[dpdk-dev] [PATCH v5] arch/arm: optimization for memcpy on AArch64

Thomas Monjalon thomas at monjalon.net
Mon Jan 15 12:37:35 CET 2018


15/01/2018 11:57, Herbert Guan:
> Hi Thomas,
> 
> From: Thomas Monjalon [mailto:thomas at monjalon.net]
> > Hi,
> > 
> > All the code is using ARM64, but the title suggests AArch64.
> > What is the difference between AArch64 and ARM64 (and ARMv8)?
> 
> AArch64 and ARM64 refer to the same thing.  AArch64 refers to the 64-bit architecture introduced since ARMv8-A.  But the Linux kernel community calls it as ARM64.  As to DPDK, in most existing compile flags, ARM64 is used.  So this patch keeps the ARM64 naming in newly added compile options.

So please let's continue to call it ARM64.

> > 04/01/2018 11:20, Herbert Guan:
> > > +#define RTE_ARM64_MEMCPY_IS_UNALIGNED_COPY(dst, src) \
> > > +	((uintptr_t)(dst) & RTE_ARM64_MEMCPY_ALIGN_MASK) #else
> > > +/* Both dst and src unalignment will be treated as unaligned copy */
> > > +#define RTE_ARM64_MEMCPY_IS_UNALIGNED_COPY(dst, src) \
> > > +	(((uintptr_t)(dst) | (uintptr_t)(src)) &
> > > +RTE_ARM64_MEMCPY_ALIGN_MASK) #endif
> > > +
> > > +
> > > +/*
> > > + * If copy size is larger than threshold, memcpy() will be used.
> > > + * Run "memcpy_perf_autotest" to determine the proper threshold.
> > > + */
> > > +#define RTE_ARM64_MEMCPY_ALIGNED_THRESHOLD
> > ((size_t)(0xffffffff))
> > > +#define RTE_ARM64_MEMCPY_UNALIGNED_THRESHOLD
> > ((size_t)(0xffffffff))
> > > +
> > > +/*
> > > + * The logic of USE_RTE_MEMCPY() can also be modified to best fit
> > platform.
> > > + */
> > > +#define USE_RTE_MEMCPY(dst, src, n) \
> > > +((!RTE_ARM64_MEMCPY_IS_UNALIGNED_COPY(dst, src) && \ n <=
> > > +RTE_ARM64_MEMCPY_ALIGNED_THRESHOLD) \
> > > +|| (RTE_ARM64_MEMCPY_IS_UNALIGNED_COPY(dst, src) && \
> > > +n <= RTE_ARM64_MEMCPY_UNALIGNED_THRESHOLD))
> > > +
> > > +/**************************************
> > > + * End of customization section
> > > + **************************************/
> > 
> > Modifying the code to asjust the platform is not easy for deployment.
> > Can we move some customization variables inside the configuration file?
> 
> RTE_ARM64_MEMCPY_ALIGNED_THRESHOLD and RTE_ARM64_MEMCPY_UNALIGNED_THRESHOLD are the 2 parameters can be configured during build-time.  The values can be specified with the best values for the target platform.  Usually it's not necessary to change the expression, the comment added in the code is just to raise the hint that this code piece can be modified.

The build time configuration must be set in the config file
(config/common_armv8a_linuxapp).
v6 please?


More information about the dev mailing list