SEC10.hss 43 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962
  1. [Main]
  2. Title=Options That Control Optimization
  3. [Top]
  4. These options control various sorts of optimizations.
  5. <BR><BR>
  6. Without any optimization option, the compiler's goal is to reduce the
  7. cost of compilation and to make debugging produce the expected
  8. results. Statements are independent: if you stop the program with a
  9. breakpoint between statements, you can then assign a new value to any
  10. variable or change the program counter to any other statement in the
  11. function and get exactly the results you would expect from the source
  12. code.
  13. <BR><BR>
  14. Turning on optimization flags makes the compiler attempt to improve
  15. the performance and/or code size at the expense of compilation time
  16. and possibly the ability to debug the program.
  17. <BR><BR>
  18. Not all optimizations are controlled directly by a flag. Only
  19. optimizations that have a flag are listed.
  20. <DL>
  21. <DT><B>-O</B>
  22. <BR><B>-O1</B>
  23. <DD>Optimize. Optimizing compilation takes somewhat more time, and a lot
  24. more memory for a large function.
  25. <BR><BR>
  26. With <B>'-O'</B>, the compiler tries to reduce code size and execution
  27. time, without performing any optimizations that take a great deal of
  28. compilation time.
  29. <BR><BR>
  30. <B>'-O'</B> turns on the following optimization flags:
  31. -fdefer-pop -fmerge-constants -fthread-jumps -floop-optimize -fcrossjumping -fif-conversion -fif-conversion2 -fdelayed-branch -fguess-branch-probability -fcprop-registers
  32. <BR><BR>
  33. <B>'-O'</B> also turns on <B>'-fomit-frame-pointer'</B> on machines
  34. where doing so does not interfere with debugging.
  35. <BR><BR>
  36. <DT><B>-O2</B>
  37. <DD>Optimize even more. GCC performs nearly all supported optimizations
  38. that do not involve a space-speed tradeoff. The compiler does not
  39. perform loop unrolling or function inlining when you specify <B>'-O2'</B>.
  40. As compared to <B>'-O'</B>, this option increases both compilation time
  41. and the performance of the generated code.
  42. <BR><BR>
  43. <B>'-O2'</B> turns on all optimization flags specified by <B>'-O'</B>. It
  44. also turns on the following optimization flags:
  45. <BR>
  46. <CODE>
  47. -fforce-mem -foptimize-sibling-calls -fstrength-reduce -fcse-follow-jumps -fcse-skip-blocks -frerun-cse-after-loop -frerun-loop-opt -fgcse -fgcse-lm -fgcse-sm -fdelete-null-pointer-checks -fexpensive-optimizations -fregmove -fschedule-insns -fschedule-insns2 -fsched-interblock -fsched-spec -fcaller-saves -fpeephole2 -freorder-blocks -freorder-functions -fstrict-aliasing -falign-functions -falign-jumps -falign-loops -falign-labels
  48. </CODE>
  49. <BR><BR>
  50. Please note the warning under <B>'-fgcse'</B> about
  51. invoking <B>'-O2'</B> on programs that use computed gotos.
  52. <BR><BR>
  53. <DT><B>-O3</B>
  54. <DD>Optimize yet more. <B>'-O3'</B> turns on all optimizations specified by
  55. <B>'-O2'</B> and also turns on the <B>'-finline-functions'</B> and
  56. <B>'-frename-registers'</B> options.
  57. <BR><BR>
  58. <DT><B>-O0</B>
  59. <DD>Do not optimize. This is the default.
  60. <BR><BR>
  61. <DT><B>-Os</B>
  62. <DD>Optimize for size. <B>'-Os'</B> enables all <B>'-O2'</B> optimizations that
  63. do not typically increase code size. It also performs further
  64. optimizations designed to reduce code size.
  65. <BR><BR>
  66. <B>'-Os'</B> disables the following optimization flags:
  67. -falign-functions -falign-jumps -falign-loops -falign-labels -freorder-blocks -fprefetch-loop-arrays
  68. <BR><BR>
  69. If you use multiple <B>'-O'</B> options, with or without level numbers,
  70. the last such option is the one that is effective.
  71. </DL>
  72. Options of the form <B>'-f<I>flag</I>'</B> specify machine-independent
  73. flags. Most flags have both positive and negative forms; the negative
  74. form of <B>'-ffoo'</B> would be <B>'-fno-foo'</B>. In the table
  75. below, only one of the forms is listed: the one you typically will
  76. use. You can figure out the other form by either removing <B>'no-'</B>
  77. or adding it.
  78. <BR><BR>
  79. The following options control specific optimizations. They are either
  80. activated by <B>'-O'</B> options or are related to ones that are. You
  81. can use the following flags in the rare cases when &quot;fine-tuning&quot; of
  82. optimizations to be performed is desired.
  83. <DL>
  84. <DT><B>-fno-defer-pop</B>
  85. <DD>Always pop the arguments to each function call as soon as that function
  86. returns. For machines which must pop arguments after a function call,
  87. the compiler normally lets arguments accumulate on the stack for several
  88. function calls and pops them all at once.
  89. <BR><BR>
  90. Disabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  91. <BR><BR>
  92. <DT><B>-fforce-mem</B>
  93. <DD>Force memory operands to be copied into registers before doing
  94. arithmetic on them. This produces better code by making all memory
  95. references potential common subexpressions. When they are not common
  96. subexpressions, instruction combination should eliminate the separate
  97. register-load.
  98. <BR><BR>
  99. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  100. <BR><BR>
  101. <DT><B>-fforce-addr</B>
  102. <DD>Force memory address constants to be copied into registers before
  103. doing arithmetic on them. This may produce better code just as
  104. <B>'-fforce-mem'</B> may.
  105. <BR><BR>
  106. <DT><B>-fomit-frame-pointer</B>
  107. <DD>Don't keep the frame pointer in a register for functions that
  108. don't need one. This avoids the instructions to save, set up and
  109. restore frame pointers; it also makes an extra register available
  110. in many functions. It also makes debugging impossible on
  111. some machines (not on TIGCC though, thanks to unwinding tables).
