1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2003
7 * To run an STG function from C land, call
9 * rv = StgRun(f,BaseReg);
11 * where "f" is the STG function to call, and BaseReg is the address of the
12 * RegTable for this run (we might have separate RegTables if we're running
13 * multiple threads on an SMP machine).
15 * In the end, "f" must JMP to StgReturn (defined below),
16 * passing the return-value "rv" in R1,
17 * to return to the caller of StgRun returning "rv" in
18 * the whatever way C returns a value.
20 * NOTE: StgRun/StgReturn do *NOT* load or store Hp or any
21 * other registers (other than saving the C callee-saves
22 * registers). Instead, the called function "f" must do that
25 * GCC will have assumed that pushing/popping of C-stack frames is
26 * going on when it generated its code, and used stack space
27 * accordingly. However, we actually {\em post-process away} all
28 * such stack-framery (see \tr{ghc/driver/ghc-asm.lprl}). Things will
29 * be OK however, if we initially make sure there are
30 * @RESERVED_C_STACK_BYTES@ on the C-stack to begin with, for local
33 * -------------------------------------------------------------------------- */
35 #include "PosixSource.h"
39 * We define the following (unused) global register variables, because for
40 * some reason gcc generates sub-optimal code for StgRun() on the Alpha
41 * (unnecessarily saving extra registers on the stack) if we don't.
43 * Why do it at the top of this file, rather than near StgRun() below? Because
44 * gcc doesn't let us define global register variables after any function
45 * definition has been read. Any point after #include "Stg.h" would be too
48 * We define alpha_EXTRA_CAREFUL here to save $s6, $f8 and $f9 -- registers
49 * that we don't use but which are callee-save registers. The __divq() routine
50 * in libc.a clobbers $s6.
52 #include "ghcconfig.h"
53 #ifndef USE_MINIINTERPRETER
54 #ifdef alpha_HOST_ARCH
55 #define alpha_EXTRA_CAREFUL
56 register long fake_ra __asm__("$26");
57 register long fake_gp __asm__("$29");
58 #ifdef alpha_EXTRA_CAREFUL
59 register long fake_s6 __asm__("$15");
60 register double fake_f8 __asm__("$f8");
61 register double fake_f9 __asm__("$f9");
66 /* include Stg.h first because we want real machine regs in here: we
67 * have to get the value of R1 back from Stg land to C land intact.
73 #include "OSThreads.h"
74 #include "Capability.h"
81 #ifdef USE_MINIINTERPRETER
83 /* -----------------------------------------------------------------------------
84 any architecture (using miniinterpreter)
85 -------------------------------------------------------------------------- */
87 StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg STG_UNUSED)
90 /* XXX Disabled due to RtsFlags[]/RtsFlags mismatch
92 debugBelch("Jumping to ");
93 printPtr((P_)f); fflush(stdout);
97 f = (StgFunPtr) (f)();
99 return (StgRegTable *)R1.p;
102 StgFunPtr StgReturn(void)
107 #else /* !USE_MINIINTERPRETER */
109 #ifdef LEADING_UNDERSCORE
110 #define STG_RUN "_StgRun"
112 #define STG_RUN "StgRun"
115 #ifdef LEADING_UNDERSCORE
116 #define STG_RETURN "_StgReturn"
118 #define STG_RETURN "StgReturn"
121 /* -----------------------------------------------------------------------------
123 -------------------------------------------------------------------------- */
125 #ifdef i386_HOST_ARCH
127 #ifdef darwin_TARGET_OS
128 #define STG_GLOBAL ".globl "
130 #define STG_GLOBAL ".global "
134 StgRun(StgFunPtr f, StgRegTable *basereg) {
136 unsigned char space[ RESERVED_C_STACK_BYTES + 4*sizeof(void *) ];
141 * save callee-saves registers on behalf of the STG code.
143 "movl %%esp, %%eax\n\t"
145 "movl %%ebx,0(%%eax)\n\t"
146 "movl %%esi,4(%%eax)\n\t"
147 "movl %%edi,8(%%eax)\n\t"
148 "movl %%ebp,12(%%eax)\n\t"
154 * grab the function argument from the stack
160 * The stack pointer has to be aligned to a multiple of 16 bytes at
161 * this point. This works out correctly with gcc 4.0.1, but it might
162 * break at any time in the future. TODO: Make this future-proof.
