Hello,
I have the following 3 functions that utilize NEON instruction set:
function pixel_avg2_w8_neon, export=1 1: subs w5, w5, #2 ld1 {v0.8b}, [x2], x3 ld1 {v2.8b}, [x4], x3 urhadd v0.8b, v0.8b, v2.8b ld1 {v1.8b}, [x2], x3 ld1 {v3.8b}, [x4], x3 urhadd v1.8b, v1.8b, v3.8b st1 {v0.8b}, [x0], x1 st1 {v1.8b}, [x0], x1 b.gt 1b ret endfunc
function pixel_avg2_w16_neon, export=1 1: subs w5, w5, #2 ld1 {v0.16b}, [x2], x3 ld1 {v2.16b}, [x4], x3 urhadd v0.16b, v0.16b, v2.16b ld1 {v1.16b}, [x2], x3 ld1 {v3.16b}, [x4], x3 urhadd v1.16b, v1.16b, v3.16b st1 {v0.16b}, [x0], x1 st1 {v1.16b}, [x0], x1 b.gt 1b ret endfunc
function pixel_sad_\h\()_neon, export=1 ld1 {v1.16b}, [x2], x3 ld1 {v0.16b}, [x0], x1 ld1 {v3.16b}, [x2], x3 ld1 {v2.16b}, [x0], x1 uabdl v16.8h, v0.8b, v1.8b uabdl2 v17.8h, v0.16b, v1.16b uabal v16.8h, v2.8b, v3.8b uabal2 v17.8h, v2.16b, v3.16b .rept \h / 2 - 1 ld1 {v1.16b}, [x2], x3 ld1 {v0.16b}, [x0], x1 ld1 {v3.16b}, [x2], x3 ld1 {v2.16b}, [x0], x1 uabal v16.8h, v0.8b, v1.8b uabal2 v17.8h, v0.16b, v1.16b uabal v16.8h, v2.8b, v3.8b uabal2 v17.8h, v2.16b, v3.16b .endr add v16.8h, v16.8h, v17.8h uaddlv s0, v16.8h fmov w0, s0 ret endfunc
I want to use SVE/SVE2 instructions set to improve the performance of these functions. My testbed is Alibaba Yitian 710 (vector size=128 bits).
For the first 2, I couldn't find a way to improve the performance. For the latter, I wrote the following function:
function pixel_sad_\h\()_sve, export=1 ptrue p0.h, vl8 ld1b {z1.h}, p0/z, [x2] ld1b {z4.h}, p0/z, [x2, #1, mul vl] add x2, x2, x3 ld1b {z3.h}, p0/z, [x2] ld1b {z6.h}, p0/z, [x2, #1, mul vl] add x2, x2, x3 ld1b {z0.h}, p0/z, [x0] ld1b {z5.h}, p0/z, [x0, #1, mul vl] add x0, x0, x1 ld1b {z2.h}, p0/z, [x0] ld1b {z7.h}, p0/z, [x0, #1, mul vl] add x0, x0, x1 uabd v16.8h, v0.8h, v1.8h uabd v17.8h, v4.8h, v5.8h uaba v16.8h, v2.8h, v3.8h uaba v17.8h, v7.8h, v6.8h .rept \h / 2 - 1 ld1b {z1.h}, p0/z, [x2] ld1b {z4.h}, p0/z, [x2, #1, mul vl] add x2, x2, x3 ld1b {z3.h}, p0/z, [x2] ld1b {z6.h}, p0/z, [x2, #1, mul vl] add x2, x2, x3 ld1b {z0.h}, p0/z, [x0] ld1b {z5.h}, p0/z, [x0, #1, mul vl] add x0, x0, x1 ld1b {z2.h}, p0/z, [x0] ld1b {z7.h}, p0/z, [x0, #1, mul vl] add x0, x0, x1 uaba v16.8h, v0.8h, v1.8h uaba v17.8h, v4.8h, v5.8h uaba v16.8h, v2.8h, v3.8h uaba v17.8h, v7.8h, v6.8h .endr add v16.8h, v16.8h, v17.8h uaddlv s0, v16.8h fmov w0, s0 ret endfunc
However, this degrades the performance instead of improving it.
Can someone help me?
Thank you in advance,
Akis
Hi George,
For sub8x8_dct8_neon, I applied your suggestion and everything worked fine. Thanks!
For the copy functions, as you said, there is no much left to do for improving the performance.
For mbtree_propagate_list_internal_neon, I applied your suggestion. Thanks!
