 330: `PREFERRED_OUTPUT_RELOAD_CLASS (X, CLASS)' 331: Like `PREFERRED_RELOAD_CLASS', but for output reloads instead of 332: input reloads. If you don't define this macro, the default is to 333: use CLASS, unchanged. 334: 335: `LIMIT_RELOAD_CLASS (MODE, CLASS)' 336: A C expression that places additional restrictions on the register 337: class to use when it is necessary to be able to hold a value of 338: mode MODE in a reload register for which class CLASS would 339: ordinarily be used. 340: 341: Unlike `PREFERRED_RELOAD_CLASS', this macro should be used when 342: there are certain modes that simply can't go in certain reload 343: classes. 344: 345: The value is a register class; perhaps CLASS, or perhaps another, 346: smaller class. 347: 348: Don't define this macro unless the target machine has limitations 349: which require the macro to do something nontrivial. 350: 351: `SECONDARY_RELOAD_CLASS (CLASS, MODE, X)' 352: `SECONDARY_INPUT_RELOAD_CLASS (CLASS, MODE, X)' 353: `SECONDARY_OUTPUT_RELOAD_CLASS (CLASS, MODE, X)' 354: Many machines have some registers that cannot be copied directly 355: to or from memory or even from other types of registers. An 356: example is the `MQ' register, which on most machines, can only be 357: copied to or from general registers, but not memory. Some 358: machines allow copying all registers to and from memory, but 359: require a scratch register for stores to some memory locations 360: (e.g., those with symbolic address on the RT, and those with 361: certain symbolic address on the Sparc when compiling PIC). In 362: some cases, both an intermediate and a scratch register are 363: required. 364: 365: You should define these macros to indicate to the reload phase 366: that it may need to allocate at least one register for a reload in 367: addition to the register to contain the data. Specifically, if 368: copying X to a register CLASS in MODE requires an intermediate 369: register, you should define `SECONDARY_INPUT_RELOAD_CLASS' to 370: return the largest register class all of whose registers can be 371: used as intermediate registers or scratch registers. 372: 373: If copying a register CLASS in MODE to X requires an intermediate 374: or scratch register, `SECONDARY_OUTPUT_RELOAD_CLASS' should be 375: defined to return the largest register class required. If the 376: requirements for input and output reloads are the same, the macro 377: `SECONDARY_RELOAD_CLASS' should be used instead of defining both 378: macros identically. 379: 380: The values returned by these macros are often `GENERAL_REGS'. 381: Return `NO_REGS' if no spare register is needed; i.e., if X can be 382: directly copied to or from a register of CLASS in MODE without 383: requiring a scratch register. Do not define this macro if it 384: would always return `NO_REGS'. 385: 386: If a scratch register is required (either with or without an 387: intermediate register), you should define patterns for 388: `reload_inM' or `reload_outM', as required (*note Standard 389: Names::.. These patterns, which will normally be implemented with 390: a `define_expand', should be similar to the `movM' patterns, 391: except that operand 2 is the scratch register. 392: 393: Define constraints for the reload register and scratch register 394: that contain a single register class. If the original reload 395: register (whose class is CLASS) can meet the constraint given in 396: the pattern, the value returned by these macros is used for the 397: class of the scratch register. Otherwise, two additional reload 398: registers are required. Their classes are obtained from the 399: constraints in the insn pattern. 400: 401: X might be a pseudo-register or a `subreg' of a pseudo-register, 402: which could either be in a hard register or in memory. Use 403: `true_regnum' to find out; it will return -1 if the pseudo is in 404: memory and the hard register number if it is in a register. 405: 406: These macros should not be used in the case where a particular 407: class of registers can only be copied to memory and not to another 408: class of registers. In that case, secondary reload registers are 409: not needed and would not be helpful. Instead, a stack location 410: must be used to perform the copy and the `movM' pattern should use 411: memory as a intermediate storage. This case often occurs between 412: floating-point and general registers. 