170 STG_GLOBAL STG_RETURN "\n"
173 "movl %%esi, %%eax\n\t" /* Return value in R1 */
176 * restore callee-saves registers. (Don't stomp on %%eax!)
178 "movl %%esp, %%edx\n\t"
180 "movl 0(%%edx),%%ebx\n\t" /* restore the registers saved above */
181 "movl 4(%%edx),%%esi\n\t"
182 "movl 8(%%edx),%%edi\n\t"
183 "movl 12(%%edx),%%ebp\n\t"
185 : "=&a" (r), "=m" (space)
186 : "m" (f), "m" (basereg), "i" (RESERVED_C_STACK_BYTES)
187 : "edx" /* stomps on %edx */
195 /* ----------------------------------------------------------------------------
196 x86-64 is almost the same as plain x86.
198 I've done it using entirely inline assembler, because I couldn't
199 get gcc to generate the correct subtraction from %rsp by using
200 the local array variable trick. It didn't seem to reserve
201 enough space. Oh well, it's not much harder this way.
203 ------------------------------------------------------------------------- */
205 #ifdef x86_64_HOST_ARCH
207 extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg);
209 static void GNUC3_ATTRIBUTE(used)
210 StgRunIsImplementedInAssembler(void)
214 * save callee-saves registers on behalf of the STG code.
216 ".globl " STG_RUN "\n"
219 "movq %%rsp, %%rax\n\t"
220 "addq %0-48, %%rax\n\t"
221 "movq %%rbx,0(%%rax)\n\t"
222 "movq %%rbp,8(%%rax)\n\t"
223 "movq %%r12,16(%%rax)\n\t"
224 "movq %%r13,24(%%rax)\n\t"
225 "movq %%r14,32(%%rax)\n\t"
226 "movq %%r15,40(%%rax)\n\t"
230 "movq %%rsi,%%r13\n\t"
232 * grab the function argument from the stack, and jump to it.
234 "movq %%rdi,%%rax\n\t"
237 ".globl " STG_RETURN "\n"
240 "movq %%rbx, %%rax\n\t" /* Return value in R1 */
243 * restore callee-saves registers. (Don't stomp on %%rax!)
245 "movq %%rsp, %%rdx\n\t"
246 "addq %0-48, %%rdx\n\t"
247 "movq 0(%%rdx),%%rbx\n\t" /* restore the registers saved above */
248 "movq 8(%%rdx),%%rbp\n\t"
249 "movq 16(%%rdx),%%r12\n\t"
250 "movq 24(%%rdx),%%r13\n\t"
251 "movq 32(%%rdx),%%r14\n\t"
252 "movq 40(%%rdx),%%r15\n\t"
256 : : "i"(RESERVED_C_STACK_BYTES+48+8 /*stack frame size*/));
260 The x86_64 ABI specifies that on a procedure call, %rsp is
261 aligned on a 16-byte boundary + 8. That is, the first
262 argument on the stack after the return address will be
265 Which should be fine: RESERVED_C_STACK_BYTES+48 is a multiple
268 BUT... when we do a C-call from STG land, gcc likes to put the
269 stack alignment adjustment in the prolog. eg. if we're calling
270 a function with arguments in regs, gcc will insert 'subq $8,%rsp'
271 in the prolog, to keep %rsp aligned (the return address is 8
272 bytes, remember). The mangler throws away the prolog, so we
273 lose the stack alignment.
275 The hack is to add this extra 8 bytes to our %rsp adjustment
276 here, so that throughout STG code, %rsp is 16-byte aligned,
279 A quick way to see if this is wrong is to compile this code:
281 main = System.Exit.exitWith ExitSuccess
283 And run it with +RTS -sstderr. The stats code in the RTS, in
284 particular statsPrintf(), relies on the stack alignment because
285 it saves the %xmm regs on the stack, so it'll fall over if the
286 stack isn't aligned, and calling exitWith from Haskell invokes
287 shutdownHaskellAndExit using a C call.
