testing new AVX-512 support for vector indexing

[matthias314]

If you have a computer with AVX-512 (more precisely: AVX512BW), then you can help test the new code for vector indexing. The code supports AVX, AVX2 and AVX-512. However, I don't have access to a computer with AVX-512. On Linux, you can check if you have it by looking at the output of cat /proc/cpuinfo.

To test the AVX-512 code, install the package (v0.5.0) and then say

julia> using SmallCollections

julia> M = 32; v = -FixedVector{M,Int8}(1:M); p = ntuple(i -> Int8(i-1), M);

julia> SmallCollections.shuffle(Val(64), v, p)

The result should be equal to v. Once this works, we can compare it with AVX2 shuffles, for example via

julia> using Chairmarks

julia> @b SmallCollections.shuffle(Val(32), $v, $p), SmallCollections.shuffle(Val(64), $v, $p)

Your help is appreciated!

[oameye]

processor       : 53
vendor_id       : GenuineIntel
cpu family      : 6
model           : 85
model name      : Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz
stepping        : 7
microcode       : 0x5003707
cpu MHz         : 2497.295
cache size      : 28160 KB
physical id     : 0
siblings        : 40
core id         : 19
cpu cores       : 20
apicid          : 39
initial apicid  : 39
fpu             : yes
fpu_exception   : yes
cpuid level     : 22
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts vnmi pku ospke avx512_vnni md_clear flush_l1d arch_capabilities
vmx flags       : vnmi preemption_timer posted_intr invvpid ept_x_only ept_ad ept_1gb flexpriority apicv tsc_offset vtpr mtf vapic ept vpid unrestricted_guest vapic_reg vid ple shadow_vmcs pml ept_mode_based_exec tsc_scaling
bugs            : spectre_v1 spectre_v2 spec_store_bypass swapgs taa itlb_multihit mmio_stale_data retbleed eibrs_pbrsb gds bhi
bogomips        : 5000.00
clflush size    : 64
cache_alignment : 64
address sizes   : 46 bits physical, 48 bits virtual
power management:

using SmallCollections

M = 32; v = -FixedVector{M,Int8}(1:M); p = ntuple(i -> Int8(i-1), M);

SmallCollections.shuffle(Val(64), v, p)

32-element FixedVector{32, Int8}:
  -1
  -2
  -3
  -4
  -5
  -6
  -7
  -8
  -9
 -10
 -11
 -12
 -13
 -14
 -15
 -16
 -17
 -18
 -19
 -20
 -21
 -22
 -23
 -24
 -25
 -26
 -27
 -28
 -29
 -30
 -31
 -32

using Chairmarks
@b SmallCollections.shuffle(Val(32), $v, $p), SmallCollections.shuffle(Val(64), $v, $p)

(3.245 ns, 2.949 ns)

[matthias314]

Great, thanks a lot!

What about this benchmark (two outputs!):

using SmallCollections, Chairmarks
using SmallCollections: shuffle
M = 1000
a = [rand(FixedVector{32,Int8}) for _ in 1:M]
b = [NTuple{32,Int8}(rand(0:31, 32)) for _ in 1:M]
@b similar(a) map!((v, p) -> shuffle(Val(32), v, p), _, $a, $b)
@b similar(a) map!((v, p) -> shuffle(Val(64), v, p), _, $a, $b)

Could you also run the tests to make sure that there are no errors?

pkg> test SmallCollections

Thanks again!

[oameye]

using SmallCollections, Chairmarks
using SmallCollections: shuffle
M = 1000
a = [rand(FixedVector{32,Int8}) for _ in 1:M]

ERROR: MethodError: no method matching Random.Sampler(::Type{Random.TaskLocalRNG}, ::Random.SamplerType{FixedVector{32, Int8}}, ::Val{1})

Closest candidates are:
  Random.Sampler(::Type{<:Random.AbstractRNG}, ::Random.Sampler, ::Union{Val{1}, Val{Inf}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:147
  Random.Sampler(::Type{<:Random.AbstractRNG}, ::Any, ::Union{Val{1}, Val{Inf}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:183
  Random.Sampler(::Type{<:Random.AbstractRNG}, ::BitSet, ::Union{Val{1}, Val{Inf}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/generation.jl:450
  ...

Stacktrace:
 [1] Random.Sampler(T::Type{Random.TaskLocalRNG}, sp::Random.SamplerType{FixedVector{32, Int8}}, r::Val{1})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:147
 [2] Random.Sampler(rng::Random.TaskLocalRNG, x::Random.SamplerType{FixedVector{32, Int8}}, r::Val{1})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:139
 [3] rand(rng::Random.TaskLocalRNG, X::Random.SamplerType{FixedVector{32, Int8}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:255
 [4] rand(rng::Random.TaskLocalRNG, ::Type{FixedVector{32, Int8}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:258
 [5] rand(::Type{FixedVector{32, Int8}})
   @ Random ~/.julia/juliaup/julia-1.10.10+0.x64.linux.gnu/share/julia/stdlib/v1.10/Random/src/Random.jl:261
 [6] (::var"#17#18")(::Int64)
   @ Main ./none:0
 [7] iterate
   @ ./generator.jl:47 [inlined]
 [8] collect(itr::Base.Generator{UnitRange{Int64}, var"#17#18"})
   @ Base ./array.jl:834
 [9] top-level scope

[matthias314]

Which Julia version are you using? rand(FixedVector{32,Int8}) needs Julia 1.11. For Julia 1.10, you can say

a = [FixedVector{32,Int8}(rand(Int8, 32)) for _ in 1:M]

[oameye]

On julia 1.11:

using SmallCollections, Chairmarks
using SmallCollections: shuffle
M = 1000
a = [rand(FixedVector{32,Int8}) for _ in 1:M]
b = [NTuple{32,Int8}(rand(0:31, 32)) for _ in 1:M]
@b similar(a) map!((v, p) -> shuffle(Val(32), v, p), _, $a, $b)
@b similar(a) map!((v, p) -> shuffle(Val(64), v, p), _, $a, $b)

[oameye]

The test still fail with the same error

[matthias314]

OK, thanks! I'm going to look at the error soon.