413: 414: `SECONDARY_MEMORY_NEEDED (CLASS1, CLASS2, M)' 415: Certain machines have the property that some registers cannot be 416: copied to some other registers without using memory. Define this 417: macro on those machines to be a C expression that is non-zero if 418: objects of mode M in registers of CLASS1 can only be copied to 419: registers of class CLASS2 by storing a register of CLASS1 into 420: memory and loading that memory location into a register of CLASS2. 421: 422: Do not define this macro if its value would always be zero. 423: 424: `SECONDARY_MEMORY_NEEDED_RTX (MODE)' 425: Normally when `SECONDARY_MEMORY_NEEDED' is defined, the compiler 426: allocates a stack slot for a memory location needed for register 427: copies. If this macro is defined, the compiler instead uses the 428: memory location defined by this macro. 429: 430: Do not define this macro if you do not define 431: `SECONDARY_MEMORY_NEEDED'. 432: 433: `SECONDARY_MEMORY_NEEDED_MODE (MODE)' 434: When the compiler needs a secondary memory location to copy 435: between two registers of mode MODE, it normally allocates 436: sufficient memory to hold a quantity of `BITS_PER_WORD' bits and 437: performs the store and load operations in a mode that many bits 438: wide and whose class is the same as that of MODE. 439: 440: This is right thing to do on most machines because it ensures that 441: all bits of the register are copied and prevents accesses to the 442: registers in a narrower mode, which some machines prohibit for 443: floating-point registers. 444: 445: However, this default behavior is not correct on some machines, 446: such as the DEC Alpha, that store short integers in floating-point 447: registers differently than in integer registers. On those 448: machines, the default widening will not work correctly and you 449: must define this macro to suppress that widening in some cases. 450: See the file `alpha.h' for details. 451: 452: Do not define this macro if you do not define 453: `SECONDARY_MEMORY_NEEDED' or if widening MODE to a mode that is 454: `BITS_PER_WORD' bits wide is correct for your machine. 455: 456: `SMALL_REGISTER_CLASSES' 457: Normally the compiler avoids choosing registers that have been 458: explicitly mentioned in the rtl as spill registers (these 459: registers are normally those used to pass parameters and return 460: values). However, some machines have so few registers of certain 461: classes that there would not be enough registers to use as spill 462: registers if this were done. 463: 464: Define `SMALL_REGISTER_CLASSES' on these machines. When it is 465: defined, the compiler allows registers explicitly used in the rtl 466: to be used as spill registers but avoids extending the lifetime of 467: these registers. 468: 469: It is always safe to define this macro, but if you unnecessarily 470: define it, you will reduce the amount of optimizations that can be 471: performed in some cases. If you do not define this macro when it 472: is required, the compiler will run out of spill registers and 473: print a fatal error message. For most machines, you should not 474: define this macro. 475: 476: `CLASS_LIKELY_SPILLED_P (CLASS)' 477: A C expression whose value is nonzero if pseudos that have been 478: assigned to registers of class CLASS would likely be spilled 479: because registers of CLASS are needed for spill registers. 480: 481: The default value of this macro returns 1 if CLASS has exactly one 482: register and zero otherwise. On most machines, this default 483: should be used. Only define this macro to some other expression 484: if pseudo allocated by `local-alloc.c' end up in memory because 485: their hard registers were needed for spill registers. If this 486: macro returns nonzero for those classes, those pseudos will only 487: be allocated by `global.c', which knows how to reallocate the 488: pseudo to another register. If there would not be another 489: register available for reallocation, you should not change the 490: definition of this macro since the only effect of such a 491: definition would be to slow down register allocation. 492: 493: `CLASS_MAX_NREGS (CLASS, MODE)' 494: A C expression for the maximum number of consecutive registers of 495: class CLASS needed to hold a value of mode MODE. 496: 497: This is closely related to the macro `HARD_REGNO_NREGS'. In fact, 498: the value of the macro `CLASS_MAX_NREGS (CLASS, MODE)' should be 499: the maximum value of `HARD_REGNO_NREGS (REGNO, MODE)' for all 500: REGNO values in the class CLASS. 