  112. <BR><BR>
  113. This option now works with floating point arithmetic.
  114. <BR><BR>
  115. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  116. <BR><BR>
  117. <DT><B>-foptimize-sibling-calls</B>
  118. <DD>Optimize sibling and tail recursive calls.
  119. <BR><BR>
  120. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  121. <BR><BR>
  122. <DT><B>-fno-inline</B>
  123. <DD>Don't pay attention to the <CODE>inline</CODE> keyword. Normally this option
  124. is used to keep the compiler from expanding any functions inline.
  125. Note that if you are not optimizing, no functions can be expanded inline.
  126. <BR><BR>
  127. <DT><B>-finline-functions</B>
  128. <DD>Integrate all simple functions into their callers. The compiler
  129. heuristically decides which functions are simple enough to be worth
  130. integrating in this way.
  131. <BR><BR>
  132. If all calls to a given function are integrated, and the function is
  133. declared <CODE>static</CODE>, then the function is normally not output as
  134. assembler code in its own right.
  135. <BR><BR>
  136. Enabled at level <B>'-O3'</B>.
  137. <BR><BR>
  138. <DT><B>-finline-limit=<I>n</I></B>
  139. <DD>By default, gcc limits the size of functions that can be inlined. This flag
  140. allows the control of this limit for functions that are explicitly marked as
  141. inline (i.e., marked with the inline keyword).
  142. <I>n</I> is the size of functions that can be inlined in
  143. number of pseudo instructions (not counting parameter handling). The default
  144. value of <I>n</I> is 600.
  145. Increasing this value can result in more inlined code at
  146. the cost of compilation time and memory consumption. Decreasing usually makes
  147. the compilation faster and less code will be inlined (which presumably
  148. means slower programs). This option is particularly useful for programs that
  149. use inlining heavily.
  150. <BR><BR>
  151. Inlining is actually controlled by a number of parameters, which may be
  152. specified individually by using <B>'--param <I>name</I>=<I>value</I>'</B>.
  153. The <B>'-finline-limit=<I>n</I>'</B> option sets some of these parameters
  154. as follows:
  155. <BR><BR>
  156. <BR><BR><DL>
  157. <DT><B>max-inline-insns</B>
  158. <DD> is set to <I>n</I>.
  159. <DT><B>max-inline-insns-single</B>
  160. <DD> is set to <I>n</I>/2.
  161. <DT><B>max-inline-insns-auto</B>
  162. <DD> is set to <I>n</I>/2.
  163. <DT><B>min-inline-insns</B>
  164. <DD> is set to 130 or <I>n</I>/4, whichever is smaller.
  165. <DT><B>max-inline-insns-rtl</B>
  166. <DD> is set to <I>n</I>.
  167. </DL><BR>
  168. Using <B>'-finline-limit=600'</B> thus results in the default settings
  169. for these parameters. See below for a documentation of the individual
  170. parameters controlling inlining.
  171. <BR><BR>
  172. <B>Note:</B> pseudo instruction represents, in this particular context, an
  173. abstract measurement of function's size. In no way, it represents a count
  174. of assembly instructions and as such its exact meaning might change from one
  175. release to an another.
  176. <BR><BR>
  177. <DT><B>-fkeep-inline-functions</B>
  178. <DD>Even if all calls to a given function are integrated, and the function
  179. is declared <CODE>static</CODE>, nevertheless output a separate run-time
  180. callable version of the function. This switch does not affect
  181. <CODE>extern&nbsp;inline</CODE> functions.
  182. <BR><BR>
  183. <DT><B>-fkeep-static-consts</B>
  184. <DD>Emit variables declared <CODE><A HREF="$$INFOLINK(keywords/static)">static</A>&nbsp;<A HREF="$$INFOLINK(keywords/const)">const</A></CODE> when optimization isn't turned
  185. on, even if the variables aren't referenced.
  186. <BR><BR>
  187. GCC enables this option by default. If you want to force the compiler to
  188. check if the variable was referenced, regardless of whether or not
  189. optimization is turned on, use the <B>'-fno-keep-static-consts'</B> option.
  190. <BR><BR>
  191. <DT><B>-fmerge-constants</B>
  192. <DD>Attempt to merge identical constants (string constants and floating point
  193. constants) across compilation units.
  194. <BR><BR>
  195. This option is the default for optimized compilation if the assembler and
  196. linker support it. Use <B>'-fno-merge-constants'</B> to inhibit this
  197. behavior.
  198. <BR><BR>
  199. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  200. <BR><BR>
  201. <DT><B>-fmerge-all-constants</B>
  202. <DD>Attempt to merge identical constants and identical variables.
  203. <BR><BR>
  204. This option implies <B>'-fmerge-constants'</B>. In addition to
  205. <B>'-fmerge-constants'</B> this considers e.g. even constant initialized
  206. arrays or initialized constant variables with integral or floating point
  207. types. C requires each non-automatic variable to have a distinct location,
  208. so using this option will result in non-conforming behavior.
  209. <BR><BR>
  210. <DT><B>-fno-branch-count-reg</B>
  211. <DD>Do not use &quot;decrement and branch&quot; instructions on a count register,
  212. but instead generate a sequence of instructions that decrement a
  213. register, compare it against zero, then branch based upon the result.
  214. This option is only meaningful on architectures that support such
  215. instructions, which include x86, PowerPC, IA-64 and S/390.
  216. <BR><BR>
  217. The default is <B>'-fbranch-count-reg'</B>, enabled when
  218. <B>'-fstrength-reduce'</B> is enabled.
  219. <BR><BR>
  220. <DT><B>-fno-function-cse</B>
  221. <DD>Do not put function addresses in registers; make each instruction that
  222. calls a constant function contain the function's address explicitly.
  223. <BR><BR>
  224. This option results in less efficient code, but some strange hacks
  225. that alter the assembler output may be confused by the optimizations
  226. performed when this option is not used.
  227. <BR><BR>
  228. The default is <B>'-ffunction-cse'</B>
  229. <BR><BR>
  230. <DT><B>-fno-zero-initialized-in-bss</B>
  231. <DD>If the target supports a BSS section
  232. (which is always the case for TIGCC now), GCC by default puts variables that
  233. are initialized to zero into BSS. This can save space in the resulting
  234. code.