289 Future gcc releases will almost certainly break this hack...
295 /* -----------------------------------------------------------------------------
299 OLD COMMENT from GHC-3.02:
301 We want tailjumps to be calls, because `call xxx' is the only Sparc
302 branch that allows an arbitrary label as a target. (Gcc's ``goto
303 *target'' construct ends up loading the label into a register and
304 then jumping, at the cost of two extra instructions for the 32-bit
307 When entering the threaded world, we stash our return address in a
308 known location so that \tr{%i7} is available as an extra
309 callee-saves register. Of course, we have to restore this when
310 coming out of the threaded world.
312 I hate this god-forsaken architecture. Since the top of the
313 reserved stack space is used for globals and the bottom is reserved
314 for outgoing arguments, we have to stick our return address
315 somewhere in the middle. Currently, I'm allowing 100 extra
316 outgoing arguments beyond the first 6. --JSM
318 Updated info (GHC 4.06): we don't appear to use %i7 any more, so
319 I'm not sure whether we still need to save it. Incedentally, what
320 does the last paragraph above mean when it says "the top of the
321 stack is used for globals"? What globals? --SDM
323 Updated info (GHC 4.08.2): not saving %i7 any more (see below).
324 -------------------------------------------------------------------------- */
326 #ifdef sparc_HOST_ARCH
329 StgRun(StgFunPtr f, StgRegTable *basereg) {
331 unsigned char space[RESERVED_C_STACK_BYTES];
333 register void *i7 __asm__("%i7");
334 ((void **)(space))[100] = i7;
339 ".global " STG_RETURN "\n"
341 : : "p" (space) : "l0","l1","l2","l3","l4","l5","l6","l7");
342 /* we tell the C compiler that l0-l7 are clobbered on return to
343 * StgReturn, otherwise it tries to use these to save eg. the
344 * address of space[100] across the call. The correct thing
345 * to do would be to save all the callee-saves regs, but we
346 * can't be bothered to do that.
348 * We also explicitly mark space as used since gcc eliminates it
351 * The code that gcc generates for this little fragment is now
352 * terrible. We could do much better by coding it directly in
356 /* updated 4.08.2: we don't save %i7 in the middle of the reserved
357 * space any more, since gcc tries to save its address across the
358 * call to f(), this gets clobbered in STG land and we end up
359 * dereferencing a bogus pointer in StgReturn.
361 __asm__ volatile ("ld %1,%0"
362 : "=r" (i7) : "m" (((void **)(space))[100]));
364 return (StgRegTable *)R1.i;
369 /* -----------------------------------------------------------------------------
372 "The stack pointer (SP) must at all times denote an address that has octaword
373 alignment. (This restriction has the side effect that the in-memory portion
374 of the argument list, if any, will start on an octaword boundary.) Note that
375 the stack grows toward lower addresses. During a procedure invocation, SP
376 can never be set to a value that is higher than the value of SP at entry to
377 that procedure invocation.
379 "The contents of the stack, located above the portion of the argument list
380 (if any) that is passed in memory, belong to the calling procedure. Because
381 they are part of the calling procedure, they should not be read or written
382 by the called procedure, except as specified by indirect arguments or
383 language-controlled up-level references.
