/* -----------------------------------------------------------------------------
- * $Id: StgCRun.c,v 1.11 2000/02/15 13:16:20 sewardj Exp $
*
- * (c) The GHC Team, 1998-1999
+ * (c) The GHC Team, 1998-2003
*
- * STG-to-C glue. Some architectures have this code written in
- * straight assembler (see StgRun.S), some in C.
+ * STG-to-C glue.
+ *
+ * To run an STG function from C land, call
+ *
+ * rv = StgRun(f,BaseReg);
+ *
+ * where "f" is the STG function to call, and BaseReg is the address of the
+ * RegTable for this run (we might have separate RegTables if we're running
+ * multiple threads on an SMP machine).
+ *
+ * In the end, "f" must JMP to StgReturn (defined below),
+ * passing the return-value "rv" in R1,
+ * to return to the caller of StgRun returning "rv" in
+ * the whatever way C returns a value.
+ *
+ * NOTE: StgRun/StgReturn do *NOT* load or store Hp or any
+ * other registers (other than saving the C callee-saves
+ * registers). Instead, the called function "f" must do that
+ * in STG land.
+ *
+ * GCC will have assumed that pushing/popping of C-stack frames is
+ * going on when it generated its code, and used stack space
+ * accordingly. However, we actually {\em post-process away} all
+ * such stack-framery (see \tr{ghc/driver/ghc-asm.lprl}). Things will
+ * be OK however, if we initially make sure there are
+ * @RESERVED_C_STACK_BYTES@ on the C-stack to begin with, for local
+ * variables.
*
* -------------------------------------------------------------------------- */
+#include "PosixSource.h"
+
+
+/*
+ * We define the following (unused) global register variables, because for
+ * some reason gcc generates sub-optimal code for StgRun() on the Alpha
+ * (unnecessarily saving extra registers on the stack) if we don't.
+ *
+ * Why do it at the top of this file, rather than near StgRun() below? Because
+ * gcc doesn't let us define global register variables after any function
+ * definition has been read. Any point after #include "Stg.h" would be too
+ * late.
+ *
+ * We define alpha_EXTRA_CAREFUL here to save $s6, $f8 and $f9 -- registers
+ * that we don't use but which are callee-save registers. The __divq() routine
+ * in libc.a clobbers $s6.
+ */
+#include "ghcconfig.h"
+#ifdef alpha_TARGET_ARCH
+#define alpha_EXTRA_CAREFUL
+register long fake_ra __asm__("$26");
+register long fake_gp __asm__("$29");
+#ifdef alpha_EXTRA_CAREFUL
+register long fake_s6 __asm__("$15");
+register double fake_f8 __asm__("$f8");
+register double fake_f9 __asm__("$f9");
+#endif
+#endif
+
/* include Stg.h first because we want real machine regs in here: we
* have to get the value of R1 back from Stg land to C land intact.
*/
#include "Stg.h"
#include "Rts.h"
#include "StgRun.h"
+#include "RtsFlags.h"
+#include "OSThreads.h"
+#include "Capability.h"
#ifdef DEBUG
-#include "RtsFlags.h"
#include "RtsUtils.h"
#include "Printer.h"
#endif
/* -----------------------------------------------------------------------------
any architecture (using miniinterpreter)
-------------------------------------------------------------------------- */
-
-/* The static @jmp_environment@ variable allows @miniInterpret@ to
- * communicate with @StgReturn@.
- *
- * Because @StgRun@ may be used recursively, we carefully
- * save and restore the whole of @jmp_environment@.