  235. <BR><BR>
  236. This option turns off this behavior because some programs explicitly
  237. rely on variables going to the data section. E.g., so that the
  238. resulting executable can find the beginning of that section and/or make
  239. assumptions based on that.
  240. <BR><BR>
  241. The default is <B>'-fzero-initialized-in-bss'</B>.
  242. <BR><BR>
  243. <DT><B>-fstrength-reduce</B>
  244. <DD>Perform the optimizations of loop strength reduction and
  245. elimination of iteration variables.
  246. <BR><BR>
  247. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  248. <BR><BR>
  249. <DT><B>-fthread-jumps</B>
  250. <DD>Perform optimizations where we check to see if a jump branches to a
  251. location where another comparison subsumed by the first is found. If
  252. so, the first branch is redirected to either the destination of the
  253. second branch or a point immediately following it, depending on whether
  254. the condition is known to be true or false.
  255. <BR><BR>
  256. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  257. <BR><BR>
  258. <DT><B>-fcse-follow-jumps</B>
  259. <DD>In common subexpression elimination, scan through jump instructions
  260. when the target of the jump is not reached by any other path. For
  261. example, when CSE encounters an <CODE>if</CODE> statement with an
  262. <CODE>else</CODE> clause, CSE will follow the jump when the condition
  263. tested is false.
  264. <BR><BR>
  265. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  266. <BR><BR>
  267. <DT><B>-fcse-skip-blocks</B>
  268. <DD>This is similar to <B>'-fcse-follow-jumps'</B>, but causes CSE to
  269. follow jumps which conditionally skip over blocks. When CSE
  270. encounters a simple <CODE>if</CODE> statement with no else clause,
  271. <B>'-fcse-skip-blocks'</B> causes CSE to follow the jump around the
  272. body of the <CODE>if</CODE>.
  273. <BR><BR>
  274. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  275. <BR><BR>
  276. <DT><B>-frerun-cse-after-loop</B>
  277. <DD>Re-run common subexpression elimination after loop optimizations have been
  278. performed.
  279. <BR><BR>
  280. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  281. <BR><BR>
  282. <DT><B>-frerun-loop-opt</B>
  283. <DD>Run the loop optimizer twice.
  284. <BR><BR>
  285. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  286. <BR><BR>
  287. <DT><B>-fgcse</B>
  288. <DD>Perform a global common subexpression elimination pass.
  289. This pass also performs global constant and copy propagation.
  290. <BR><BR>
  291. <B>Note:</B> When compiling a program using computed gotos, a GCC
  292. extension, you may get better runtime performance if you disable
  293. the global common subexpression elimination pass by adding
  294. <B>'-fno-gcse'</B> to the command line.
  295. <BR><BR>
  296. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  297. <BR><BR>
  298. <DT><B>-fgcse-lm</B>
  299. <DD>When <B>'-fgcse-lm'</B> is enabled, global common subexpression elimination will
  300. attempt to move loads which are only killed by stores into themselves. This
  301. allows a loop containing a load/store sequence to be changed to a load outside
  302. the loop, and a copy/store within the loop.
  303. <BR><BR>
  304. Enabled by default when gcse is enabled.
  305. <BR><BR>
  306. <DT><B>-fgcse-sm</B>
  307. <DD>When <B>'-fgcse-sm'</B> is enabled, A store motion pass is run after global common
  308. subexpression elimination. This pass will attempt to move stores out of loops.
  309. When used in conjunction with <B>'-fgcse-lm'</B>, loops containing a load/store sequence
  310. can be changed to a load before the loop and a store after the loop.
  311. <BR><BR>
  312. Enabled by default when gcse is enabled.
  313. <BR><BR>
  314. <DT><B>-floop-optimize</B>
  315. <DD>Perform loop optimizations: move constant expressions out of loops, simplify
  316. exit test conditions and optionally do strength-reduction and loop unrolling as
  317. well.
  318. <BR><BR>
  319. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  320. <BR><BR>
  321. <DT><B>-fcrossjumping</B>
  322. <DD>Perform cross-jumping transformation. This transformation unifies equivalent code and save code size. The
  323. resulting code may or may not perform better than without cross-jumping.
  324. <BR><BR>
  325. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  326. <BR><BR>
  327. <DT><B>-fif-conversion</B>
  328. <DD>Attempt to transform conditional jumps into branch-less equivalents. This
  329. include use of conditional moves, min, max, set flags and abs instructions, and
  330. some tricks doable by standard arithmetics. The use of conditional execution
  331. on chips where it is available is controlled by <CODE>if-conversion2</CODE>.
  332. <BR><BR>
  333. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  334. <BR><BR>
  335. <DT><B>-fif-conversion2</B>
  336. <DD>Use conditional execution (where available) to transform conditional jumps into
  337. branch-less equivalents.
  338. <BR><BR>
  339. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  340. <BR><BR>
  341. <DT><B>-fdelete-null-pointer-checks</B>
  342. <DD>Use global dataflow analysis to identify and eliminate useless checks
  343. for null pointers. The compiler assumes that dereferencing a null
  344. pointer would have halted the program. If a pointer is checked after
  345. it has already been dereferenced, it cannot be null.
  346. <BR><BR>
  347. In some environments, this assumption is not true, and programs can
  348. safely dereference null pointers. Use
  349. <B>'-fno-delete-null-pointer-checks'</B> to disable this optimization
  350. for programs which depend on that behavior.
  351. <BR><BR>
  352. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  353. <BR><BR>
  354. <DT><B>-fexpensive-optimizations</B>
  355. <DD>Perform a number of minor optimizations that are relatively expensive.
  356. <BR><BR>
  357. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  358. <BR><BR>
  359. <DT><B>-foptimize-register-move</B>
  360. <BR><B>-fregmove</B>
  361. <DD>Attempt to reassign register numbers in move instructions and as
  362. operands of other simple instructions in order to maximize the amount of
  363. register tying. This is especially helpful on machines with two-operand
  364. instructions.
