Messages in this thread | | | From | Ard Biesheuvel <> | Date | Thu, 22 Nov 2018 09:55:18 +0100 | Subject | Re: Re: [PATCH] arm64: crc: accelerated-crc32-by-64bytes |
| |
On Thu, 22 Nov 2018 at 02:50, sunrui <sunrui26@huawei.com> wrote: > > > > On Sun, 18 Nov 2018 at 23:30, Rui Sun <sunrui26@huawei.com> wrote: > > > > > > add 64 bytes loop to acceleration calculation > > > > > > > Can you share some performance numbers please? > > > > Also, we don't need 64 byte, 32 byte and 16 byte code paths: just make the 8 byte one a loop as well, and drop the 32 byte and 16 byte ones. > > > > -- > > > > Consider of some processor has instruction N-way parallel function, with the increase of the data buf’s size, 64B loop will performance better than 16B loop. > > > > On the other hand, in the same environment I tested the 8B loop, which is worse than the 16-byte loop. > > > > The test result is shown in the fellow excel(crc test result.xlsx) sheet1(64B loop) and sheet2(8B loop) >
Maybe I phrased that wrong: if we add the 64-byte loop, there is no need for a 32-byte block, a 16 byte block and a 8 byte block, since they all use the same crc32x instruction. After the 64-byte loop, just loop in the 8-byte sequence until the remaining data is less than 8 bytes.
> > > -- > > > > > Signed-off-by: Rui Sun <sunrui26@huawei.com> > > > --- > > > arch/arm64/lib/crc32.S | 54 > > > ++++++++++++++++++++++++++++++++++++++++++++++---- > > > 1 file changed, 50 insertions(+), 4 deletions(-) > > > > > > diff --git a/arch/arm64/lib/crc32.S b/arch/arm64/lib/crc32.S index > > > 5bc1e85..2b37009 100644 > > > --- a/arch/arm64/lib/crc32.S > > > +++ b/arch/arm64/lib/crc32.S > > > @@ -15,15 +15,61 @@ > > > .cpu generic+crc > > > > > > .macro __crc32, c > > > -0: subs x2, x2, #16 > > > - b.mi 8f > > > + > > > +64: cmp x2, #64 > > > + b.lt 32f > > > + > > > + adds x11, x1, #16 > > > + adds x12, x1, #32 > > > + adds x13, x1, #48 > > > + > > > +0 : subs x2, x2, #64 > > > + b.mi 32f > > > + > > > + ldp x3, x4, [x1], #64 > > > + ldp x5, x6, [x11], #64 > > > + ldp x7, x8, [x12], #64 > > > + ldp x9, x10,[x13], #64 > > > + > > > > Can we do this instead, and get rid of the temp variables? > > > > ldp x3, x4, [x1], #64 > > ldp x5, x6, [x1, #-48] > > ldp x7, x8, [x1, #-32] > > ldp x9, x10,[x1, #-16] > > > > > + CPU_BE( rev x3, x3 ) > > > + CPU_BE( rev x4, x4 ) > > > + CPU_BE( rev x5, x5 ) > > > + CPU_BE( rev x6, x6 ) > > > + CPU_BE( rev x7, x7 ) > > > + CPU_BE( rev x8, x8 ) > > > + CPU_BE( rev x9, x9 ) > > > + CPU_BE( rev x10,x10 ) > > > + > > > + crc32\c\()x w0, w0, x3 > > > + crc32\c\()x w0, w0, x4 > > > + crc32\c\()x w0, w0, x5 > > > + crc32\c\()x w0, w0, x6 > > > + crc32\c\()x w0, w0, x7 > > > + crc32\c\()x w0, w0, x8 > > > + crc32\c\()x w0, w0, x9 > > > + crc32\c\()x w0, w0, x10 > > > + > > > + b.ne 0b > > > + ret > > > + > > > +32: tbz x2, #5, 16f > > > + ldp x3, x4, [x1], #16 > > > + ldp x5, x6, [x1], #16 > > > +CPU_BE( rev x3, x3 ) > > > +CPU_BE( rev x4, x4 ) > > > +CPU_BE( rev x5, x5 ) > > > +CPU_BE( rev x6, x6 ) > > > + crc32\c\()x w0, w0, x3 > > > + crc32\c\()x w0, w0, x4 > > > + crc32\c\()x w0, w0, x5 > > > + crc32\c\()x w0, w0, x6 > > > + > > > +16: tbz x2, #4, 8f > > > ldp x3, x4, [x1], #16 > > > CPU_BE( rev x3, x3 ) > > > CPU_BE( rev x4, x4 ) > > > crc32\c\()x w0, w0, x3 > > > crc32\c\()x w0, w0, x4 > > > - b.ne 0b > > > - ret > > > > > > 8: tbz x2, #3, 4f > > > ldr x3, [x1], #8 > > > -- > > > 1.8.3.1 > > > > >
| |