- */
-#include <setjmp.h>
-#include <string.h> /* for memcpy */
-
-static jmp_buf jmp_environment;
-
-#if 0
-extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg)
+StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg STG_UNUSED)
{
- jmp_buf save_buf;
- /* Save jmp_environment for previous call to miniInterpret */
- memcpy((void *) jmp_environment, (void *) save_buf, sizeof(jmp_buf));
- if (setjmp(jmp_environment) == 0) {
- while ( 1 ) {
- IF_DEBUG(evaluator,
- fprintf(stderr,"Jumping to ");
- printPtr((P_)f);
- fprintf(stderr,"\n");
- );
- f = (StgFunPtr) (f)();
+ while (f) {
+ if (RtsFlags[0].DebugFlags.interpreter) {
+ debugBelch("Jumping to ");
+ printPtr((P_)f); fflush(stdout);
+ debugBelch("\n");
}
+ f = (StgFunPtr) (f)();
}
- /* Restore jmp_environment for previous call */
- memcpy((void*) save_buf, (void*) jmp_environment, sizeof(jmp_buf));
-
return (StgThreadReturnCode)R1.i;
}
-EXTFUN(StgReturn)
+StgFunPtr StgReturn(void)
{
- longjmp(jmp_environment, 1);
+ return 0;
}
+#else /* !USE_MINIINTERPRETER */
+
+#ifdef LEADING_UNDERSCORE
+#define STG_RETURN "_StgReturn"
#else
+#define STG_RETURN "StgReturn"
+#endif
-static void scanStackSeg ( W_* ptr, int nwords )
-{
- W_ w;
- int nwords0 = nwords;
- while (nwords > 0) {
- w = *ptr;
- if (IS_ARG_TAG(w)) {
- fprintf ( stderr, "%d",w ); nwords--; ptr++;
- while (w) { fprintf(stderr, "_"); w--; nwords--; ptr++; }
- }
- else {
- fprintf(stderr, "p");
- nwords--; ptr++;
- }
- }
- if (nwords < 0) fprintf(stderr, "erk: nwords < 0\n");
- checkStackChunk ( ptr, ptr-nwords0 );
-}
+/* -----------------------------------------------------------------------------
+ x86 architecture
+ -------------------------------------------------------------------------- */
+#ifdef i386_TARGET_ARCH
-extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg)
-{
- char* nm;
- while (1) {
-
-#define STACK_DETAILS 0
-
-#if STACK_DETAILS
- {
- int i;
- StgWord* sp = basereg->rSp;
- StgWord* su = basereg->rSu;
- StgTSO* tso = basereg->rCurrentTSO;
- StgWord* sb = tso->stack + tso->stack_size;
- int ws;
-
- fprintf(stderr, "== SP = %p SU = %p\n", sp,su);
-
- if (su >= sb) goto postloop;
- if (!sp || !su) goto postloop;
-
- //printStack ( sp, sb, su);
-
- while (1) {
- ws = su - sp;
- switch (get_itbl((StgClosure*)su)->type) {
- case STOP_FRAME:
- scanStackSeg(sp,ws);
- fprintf(stderr, "S%d ",ws);
- fprintf(stderr, "\n");
- goto postloop;
- case UPDATE_FRAME:
- scanStackSeg(sp,ws);
- fprintf(stderr,"U%d ",ws);
- sp = su + sizeofW(StgUpdateFrame);
- su = ((StgUpdateFrame*)su)->link;
- break;
- case SEQ_FRAME:
- scanStackSeg(sp,ws);
- fprintf(stderr,"Q%d ",ws);
- sp = su + sizeofW(StgSeqFrame);
- su = ((StgSeqFrame*)su)->link;
- break;
- case CATCH_FRAME:
- scanStackSeg(sp,ws);
- fprintf(stderr,"C%d ",ws);
- sp = su + sizeofW(StgCatchFrame);
- su = ((StgCatchFrame*)su)->link;
- break;
- default:
- fprintf(stderr, "?\nweird record on stack\n");
- goto postloop;
- }
- }
- postloop:
- }
-#endif
-
-#if STACK_DETAILS
- fprintf(stderr,"\n");
-#endif
- fprintf(stderr,"-- enter: ");
- nm = nameFromOPtr ( f );
- if (nm)
- fprintf(stderr, "%s (%p)", nm, f); else
- printPtr((P_)f);
- fprintf ( stderr, "\n");
-#if STACK_DETAILS
- fprintf(stderr,"\n");
-#endif
- f = (StgFunPtr) (f)();
- if (!f) break;
- }
- fprintf (stderr, "miniInterpreter: bye!\n\n" );
- return (StgThreadReturnCode)R1.i;
-}
+StgThreadReturnCode
+StgRun(StgFunPtr f, StgRegTable *basereg) {
-EXTFUN(StgReturn)
-{
- return 0;
+ unsigned char space[ RESERVED_C_STACK_BYTES + 4*sizeof(void *) ];
+ StgThreadReturnCode r;
+
+ __asm__ volatile (
+ /*
+ * save callee-saves registers on behalf of the STG code.