  365. <BR><BR>
  366. Note <B>'-fregmove'</B> and <B>'-foptimize-register-move'</B> are the same
  367. optimization.
  368. <BR><BR>
  369. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  370. <BR><BR>
  371. <DT><B>-fdelayed-branch</B>
  372. <DD>If supported for the target machine, attempt to reorder instructions
  373. to exploit instruction slots available after delayed branch
  374. instructions.
  375. <BR><BR>
  376. Enabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  377. <BR><BR>
  378. <DT><B>-fschedule-insns</B>
  379. <DD>If supported for the target machine, attempt to reorder instructions to
  380. eliminate execution stalls due to required data being unavailable. This
  381. helps machines that have slow floating point or memory load instructions
  382. by allowing other instructions to be issued until the result of the load
  383. or floating point instruction is required.
  384. <BR><BR>
  385. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  386. <BR><BR>
  387. <DT><B>-fschedule-insns2</B>
  388. <DD>Similar to <B>'-fschedule-insns'</B>, but requests an additional pass of
  389. instruction scheduling after register allocation has been done. This is
  390. especially useful on machines with a relatively small number of
  391. registers and where memory load instructions take more than one cycle.
  392. <BR><BR>
  393. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  394. <BR><BR>
  395. <DT><B>-fno-sched-interblock</B>
  396. <DD>Don't schedule instructions across basic blocks. This is normally
  397. enabled by default when scheduling before register allocation, i.e.
  398. with <B>'-fschedule-insns'</B> or at <B>'-O2'</B> or higher.
  399. <BR><BR>
  400. <DT><B>-fno-sched-spec</B>
  401. <DD>Don't allow speculative motion of non-load instructions. This is normally
  402. enabled by default when scheduling before register allocation, i.e.
  403. with <B>'-fschedule-insns'</B> or at <B>'-O2'</B> or higher.
  404. <BR><BR>
  405. <DT><B>-fsched-spec-load</B>
  406. <DD>Allow speculative motion of some load instructions. This only makes
  407. sense when scheduling before register allocation, i.e. with
  408. <B>'-fschedule-insns'</B> or at <B>'-O2'</B> or higher.
  409. <BR><BR>
  410. <DT><B>-fsched-spec-load-dangerous</B>
  411. <DD>Allow speculative motion of more load instructions. This only makes
  412. sense when scheduling before register allocation, i.e. with
  413. <B>'-fschedule-insns'</B> or at <B>'-O2'</B> or higher.
  414. <BR><BR>
  415. <DT><B>-fcaller-saves</B>
  416. <DD>Enable values to be allocated in registers that will be clobbered by
  417. function calls, by emitting extra instructions to save and restore the
  418. registers around such calls. Such allocation is done only when it
  419. seems to result in better code than would otherwise be produced.
  420. <BR><BR>
  421. This option is always enabled by default on certain machines, usually
  422. those which have no call-preserved registers to use instead.
  423. <BR><BR>
  424. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  425. <BR><BR>
  426. <DT><B>-fmove-all-movables</B>
  427. <DD>Forces all invariant computations in loops to be moved
  428. outside the loop.
  429. <BR><BR>
  430. <DT><B>-freduce-all-givs</B>
  431. <DD>Forces all general-induction variables in loops to be
  432. strength-reduced.
  433. <BR><BR>
  434. These options may generate better or worse code; results are highly
  435. dependent on the structure of loops within the source code.
  436. <BR><BR>
  437. These two options are intended to be removed someday, once
  438. they have helped determine the efficacy of various
  439. approaches to improving loop optimizations.
  440. <BR><BR>
  441. <DT><B>-fno-peephole</B>
  442. <BR><B>-fno-peephole2</B>
  443. <DD>Disable any machine-specific peephole optimizations. The difference
  444. between <B>'-fno-peephole'</B> and <B>'-fno-peephole2'</B> is in how they
  445. are implemented in the compiler; some targets use one, some use the
  446. other, a few use both.
  447. <BR><BR>
  448. <B>'-fpeephole'</B> is enabled by default.
  449. <B>'-fpeephole2'</B> enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  450. <BR><BR>
  451. <DT><B>-fbranch-probabilities</B>
  452. <BR><B>-fno-guess-branch-probability</B>
  453. <DD>Do not guess branch probabilities using a randomized model.
  454. <BR><BR>
  455. Sometimes gcc will opt to use a randomized model to guess branch
  456. probabilities, when none are available from either profiling feedback
  457. (<B>'-fprofile-arcs'</B>) or <CODE>__builtin_expect</CODE>. This means that
  458. different runs of the compiler on the same program may produce different
  459. object code.
  460. <BR><BR>
  461. In a hard real-time system, people don't want different runs of the
  462. compiler to produce code that has different behavior; minimizing
  463. non-determinism is of paramount importance. This switch allows users to
  464. reduce non-determinism, possibly at the expense of inferior
  465. optimization.
  466. <BR><BR>
  467. The default is <B>'-fguess-branch-probability'</B> at levels
  468. <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  469. <BR><BR>
  470. <DT><B>-freorder-blocks</B>
  471. <DD>Reorder basic blocks in the compiled function in order to reduce number of
  472. taken branches and improve code locality.
  473. <BR><BR>
  474. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>.
  475. <BR><BR>
  476. <DT><B>-freorder-functions</B>
  477. <DD>Reorder basic blocks in the compiled function in order to reduce number of
  478. taken branches and improve code locality. This is implemented by using special
  479. subsections <CODE>text.hot</CODE> for most frequently executed functions and
  480. <CODE>text.unlikely</CODE> for unlikely executed functions. Reordering is done by
  481. the linker so object file format must support named sections and linker must
  482. place them in a reasonable way.
  483. <BR><BR>
  484. Also profile feedback must be available in to make this option effective. See
  485. <B>'-fprofile-arcs'</B> for details.