501: 502: This macro helps control the handling of multiple-word values in 503: the reload pass. 504: 505: `CLASS_CANNOT_CHANGE_SIZE' 506: If defined, a C expression for a class that contains registers 507: which the compiler must always access in a mode that is the same 508: size as the mode in which it loaded the register. 509: 510: For the example, loading 32-bit integer or floating-point objects 511: into floating-point registers on the Alpha extends them to 64-bits. 512: Therefore loading a 64-bit object and then storing it as a 32-bit 513: object does not store the low-order 32-bits, as would be the case 514: for a normal register. Therefore, `alpha.h' defines this macro as 515: `FLOAT_REGS'. 516: 517: Three other special macros describe which operands fit which 518: constraint letters. 519: 520: `CONST_OK_FOR_LETTER_P (VALUE, C)' 521: A C expression that defines the machine-dependent operand 522: constraint letters that specify particular ranges of integer 523: values. If C is one of those letters, the expression should check 524: that VALUE, an integer, is in the appropriate range and return 1 525: if so, 0 otherwise. If C is not one of those letters, the value 526: should be 0 regardless of VALUE. 527: 528: `CONST_DOUBLE_OK_FOR_LETTER_P (VALUE, C)' 529: A C expression that defines the machine-dependent operand 530: constraint letters that specify particular ranges of 531: `const_double' values. 532: 533: If C is one of those letters, the expression should check that 534: VALUE, an RTX of code `const_double', is in the appropriate range 535: and return 1 if so, 0 otherwise. If C is not one of those 536: letters, the value should be 0 regardless of VALUE. 537: 538: `const_double' is used for all floating-point constants and for 539: `DImode' fixed-point constants. A given letter can accept either 540: or both kinds of values. It can use `GET_MODE' to distinguish 541: between these kinds. 542: 543: `EXTRA_CONSTRAINT (VALUE, C)' 544: A C expression that defines the optional machine-dependent 545: constraint letters that can be used to segregate specific types of 546: operands, usually memory references, for the target machine. 547: Normally this macro will not be defined. If it is required for a 548: particular target machine, it should return 1 if VALUE corresponds 549: to the operand type represented by the constraint letter C. If C 550: is not defined as an extra constraint, the value returned should 551: be 0 regardless of VALUE. 552: 553: For example, on the ROMP, load instructions cannot have their 554: output in r0 if the memory reference contains a symbolic address. 555: Constraint letter `Q' is defined as representing a memory address 556: that does *not* contain a symbolic address. An alternative is 557: specified with a `Q' constraint on the input and `r' on the 558: output. The next alternative specifies `m' on the input and a 559: register class that does not include r0 on the output. 1.1 root 560: 561: 1.1.1.4 root 562: File: gcc.info, Node: Stack and Calling, Next: Varargs, Prev: Register Classes, Up: Target Macros 1.1 root 563: 1.1.1.4 root 564: Stack Layout and Calling Conventions 565: ==================================== 1.1 root 566: 1.1.1.4 root 567: This describes the stack layout and calling conventions. 1.1 root 568: 1.1.1.4 root 569: * Menu: 1.1 root 570: 1.1.1.4 root 571: * Frame Layout:: 572: * Frame Registers:: 573: * Elimination:: 574: * Stack Arguments:: 575: * Register Arguments:: 576: * Scalar Return:: 577: * Aggregate Return:: 578: * Caller Saves:: 579: * Function Entry:: 580: * Profiling:: 1.1 root 581: 1.1.1.4 root 582: 583: File: gcc.info, Node: Frame Layout, Next: Frame Registers, Up: Stack and Calling 1.1 root 584: 1.1.1.4 root 585: Basic Stack Layout 586: ------------------ 1.1.1.3 root 587: 1.1.1.4 root 588: Here is the basic stack layout. 1.1.1.3 root 589: 1.1.1.4 root 590: `STACK_GROWS_DOWNWARD' 591: Define this macro if pushing a word onto the stack moves the stack 592: pointer to a smaller address. 593: 594: When we say, "define this macro if ...," it means that the 595: compiler checks this macro only with `#ifdef' so the precise 596: definition used does not matter. 597: 598: `FRAME_GROWS_DOWNWARD' 599: Define this macro if the addresses of local variable slots are at 600: negative offsets from the frame pointer. 601: 602: `ARGS_GROW_DOWNWARD' 603: Define this macro if successive arguments to a function occupy 604: decreasing addresses on the stack. 605: 606: `STARTING_FRAME_OFFSET' 607: Offset from the frame pointer to the first local variable slot to 608: be allocated. 