--- cvsroot/libcoro/coro.c	2006/11/30 18:21:14	1.23
+++ cvsroot/libcoro/coro.c	2007/05/02 05:53:26	1.27
@@ -53,7 +53,7 @@
 # include <stddef.h>
 #endif
 
-#if CORO_SJLJ || CORO_LOSER || CORO_LINUX || CORO_IRIX
+#if CORO_SJLJ || CORO_LOSER || CORO_LINUX || CORO_IRIX || CORO_ASM
 
 #include <stdlib.h>
 
@@ -67,6 +67,12 @@
 static volatile void *coro_init_arg;
 static volatile coro_context *new_coro, *create_coro;
 
+/* what we really want to detect here is wether we use a new-enough version of GAS */
+/* instead, check for gcc 3 and ELF and hope for the best */
+#if __GNUC__ >= 3 && __ELF__
+# define HAVE_CFI 1
+#endif
+
 static void
 coro_init (void)
 {
@@ -90,7 +96,15 @@
 trampoline (int sig)
 {
   if (setjmp (((coro_context *)new_coro)->env))
-    coro_init (); /* start it */
+    {
+#if HAVE_CFI
+      asm (".cfi_startproc");
+#endif
+      coro_init (); /* start it */
+#if HAVE_CFI
+      asm (".cfi_endproc");
+#endif
+    }
   else
     trampoline_count++;
 }
@@ -99,6 +113,46 @@
 
 #endif
 
+#if CORO_ASM
+void __attribute__((__noinline__, __fastcall__))
+coro_transfer (struct coro_context *prev, struct coro_context *next)
+{
+  asm volatile (
+#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"
+#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"
+#else
+# error unsupported architecture
+#endif
+     : "=m" (prev->sp)
+     : "m" (next->sp)
+  );
+}
+#endif
+
 /* initialize a machine state */
 void coro_create (coro_context *ctx,
                   coro_func coro, void *arg,
@@ -115,7 +169,7 @@
 
   makecontext (&(ctx->uc), (void (*)()) coro, 1, arg);
 
-#elif CORO_SJLJ || CORO_LOSER || CORO_LINUX || CORO_IRIX
+#elif CORO_SJLJ || CORO_LOSER || CORO_LINUX || CORO_IRIX || CORO_ASM
 
 # if CORO_SJLJ
   stack_t ostk, nstk;
@@ -203,36 +257,41 @@
 # elif CORO_LINUX
 
   _setjmp (ctx->env);
-#if defined(__GLIBC__) && defined(__GLIBC_MINOR__) \
-    && __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 0 && defined(JB_PC) && defined(JB_SP)
+#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 0 && defined (JB_PC) && defined (JB_SP)
   ctx->env[0].__jmpbuf[JB_PC] = (long)coro_init;
-  ctx->env[0].__jmpbuf[JB_SP] = (long)STACK_ADJUST_PTR (sptr,ssize);
-#elif defined(__GLIBC__) && defined(__GLIBC_MINOR__) \
-    && __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 0 && defined(__mc68000__)
+  ctx->env[0].__jmpbuf[JB_SP] = (long)STACK_ADJUST_PTR (sptr, ssize);
+#elif __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 0 && defined (__mc68000__)
   ctx->env[0].__jmpbuf[0].__aregs[0] = (long int)coro_init;
   ctx->env[0].__jmpbuf[0].__sp = (int *)((char *)sptr + ssize);
-#elif defined(__GNU_LIBRARY__) && defined(__i386__)
+#elif defined (__GNU_LIBRARY__) && defined (__i386__)
   ctx->env[0].__jmpbuf[0].__pc = (char *)coro_init;
   ctx->env[0].__jmpbuf[0].__sp = (void *)((char *)sptr + ssize);
-#elif defined(__GNU_LIBRARY__) && defined(__amd64__)
+#elif defined (__GNU_LIBRARY__) && defined (__amd64__)
   ctx->env[0].__jmpbuf[JB_PC]  = (long)coro_init;
-  ctx->env[0].__jmpbuf[JB_RSP] = (long)STACK_ADJUST_PTR (sptr,ssize);
+  ctx->env[0].__jmpbuf[JB_RSP] = (long)STACK_ADJUST_PTR (sptr, ssize);
 #else
-#error "linux libc or architecture not supported"
+# error "linux libc or architecture not supported"
 #endif
 
 # elif CORO_IRIX
 
   setjmp (ctx->env);
   ctx->env[JB_PC] = (__uint64_t)coro_init;
-  ctx->env[JB_SP] = (__uint64_t)STACK_ADJUST_PTR (sptr,ssize);
+  ctx->env[JB_SP] = (__uint64_t)STACK_ADJUST_PTR (sptr, ssize);
+
+# elif CORO_ASM
+
+  ctx->sp = (volatile void **)(ssize + (char *)sptr);
+  *--ctx->sp = (void *)coro_init;
+  *--ctx->sp = (void *)coro_init; // this is needed when the prologue saves ebp
+  ctx->sp -= NUM_CLOBBERED;
 
 # endif
 
   coro_transfer ((coro_context *)create_coro, (coro_context *)new_coro);
 
 #else
-error unsupported architecture
+# error unsupported architecture
 #endif
 }