  486. <BR><BR>
  487. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  488. <BR><BR>
  489. <DT><B>-fstrict-aliasing</B>
  490. <DD>Allows the compiler to assume the strictest aliasing rules applicable to
  491. the language being compiled. For C, this activates
  492. optimizations based on the type of expressions. In particular, an
  493. object of one type is assumed never to reside at the same address as an
  494. object of a different type, unless the types are almost the same. For
  495. example, an <CODE>unsigned&nbsp;int</CODE> can alias an <CODE>int</CODE>, but not a
  496. <CODE>void*</CODE> or a <CODE>double</CODE>. A character type may alias any other
  497. type.
  498. <BR><BR>
  499. Pay special attention to code like this:
  500. <PRE>union a_union {
  501. int i;
  502. double d;
  503. };
  504. int f() {
  505. a_union t;
  506. t.d = 3.0;
  507. return t.i;
  508. }
  509. </PRE>
  510. The practice of reading from a different union member than the one most
  511. recently written to (called &quot;type-punning&quot;) is common. Even with
  512. <B>'-fstrict-aliasing'</B>, type-punning is allowed, provided the memory
  513. is accessed through the union type. So, the code above will work as
  514. expected. However, this code might not:
  515. <PRE>int f() {
  516. a_union t;
  517. int* ip;
  518. t.d = 3.0;
  519. ip = &amp;t.i;
  520. return *ip;
  521. }
  522. </PRE>
  523. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  524. <BR><BR>
  525. <DT><B>-falign-functions</B>
  526. <BR><B>-falign-functions=<I>n</I></B>
  527. <DD>Align the start of functions to the next power-of-two greater than
  528. <I>n</I>, skipping up to <I>n</I> bytes. For instance,
  529. <B>'-falign-functions=32'</B> aligns functions to the next 32-byte
  530. boundary, but <B>'-falign-functions=24'</B> would align to the next
  531. 32-byte boundary only if this can be done by skipping 23 bytes or less.
  532. <BR><BR>
  533. <B>'-fno-align-functions'</B> and <B>'-falign-functions=1'</B> are
  534. equivalent and mean that functions will not be aligned.
  535. <BR><BR>
  536. Some assemblers only support this flag when <I>n</I> is a power of two;
  537. in that case, it is rounded up.
  538. <BR><BR>
  539. If <I>n</I> is not specified, use a machine-dependent default.
  540. <BR><BR>
  541. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>.
  542. <BR><BR>
  543. <DT><B>-falign-labels</B>
  544. <BR><B>-falign-labels=<I>n</I></B>
  545. <DD>Align all branch targets to a power-of-two boundary, skipping up to
  546. <I>n</I> bytes like <B>'-falign-functions'</B>. This option can easily
  547. make code slower, because it must insert dummy operations for when the
  548. branch target is reached in the usual flow of the code.
  549. <BR><BR>
  550. If <B>'-falign-loops'</B> or <B>'-falign-jumps'</B> are applicable and
  551. are greater than this value, then their values are used instead.
  552. <BR><BR>
  553. If <I>n</I> is not specified, use a machine-dependent default which is
  554. very likely to be <CODE>1</CODE>, meaning no alignment.
  555. <BR><BR>
  556. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>.
  557. <BR><BR>
  558. <DT><B>-falign-loops</B>
  559. <BR><B>-falign-loops=<I>n</I></B>
  560. <DD>Align loops to a power-of-two boundary, skipping up to <I>n</I> bytes
  561. like <B>'-falign-functions'</B>. The hope is that the loop will be
  562. executed many times, which will make up for any execution of the dummy
  563. operations.
  564. <BR><BR>
  565. If <I>n</I> is not specified, use a machine-dependent default.
  566. <BR><BR>
  567. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>.
  568. <BR><BR>
  569. <DT><B>-falign-jumps</B>
  570. <BR><B>-falign-jumps=<I>n</I></B>
  571. <DD>Align branch targets to a power-of-two boundary, for branch targets
  572. where the targets can only be reached by jumping, skipping up to <I>n</I>
  573. bytes like <B>'-falign-functions'</B>. In this case, no dummy operations
  574. need be executed.
  575. <BR><BR>
  576. If <I>n</I> is not specified, use a machine-dependent default.
  577. <BR><BR>
  578. Enabled at levels <B>'-O2'</B>, <B>'-O3'</B>.
  579. <BR><BR>
  580. <DT><B>-frename-registers</B>
  581. <DD>Attempt to avoid false dependencies in scheduled code by making use
  582. of registers left over after register allocation. This optimization
  583. will most benefit processors with lots of registers. It can, however,
  584. make debugging impossible, since variables will no longer stay in
  585. a &quot;home register&quot;.
  586. <BR><BR>
  587. Enabled at levels <B>'-O3'</B>.
  588. <BR><BR>
  589. <DT><B>-fno-cprop-registers</B>
  590. <DD>After register allocation and post-register allocation instruction splitting,
  591. we perform a copy-propagation pass to try to reduce scheduling dependencies
  592. and occasionally eliminate the copy.
  593. <BR><BR>
  594. Disabled at levels <B>'-O'</B>, <B>'-O2'</B>, <B>'-O3'</B>, <B>'-Os'</B>.
  595. </DL>
  596. The following options control compiler behavior regarding floating
  597. point arithmetic. These options trade off between speed and
  598. correctness. All must be specifically enabled.
  599. <DL>
  600. <DT><B>-ffloat-store</B>
  601. <DD>Do not store floating point variables in registers, and inhibit other
  602. options that might change whether a floating point value is taken from a
  603. register or memory.
  604. <BR><BR>
  605. This option prevents undesirable excess precision on machines such as
  606. the 68000 where the floating registers (of the 68881) keep more
  607. precision than a <CODE><A HREF="$$INFOLINK(keywords/double)">double</A></CODE> is supposed to have. Similarly for the
  608. x86 architecture. For most programs, the excess precision does only
  609. good, but a few programs rely on the precise definition of IEEE floating
  610. point. Use <B>'-ffloat-store'</B> for such programs, after modifying
  611. them to store all pertinent intermediate computations into variables.
  612. <BR><BR>
  613. This option is probably useless in TIGCC, except as a workaround for floating
  614. point arithmetic errors.