609: 610: If `FRAME_GROWS_DOWNWARD', find the next slot's offset by 611: subtracting the first slot's length from `STARTING_FRAME_OFFSET'. 612: Otherwise, it is found by adding the length of the first slot to 613: the value `STARTING_FRAME_OFFSET'. 614: 615: `STACK_POINTER_OFFSET' 616: Offset from the stack pointer register to the first location at 617: which outgoing arguments are placed. If not specified, the 618: default value of zero is used. This is the proper value for most 619: machines. 620: 621: If `ARGS_GROW_DOWNWARD', this is the offset to the location above 622: the first location at which outgoing arguments are placed. 623: 624: `FIRST_PARM_OFFSET (FUNDECL)' 625: Offset from the argument pointer register to the first argument's 626: address. On some machines it may depend on the data type of the 627: function. 628: 629: If `ARGS_GROW_DOWNWARD', this is the offset to the location above 630: the first argument's address. 631: 632: `STACK_DYNAMIC_OFFSET (FUNDECL)' 633: Offset from the stack pointer register to an item dynamically 634: allocated on the stack, e.g., by `alloca'. 635: 636: The default value for this macro is `STACK_POINTER_OFFSET' plus the 637: length of the outgoing arguments. The default is correct for most 638: machines. See `function.c' for details. 639: 640: `DYNAMIC_CHAIN_ADDRESS (FRAMEADDR)' 641: A C expression whose value is RTL representing the address in a 642: stack frame where the pointer to the caller's frame is stored. 643: Assume that FRAMEADDR is an RTL expression for the address of the 644: stack frame itself. 645: 646: If you don't define this macro, the default is to return the value 647: of FRAMEADDR--that is, the stack frame address is also the address 648: of the stack word that points to the previous frame. 649: 650: `SETUP_FRAME_ADDRESSES ()' 651: If defined, a C expression that produces the machine-specific code 652: to setup the stack so that arbitrary frames can be accessed. For 653: example, on the Sparc, we must flush all of the register windows 654: to the stack before we can access arbitrary stack frames. This 655: macro will seldom need to be defined. 656: 657: `RETURN_ADDR_RTX (COUNT, FRAMEADDR)' 658: A C expression whose value is RTL representing the value of the 659: return address for the frame COUNT steps up from the current frame. 1.1.1.5 ! root 660: FRAMEADDR is the frame pointer of the COUNT frame, or the frame 1.1.1.4 root 661: pointer of the COUNT - 1 frame if `RETURN_ADDR_IN_PREVIOUS_FRAME' 662: is defined. 663: 664: `RETURN_ADDR_IN_PREVIOUS_FRAME' 665: Define this if the return address of a particular stack frame is 666: accessed from the frame pointer of the previous stack frame. 1.1 root 667: 1.1.1.2 root 668: 1.1.1.4 root 669: File: gcc.info, Node: Frame Registers, Next: Elimination, Prev: Frame Layout, Up: Stack and Calling 1.1 root 670: 1.1.1.4 root 671: Registers That Address the Stack Frame 672: -------------------------------------- 1.1 root 673: 1.1.1.4 root 674: This discusses registers that address the stack frame. 1.1 root 675: 1.1.1.4 root 676: `STACK_POINTER_REGNUM' 677: The register number of the stack pointer register, which must also 678: be a fixed register according to `FIXED_REGISTERS'. On most 679: machines, the hardware determines which register this is. 680: 681: `FRAME_POINTER_REGNUM' 682: The register number of the frame pointer register, which is used to 683: access automatic variables in the stack frame. On some machines, 684: the hardware determines which register this is. On other 685: machines, you can choose any register you wish for this purpose. 686: 687: `HARD_FRAME_POINTER_REGNUM' 688: On some machines the offset between the frame pointer and starting 689: offset of the automatic variables is not known until after register 690: allocation has been done (for example, because the saved registers 691: are between these two locations). On those machines, define 692: `FRAME_POINTER_REGNUM' the number of a special, fixed register to 693: be used internally until the offset is known, and define 694: `HARD_FRAME_POINTER_REGNUM' to be actual the hard register number 695: used for the frame pointer. 696: 697: You should define this macro only in the very rare circumstances 698: when it is not possible to calculate the offset between the frame 699: pointer and the automatic variables until after register 700: allocation has been completed. When this macro is defined, you 701: must also indicate in your definition of `ELIMINABLE_REGS' how to 702: eliminate `FRAME_POINTER_REGNUM' into either 703: `HARD_FRAME_POINTER_REGNUM' or `STACK_POINTER_REGNUM'. 