+ */
+ "movl %%esp, %%eax\n\t"
+ "addl %4, %%eax\n\t"
+ "movl %%ebx,0(%%eax)\n\t"
+ "movl %%esi,4(%%eax)\n\t"
+ "movl %%edi,8(%%eax)\n\t"
+ "movl %%ebp,12(%%eax)\n\t"
+ /*
+ * Set BaseReg
+ */
+ "movl %3,%%ebx\n\t"
+ /*
+ * grab the function argument from the stack, and jump to it.
+ */
+ "movl %2,%%eax\n\t"
+ "jmp *%%eax\n\t"
+
+ ".global " STG_RETURN "\n"
+ STG_RETURN ":\n\t"
+
+ "movl %%esi, %%eax\n\t" /* Return value in R1 */
+
+ /*
+ * restore callee-saves registers. (Don't stomp on %%eax!)
+ */
+ "movl %%esp, %%edx\n\t"
+ "addl %4, %%edx\n\t"
+ "movl 0(%%edx),%%ebx\n\t" /* restore the registers saved above */
+ "movl 4(%%edx),%%esi\n\t"
+ "movl 8(%%edx),%%edi\n\t"
+ "movl 12(%%edx),%%ebp\n\t"
+
+ : "=&a" (r), "=m" (space)
+ : "m" (f), "m" (basereg), "i" (RESERVED_C_STACK_BYTES)
+ : "edx" /* stomps on %edx */
+ );
+
+ return r;
}
+
#endif
+/* ----------------------------------------------------------------------------
+ x86-64 is almost the same as plain x86.
+ I've done it using entirely inline assembler, because I couldn't
+ get gcc to generate the correct subtraction from %rsp by using
+ the local array variable trick. It didn't seem to reserve
+ enough space. Oh well, it's not much harder this way.
-#else /* !USE_MINIINTERPRETER */
+ ------------------------------------------------------------------------- */
-#ifdef LEADING_UNDERSCORE
-#define STG_RETURN "_StgReturn"
-#else
-#define STG_RETURN "StgReturn"
-#endif
+#ifdef x86_64_TARGET_ARCH
+
+extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg);
+
+static void StgRunIsImplementedInAssembler(void)
+{
+ __asm__ volatile (
+ /*
+ * save callee-saves registers on behalf of the STG code.
+ */
+ ".globl StgRun\n"
+ "StgRun:\n\t"
+ "subq %0, %%rsp\n\t"
+ "movq %%rsp, %%rax\n\t"
+ "addq %0-48, %%rax\n\t"
+ "movq %%rbx,0(%%rax)\n\t"
+ "movq %%rbp,8(%%rax)\n\t"
+ "movq %%r12,16(%%rax)\n\t"
+ "movq %%r13,24(%%rax)\n\t"
+ "movq %%r14,32(%%rax)\n\t"
+ "movq %%r15,40(%%rax)\n\t"
+ /*
+ * Set BaseReg
+ */
+ "movq %%rsi,%%rbx\n\t"
+ /*
+ * grab the function argument from the stack, and jump to it.