  615. <BR><BR>
  616. <DT><B>-ffast-math</B>
  617. <DD>Sets <B>'-fno-math-errno'</B>, <B>'-funsafe-math-optimizations'</B>,
  618. <B>'-fno-trapping-math'</B>, <B>'-ffinite-math-only'</B> and
  619. <B>'-fno-signaling-nans'</B>.
  620. <BR><BR>
  621. This option causes the preprocessor macro <CODE>__FAST_MATH__</CODE> to be defined.
  622. <BR><BR>
  623. This option should never be turned on by any <B>'-O'</B> option since
  624. it can result in incorrect output for programs which depend on
  625. an exact implementation of IEEE or ISO rules/specifications for
  626. math functions.
  627. <BR><BR>
  628. This option is probably useless in TIGCC.
  629. <BR><BR>
  630. <DT><B>-fno-math-errno</B>
  631. <DD>Do not set ERRNO after calling math functions that are executed
  632. with a single instruction, e.g., sqrt. A program that relies on
  633. IEEE exceptions for math error handling may want to use this flag
  634. for speed while maintaining IEEE arithmetic compatibility.
  635. <BR><BR>
  636. This option should never be turned on by any <B>'-O'</B> option since
  637. it can result in incorrect output for programs which depend on
  638. an exact implementation of IEEE or ISO rules/specifications for
  639. math functions.
  640. <BR><BR>
  641. This option is probably useless in TIGCC.
  642. <BR><BR>
  643. The default is <B>'-fmath-errno'</B>.
  644. <BR><BR>
  645. <DT><B>-funsafe-math-optimizations</B>
  646. <DD>Allow optimizations for floating-point arithmetic that (a) assume
  647. that arguments and results are valid and (b) may violate IEEE or
  648. ANSI standards. When used at link-time, it may include libraries
  649. or startup files that change the default FPU control word or other
  650. similar optimizations.
  651. <BR><BR>
  652. This option should never be turned on by any <B>'-O'</B> option since
  653. it can result in incorrect output for programs which depend on
  654. an exact implementation of IEEE or ISO rules/specifications for
  655. math functions.
  656. <BR><BR>
  657. This option is probably useless in TIGCC.
  658. <BR><BR>
  659. The default is <B>'-fno-unsafe-math-optimizations'</B>.
  660. <BR><BR>
  661. <DT><B>-ffinite-math-only</B>
  662. <DD>Allow optimizations for floating-point arithmetic that assume
  663. that arguments and results are not NaNs or +-Infs.
  664. <BR><BR>
  665. This option should never be turned on by any <B>'-O'</B> option since
  666. it can result in incorrect output for programs which depend on
  667. an exact implementation of IEEE or ISO rules/specifications.
  668. <BR><BR>
  669. The default is <B>'-fno-finite-math-only'</B>.
  670. <BR><BR>
  671. <DT><B>-fno-trapping-math</B>
  672. <DD>Compile code assuming that floating-point operations cannot generate
  673. user-visible traps. These traps include division by zero, overflow,
  674. underflow, inexact result and invalid operation. This option implies
  675. <B>'-fno-signaling-nans'</B>. Setting this option may allow faster
  676. code if one relies on &quot;non-stop&quot; IEEE arithmetic, for example.
  677. <BR><BR>
  678. This option should never be turned on by any <B>'-O'</B> option since
  679. it can result in incorrect output for programs which depend on
  680. an exact implementation of IEEE or ISO rules/specifications for
  681. math functions.
  682. <BR><BR>
  683. This option is probably useless in TIGCC.
  684. <BR><BR>
  685. The default is <B>'-ftrapping-math'</B>.
  686. <BR><BR>
  687. <DT><B>-fsignaling-nans</B>
  688. <DD>Compile code assuming that IEEE signaling NaNs may generate user-visible
  689. traps during floating-point operations. Setting this option disables
  690. optimizations that may change the number of exceptions visible with
  691. signaling NaNs. This option implies <B>'-ftrapping-math'</B>.
  692. <BR><BR>
  693. This option causes the preprocessor macro <CODE>__SUPPORT_SNAN__</CODE> to
  694. be defined.
  695. <BR><BR>
  696. The default is <B>'-fno-signaling-nans'</B>.
  697. <BR><BR>
  698. This option is experimental and does not currently guarantee to
  699. disable all GCC optimizations that affect signaling NaN behavior.
  700. <BR><BR>
  701. <DT><B>-fsingle-precision-constant</B>
  702. <DD>Treat floating point constant as single precision constant instead of
  703. implicitly converting it to double precision constant.
  704. </DL>
  705. The following options control optimizations that may improve
  706. performance, but are not enabled by any <B>'-O'</B> options. This
  707. section includes experimental options that may produce broken code.
  708. <DL>
  709. <DT><B>-fbranch-probabilities</B>
  710. <DD>After running a program compiled with <B>'-fprofile-arcs'</B>
  711. (see <A HREF="$$LINK(SEC9)">Options for Debugging Your Program or GCC</A>),
  712. you can compile it a second time using
  713. <B>'-fbranch-probabilities'</B>, to improve optimizations based on
  714. the number of times each branch was taken. When the program
  715. compiled with <B>'-fprofile-arcs'</B> exits, it saves arc execution
  716. counts to a file called <CODE><I>sourcename</I>.da</CODE> for each source
  717. file. The information in this data file is very dependent on the
  718. structure of the generated code, so you must use the same source code
  719. and the same optimization options for both compilations.
  720. <BR><BR>
  721. With <B>'-fbranch-probabilities'</B>, GCC puts a
  722. <CODE>REG_BR_PROB</CODE> note on each <CODE>JUMP_INSN</CODE> and <CODE>CALL_INSN</CODE>.
  723. These can be used to improve optimization. Currently, they are only
  724. used in one place: in <CODE>reorg.c</CODE>, instead of guessing which path a
  725. branch is mostly to take, the <CODE>REG_BR_PROB</CODE> values are used to
  726. exactly determine which path is taken more often.
  727. <BR><BR>
  728. <DT><B>-fnew-ra</B>
  729. <DD>Use a graph coloring register allocator. Currently this option is meant
  730. for testing, so we are interested to hear about miscompilations with
  731. <B>'-fnew-ra'</B>.
