1 /* -----------------------------------------------------------------------------
2 * $Id: TailCalls.h,v 1.17 2004/11/21 22:25:24 desrt Exp $
4 * (c) The GHC Team, 1998-1999
6 * Stuff for implementing proper tail jumps.
8 * ---------------------------------------------------------------------------*/
13 /* -----------------------------------------------------------------------------
14 Unmangled tail-jumping: use the mini interpretter.
15 -------------------------------------------------------------------------- */
17 #ifdef USE_MINIINTERPRETER
19 #define JMP_(cont) return((StgFunPtr)(cont))
25 extern void __DISCARD__(void);
27 /* -----------------------------------------------------------------------------
29 -------------------------------------------------------------------------- */
33 /* Note about discard: possibly there to fool GCC into clearing up
34 before we do the jump eg. if there are some arguments left on the C
35 stack that GCC hasn't popped yet. Also possibly to fool any
36 optimisations (a function call often acts as a barrier). Not sure
37 if any of this is necessary now -- SDM
39 Comment to above note: I don't think the __DISCARD__() in JMP_ is
40 necessary. Arguments should be popped from the C stack immediately
41 after returning from a function, as long as we pass -fno-defer-pop
42 to gcc. Moreover, a goto to a first-class label acts as a barrier
43 for optimisations in the same way a function call does.
47 /* The goto here seems to cause gcc -O2 to delete all the code after
48 it - including the FE_ marker and the epilogue code - exactly what
56 __target = (void *)(cont); \
60 #endif /* i386_TARGET_ARCH */
62 /* -----------------------------------------------------------------------------
63 Tail calling on x86_64
64 -------------------------------------------------------------------------- */
66 #if x86_64_TARGET_ARCH
71 __target = (void *)(cont); \
75 #endif /* x86_64_TARGET_ARCH */
77 /* -----------------------------------------------------------------------------
79 -------------------------------------------------------------------------- */
81 #ifdef sparc_TARGET_ARCH
83 #define JMP_(cont) ((F_) (cont))()
84 /* Oh so happily, the above turns into a "call" instruction,
85 which, on a SPARC, is nothing but a "jmpl" with the
86 return address in %o7 [which we don't care about].
89 /* Don't need these for sparc mangling */
93 #endif /* sparc_TARGET_ARCH */
95 /* -----------------------------------------------------------------------------
97 -------------------------------------------------------------------------- */
99 #ifdef alpha_TARGET_ARCH
102 register void *_procedure __asm__("$27");
106 do { _procedure = (void *)(cont); \
111 /* Don't need these for alpha mangling */
115 #endif /* alpha_TARGET_ARCH */
117 /* -----------------------------------------------------------------------------
120 Description of HP's weird procedure linkage, many thanks to Andy Bennet
121 <andy_bennett@hp.com>:
123 I've been digging a little further into the problem of how HP-UX does
124 dynamic procedure calls. My solution in the last e-mail inserting an extra
125 'if' statement into the JMP_ I think is probably the best general solution I
126 can come up with. There are still a few problems with it however: It wont
127 work, if JMP_ ever has to call anything in a shared library, if this is
128 likely to be required it'll need something more elaborate. It also wont work
129 with PA-RISC 2.0 wide mode (64-bit) which uses a different format PLT.
131 I had some feedback from someone in HP's compiler lab and the problem
132 relates to the linker on HP-UX, not gcc as I first suspected. The reason the
133 'hsc' executable works is most likely due to a change in 'ld's behaviour for
134 performance reasons between your revision and mine.
136 The major issue relating to this is shared libraries and how they are
137 implented under HP-UX. The whole point of the Procedure Label Table (PLT) is
138 to allow a function pointer to hold the address of the function and a
139 pointer to the library's global data lookup table (DLT) used by position
140 independent code (PIC). This makes the PLT absolutely essential for shared
141 library calls. HP has two linker introduced assembly functions for dealing
142 with dynamic calls, $$dyncall and $$dyncall_external. The former does a
143 check to see if the address is a PLT pointer and dereferences if necessary
144 or just calls the address otherwise; the latter skips the check and just
145 does the indirect jump no matter what.
147 Since $$dyncall_external runs faster due to its not having the test, the
148 linker nowadays prefers to generate calls to that, rather than $$dyncall. It
149 makes this decision based on the presence of any shared library. If it even
150 smells an sl's existence at link time, it rigs the runtime system to
151 generate PLT references for everything on the assumption that the result
152 will be slightly more efficient. This is what is crashing GHC since the
153 calls it is generating have no understanding of the procedure label proper.
154 The only way to get real addresses is to link everything archive, including
155 system libraries, at which point it assumes you probably are going to be
156 using calls similar to GHC's (its rigged for HP's +ESfic compiler option)
157 but uses $$dyncall if necessary to cope, just in case you aren't.
159 -------------------------------------------------------------------------- */
161 #ifdef hppa1_1_hp_hpux_TARGET
164 do { void *_procedure = (void *)(cont); \
165 if (((int) _procedure) & 2) \
166 _procedure = (void *)(*((int *) (_procedure - 2))); \
170 #endif /* hppa1_1_hp_hpux_TARGET */
172 /* -----------------------------------------------------------------------------
173 Tail calling on PowerPC
174 -------------------------------------------------------------------------- */
176 #ifdef powerpc_TARGET_ARCH
181 target = (void *)(cont); \
187 The __DISCARD__ is there because Apple's April 2002 Beta of GCC 3.1
188 sometimes generates incorrect code otherwise.
189 It tends to "forget" to update global register variables in the presence
190 of decrement/increment operators:
191 JMP_(*(--Sp)) is wrongly compiled as JMP_(Sp[-1]).
192 Calling __DISCARD__ in between works around this problem.
196 I would _love_ to use the following instead,
197 but some versions of Apple's GCC fail to generate code for it
198 if it is called for a casted data pointer - which is exactly what
199 we are going to do...
201 #define JMP_(cont) ((F_) (cont))()
204 #endif /* powerpc_TARGET_ARCH */
206 #ifdef powerpc64_TARGET_ARCH
207 #define JMP_(cont) ((F_) (cont))()
210 /* -----------------------------------------------------------------------------
212 -------------------------------------------------------------------------- */
214 #ifdef ia64_TARGET_ARCH
216 /* The compiler can more intelligently decide how to do this. We therefore
217 * implement it as a call and optimise to a jump at mangle time. */
218 #define JMP_(cont) ((F_) (cont))(); __asm__ volatile ("--- TAILCALL ---");
220 /* Don't emit calls to __DISCARD__ as this causes hassles */
221 #define __DISCARD__()
225 /* -----------------------------------------------------------------------------
228 These are markers indicating the start and end of Real Code in a
229 function. All instructions between the actual start and end of the
230 function and these markers is shredded by the mangler.
231 -------------------------------------------------------------------------- */
233 /* The following __DISCARD__() has become necessary with gcc 2.96 on x86.
234 * It prevents gcc from moving stack manipulation code from the function
235 * body (aka the Real Code) into the function prologue, ie, from moving it
236 * over the --- BEGIN --- marker. It should be noted that (like some
237 * other black magic in GHC's code), there is no essential reason why gcc
238 * could not move some stack manipulation code across the __DISCARD__() -
239 * it just doesn't choose to do it at the moment.
244 #define FB_ __asm__ volatile ("--- BEGIN ---"); __DISCARD__ ();
248 #define FE_ __asm__ volatile ("--- END ---");
251 #endif /* !USE_MINIINTERPRETER */
253 #endif /* TAILCALLS_H */