Package: kwave
Version: 0.7.3-4
Severity: important
Tags: patch
Hi,
It seems that #288781 is back, let's try to fix this for good
now.
Using push/pop on an int isn't going to work, we need something
that is 64 bit. So I've changed it to use a int64_t instead of
an int, which should hopefully work as expected.
All the registers probably don't need to be 64 bit, eax was
probably enough.
At startup, I'm seeing this:
Benchmarking memcpy methods (smaller is better):
libc memcpy() : 248273109
linux kernel memcpy() : 269018605
MMX optimized memcpy() : -4046833690
MMXEXT optimized memcpy() : 157412020
using -> 'MMXEXT optimized memcpy()'
And have to wonder why mmx has a negative number there.
Kurt
--- libkwave/cputest.c.old 2005-09-10 16:38:49.231573224 +0200
+++ libkwave/cputest.c 2005-09-10 16:44:48.560946856 +0200
@@ -58,20 +58,20 @@
int mm_support(void)
{
int rval;
- int eax, ebx, ecx, edx;
#if defined(ARCH_X86_64)
/* use 64bit pushq / popq */
+ int64_t eax, ebx, ecx, edx;
__asm__ __volatile__ (
/* See if CPUID instruction is supported ... */
/* ... Get copies of EFLAGS into eax and ecx */
"pushf\n\t"
"popq %0\n\t"
- "movl %0, %1\n\t"
+ "movq %0, %1\n\t"
/* ... Toggle the ID bit in one copy and store */
/* to the EFLAGS reg */
- "xorl $0x200000, %0\n\t"
+ "xorq $0x200000, %0\n\t"
"pushq %0\n\t"
"popf\n\t"
@@ -84,6 +84,7 @@
);
#else
/* use 32bit push / pop */
+ int eax, ebx, ecx, edx;
__asm__ __volatile__ (
/* See if CPUID instruction is supported ... */
/* ... Get copies of EFLAGS into eax and ecx */
