[Sbcl-commits] CVS: sbcl/src/runtime ppc-linux-os.c,1.1,1.2

[Christophe R. <cr...@us...>]

Update of /cvsroot/sbcl/sbcl/src/runtime
In directory usw-pr-cvs1:/tmp/cvs-serv29652/src/runtime

Modified Files:
	ppc-linux-os.c 
Log Message:
0.7.5.10:
	Begin to fix floating point exceptions on PPC
	... write an os_context_fp_control function
	... use it in CONTEXT-FLOATING-POINT-MODES
	(this, unlike on the SPARC and Alpha ports, doesn't fix all the
		normal-user problems.  After the first return from the Lisp
		debugger on the PPC, the FLOATING-POINT-MODES are (almost)
		cleared, which means that the second bad floating point
		operation is not trapped).


Index: ppc-linux-os.c
===================================================================
RCS file: /cvsroot/sbcl/sbcl/src/runtime/ppc-linux-os.c,v
retrieving revision 1.1
retrieving revision 1.2
diff -u -d -r1.1 -r1.2
--- ppc-linux-os.c	18 Mar 2002 17:59:37 -0000	1.1
+++ ppc-linux-os.c	11 Jul 2002 15:17:01 -0000	1.2
@@ -67,6 +67,17 @@
     return &context->uc_sigmask;
 }
 
+unsigned long
+os_context_fp_control(os_context_t *context)
+{
+    /* So this may look like nice, well behaved code. However, closer
+       inspection reveals that gpr is simply the general purpose
+       registers, and PT_FPSCR is an offset that is larger than 32
+       (the number of ppc registers), but that happens to get the
+       right answer. -- CSR, 2002-07-11 */
+    return &((context->uc_mcontext.regs)->gpr[PT_FPSCR]); 
+}
+
 void os_flush_icache(os_vm_address_t address, os_vm_size_t length)
 {
     /* see ppc-arch.c */