704: 705: Do not define this macro if it would be the same as 706: `FRAME_POINTER_REGNUM'. 707: 708: `ARG_POINTER_REGNUM' 709: The register number of the arg pointer register, which is used to 710: access the function's argument list. On some machines, this is 711: the same as the frame pointer register. On some machines, the 712: hardware determines which register this is. On other machines, 713: you can choose any register you wish for this purpose. If this is 714: not the same register as the frame pointer register, then you must 715: mark it as a fixed register according to `FIXED_REGISTERS', or 716: arrange to be able to eliminate it (*note Elimination::.). 717: 718: `STATIC_CHAIN_REGNUM' 719: `STATIC_CHAIN_INCOMING_REGNUM' 720: Register numbers used for passing a function's static chain 721: pointer. If register windows are used, the register number as 722: seen by the called function is `STATIC_CHAIN_INCOMING_REGNUM', 723: while the register number as seen by the calling function is 724: `STATIC_CHAIN_REGNUM'. If these registers are the same, 725: `STATIC_CHAIN_INCOMING_REGNUM' need not be defined. 726: 727: The static chain register need not be a fixed register. 728: 729: If the static chain is passed in memory, these macros should not be 730: defined; instead, the next two macros should be defined. 731: 732: `STATIC_CHAIN' 733: `STATIC_CHAIN_INCOMING' 734: If the static chain is passed in memory, these macros provide rtx 735: giving `mem' expressions that denote where they are stored. 736: `STATIC_CHAIN' and `STATIC_CHAIN_INCOMING' give the locations as 737: seen by the calling and called functions, respectively. Often the 738: former will be at an offset from the stack pointer and the latter 739: at an offset from the frame pointer. 740: 741: The variables `stack_pointer_rtx', `frame_pointer_rtx', and 742: `arg_pointer_rtx' will have been initialized prior to the use of 743: these macros and should be used to refer to those items. 1.1 root 744: 1.1.1.4 root 745: If the static chain is passed in a register, the two previous 746: macros should be defined instead. 1.1 root 747: 1.1.1.4 root 748: 749: File: gcc.info, Node: Elimination, Next: Stack Arguments, Prev: Frame Registers, Up: Stack and Calling 1.1 root 750: 1.1.1.4 root 751: Eliminating Frame Pointer and Arg Pointer 752: ----------------------------------------- 1.1 root 753: 1.1.1.4 root 754: This is about eliminating the frame pointer and arg pointer. 1.1 root 755: 1.1.1.4 root 756: `FRAME_POINTER_REQUIRED' 757: A C expression which is nonzero if a function must have and use a 758: frame pointer. This expression is evaluated in the reload pass. 759: If its value is nonzero the function will have a frame pointer. 760: 761: The expression can in principle examine the current function and 762: decide according to the facts, but on most machines the constant 0 763: or the constant 1 suffices. Use 0 when the machine allows code to 764: be generated with no frame pointer, and doing so saves some time 765: or space. Use 1 when there is no possible advantage to avoiding a 766: frame pointer. 767: 768: In certain cases, the compiler does not know how to produce valid 769: code without a frame pointer. The compiler recognizes those cases 770: and automatically gives the function a frame pointer regardless of 771: what `FRAME_POINTER_REQUIRED' says. You don't need to worry about 772: them. 773: 774: In a function that does not require a frame pointer, the frame 775: pointer register can be allocated for ordinary usage, unless you 776: mark it as a fixed register. See `FIXED_REGISTERS' for more 777: information. 778: 779: `INITIAL_FRAME_POINTER_OFFSET (DEPTH-VAR)' 780: A C statement to store in the variable DEPTH-VAR the difference 781: between the frame pointer and the stack pointer values immediately 782: after the function prologue. The value would be computed from 783: information such as the result of `get_frame_size ()' and the 784: tables of registers `regs_ever_live' and `call_used_regs'. 785: 786: If `ELIMINABLE_REGS' is defined, this macro will be not be used and 787: need not be defined. Otherwise, it must be defined even if 788: `FRAME_POINTER_REQUIRED' is defined to always be true; in that 789: case, you may set DEPTH-VAR to anything. 790: 791: `ELIMINABLE_REGS' 792: If defined, this macro specifies a table of register pairs used to 793: eliminate unneeded registers that point into the stack frame. If 794: it is not defined, the only elimination attempted by the compiler 795: is to replace references to the frame pointer with references to 796: the stack pointer. 