+ */
+ "movq %%rdi,%%rax\n\t"
+ "jmp *%%rax\n\t"
+
+ ".global " STG_RETURN "\n"
+ STG_RETURN ":\n\t"
+
+ "movq %%r13, %%rax\n\t" /* Return value in R1 */
+
+ /*
+ * restore callee-saves registers. (Don't stomp on %%rax!)
+ */
+ "movq %%rsp, %%rdx\n\t"
+ "addq %0-48, %%rdx\n\t"
+ "movq 0(%%rdx),%%rbx\n\t" /* restore the registers saved above */
+ "movq 8(%%rdx),%%rbp\n\t"
+ "movq 16(%%rdx),%%r12\n\t"
+ "movq 24(%%rdx),%%r13\n\t"
+ "movq 32(%%rdx),%%r14\n\t"
+ "movq 40(%%rdx),%%r15\n\t"
+ "addq %0, %%rsp\n\t"
+ "retq"
+
+ : : "i"(RESERVED_C_STACK_BYTES+48 /*stack frame size*/));
+}
+
+#endif /* x86-64 */
/* -----------------------------------------------------------------------------
- sparc architecture
+ Sparc architecture
+
+ --
+ OLD COMMENT from GHC-3.02:
+
+ We want tailjumps to be calls, because `call xxx' is the only Sparc
+ branch that allows an arbitrary label as a target. (Gcc's ``goto
+ *target'' construct ends up loading the label into a register and
+ then jumping, at the cost of two extra instructions for the 32-bit
+ load.)
+
+ When entering the threaded world, we stash our return address in a
+ known location so that \tr{%i7} is available as an extra
+ callee-saves register. Of course, we have to restore this when
+ coming out of the threaded world.
+
+ I hate this god-forsaken architecture. Since the top of the
+ reserved stack space is used for globals and the bottom is reserved
+ for outgoing arguments, we have to stick our return address
+ somewhere in the middle. Currently, I'm allowing 100 extra
+ outgoing arguments beyond the first 6. --JSM
+
+ Updated info (GHC 4.06): we don't appear to use %i7 any more, so
+ I'm not sure whether we still need to save it. Incedentally, what
+ does the last paragraph above mean when it says "the top of the
+ stack is used for globals"? What globals? --SDM
+
+ Updated info (GHC 4.08.2): not saving %i7 any more (see below).
-------------------------------------------------------------------------- */
-
+
#ifdef sparc_TARGET_ARCH
StgThreadReturnCode
StgRun(StgFunPtr f, StgRegTable *basereg) {
- StgChar space[RESERVED_C_STACK_BYTES+sizeof(void *)];
+ unsigned char space[RESERVED_C_STACK_BYTES];
+#if 0
register void *i7 __asm__("%i7");
((void **)(space))[100] = i7;
+#endif
f();
- __asm__ volatile (".align 4\n"
+ __asm__ volatile (
+ ".align 4\n"
".global " STG_RETURN "\n"
- STG_RETURN ":\n"
- "\tld %1,%0" : "=r" (i7) : "m" (((void **)(space))[100]));
+ STG_RETURN ":"
+ : : : "l0","l1","l2","l3","l4","l5","l6","l7");
+ /* we tell the C compiler that l0-l7 are clobbered on return to
+ * StgReturn, otherwise it tries to use these to save eg. the
+ * address of space[100] across the call. The correct thing
+ * to do would be to save all the callee-saves regs, but we
+ * can't be bothered to do that.
+ *
+ * The code that gcc generates for this little fragment is now
+ * terrible. We could do much better by coding it directly in
+ * assembler.
+ */
+#if 0
+ /* updated 4.08.2: we don't save %i7 in the middle of the reserved
+ * space any more, since gcc tries to save its address across the
+ * call to f(), this gets clobbered in STG land and we end up
+ * dereferencing a bogus pointer in StgReturn.
