Fast 64-bit atomic writes (SSE?)

From: Roland Dreier (roland_at_digitalvampire.org)
Date: 03/20/04

  • Next message: Jeff Garzik: "Re: [PATCH] barrier patch set" 
    To: linux-kernel@vger.kernel.org
Date:	19 Mar 2004 16:39:17 -0800

    Hi, I'm trying to find the best (fastest) way to write a 64-bit
    quantity atomically to a device (on i386). Taking a spinlock around
    two 32-bit accesses seems to be rather expensive. I'm mostly
    concerned about newish CPUs, so I'm willing to use SSE or SSE2
    instructions (of course falling back to a slower locked path if the
    kernel is built for an old CPU).

    I guess I have two questions, a general one and a specific one.

    General question:
      What's the best way to do this?

    Specific question:
      I've come up with the below function. However, you may notice that
    I'm forced to use movdqu (instead of movdqa, which is presumably
    faster). This is because even with the __attribute__((aligned(16))),
    my xmmsave array is not guaranteed to be aligned to 16 bytes. I could
    just allocate 31 bytes for xmmsave and align it to 16 bytes by hand,
    but that seems a little ugly. Is there some magic I'm missing?

    Thanks,
      Roland

    static inline void mywrite64(u32 *val, void *dest)
    {
            u8 xmmsave[16] __attribute__((aligned(16)));

    /* The #ifs are a hack to deal with 2.4 kernels without preempt. */
    #if CONFIG_PREEMPT
            preempt_disable();
    #endif

            /* We use movdqu for the moment, because even
               __attribute__((aligned(16))) doesn't seem to guarantee
               xmmsave is aligned to a 16 byte boundary. */
            __asm__ __volatile__ (
                    "movdqu %%xmm0,(%0); \n\t"
                    "movq (%1),%%xmm0; \n\t"
                    "movq %%xmm0,(%2); \n\t"
                    "movdqu (%0),%%xmm0; \n\t"
                    :
                    : "r" (xmmsave), "r" (val), "r" (dest)
                    : "memory" );

    #if CONFIG_PREEMPT
            preempt_enable();
    #endif
    }
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Jeff Garzik: "Re: [PATCH] barrier patch set"

    Relevant Pages