385 "The SP value might be used by the hardware when raising exceptions and
386 asynchronous interrupts. It must be assumed that the contents of the stack
387 below the current SP value and within the stack for the current thread are
388 continually and unpredictably modified, as specified in the _Alpha
389 Architecture Reference Manual_, and as a result of asynchronous software
392 -- Compaq Computer Corporation, Houston. Tru64 UNIX Calling Standard for
393 Alpha Systems, 5.1 edition, August 2000, section 3.2.1. http://www.
394 tru64unix.compaq.com/docs/base_doc/DOCUMENTATION/V51_PDF/ARH9MBTE.PDF
395 -------------------------------------------------------------------------- */
397 #ifdef alpha_HOST_ARCH
400 StgRun(StgFunPtr f, StgRegTable *basereg)
402 register long real_ra __asm__("$26"); volatile long save_ra;
403 register long real_gp __asm__("$29"); volatile long save_gp;
405 register long real_s0 __asm__("$9" ); volatile long save_s0;
406 register long real_s1 __asm__("$10"); volatile long save_s1;
407 register long real_s2 __asm__("$11"); volatile long save_s2;
408 register long real_s3 __asm__("$12"); volatile long save_s3;
409 register long real_s4 __asm__("$13"); volatile long save_s4;
410 register long real_s5 __asm__("$14"); volatile long save_s5;
411 #ifdef alpha_EXTRA_CAREFUL
412 register long real_s6 __asm__("$15"); volatile long save_s6;
415 register double real_f2 __asm__("$f2"); volatile double save_f2;
416 register double real_f3 __asm__("$f3"); volatile double save_f3;
417 register double real_f4 __asm__("$f4"); volatile double save_f4;
418 register double real_f5 __asm__("$f5"); volatile double save_f5;
419 register double real_f6 __asm__("$f6"); volatile double save_f6;
420 register double real_f7 __asm__("$f7"); volatile double save_f7;
421 #ifdef alpha_EXTRA_CAREFUL
422 register double real_f8 __asm__("$f8"); volatile double save_f8;
423 register double real_f9 __asm__("$f9"); volatile double save_f9;
426 register StgFunPtr real_pv __asm__("$27");
439 #ifdef alpha_EXTRA_CAREFUL
449 #ifdef alpha_EXTRA_CAREFUL
456 __asm__ volatile( "lda $30,-%0($30)" "\n"
457 "\t" "jmp ($27)" "\n"
459 ".globl " STG_RETURN "\n"
461 "\t" "lda $30,%0($30)" "\n"
462 : : "K" (RESERVED_C_STACK_BYTES));
472 #ifdef alpha_EXTRA_CAREFUL
482 #ifdef alpha_EXTRA_CAREFUL
493 #endif /* alpha_HOST_ARCH */
495 /* -----------------------------------------------------------------------------
497 -------------------------------------------------------------------------- */
499 #ifdef hppa1_1_HOST_ARCH
502 StgRun(StgFunPtr f, StgRegTable *basereg)
504 StgChar space[RESERVED_C_STACK_BYTES+16*sizeof(long)+10*sizeof(double)];
507 __asm__ volatile ("ldo %0(%%r30),%%r19\n"
508 "\tstw %%r3, 0(0,%%r19)\n"
509 "\tstw %%r4, 4(0,%%r19)\n"
510 "\tstw %%r5, 8(0,%%r19)\n"
511 "\tstw %%r6,12(0,%%r19)\n"
512 "\tstw %%r7,16(0,%%r19)\n"
513 "\tstw %%r8,20(0,%%r19)\n"
514 "\tstw %%r9,24(0,%%r19)\n"
515 "\tstw %%r10,28(0,%%r19)\n"
516 "\tstw %%r11,32(0,%%r19)\n"
517 "\tstw %%r12,36(0,%%r19)\n"
518 "\tstw %%r13,40(0,%%r19)\n"
519 "\tstw %%r14,44(0,%%r19)\n"
520 "\tstw %%r15,48(0,%%r19)\n"
521 "\tstw %%r16,52(0,%%r19)\n"
522 "\tstw %%r17,56(0,%%r19)\n"
523 "\tstw %%r18,60(0,%%r19)\n"
524 "\tldo 80(%%r19),%%r19\n"
525 "\tfstds %%fr12,-16(0,%%r19)\n"
526 "\tfstds %%fr13, -8(0,%%r19)\n"
527 "\tfstds %%fr14, 0(0,%%r19)\n"
528 "\tfstds %%fr15, 8(0,%%r19)\n"
529 "\tldo 32(%%r19),%%r19\n"
