Hi Ivan,
On Wed, Feb 11, 2009 at 10:48 AM, Ivan Kokshaysky
<ink@xxxxxxxxxxxxxxxxxxxx> wrote:
> Well, I've got slightly improved version, which saves one instruction
> for register-register move. Another instruction can be saved if we
> define DRM_CAS_RESULT(_result) as "long" (no need for sign extending then).
>
> Can you test this please?
>
> Ivan.
>
> --- libdrm-2.4.4/libdrm/xf86drm.h.orig Wed Dec 17 21:28:24 2008
> +++ libdrm-2.4.4/libdrm/xf86drm.h Wed Feb 11 03:07:05 2009
> @@ -325,28 +325,28 @@ typedef struct _drmSetVersion {
>
> #elif defined(__alpha__)
>
> -#define DRM_CAS(lock, old, new, ret) \
> - do { \
> - int old32; \
> - int cur32; \
> - __asm__ __volatile__( \
> - " mb\n" \
> - " zap %4, 0xF0, %0\n" \
> - " ldl_l %1, %2\n" \
> - " zap %1, 0xF0, %1\n" \
> - " cmpeq %0, %1, %1\n" \
> - " beq %1, 1f\n" \
> - " bis %5, %5, %1\n" \
> - " stl_c %1, %2\n" \
> - "1: xor %1, 1, %1\n" \
> - " stl %1, %3" \
> - : "=r" (old32), \
> - "=&r" (cur32), \
> - "=m" (__drm_dummy_lock(lock)),\
> - "=m" (ret) \
> - : "r" (old), \
> - "r" (new)); \
> - } while(0)
> +#define DRM_CAS(lock, old, new, ret) \
> + do { \
> + int tmp, old32; \
> + __asm__ __volatile__( \
> + " addl $31, %5, %3\n" \
> + "1: ldl_l %0, %2\n" \
> + " cmpeq %0, %3, %1\n" \
> + " beq %1, 2f\n" \
> + " mov %4, %0\n" \
> + " stl_c %0, %2\n" \
> + " beq %0, 3f\n" \
> + " mb\n" \
> + "2: cmpeq %1, 0, %1\n" \
> + ".subsection 2\n" \
> + "3: br 1b\n" \
> + ".previous" \
> + : "=&r"(tmp), "=&r"(ret), \
> + "=m"(__drm_dummy_lock(lock)), \
> + "=&r"(old32) \
> + : "r"(new), "r"(old) \
> + : "memory"); \
> + } while (0)
>
> #elif defined(__sparc__)
>
> @@ -429,7 +429,9 @@ do { register unsigned int __old __asm("
> #define DRM_CAS(lock,old,new,ret) do { ret=1; } while (0) /* FAST LOCK FAILS */
> #endif
>
> -#if defined(__alpha__) || defined(__powerpc__)
> +#if defined(__alpha__)
> +#define DRM_CAS_RESULT(_result) long _result
> +#elif defined(__powerpc__)
> #define DRM_CAS_RESULT(_result) int _result
> #else
> #define DRM_CAS_RESULT(_result) char _result
>
I changed 'old32' to 'old' for the %3 output argument, but it fails to
compile with
> error: lvalue required in asm statement
> error: invalid lvalue in asm output 3
I saw this error with the first patch as well, but it was easily
avoided by changing old to an input. I'd fix it, but I don't know what
the appropriate thing is.
Thanks,
Matt Turner
--
To unsubscribe from this list: send the line "unsubscribe linux-alpha" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