  732. <BR><BR>
  733. <DT><B>-ftracer</B>
  734. <DD>Perform tail duplication to enlarge superblock size. This transformation
  735. simplifies the control flow of the function allowing other optimizations to do
  736. better job.
  737. <BR><BR>
  738. <DT><B>-funroll-loops</B>
  739. <DD>Unroll loops whose number of iterations can be determined at compile
  740. time or upon entry to the loop. <B>'-funroll-loops'</B> implies both
  741. <B>'-fstrength-reduce'</B> and <B>'-frerun-cse-after-loop'</B>. This
  742. option makes code larger, and may or may not make it run faster.
  743. <BR><BR>
  744. <DT><B>-funroll-all-loops</B>
  745. <DD>Unroll all loops, even if their number of iterations is uncertain when
  746. the loop is entered. This usually makes programs run more slowly.
  747. <B>'-funroll-all-loops'</B> implies the same options as
  748. <B>'-funroll-loops'</B>,
  749. <BR><BR>
  750. <DT><B>-fprefetch-loop-arrays</B>
  751. <DD>If supported by the target machine, generate instructions to prefetch
  752. memory to improve the performance of loops that access large arrays.
  753. <BR><BR>
  754. Disabled at level <B>'-Os'</B>.
  755. <BR><BR>
  756. <DT><B>-ffunction-sections</B>
  757. <BR><B>-fdata-sections</B>
  758. <DD>Place each function or data item into its own section in the output
  759. file if the target supports arbitrary sections. The name of the
  760. function or the name of the data item determines the section's name
  761. in the output file.
  762. <BR><BR>
  763. Use these options on systems where the linker can perform optimizations
  764. to improve locality of reference in the instruction space. Most systems
  765. using the ELF object format and SPARC processors running Solaris 2 have
  766. linkers with such optimizations. AIX may have these optimizations in
  767. the future.
  768. <BR><BR>
  769. Only use these options when there are significant benefits from doing
  770. so. When you specify these options, the assembler and linker will
  771. create larger object and executable files and will also be slower.
  772. You will not be able to use <CODE>gprof</CODE> on all systems if you
  773. specify this option and you may have problems with debugging if
  774. you specify both this option and <B>'-g'</B>.
  775. <BR><BR>
  776. <DT><B>-fssa</B>
  777. <DD>Perform optimizations in static single assignment form. Each function's
  778. flow graph is translated into SSA form, optimizations are performed, and
  779. the flow graph is translated back from SSA form. Users should not
  780. specify this option, since it is not yet ready for production use.
  781. <BR><BR>
  782. <DT><B>-fssa-ccp</B>
  783. <DD>Perform Sparse Conditional Constant Propagation in SSA form. Requires
  784. <B>'-fssa'</B>. Like <B>'-fssa'</B>, this is an experimental feature.
  785. <BR><BR>
  786. <DT><B>-fssa-dce</B>
  787. <DD>Perform aggressive dead-code elimination in SSA form. Requires <B>'-fssa'</B>.
  788. Like <B>'-fssa'</B>, this is an experimental feature.
  789. <BR><BR>
  790. <DT><B>--param <I>name</I>=<I>value</I></B>
  791. <DD>In some places, GCC uses various constants to control the amount of
  792. optimization that is done. For example, GCC will not inline functions
  793. that contain more that a certain number of instructions. You can
  794. control some of these constants on the command-line using the
  795. <B>'--param'</B> option.
  796. <BR><BR>
  797. In each case, the <I>value</I> is an integer. The allowable choices for
  798. <I>name</I> are given in the following table:
  799. <BR><BR><DL>
  800. <DT><B>max-crossjump-edges</B>
  801. <DD>The maximum number of incoming edges to consider for crossjumping.
  802. The algorithm used by <B>'-fcrossjumping'</B> is <CODE>O(N^2)</CODE> in
  803. the number of edges incoming to each block. Increasing values mean
  804. more aggressive optimization, making the compile time increase with
  805. probably small improvement in executable size.
  806. <BR><BR>
  807. <DT><B>max-delay-slot-insn-search</B>
  808. <DD>The maximum number of instructions to consider when looking for an
  809. instruction to fill a delay slot. If more than this arbitrary number of
  810. instructions is searched, the time savings from filling the delay slot
  811. will be minimal so stop searching. Increasing values mean more
  812. aggressive optimization, making the compile time increase with probably
  813. small improvement in executable run time.
  814. <BR><BR>
  815. <DT><B>max-delay-slot-live-search</B>
  816. <DD>When trying to fill delay slots, the maximum number of instructions to
  817. consider when searching for a block with valid live register
  818. information. Increasing this arbitrarily chosen value means more
  819. aggressive optimization, increasing the compile time. This parameter
  820. should be removed when the delay slot code is rewritten to maintain the
  821. control-flow graph.
  822. <BR><BR>
  823. <DT><B>max-gcse-memory</B>
  824. <DD>The approximate maximum amount of memory that will be allocated in
  825. order to perform the global common subexpression elimination
  826. optimization. If more memory than specified is required, the
  827. optimization will not be done.
  828. <BR><BR>
  829. <DT><B>max-gcse-passes</B>
  830. <DD>The maximum number of passes of GCSE to run.
  831. <BR><BR>
  832. <DT><B>max-pending-list-length</B>
  833. <DD>The maximum number of pending dependencies scheduling will allow
  834. before flushing the current state and starting over. Large functions
  835. with few branches or calls can create excessively large lists which
  836. needlessly consume memory and resources.
  837. <BR><BR>
  838. <DT><B>max-inline-insns-single</B>
  839. <DD>Several parameters control the tree inliner used in gcc.
  840. This number sets the maximum number of instructions (counted in gcc's
  841. internal representation) in a single function that the tree inliner
  842. will consider for inlining. This only affects functions declared
  843. inline.
  844. The default value is 300.
  845. <BR><BR>
  846. <DT><B>max-inline-insns-auto</B>
  847. <DD>When you use <B>'-finline-functions'</B> (included in <B>'-O3'</B>),
  848. a lot of functions that would otherwise not be considered for inlining
  849. by the compiler will be investigated. To those functions, a different
  850. (more restrictive) limit compared to functions declared inline can
  851. be applied.