+ */
+ __asm__ volatile ("ld %1,%0"
+ : "=r" (i7) : "m" (((void **)(space))[100]));
+#endif
return (StgThreadReturnCode)R1.i;
}
/* -----------------------------------------------------------------------------
alpha architecture
+
+ "The stack pointer (SP) must at all times denote an address that has octaword
+ alignment. (This restriction has the side effect that the in-memory portion
+ of the argument list, if any, will start on an octaword boundary.) Note that
+ the stack grows toward lower addresses. During a procedure invocation, SP
+ can never be set to a value that is higher than the value of SP at entry to
+ that procedure invocation.
+
+ "The contents of the stack, located above the portion of the argument list
+ (if any) that is passed in memory, belong to the calling procedure. Because
+ they are part of the calling procedure, they should not be read or written
+ by the called procedure, except as specified by indirect arguments or
+ language-controlled up-level references.
+
+ "The SP value might be used by the hardware when raising exceptions and
+ asynchronous interrupts. It must be assumed that the contents of the stack
+ below the current SP value and within the stack for the current thread are
+ continually and unpredictably modified, as specified in the _Alpha
+ Architecture Reference Manual_, and as a result of asynchronous software
+ actions."
+
+ -- Compaq Computer Corporation, Houston. Tru64 UNIX Calling Standard for
+ Alpha Systems, 5.1 edition, August 2000, section 3.2.1. http://www.
+ tru64unix.compaq.com/docs/base_doc/DOCUMENTATION/V51_PDF/ARH9MBTE.PDF
-------------------------------------------------------------------------- */
#ifdef alpha_TARGET_ARCH
StgThreadReturnCode
-StgRun(StgFunPtr f, StgRegTable *basereg)
+StgRun(StgFunPtr f, StgRegTable *basereg)
{
+ register long real_ra __asm__("$26"); volatile long save_ra;
+ register long real_gp __asm__("$29"); volatile long save_gp;
+
+ register long real_s0 __asm__("$9" ); volatile long save_s0;
+ register long real_s1 __asm__("$10"); volatile long save_s1;
+ register long real_s2 __asm__("$11"); volatile long save_s2;
+ register long real_s3 __asm__("$12"); volatile long save_s3;
+ register long real_s4 __asm__("$13"); volatile long save_s4;
+ register long real_s5 __asm__("$14"); volatile long save_s5;
+#ifdef alpha_EXTRA_CAREFUL
+ register long real_s6 __asm__("$15"); volatile long save_s6;
+#endif
+
+ register double real_f2 __asm__("$f2"); volatile double save_f2;
+ register double real_f3 __asm__("$f3"); volatile double save_f3;
+ register double real_f4 __asm__("$f4"); volatile double save_f4;
+ register double real_f5 __asm__("$f5"); volatile double save_f5;
+ register double real_f6 __asm__("$f6"); volatile double save_f6;
+ register double real_f7 __asm__("$f7"); volatile double save_f7;
+#ifdef alpha_EXTRA_CAREFUL
+ register double real_f8 __asm__("$f8"); volatile double save_f8;
+ register double real_f9 __asm__("$f9"); volatile double save_f9;
