Asm hack for GCC 3.4 changes

Asm hack for GCC 3.4 changes

[Clark Williams]

[-- Attachment #1: Type: text/plain, Size: 486 bytes --]

Ralf,

This is a patch I use on MIPS trees to deal with the changes in the way
GCC 3.4 handles the 'accum' pseudo register in inline asms. I originally
just applied it to time.c, but then I noticed that cpu-bugs64.c had an
inline asm that referenced 'accum', so I lifted the conditional macro
def to cpu.h, which was the only common header between time.c and
cpu-bugs64.c. Not really sure it's appropriate for cpu.h, but I didn't
want to create a new header.

Hope it's useful.

Clark


[-- Attachment #2: mips_gcc34.patch --]
[-- Type: text/x-patch, Size: 2014 bytes --]

--- include/asm-mips/cpu.h.orig	2004-01-19 12:23:16.000000000 -0600
+++ include/asm-mips/cpu.h	2004-10-06 09:48:58.415989000 -0500
@@ -219,4 +219,14 @@
 #define MIPS_CPU_SUBSET_CACHES	0x00020000 /* P-cache subset enforced */
 #define MIPS_CPU_PREFETCH	0x00040000 /* CPU has usable prefetch */
 
+// hack for GCC 3.4 inline asm change
+
+#if __GNUC__ == 3 && __GNUC_MINOR__ < 4
+#define CLOBBER_LO "lo","accum"
+#define CLOBBER_HILO "hi", "lo", "accum"
+#else
+#define CLOBBER_LO "lo"
+#define CLOBBER_HILO "hi", "lo"
+#endif
+
 #endif /* _ASM_CPU_H */
--- arch/mips/kernel/time.c.orig	2004-09-19 07:30:04.000000000 -0500
+++ arch/mips/kernel/time.c	2004-10-06 09:49:13.360990000 -0500
@@ -277,7 +277,7 @@
 	__asm__("multu	%1,%2"
 		: "=h" (res)
 		: "r" (count), "r" (sll32_usecs_per_cycle)
-		: "lo", "accum");
+		: CLOBBER_LO);
 
 	/*
 	 * Due to possible jiffies inconsistencies, we need to check
@@ -332,7 +332,7 @@
 	__asm__("multu  %1,%2"
 		: "=h" (res)
 		: "r" (count), "r" (quotient)
-		: "lo", "accum");
+		: CLOBBER_LO);
 
 	/*
 	 * Due to possible jiffies inconsistencies, we need to check
@@ -374,7 +374,7 @@
 				: "r" (timerhi), "m" (timerlo),
 				  "r" (tmp), "r" (USECS_PER_JIFFY),
 				  "r" (USECS_PER_JIFFY_FRAC)
-				: "hi", "lo", "accum");
+				: CLOBBER_HILO);
 			cached_quotient = quotient;
 		}
 	}
@@ -388,7 +388,7 @@
 	__asm__("multu	%1,%2"
 		: "=h" (res)
 		: "r" (count), "r" (quotient)
-		: "lo", "accum");
+		: CLOBBER_LO);
 
 	/*
 	 * Due to possible jiffies inconsistencies, we need to check
--- arch/mips/kernel/cpu-bugs64.c.orig	2004-10-06 09:50:23.396989000 -0500
+++ arch/mips/kernel/cpu-bugs64.c	2004-10-06 09:50:41.006989000 -0500
@@ -82,7 +82,7 @@
 		".set	pop"
 		: "=&r" (lv1), "=r" (lw)
 		: "r" (m1), "r" (m2), "r" (s), "I" (0)
-		: "hi", "lo", "accum");
+		: CLOBBER_HILO);
 	/* We have to use single integers for m1 and m2 and a double
 	 * one for p to be sure the mulsidi3 gcc's RTL multiplication
 	 * instruction has the workaround applied.  Older versions of

Re: Asm hack for GCC 3.4 changes

[Ralf Baechle]

On Wed, Oct 06, 2004 at 10:28:06AM -0500, Clark Williams wrote:

> This is a patch I use on MIPS trees to deal with the changes in the way
> GCC 3.4 handles the 'accum' pseudo register in inline asms. I originally
> just applied it to time.c, but then I noticed that cpu-bugs64.c had an
> inline asm that referenced 'accum', so I lifted the conditional macro
> def to cpu.h, which was the only common header between time.c and
> cpu-bugs64.c. Not really sure it's appropriate for cpu.h, but I didn't
> want to create a new header.
> 
> Hope it's useful.

I was actually waiting for somebody else to commit a similar but more
complete patch into CVS.  His patch was a little more complete but your's
is a bit more stylish ...

Thanks,

  Ralf

Re: Asm hack for GCC 3.4 changes

[Clark Williams]

On Wed, 2004-10-06 at 15:39, Ralf Baechle wrote:
> > Hope it's useful.
> 
> I was actually waiting for somebody else to commit a similar but more
> complete patch into CVS.  His patch was a little more complete but your's
> is a bit more stylish ...

So, what you're saying is that the German judge scored me big on
artistic merit, but my technical score wasn't high enough for a gold
medal?

:)

Clark