530 "\tfstds %%fr16,-16(0,%%r19)\n"
531 "\tfstds %%fr17, -8(0,%%r19)\n"
532 "\tfstds %%fr18, 0(0,%%r19)\n"
533 "\tfstds %%fr19, 8(0,%%r19)\n"
534 "\tldo 32(%%r19),%%r19\n"
535 "\tfstds %%fr20,-16(0,%%r19)\n"
536 "\tfstds %%fr21, -8(0,%%r19)\n" : :
537 "n" (-(116 * sizeof(long) + 10 * sizeof(double))) : "%r19"
542 __asm__ volatile (".align 4\n"
543 "\t.EXPORT " STG_RETURN ",CODE\n"
544 "\t.EXPORT " STG_RETURN ",ENTRY,PRIV_LEV=3\n"
546 /* "\tldo %0(%%r3),%%r19\n" */
547 "\tldo %1(%%r30),%%r19\n"
548 "\tcopy %%r11, %0\n" /* save R1 */
549 "\tldw 0(0,%%r19),%%r3\n"
550 "\tldw 4(0,%%r19),%%r4\n"
551 "\tldw 8(0,%%r19),%%r5\n"
552 "\tldw 12(0,%%r19),%%r6\n"
553 "\tldw 16(0,%%r19),%%r7\n"
554 "\tldw 20(0,%%r19),%%r8\n"
555 "\tldw 24(0,%%r19),%%r9\n"
556 "\tldw 28(0,%%r19),%%r10\n"
557 "\tldw 32(0,%%r19),%%r11\n"
558 "\tldw 36(0,%%r19),%%r12\n"
559 "\tldw 40(0,%%r19),%%r13\n"
560 "\tldw 44(0,%%r19),%%r14\n"
561 "\tldw 48(0,%%r19),%%r15\n"
562 "\tldw 52(0,%%r19),%%r16\n"
563 "\tldw 56(0,%%r19),%%r17\n"
564 "\tldw 60(0,%%r19),%%r18\n"
565 "\tldo 80(%%r19),%%r19\n"
566 "\tfldds -16(0,%%r19),%%fr12\n"
567 "\tfldds -8(0,%%r19),%%fr13\n"
568 "\tfldds 0(0,%%r19),%%fr14\n"
569 "\tfldds 8(0,%%r19),%%fr15\n"
570 "\tldo 32(%%r19),%%r19\n"
571 "\tfldds -16(0,%%r19),%%fr16\n"
572 "\tfldds -8(0,%%r19),%%fr17\n"
573 "\tfldds 0(0,%%r19),%%fr18\n"
574 "\tfldds 8(0,%%r19),%%fr19\n"
575 "\tldo 32(%%r19),%%r19\n"
576 "\tfldds -16(0,%%r19),%%fr20\n"
577 "\tfldds -8(0,%%r19),%%fr21\n"
579 : "n" (-(116 * sizeof(long) + 10 * sizeof(double)))
586 #endif /* hppa1_1_HOST_ARCH */
588 /* -----------------------------------------------------------------------------
591 Everything is in assembler, so we don't have to deal with GCC...
593 -------------------------------------------------------------------------- */
595 #ifdef powerpc_HOST_ARCH
597 extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg);
599 #ifdef darwin_HOST_OS
600 void StgRunIsImplementedInAssembler(void)
602 #if HAVE_SUBSECTIONS_VIA_SYMBOLS
603 // if the toolchain supports deadstripping, we have to
604 // prevent it here (it tends to get confused here).
605 __asm__ volatile (".no_dead_strip _StgRunIsImplementedInAssembler");
611 "\tbl saveFP # f14\n"
612 "\tstmw r13,-220(r1)\n"
613 "\tstwu r1,-%0(r1)\n"
614 "\tmr r27,r4\n" // BaseReg == r27
618 ".globl _StgReturn\n"
622 "\tlmw r13,-220(r1)\n"
624 : : "i"(RESERVED_C_STACK_BYTES+224 /*stack frame size*/));
628 // This version is for PowerPC Linux.
630 // Differences from the Darwin/Mac OS X version:
631 // *) Different Assembler Syntax
632 // *) Doesn't use Register Saving Helper Functions (although they exist somewhere)
633 // *) We may not access positive stack offsets
634 // (no "Red Zone" as in the Darwin ABI)
635 // *) The Link Register is saved to a different offset in the caller's stack frame
636 // (Linux: 4(r1), Darwin 8(r1))
638 static void GNUC3_ATTRIBUTE(used)
639 StgRunIsImplementedInAssembler(void)
643 "\t.type StgRun,@function\n"
649 "\tstmw 13,-220(5)\n"
650 "\tstfd 14,-144(5)\n"
651 "\tstfd 15,-136(5)\n"
652 "\tstfd 16,-128(5)\n"
653 "\tstfd 17,-120(5)\n"
654 "\tstfd 18,-112(5)\n"
655 "\tstfd 19,-104(5)\n"
668 "\tmr 27,4\n" // BaseReg == r27
673 "\t.type StgReturn,@function\n"
700 : : "i"(RESERVED_C_STACK_BYTES+224 /*stack frame size*/));
706 /* -----------------------------------------------------------------------------