For pixel_var2_8x\h\()_neon, I used the udot instruction, but it doesn't work. It seems that some vectors (for example v0.8h, v1.8h, v6.8h and v7.8h) are still needed after widening instructions. I developed the following function:
function pixel_var2_8x\h\()_sve, export=1 movi v30.4s, #0 movi v31.4s, #0 mov x3, #16 ld1 {v16.8b}, [x0], #8 ld1 {v18.8b}, [x1], x3 ld1 {v17.8b}, [x0], #8 ld1 {v19.8b}, [x1], x3 mov x5, \h - 2 uabd v28.8b, v16.8b, v18.8b usubl v0.8h, v16.8b, v18.8b uabd v29.8b, v17.8b, v19.8b usubl v1.8h, v17.8b, v19.8b ld1 {v16.8b}, [x0], #8 ld1 {v18.8b}, [x1], x3 udot v30.2s, v28.8b, v28.8b udot v31.2s, v29.8b, v29.8b uabd v28.8b, v16.8b, v18.8b usubl v6.8h, v16.8b, v18.8b 1: subs x5, x5, #1 ld1 {v17.8b}, [x0], #8 ld1 {v19.8b}, [x1], x3 udot v30.2s, v28.8b, v28.8b uabd v29.8b, v17.8b, v19.8b usubl v7.8h, v17.8b, v19.8b add v0.8h, v0.8h, v6.8h ld1 {v16.8b}, [x0], #8 ld1 {v18.8b}, [x1], x3 udot v31.2s, v29.8b, v29.8b uabd v28.8b, v16.8b, v18.8b usubl v6.8h, v16.8b, v18.8b add v1.8h, v1.8h, v7.8h b.gt 1b ld1 {v17.8b}, [x0], #8 ld1 {v19.8b}, [x1], x3 udot v30.2s, v6.8b, v6.8b uabd v29.8b, v17.8b, v19.8b usubl v7.8h, v17.8b, v19.8b add v0.8h, v0.8h, v6.8h udot v31.2s, v29.8b, v29.8b add v1.8h, v1.8h, v7.8h saddlv s0, v0.8h saddlv s1, v1.8h mov w0, v0.s[0] mov w1, v1.s[0] addv s2, v30.4s addv s4, v31.4s mul w0, w0, w0 mul w1, w1, w1 mov w3, v30.s[0] mov w4, v31.s[0] sub w0, w3, w0, lsr # 6 + (\h >> 4) sub w1, w4, w1, lsr # 6 + (\h >> 4) str w3, [x2] add w0, w0, w1 str w4, [x2, #4] ret endfunc
Unit test fail. Can you please tell me what I am doing wrong? Also, the usage of the three load merging commands instead of the initial four, degrades the performance. I do not know why.
For pixel_sad_x_h\()_neon_10, I also agree that we cannot improve it.
BR,
Hi Akis,
It's a bit hard for me to try and debug the whole code snippet. One thing I did notice though is that at the end of the function you reduce v30 and v31 as such:
addv s2, v30.4s addv s4, v31.4s ... mov w3, v30.s[0] // Should this be v2.s[0] ? mov w4, v31.s[0] // Should this be v4.s[0] ?
This seems suspicious since s2 and s4 are otherwise never used after those instructions.
With regards to still needing the USUBL, do you know if either the absolute difference (UABD) or a non-widening subtract (SUB) would work here instead? If so then we can potentially use only one of those instead since the UABD and USUBL are doing very similar things at the moment? Assuming that a non-widening approach works here you could then sum the results with UADDW or another UDOT instruction with all-1s as the other operand.
For example, instead of:
uabd v28.8b, v16.8b, v18.8b usubl v6.8h, v16.8b, v18.8b udot v30.2s, v28.8b, v28.8b add v0.8h, v0.8h, v6.8h
We could see if something like this would work instead:
uabd v28.8b, v16.8b, v18.8b // or SUB? udot v30.2s, v28.8b, v28.8b uaddw v0.8h, v0.8h, v28.8b
Using the dot product would also work here if we need to widen beyond a 16-bit accumulator for v0 since it allows us to accumulate in 32-bits by multiplying by a vector of all-1s:
mov v6.16b, #1 ... uabd v28.8b, v16.8b, v18.8b // or SUB? udot v30.2s, v28.8b, v28.8b udot v0.2s, v28.8b, v6.8b // v28.8b * 1
If an approach like that works then at that point it may be beneficial to re-try the three-load appoach since the entire computation can be moved from .8b to .16b which could be more significant than your previous attempt?
Hope that helps!
Thanks,George
after using the mov instructions you proposed, everything worked fine! Thanks!
Unfortunately, after some testing, I can neither use sub nor uabd. Unit tests fail again. So I can not use the 3 load instructions as well. But your proposed solution is very interesting and it may help me optimize other functions. Thanks!
I think we can close this thread. You gave me a lot of help. I couldn't reach up to this point without your help. Once again, thanks!
If I need further help, I will create new thread (I hope that this is OK).