+#endif
+
+ register StgFunPtr real_pv __asm__("$27");
+
StgThreadReturnCode ret;
- __asm__ volatile ("stq $9,-8($30)\n\t"
- "stq $10,-16($30)\n\t"
- "stq $11,-24($30)\n\t"
- "stq $12,-32($30)\n\t"
- "stq $13,-40($30)\n\t"
- "stq $14,-48($30)\n\t"
- "stq $15,-56($30)\n\t"
- "stt $f2,-64($30)\n\t"
- "stt $f3,-72($30)\n\t"
- "stt $f4,-80($30)\n\t"
- "stt $f5,-88($30)\n\t"
- "stt $f6,-96($30)\n\t"
- "stt $f7,-104($30)\n\t"
- "stt $f8,-112($30)\n\t"
- "stt $f9,-120($30)\n\t"
- "lda $30,-%0($30)" : :
- "K" (RESERVED_C_STACK_BYTES+
- 8*sizeof(double)+8*sizeof(long)));
+ save_ra = real_ra;
+ save_gp = real_gp;
+
+ save_s0 = real_s0;
+ save_s1 = real_s1;
+ save_s2 = real_s2;
+ save_s3 = real_s3;
+ save_s4 = real_s4;
+ save_s5 = real_s5;
+#ifdef alpha_EXTRA_CAREFUL
+ save_s6 = real_s6;
+#endif
- f();
+ save_f2 = real_f2;
+ save_f3 = real_f3;
+ save_f4 = real_f4;
+ save_f5 = real_f5;
+ save_f6 = real_f6;
+ save_f7 = real_f7;
+#ifdef alpha_EXTRA_CAREFUL
+ save_f8 = real_f8;
+ save_f9 = real_f9;
+#endif
+
+ real_pv = f;
+
+ __asm__ volatile( "lda $30,-%0($30)" "\n"
+ "\t" "jmp ($27)" "\n"
+ "\t" ".align 3" "\n"
+ ".globl " STG_RETURN "\n"
+ STG_RETURN ":" "\n"
+ "\t" "lda $30,%0($30)" "\n"
+ : : "K" (RESERVED_C_STACK_BYTES));
+
+ ret = real_s5;
+
+ real_s0 = save_s0;
+ real_s1 = save_s1;
+ real_s2 = save_s2;
+ real_s3 = save_s3;
+ real_s4 = save_s4;
+ real_s5 = save_s5;
+#ifdef alpha_EXTRA_CAREFUL
+ real_s6 = save_s6;
+#endif
- __asm__ volatile (".align 3\n"
- ".globl " STG_RETURN "\n"
- STG_RETURN ":\n\t"
- "lda %0,($14)\n\t" /* save R1 */
- "lda $30,%0($30)\n\t"
- "ldq $9,-8($30)\n\t"
- "ldq $10,-16($30)\n\t"
- "ldq $11,-24($30)\n\t"
- "ldq $12,-32($30)\n\t"
- "ldq $13,-40($30)\n\t"
- "ldq $14,-48($30)\n\t"
- "ldq $15,-56($30)\n\t"
- "ldt $f2,-64($30)\n\t"
- "ldt $f3,-72($30)\n\t"
- "ldt $f4,-80($30)\n\t"
- "ldt $f5,-88($30)\n\t"
- "ldt $f6,-96($30)\n\t"
- "ldt $f7,-104($30)\n\t"
- "ldt $f8,-112($30)\n\t"
- "ldt $f9,-120($30)"
- : "=r" (ret)
- : "K" (RESERVED_C_STACK_BYTES+
- 8*sizeof(double)+8*sizeof(long)));
+ real_f2 = save_f2;
+ real_f3 = save_f3;
+ real_f4 = save_f4;
+ real_f5 = save_f5;
+ real_f6 = save_f6;
+ real_f7 = save_f7;
+#ifdef alpha_EXTRA_CAREFUL
+ real_f8 = save_f8;
+ real_f9 = save_f9;
+#endif
+
+ real_ra = save_ra;
+ real_gp = save_gp;
return ret;
}
#ifdef hppa1_1_TARGET_ARCH
StgThreadReturnCode
-StgRun(StgFunPtr f, StgRegTable *basereg)
+StgRun(StgFunPtr f, StgRegTable *basereg)
{
StgChar space[RESERVED_C_STACK_BYTES+16*sizeof(long)+10*sizeof(double)];
StgThreadReturnCode ret;
"\tfldds 8(0,%%r19),%%fr19\n"
"\tldo 32(%%r19),%%r19\n"
"\tfldds -16(0,%%r19),%%fr20\n"
- "\tfldds -8(0,%%r19),%%fr21\n"
+ "\tfldds -8(0,%%r19),%%fr21\n"
: "=r" (ret)
: "n" (-(116 * sizeof(long) + 10 * sizeof(double)))
: "%r19"
#endif /* hppa1_1_TARGET_ARCH */
+/* -----------------------------------------------------------------------------
+ PowerPC architecture
+
+ Everything is in assembler, so we don't have to deal with GCC...
