On Thu, Mar 02, 2017 at 10:09:31AM +0200, Janne Blomqvist wrote:
> > Here's something from the new matmul_r8_avx2:
> >
> > 156c: c4 62 e5 b8 fd vfmadd231pd %ymm5,%ymm3,%ymm15
> > 1571: c4 c1 79 10 04 06 vmovupd (%r14,%rax,1),%xmm0
> > 1577: c4 62 dd b8 db vfmadd231pd %ymm3,%ymm4,%ymm11
> > 157c: c4 c3 7d 18 44 06 10 vinsertf128
> > $0x1,0x10(%r14,%rax,1),%ymm0,%ymm0
> > 1583: 01
> > 1584: c4 62 ed b8 ed vfmadd231pd %ymm5,%ymm2,%ymm13
> > 1589: c4 e2 ed b8 fc vfmadd231pd %ymm4,%ymm2,%ymm7
> > 158e: c4 e2 fd a8 ad 30 ff vfmadd213pd
> > -0x800d0(%rbp),%ymm0,%ymm5
>
> Great, looks good!
>
> > ... and here from matmul_r8_avx512f:
> >
> > 1da8: c4 a1 7b 10 14 d6 vmovsd (%rsi,%r10,8),%xmm2
> > 1dae: c4 c2 b1 b9 f0 vfmadd231sd %xmm8,%xmm9,%xmm6
> > 1db3: 62 62 ed 08 b9 e5 vfmadd231sd %xmm5,%xmm2,%xmm28
> > 1db9: 62 62 ed 08 b9 ec vfmadd231sd %xmm4,%xmm2,%xmm29
> > 1dbf: 62 62 ed 08 b9 f3 vfmadd231sd %xmm3,%xmm2,%xmm30
> > 1dc5: c4 e2 91 99 e8 vfmadd132sd %xmm0,%xmm13,%xmm5
> > 1dca: c4 e2 99 99 e0 vfmadd132sd %xmm0,%xmm12,%xmm4
> > 1dcf: c4 e2 a1 99 d8 vfmadd132sd %xmm0,%xmm11,%xmm3
> > 1dd4: c4 c2 a9 99 d1 vfmadd132sd %xmm9,%xmm10,%xmm2
> > 1dd9: c4 c2 89 99 c1 vfmadd132sd %xmm9,%xmm14,%xmm0
> > 1dde: 0f 8e d3 fe ff ff jle 1cb7
> > <matmul_r8_avx512f+0x1cb7>
>
> Good, it's using fma, but why is this using xmm registers? That would
> mean it's operating only on 128 bit blocks at a time so no better than
> plain AVX. AFAIU avx512 should use zmm registers to operate on 512 bit
> chunks.
Well, it uses sd, i.e. the scalar fma, not pd, so those are always xmm regs
and only a single double in them, this must be some scalar epilogue loop or
whatever; but matmul_r8_avx512f also has:
140c: 62 72 e5 40 98 c1 vfmadd132pd %zmm1,%zmm19,%zmm8
1412: 62 72 e5 40 98 cd vfmadd132pd %zmm5,%zmm19,%zmm9
1418: 62 72 e5 40 98 d1 vfmadd132pd %zmm1,%zmm19,%zmm10
141e: 62 72 e5 40 98 de vfmadd132pd %zmm6,%zmm19,%zmm11
1424: 62 72 e5 40 98 e1 vfmadd132pd %zmm1,%zmm19,%zmm12
142a: 62 e2 e5 40 98 c6 vfmadd132pd %zmm6,%zmm19,%zmm16
1430: 62 f2 e5 40 98 c8 vfmadd132pd %zmm0,%zmm19,%zmm1
1436: 62 f2 e5 40 98 f0 vfmadd132pd %zmm0,%zmm19,%zmm6
143c: 62 72 e5 40 98 fd vfmadd132pd %zmm5,%zmm19,%zmm15
1442: 62 72 e5 40 98 f4 vfmadd132pd %zmm4,%zmm19,%zmm14
1448: 62 72 e5 40 98 eb vfmadd132pd %zmm3,%zmm19,%zmm13
144e: 62 f2 e5 40 98 d0 vfmadd132pd %zmm0,%zmm19,%zmm2
1454: 62 b2 e5 40 98 ec vfmadd132pd %zmm20,%zmm19,%zmm5
145a: 62 b2 e5 40 98 e4 vfmadd132pd %zmm20,%zmm19,%zmm4
1460: 62 b2 e5 40 98 dc vfmadd132pd %zmm20,%zmm19,%zmm3
1466: 62 b2 e5 40 98 c4 vfmadd132pd %zmm20,%zmm19,%zmm0
etc. where 8 doubles in zmm regs are processed together.
Jakub