797: 798: The definition of this macro is a list of structure 799: initializations, each of which specifies an original and 800: replacement register. 801: 802: On some machines, the position of the argument pointer is not 803: known until the compilation is completed. In such a case, a 804: separate hard register must be used for the argument pointer. 805: This register can be eliminated by replacing it with either the 806: frame pointer or the argument pointer, depending on whether or not 807: the frame pointer has been eliminated. 808: 809: In this case, you might specify: 810: #define ELIMINABLE_REGS \ 811: {{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \ 812: {ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM}, \ 813: {FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}} 814: 815: Note that the elimination of the argument pointer with the stack 816: pointer is specified first since that is the preferred elimination. 817: 818: `CAN_ELIMINATE (FROM-REG, TO-REG)' 819: A C expression that returns non-zero if the compiler is allowed to 820: try to replace register number FROM-REG with register number 821: TO-REG. This macro need only be defined if `ELIMINABLE_REGS' is 822: defined, and will usually be the constant 1, since most of the 823: cases preventing register elimination are things that the compiler 824: already knows about. 825: 826: `INITIAL_ELIMINATION_OFFSET (FROM-REG, TO-REG, OFFSET-VAR)' 827: This macro is similar to `INITIAL_FRAME_POINTER_OFFSET'. It 828: specifies the initial difference between the specified pair of 829: registers. This macro must be defined if `ELIMINABLE_REGS' is 830: defined. 831: 832: `LONGJMP_RESTORE_FROM_STACK' 833: Define this macro if the `longjmp' function restores registers from 834: the stack frames, rather than from those saved specifically by 835: `setjmp'. Certain quantities must not be kept in registers across 836: a call to `setjmp' on such machines. 1.1 root 837: 1.1.1.4 root 838: 839: File: gcc.info, Node: Stack Arguments, Next: Register Arguments, Prev: Elimination, Up: Stack and Calling 1.1 root 840: 1.1.1.4 root 841: Passing Function Arguments on the Stack 842: --------------------------------------- 1.1.1.3 root 843: 1.1.1.4 root 844: The macros in this section control how arguments are passed on the 845: stack. See the following section for other macros that control passing 846: certain arguments in registers. 847: 848: `PROMOTE_PROTOTYPES' 849: Define this macro if an argument declared in a prototype as an 850: integral type smaller than `int' should actually be passed as an 851: `int'. In addition to avoiding errors in certain cases of 852: mismatch, it also makes for better code on certain machines. 853: 854: `PUSH_ROUNDING (NPUSHED)' 855: A C expression that is the number of bytes actually pushed onto the 856: stack when an instruction attempts to push NPUSHED bytes. 857: 858: If the target machine does not have a push instruction, do not 859: define this macro. That directs GNU CC to use an alternate 860: strategy: to allocate the entire argument block and then store the 861: arguments into it. 862: 863: On some machines, the definition 864: 865: #define PUSH_ROUNDING(BYTES) (BYTES) 866: 867: will suffice. But on other machines, instructions that appear to 868: push one byte actually push two bytes in an attempt to maintain 869: alignment. Then the definition should be 870: 871: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 872: 873: `ACCUMULATE_OUTGOING_ARGS' 874: If defined, the maximum amount of space required for outgoing 875: arguments will be computed and placed into the variable 876: `current_function_outgoing_args_size'. No space will be pushed 877: onto the stack for each call; instead, the function prologue should 878: increase the stack frame size by this amount. 879: 880: Defining both `PUSH_ROUNDING' and `ACCUMULATE_OUTGOING_ARGS' is 881: not proper. 882: 883: `REG_PARM_STACK_SPACE (FNDECL)' 884: Define this macro if functions should assume that stack space has 885: been allocated for arguments even when their values are passed in 886: registers. 1.1.1.3 root 887: 1.1.1.4 root 888: The value of this macro is the size, in bytes, of the area 889: reserved for arguments passed in registers for the function 890: represented by FNDECL. 