+
+ -------------------------------------------------------------------------- */
+
+#ifdef powerpc_TARGET_ARCH
+
+extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg);
+
+#ifdef darwin_TARGET_OS
+static void StgRunIsImplementedInAssembler(void)
+{
+ __asm__ volatile (
+ "\n.globl _StgRun\n"
+ "_StgRun:\n"
+ "\tmflr r0\n"
+ "\tbl saveFP # f14\n"
+ "\tstmw r13,-220(r1)\n"
+ "\tstwu r1,-%0(r1)\n"
+ "\tmr r27,r4\n" // BaseReg == r27
+ "\tmtctr r3\n"
+ "\tmr r12,r3\n"
+ "\tbctr\n"
+ ".globl _StgReturn\n"
+ "_StgReturn:\n"
+ "\tmr r3,r14\n"
+ "\tla r1,%0(r1)\n"
+ "\tlmw r13,-220(r1)\n"
+ "\tb restFP # f14\n"
+ : : "i"(RESERVED_C_STACK_BYTES+288 /*stack frame size*/));
+}
+#else
+
+// This version is for PowerPC Linux.
+
+// Differences from the Darwin/Mac OS X version:
+// *) Different Assembler Syntax
+// *) Doesn't use Register Saving Helper Functions (although they exist somewhere)
+// *) We may not access positive stack offsets
+// (no "Red Zone" as in the Darwin ABI)
+// *) The Link Register is saved to a different offset in the caller's stack frame
+// (Linux: 4(r1), Darwin 8(r1))
+
+static void StgRunIsImplementedInAssembler(void)
+{
+ __asm__ volatile (
+ "\t.globl StgRun\n"
+ "\t.type StgRun,@function\n"
+ "StgRun:\n"
+ "\tmflr 0\n"
+ "\tstw 0,4(1)\n"
+ "\tmr 5,1\n"
+ "\tstwu 1,-%0(1)\n"
+ "\tstmw 13,-220(5)\n"
+ "\tstfd 14,-144(5)\n"
+ "\tstfd 15,-136(5)\n"
+ "\tstfd 16,-128(5)\n"
+ "\tstfd 17,-120(5)\n"
+ "\tstfd 18,-112(5)\n"
+ "\tstfd 19,-104(5)\n"
+ "\tstfd 20,-96(5)\n"
+ "\tstfd 21,-88(5)\n"
+ "\tstfd 22,-80(5)\n"
+ "\tstfd 23,-72(5)\n"
+ "\tstfd 24,-64(5)\n"
+ "\tstfd 25,-56(5)\n"
+ "\tstfd 26,-48(5)\n"
+ "\tstfd 27,-40(5)\n"
+ "\tstfd 28,-32(5)\n"
+ "\tstfd 29,-24(5)\n"
+ "\tstfd 30,-16(5)\n"
+ "\tstfd 31,-8(5)\n"
+ "\tmr 27,4\n" // BaseReg == r27
+ "\tmtctr 3\n"
+ "\tmr 12,3\n"
+ "\tbctr\n"
+ ".globl StgReturn\n"
+ "\t.type StgReturn,@function\n"
+ "StgReturn:\n"
+ "\tmr 3,14\n"
+ "\tla 5,%0(1)\n"
+ "\tlmw 13,-220(5)\n"
+ "\tlfd 14,-144(5)\n"
+ "\tlfd 15,-136(5)\n"
+ "\tlfd 16,-128(5)\n"
+ "\tlfd 17,-120(5)\n"
+ "\tlfd 18,-112(5)\n"
+ "\tlfd 19,-104(5)\n"
+ "\tlfd 20,-96(5)\n"
+ "\tlfd 21,-88(5)\n"
+ "\tlfd 22,-80(5)\n"
+ "\tlfd 23,-72(5)\n"
+ "\tlfd 24,-64(5)\n"
+ "\tlfd 25,-56(5)\n"
+ "\tlfd 26,-48(5)\n"
+ "\tlfd 27,-40(5)\n"
+ "\tlfd 28,-32(5)\n"
+ "\tlfd 29,-24(5)\n"
+ "\tlfd 30,-16(5)\n"
+ "\tlfd 31,-8(5)\n"
+ "\tmr 1,5\n"
+ "\tlwz 0,4(1)\n"
+ "\tmtlr 0\n"
+ "\tblr\n"
+ : : "i"(RESERVED_C_STACK_BYTES+288 /*stack frame size*/));
+}
+#endif
+
+#endif
+
+/* -----------------------------------------------------------------------------
+ IA64 architecture
+
+ Again, in assembler - so we can fiddle with the register stack, and because
+ gcc doesn't handle asm-clobbered callee-saves correctly.