707 PowerPC 64 architecture
709 Everything is in assembler, so we don't have to deal with GCC...
711 -------------------------------------------------------------------------- */
713 #ifdef powerpc64_HOST_ARCH
716 extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg);
718 static void GNUC3_ATTRIBUTE(used)
719 StgRunIsImplementedInAssembler(void)
723 // r2 toc - needs to be saved
724 // r3-r10 argument passing, volatile
725 // r11, r12 very volatile (not saved across cross-module calls)
726 // r13 thread local state (never modified, don't need to save)
727 // r14-r31 callee-save
729 ".section \".opd\",\"aw\"\n"
733 "\t.quad\t.StgRun,.TOC.@tocbase,0\n"
734 "\t.size StgRun,24\n"
737 "\t.quad\t.StgReturn,.TOC.@tocbase,0\n"
738 "\t.size StgReturn,24\n"
741 ".type .StgRun,@function\n"
748 "\tstd 14, -288(5)\n"
749 "\tstd 15, -280(5)\n"
750 "\tstd 16, -272(5)\n"
751 "\tstd 17, -264(5)\n"
752 "\tstd 18, -256(5)\n"
753 "\tstd 19, -248(5)\n"
754 "\tstd 20, -240(5)\n"
755 "\tstd 21, -232(5)\n"
756 "\tstd 22, -224(5)\n"
757 "\tstd 23, -216(5)\n"
758 "\tstd 24, -208(5)\n"
759 "\tstd 25, -200(5)\n"
760 "\tstd 26, -192(5)\n"
761 "\tstd 27, -184(5)\n"
762 "\tstd 28, -176(5)\n"
763 "\tstd 29, -168(5)\n"
764 "\tstd 30, -160(5)\n"
765 "\tstd 31, -152(5)\n"
766 "\tstfd 14, -144(5)\n"
767 "\tstfd 15, -136(5)\n"
768 "\tstfd 16, -128(5)\n"
769 "\tstfd 17, -120(5)\n"
770 "\tstfd 18, -112(5)\n"
771 "\tstfd 19, -104(5)\n"
772 "\tstfd 20, -96(5)\n"
773 "\tstfd 21, -88(5)\n"
774 "\tstfd 22, -80(5)\n"
775 "\tstfd 23, -72(5)\n"
776 "\tstfd 24, -64(5)\n"
777 "\tstfd 25, -56(5)\n"
778 "\tstfd 26, -48(5)\n"
779 "\tstfd 27, -40(5)\n"
780 "\tstfd 28, -32(5)\n"
781 "\tstfd 29, -24(5)\n"
782 "\tstfd 30, -16(5)\n"
784 "\tmr 27, 4\n" // BaseReg == r27
789 ".globl .StgReturn\n"
790 ".type .StgReturn,@function\n"
793 "\tla 5, %0(1)\n" // load address == addi r5, r1, %0
813 "\tlfd 14, -144(5)\n"
814 "\tlfd 15, -136(5)\n"
815 "\tlfd 16, -128(5)\n"
816 "\tlfd 17, -120(5)\n"
817 "\tlfd 18, -112(5)\n"
818 "\tlfd 19, -104(5)\n"
835 : : "i"(RESERVED_C_STACK_BYTES+304 /*stack frame size*/));
837 #else // linux_HOST_OS
838 #error Only linux support for power64 right now.
843 /* -----------------------------------------------------------------------------
846 Again, in assembler - so we can fiddle with the register stack, and because
847 gcc doesn't handle asm-clobbered callee-saves correctly.