891: 892: This space can be allocated by the caller, or be a part of the 893: machine-dependent stack frame: `OUTGOING_REG_PARM_STACK_SPACE' says 894: which. 895: 896: `MAYBE_REG_PARM_STACK_SPACE' 897: `FINAL_REG_PARM_STACK_SPACE (CONST_SIZE, VAR_SIZE)' 898: Define these macros in addition to the one above if functions might 899: allocate stack space for arguments even when their values are 900: passed in registers. These should be used when the stack space 901: allocated for arguments in registers is not a simple constant 902: independent of the function declaration. 903: 904: The value of the first macro is the size, in bytes, of the area 905: that we should initially assume would be reserved for arguments 906: passed in registers. 907: 908: The value of the second macro is the actual size, in bytes, of the 909: area that will be reserved for arguments passed in registers. 910: This takes two arguments: an integer representing the number of 911: bytes of fixed sized arguments on the stack, and a tree 912: representing the number of bytes of variable sized arguments on 913: the stack. 914: 915: When these macros are defined, `REG_PARM_STACK_SPACE' will only be 916: called for libcall functions, the current function, or for a 917: function being called when it is known that such stack space must 918: be allocated. In each case this value can be easily computed. 919: 920: When deciding whether a called function needs such stack space, 921: and how much space to reserve, GNU CC uses these two macros 922: instead of `REG_PARM_STACK_SPACE'. 923: 924: `OUTGOING_REG_PARM_STACK_SPACE' 925: Define this if it is the responsibility of the caller to allocate 926: the area reserved for arguments passed in registers. 927: 928: If `ACCUMULATE_OUTGOING_ARGS' is defined, this macro controls 929: whether the space for these arguments counts in the value of 930: `current_function_outgoing_args_size'. 931: 932: `STACK_PARMS_IN_REG_PARM_AREA' 933: Define this macro if `REG_PARM_STACK_SPACE' is defined, but the 934: stack parameters don't skip the area specified by it. 935: 936: Normally, when a parameter is not passed in registers, it is 937: placed on the stack beyond the `REG_PARM_STACK_SPACE' area. 938: Defining this macro suppresses this behavior and causes the 939: parameter to be passed on the stack in its natural location. 940: 941: `RETURN_POPS_ARGS (FUNDECL, FUNTYPE, STACK-SIZE)' 942: A C expression that should indicate the number of bytes of its own 943: arguments that a function pops on returning, or 0 if the function 944: pops no arguments and the caller must therefore pop them all after 945: the function returns. 946: 947: FUNDECL is a C variable whose value is a tree node that describes 948: the function in question. Normally it is a node of type 949: `FUNCTION_DECL' that describes the declaration of the function. 950: From this it is possible to obtain the DECL_MACHINE_ATTRIBUTES of 951: the function. 952: 953: FUNTYPE is a C variable whose value is a tree node that describes 954: the function in question. Normally it is a node of type 955: `FUNCTION_TYPE' that describes the data type of the function. 956: From this it is possible to obtain the data types of the value and 957: arguments (if known). 958: 959: When a call to a library function is being considered, FUNTYPE 960: will contain an identifier node for the library function. Thus, if 961: you need to distinguish among various library functions, you can 962: do so by their names. Note that "library function" in this 963: context means a function used to perform arithmetic, whose name is 964: known specially in the compiler and was not mentioned in the C 965: code being compiled. 966: 967: STACK-SIZE is the number of bytes of arguments passed on the 968: stack. If a variable number of bytes is passed, it is zero, and 969: argument popping will always be the responsibility of the calling 970: function. 971: 972: On the Vax, all functions always pop their arguments, so the 973: definition of this macro is STACK-SIZE. On the 68000, using the 974: standard calling convention, no functions pop their arguments, so 975: the value of the macro is always 0 in this case. But an 976: alternative calling convention is available in which functions 977: that take a fixed number of arguments pop them but other functions 978: (such as `printf') pop nothing (the caller pops all). When this 979: convention is in use, FUNTYPE is examined to determine whether a 980: function takes a fixed number of arguments. 1.1 root 981:
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.