summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/util
diff options
context:
space:
mode:
authorJosé Fonseca <jfonseca@vmware.com>2009-10-05 16:45:38 +0100
committerJosé Fonseca <jfonseca@vmware.com>2009-10-05 16:49:21 +0100
commit6971be783b970f882e873fa40e2dccde4137201f (patch)
tree22296bb402e761791260512906d3cb3200ad56fc /src/gallium/auxiliary/util
parent75e0a376cd32b127f3168c0af12992b5c8576e92 (diff)
util: Improve the cpuid assembly.
No need to save ebx on 64bit. Use just xchgl. Refer to gcc's cpuid.h header. Thanks to Uros Bizjak for pointing this out.
Diffstat (limited to 'src/gallium/auxiliary/util')
-rw-r--r--src/gallium/auxiliary/util/u_cpu_detect.c22
1 files changed, 12 insertions, 10 deletions
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.c b/src/gallium/auxiliary/util/u_cpu_detect.c
index e26214cb91..70ce25cfcf 100644
--- a/src/gallium/auxiliary/util/u_cpu_detect.c
+++ b/src/gallium/auxiliary/util/u_cpu_detect.c
@@ -73,7 +73,7 @@
struct util_cpu_caps util_cpu_caps;
static int has_cpuid(void);
-static int cpuid(unsigned int ax, unsigned int *p);
+static int cpuid(uint32_t ax, uint32_t *p);
#if defined(PIPE_ARCH_X86)
@@ -331,18 +331,22 @@ static int has_cpuid(void)
#endif
}
+
+/**
+ * @sa cpuid.h included in gcc-4.3 onwards.
+ */
static INLINE int
-cpuid(unsigned int ax, unsigned int *p)
+cpuid(uint32_t ax, uint32_t *p)
{
int ret = -1;
#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
__asm __volatile (
- "movl %%ebx, %%esi\n\t"
+ "xchgl %%ebx, %1\n\t"
"cpuid\n\t"
- "xchgl %%ebx, %%esi"
+ "xchgl %%ebx, %1"
: "=a" (p[0]),
- "=S" (p[1]),
+ "=m" (p[1]),
"=c" (p[2]),
"=d" (p[3])
: "0" (ax)
@@ -350,11 +354,9 @@ cpuid(unsigned int ax, unsigned int *p)
ret = 0;
#elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
__asm __volatile (
- "movq %%rbx, %%rsi\n\t"
"cpuid\n\t"
- "xchgq %%rbx, %%rsi"
: "=a" (p[0]),
- "=S" (p[1]),
+ "=b" (p[1]),
"=c" (p[2]),
"=d" (p[3])
: "0" (ax)
@@ -418,8 +420,8 @@ util_cpu_detect(void)
#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
if (has_cpuid()) {
- unsigned int regs[4];
- unsigned int regs2[4];
+ uint32_t regs[4];
+ uint32_t regs2[4];
util_cpu_caps.cacheline = 32;