+
+ loc0 - loc15: preserved locals
+ loc16 - loc28: STG registers
+ loc29: saved ar.pfs
+ loc30: saved b0
+ loc31: saved gp (gcc 3.3 uses this slot)
+ -------------------------------------------------------------------------- */
+
+#ifdef ia64_TARGET_ARCH
+
+/* the memory stack is rarely used, so 16K is excessive */
+#undef RESERVED_C_STACK_BYTES
+#define RESERVED_C_STACK_BYTES 1024
+
+#if ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 3)) || (__GNUC__ > 3)
+/* gcc 3.3+: leave an extra slot for gp saves */
+#define LOCALS 32
+#else
+#define LOCALS 31
+#endif
+
+static void StgRunIsImplementedInAssembler(void)
+{
+ __asm__ volatile(
+ ".global StgRun\n"
+ "StgRun:\n"
+ "\talloc loc29 = ar.pfs, 0, %1, 8, 0\n" /* setup register frame */
+ "\tld8 r18 = [r32],8\n" /* get procedure address */
+ "\tadds sp = -%0, sp ;;\n" /* setup stack */
+ "\tld8 gp = [r32]\n" /* get procedure GP */
+ "\tadds r16 = %0-(6*16), sp\n"
+ "\tadds r17 = %0-(5*16), sp ;;\n"
+ "\tstf.spill [r16] = f16,32\n" /* spill callee-saved fp regs */
+ "\tstf.spill [r17] = f17,32\n"
+ "\tmov b6 = r18 ;;\n" /* set target address */
+ "\tstf.spill [r16] = f18,32\n"
+ "\tstf.spill [r17] = f19,32\n"
+ "\tmov loc30 = b0 ;;\n" /* save return address */
+ "\tstf.spill [r16] = f20,32\n"
+ "\tstf.spill [r17] = f21,32\n"
+ "\tbr.few b6 ;;\n" /* branch to function */
+ ".global StgReturn\n"
+ "StgReturn:\n"
+ "\tmov r8 = loc16\n" /* return value in r8 */
+ "\tadds r16 = %0-(6*16), sp\n"
+ "\tadds r17 = %0-(5*16), sp ;;\n"
+ "\tldf.fill f16 = [r16],32\n" /* start restoring fp regs */
+ "\tldf.fill f17 = [r17],32\n"
+ "\tmov ar.pfs = loc29 ;;\n" /* restore register frame */
+ "\tldf.fill f18 = [r16],32\n"
+ "\tldf.fill f19 = [r17],32\n"
+ "\tmov b0 = loc30 ;;\n" /* restore return address */
+ "\tldf.fill f20 = [r16],32\n"
+ "\tldf.fill f21 = [r17],32\n"
+ "\tadds sp = %0, sp\n" /* restore stack */
+ "\tbr.ret.sptk.many b0 ;;\n" /* return */
+ : : "i"(RESERVED_C_STACK_BYTES + 6*16), "i"(LOCALS));
+}
+
+#endif
+
#endif /* !USE_MINIINTERPRETER */
+