On Fri, Apr 05, 2024 at 02:46:28PM +0200, Morten Brørup wrote:
> When the rte_memcpy() size is 16, the same 16 bytes are copied twice.
> In the case where the size is known to be 16 at build tine, omit the
> duplicate copy.
>
> Reduced the amount of effectively copy-pasted code by using #ifdef
> inside functions instead of outside functions.
>
> Suggested-by: Stephen Hemminger <step...@networkplumber.org>
> Signed-off-by: Morten Brørup <m...@smartsharesystems.com>
> Acked-by: Bruce Richardson <bruce.richard...@intel.com>
> ---
> v3:
> * AVX2 is a superset of AVX;
> for a block of AVX code, testing for AVX suffices. (Bruce Richardson)
> * Define RTE_MEMCPY_AVX if AVX is available, to avoid copy-pasting the
> check for older GCC version. (Bruce Richardson)
> v2:
> * For GCC, version 11 is required for proper AVX handling;
> if older GCC version, treat AVX as SSE.
> Clang does not have this issue.
> Note: Original code always treated AVX as SSE, regardless of compiler.
> * Do not add copyright. (Stephen Hemminger)
> ---
> lib/eal/x86/include/rte_memcpy.h | 234 ++++++++-----------------------
> 1 file changed, 59 insertions(+), 175 deletions(-)
>
> diff --git a/lib/eal/x86/include/rte_memcpy.h
> b/lib/eal/x86/include/rte_memcpy.h
> index 72a92290e0..b56bc46713 100644
> --- a/lib/eal/x86/include/rte_memcpy.h
> +++ b/lib/eal/x86/include/rte_memcpy.h
> @@ -27,6 +27,11 @@ extern "C" {
> #pragma GCC diagnostic ignored "-Wstringop-overflow"
> #endif
>
> +/* GCC prior to version 11 doesn't compile AVX properly, so use SSE instead.
> */
> +#if defined __AVX__ && !(defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION <
> 110000))
> +#define RTE_MEMCPY_AVX
> +#endif
> +
Strictly speaking, to have the same behaviour as before, you need to check
for AVX2 also, since the issue with GCC < 11 is for (AVX && !AVX2), i.e. if
AVX2 is supported, all compilers are fine.
My suggestion:
#ifdef __AVX2__
#define RTE_MEMCPY_AVX
#elif defined __AVX__ && !(defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 110000))
#define RTE_MEMCPY_AVX
#endif
You can obviously merge the two branches if you want, but I find the split
slightly easier to follow, than a mix of && and || with brackets for
precedence.
Final alternative I see, you can change defined(RTE_MEMCPY_AVX) to
"defined(__AVX2__) || defined(RTE_MEMCPY_AVX)" each place it's used.
/Bruce