849 loc0 - loc15: preserved locals
850 loc16 - loc28: STG registers
853 loc31: saved gp (gcc 3.3 uses this slot)
856 f2 - f5: preserved floating-point registers
857 f16 - f23: preserved floating-point registers
858 -------------------------------------------------------------------------- */
860 #ifdef ia64_HOST_ARCH
862 /* the memory stack is rarely used, so 16K is excessive */
863 #undef RESERVED_C_STACK_BYTES
864 #define RESERVED_C_STACK_BYTES 1024
866 /* We don't spill all the callee-save FP registers, only the ones that
867 * gcc has been observed to use */
868 #define PRESERVED_FP_REGISTERS 12
870 /* We always allocate 34 local and 8 output registers. As long as gcc used
871 * fewer than 32 locals, the mangler will adjust the stack frame accordingly. */
874 static void GNUC3_ATTRIBUTE(used)
875 StgRunIsImplementedInAssembler(void)
880 "\talloc loc29 = ar.pfs, 0, %1, 8, 0\n" /* setup register frame */
881 "\tld8 r18 = [r32],8\n" /* get procedure address */
882 "\tadds sp = -%0, sp ;;\n" /* setup stack */
883 "\tld8 gp = [r32]\n" /* get procedure GP */
884 "\tadds r16 = %0-(%2*16), sp\n"
885 "\tadds r17 = %0-((%2-1)*16), sp ;;\n"
886 "\tstf.spill [r16] = f16,32\n" /* spill callee-saved fp regs */
887 "\tstf.spill [r17] = f17,32\n"
888 "\tmov b6 = r18 ;;\n" /* set target address */
889 "\tstf.spill [r16] = f18,32\n"
890 "\tstf.spill [r17] = f19,32\n"
891 "\tmov loc30 = b0 ;;\n" /* save return address */
892 "\tstf.spill [r16] = f20,32\n"
893 "\tstf.spill [r17] = f21,32 ;;\n"
894 "\tstf.spill [r16] = f22,32\n"
895 "\tstf.spill [r17] = f23,32\n"
896 "\tmov loc32 = ar.lc ;;\n" /* save loop counter */
897 "\tstf.spill [r16] = f2,32\n"
898 "\tstf.spill [r17] = f3,32\n"
899 "\tmov loc33 = pr ;;\n" /* save predicate registers */
900 "\tstf.spill [r16] = f4,32\n"
901 "\tstf.spill [r17] = f5,32\n"
902 "\tbr.few b6 ;;\n" /* branch to function */
903 ".global StgReturn\n"
905 "\tmov r8 = loc16\n" /* return value in r8 */
906 "\tadds r16 = %0-(%2*16), sp\n"
907 "\tadds r17 = %0-((%2-1)*16), sp ;;\n"
908 "\tldf.fill f16 = [r16],32\n" /* start restoring fp regs */
909 "\tldf.fill f17 = [r17],32\n"
910 "\tmov ar.pfs = loc29 ;;\n" /* restore register frame */
911 "\tldf.fill f18 = [r16],32\n"
912 "\tldf.fill f19 = [r17],32\n"
913 "\tmov b0 = loc30 ;;\n" /* restore return address */
914 "\tldf.fill f20 = [r16],32\n"
915 "\tldf.fill f21 = [r17],32\n"
916 "\tmov ar.lc = loc32 ;;\n" /* restore loop counter */
917 "\tldf.fill f22 = [r16],32\n"
918 "\tldf.fill f23 = [r17],32\n"
919 "\tmov pr = loc33 ;;\n" /* restore predicate registers */
920 "\tldf.fill f2 = [r16],32\n"
921 "\tldf.fill f3 = [r17],32\n"
922 "\tadds sp = %0, sp ;;\n" /* restore stack */
923 "\tldf.fill f4 = [r16],32\n"
924 "\tldf.fill f5 = [r17],32\n"
925 "\tbr.ret.sptk.many b0 ;;\n" /* return */
926 : : "i"(RESERVED_C_STACK_BYTES + PRESERVED_FP_REGISTERS*16),
928 "i"(PRESERVED_FP_REGISTERS));
933 /* -----------------------------------------------------------------------------
935 -------------------------------------------------------------------------- */
937 #ifdef mips_HOST_ARCH
940 StgRun(StgFunPtr f, StgRegTable *basereg)
942 register StgThreadReturnCode __v0 __asm__("$2");
944 __asm__ __volatile__(
949 " .globl " STG_RETURN " \n"
950 " .aent " STG_RETURN " \n"
955 : "r" (f), "r" (basereg)
956 "$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23",
958 "$f20", "$f22", "$f24", "$f26", "$f28", "$f30",
964 #endif /* mips_HOST_ARCH */
966 #endif /* !USE_MINIINTERPRETER */