--- libcoro/coro.c	2008/04/04 20:07:35	1.30
+++ libcoro/coro.c	2008/11/03 16:05:38	1.35
@@ -45,10 +45,10 @@
 # if __sgi
 #  define STACK_ADJUST_PTR(sp,ss) ((char *)(sp) + (ss) - 8)
 #  define STACK_ADJUST_SIZE(sp,ss) ((ss) - 8)
-# elif __i386__ && CORO_LINUX
+# elif (__i386__ && CORO_LINUX) || (_M_IX86 && CORO_LOSER)
 #  define STACK_ADJUST_PTR(sp,ss) ((char *)(sp) + (ss))
 #  define STACK_ADJUST_SIZE(sp,ss) (ss)
-# elif __amd64__ && CORO_LINUX
+# elif (__amd64__ && CORO_LINUX) || ((_M_AMD64 || _M_IA64) && CORO_LOSER)
 #  define STACK_ADJUST_PTR(sp,ss) ((char *)(sp) + (ss) - 8)
 #  define STACK_ADJUST_SIZE(sp,ss) (ss)
 # else
@@ -122,43 +122,42 @@
 #endif
 
 #if CORO_ASM
-void __attribute__((__noinline__, __fastcall__))
-coro_transfer (struct coro_context *prev, struct coro_context *next)
-{
-  asm volatile (
+asm (
+     ".text\n"
+     ".globl coro_transfer\n"
+     ".type coro_transfer, @function\n"
+     "coro_transfer:\n"
 #if __amd64
-# define NUM_CLOBBERED 5
-     "push %%rbx\n\t"
-     "push %%r12\n\t"
-     "push %%r13\n\t"
-     "push %%r14\n\t"
-     "push %%r15\n\t"
-     "mov  %%rsp, %0\n\t"
-     "mov  %1, %%rsp\n\t"
-     "pop  %%r15\n\t"
-     "pop  %%r14\n\t"
-     "pop  %%r13\n\t"
-     "pop  %%r12\n\t"
-     "pop  %%rbx\n\t"
+# define NUM_SAVED 5
+     "\tpush %rbx\n"
+     "\tpush %r12\n"
+     "\tpush %r13\n"
+     "\tpush %r14\n"
+     "\tpush %r15\n"
+     "\tmov  %rsp, (%rdi)\n"
+     "\tmov  (%rsi), %rsp\n"
+     "\tpop  %r15\n"
+     "\tpop  %r14\n"
+     "\tpop  %r13\n"
+     "\tpop  %r12\n"
+     "\tpop  %rbx\n"
 #elif __i386
-# define NUM_CLOBBERED 4
-     "push %%ebx\n\t"
-     "push %%esi\n\t"
-     "push %%edi\n\t"
-     "push %%ebp\n\t"
-     "mov  %%esp, %0\n\t"
-     "mov  %1, %%esp\n\t"
-     "pop  %%ebp\n\t"
-     "pop  %%edi\n\t"
-     "pop  %%esi\n\t"
-     "pop  %%ebx\n\t"
+# define NUM_SAVED 4
+     "\tpush %ebx\n"
+     "\tpush %esi\n"
+     "\tpush %edi\n"
+     "\tpush %ebp\n"
+     "\tmov  %esp, (%eax)\n"
+     "\tmov  (%edx), %esp\n"
+     "\tpop  %ebp\n"
+     "\tpop  %edi\n"
+     "\tpop  %esi\n"
+     "\tpop  %ebx\n"
 #else
 # error unsupported architecture
 #endif
-     : "=m" (prev->sp)
-     : "m" (next->sp)
-  );
-}
+     "\tret\n"
+);
 #endif
 
 #if CORO_PTHREAD
@@ -187,6 +186,8 @@
   return 0;
 }
 
+asm("");
+
 void coro_transfer(coro_context *prev, coro_context *next)
 {
   pthread_cond_init (&prev->c, 0);
@@ -287,13 +288,13 @@
   ctx->env[8] = (long)coro_init;
 #elif defined(_M_IX86)
   ((_JUMP_BUFFER *)&ctx->env)->Eip   = (long)coro_init;
-  ((_JUMP_BUFFER *)&ctx->env)->Esp   = (long)STACK_ADJUST_PTR (sptr,ssize);
+  ((_JUMP_BUFFER *)&ctx->env)->Esp   = (long)STACK_ADJUST_PTR (sptr, ssize);
 #elif defined(_M_AMD64)
   ((_JUMP_BUFFER *)&ctx->env)->Rip   = (__int64)coro_init;
-  ((_JUMP_BUFFER *)&ctx->env)->Rsp   = (__int64)STACK_ADJUST_PTR (sptr,ssize);
+  ((_JUMP_BUFFER *)&ctx->env)->Rsp   = (__int64)STACK_ADJUST_PTR (sptr, ssize);
 #elif defined(_M_IA64)
   ((_JUMP_BUFFER *)&ctx->env)->StIIP = (__int64)coro_init;
-  ((_JUMP_BUFFER *)&ctx->env)->IntSp = (__int64)STACK_ADJUST_PTR (sptr,ssize);
+  ((_JUMP_BUFFER *)&ctx->env)->IntSp = (__int64)STACK_ADJUST_PTR (sptr, ssize);
 #else
 # error "microsoft libc or architecture not supported"
 #endif
@@ -326,9 +327,13 @@
 # elif CORO_ASM
 
   ctx->sp = (volatile void **)(ssize + (char *)sptr);
+  /* we try to allow for both functions with and without frame pointers */
   *--ctx->sp = (void *)coro_init;
-  *--ctx->sp = (void *)coro_init; // this is needed when the prologue saves ebp
-  ctx->sp -= NUM_CLOBBERED;
+  {
+    int i;
+    for (i = NUM_SAVED; i--; )
+      *--ctx->sp = 0;
+  }
 
 # endif