  852. The default value is 300.
  853. <BR><BR>
  854. <DT><B>max-inline-insns</B>
  855. <DD>The tree inliner does decrease the allowable size for single functions
  856. to be inlined after we already inlined the number of instructions
  857. given here by repeated inlining. This number should be a factor of
  858. two or more larger than the single function limit.
  859. Higher numbers result in better runtime performance, but incur higher
  860. compile-time resource (CPU time, memory) requirements and result in
  861. larger binaries. Very high values are not advisable, as too large
  862. binaries may adversely affect runtime performance.
  863. The default value is 600.
  864. <BR><BR>
  865. <DT><B>max-inline-slope</B>
  866. <DD>After exceeding the maximum number of inlined instructions by repeated
  867. inlining, a linear function is used to decrease the allowable size
  868. for single functions. The slope of that function is the negative
  869. reciprocal of the number specified here.
  870. The default value is 32.
  871. <BR><BR>
  872. <DT><B>min-inline-insns</B>
  873. <DD>The repeated inlining is throttled more and more by the linear function
  874. after exceeding the limit. To avoid too much throttling, a minimum for
  875. this function is specified here to allow repeated inlining for very small
  876. functions even when a lot of repeated inlining already has been done.
  877. The default value is 130.
  878. <BR><BR>
  879. <DT><B>max-inline-insns-rtl</B>
  880. <DD>For languages that use the RTL inliner (this happens at a later stage
  881. than tree inlining), you can set the maximum allowable size (counted
  882. in RTL instructions) for the RTL inliner with this parameter.
  883. The default value is 600.
  884. <BR><BR>
  885. <DT><B>max-unrolled-insns</B>
  886. <DD>The maximum number of instructions that a loop should have if that loop
  887. is unrolled, and if the loop is unrolled, it determines how many times
  888. the loop code is unrolled.
  889. <BR><BR>
  890. <DT><B>hot-bb-count-fraction</B>
  891. <DD>Select fraction of the maximal count of repetitions of basic block in program
  892. given basic block needs to have to be considered hot.
  893. <BR><BR>
  894. <DT><B>hot-bb-frequency-fraction</B>
  895. <DD>Select fraction of the maximal frequency of executions of basic block in
  896. function given basic block needs to have to be considered hot
  897. <BR><BR>
  898. <DT><B>tracer-dynamic-coverage</B>
  899. <BR><B>tracer-dynamic-coverage-feedback</B>
  900. <DD>This value is used to limit superblock formation once the given percentage of
  901. executed instructions is covered. This limits unnecessary code size
  902. expansion.
  903. <BR><BR>
  904. The <B>'tracer-dynamic-coverage-feedback'</B> is used only when profile
  905. feedback is available. The real profiles (as opposed to statically estimated
  906. ones) are much less balanced allowing the threshold to be larger value.
  907. <BR><BR>
  908. <DT><B>tracer-max-code-growth</B>
  909. <DD>Stop tail duplication once code growth has reached given percentage. This is
  910. rather hokey argument, as most of the duplicates will be eliminated later in
  911. cross jumping, so it may be set to much higher values than is the desired code
  912. growth.
  913. <BR><BR>
  914. <DT><B>tracer-min-branch-ratio</B>
  915. <DD>Stop reverse growth when the reverse probability of best edge is less than this
  916. threshold (in percent).
  917. <BR><BR>
  918. <DT><B>tracer-min-branch-ratio</B>
  919. <BR><B>tracer-min-branch-ratio-feedback</B>
  920. <DD>Stop forward growth if the best edge do have probability lower than this
  921. threshold.
  922. <BR><BR>
  923. Similarly to <B>'tracer-dynamic-coverage'</B> two values are present, one for
  924. compilation for profile feedback and one for compilation without. The value
  925. for compilation with profile feedback needs to be more conservative (higher) in
  926. order to make tracer effective.
  927. <BR><BR>
  928. <DT><B>ggc-min-expand</B>
  929. <DD>GCC uses a garbage collector to manage its own memory allocation. This
  930. parameter specifies the minimum percentage by which the garbage
  931. collector's heap should be allowed to expand between collections.
  932. Tuning this may improve compilation speed; it has no effect on code
  933. generation.
  934. <BR><BR>
  935. The default is 30% + 70% * (RAM/1GB) with an upper bound of 100% when
  936. RAM &gt;= 1GB. If <CODE>getrlimit</CODE> is available, the notion of &quot;RAM&quot; is
  937. the smallest of actual RAM, RLIMIT_RSS, RLIMIT_DATA and RLIMIT_AS. If
  938. GCC is not able to calculate RAM on a particular platform, the lower
  939. bound of 30% is used. Setting this parameter and
  940. <B>'ggc-min-heapsize'</B> to zero causes a full collection to occur at
  941. every opportunity. This is extremely slow, but can be useful for
  942. debugging.
  943. <BR><BR>
  944. <DT><B>ggc-min-heapsize</B>
  945. <DD>Minimum size of the garbage collector's heap before it begins bothering
  946. to collect garbage. The first collection occurs after the heap expands
  947. by <B>'ggc-min-expand'</B>% beyond <B>'ggc-min-heapsize'</B>. Again,
  948. tuning this may improve compilation speed, and has no effect on code
  949. generation.
  950. <BR><BR>
  951. The default is RAM/8, with a lower bound of 4096 (four megabytes) and an
  952. upper bound of 131072 (128 megabytes). If <CODE>getrlimit</CODE> is
  953. available, the notion of &quot;RAM&quot; is the smallest of actual RAM,
  954. RLIMIT_RSS, RLIMIT_DATA and RLIMIT_AS. If GCC is not able to calculate
  955. RAM on a particular platform, the lower bound is used. Setting this
  956. parameter very large effectively disables garbage collection. Setting
  957. this parameter and <B>'ggc-min-expand'</B> to zero causes a full
  958. collection to occur at every opportunity.
  959. </DL>
  960. </DL>