Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add support for intrinsics for NTuple{VecElement} #55118

Draft
wants to merge 12 commits into
base: master
Choose a base branch
from
+567 −173
Draft

Add support for intrinsics for NTuple{VecElement} #55118

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
aacdb85
runtime works
oscardssmith Jul 13, 2024
3b5248f
IT WORKSjsnjsnjsnjsnjsn!
oscardssmith Jul 13, 2024
7b868b1
works with bitcast
oscardssmith Jul 13, 2024
2e9f793
undo array changes
oscardssmith Jul 15, 2024
ddc8890
Update src/intrinsics.cpp
oscardssmith Jul 28, 2024
e6d6f4a
implement vload/vstore! and a primitive Vec type
vchuravy Oct 8, 2024
0a0f7e0
start on preferred_width intrinsic
vchuravy Oct 9, 2024
0e95cfd
fixup! start on preferred_width intrinsic
vchuravy Nov 4, 2024
c8d0ba3
add some basic arithmetic support
vchuravy Nov 4, 2024
799e7e2
add Mask dt
vchuravy Nov 4, 2024
71dd9ab
implement select
vchuravy Nov 4, 2024
52aac7d
WIP: Implement SIMD functionality for XorishoSIMD
vchuravy Nov 5, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 23 additions & 2 deletions base/simd.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
module SIMD

import Base: VecElement, Memory, MemoryRef
import Base: VecElement, Memory, MemoryRef, IEEEFloat
import Base: @propagate_inbounds, @_propagate_inbounds_meta, @_boundscheck, @_noub_if_noinbounds_meta
import Base: memoryrefget, memoryrefnew, memoryrefset!

Expand Down Expand Up @@ -46,7 +46,6 @@ function Base.show(io::IO, v::Vec{N, T}) where {N, T}
print(io, "]")
end

import Base: +, -, *

# Mocked vload/vstore! relying on SLP

Expand Down Expand Up @@ -80,4 +79,26 @@ end
return nothing
end

import Base: +, -, *, /, muladd, promote_rule, widen
import Core.Intrinsics: add_float, sub_float, mul_float, div_float, muladd_float, neg_float

## floating point promotions ##
promote_rule(::Type{Vec{N, Float32}}, ::Type{Vec{N, Float16}}) where N = Vec{N, Float32}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is not obvious to me that these should be defined. When you are doing low level SIMD stuff you probably don't want to accidentally promote things and in case where you really want to work with different types, an explicit convert might be better for clarity?

promote_rule(::Type{Vec{N, Float64}}, ::Type{Vec{N, Float16}}) where N = Vec{N, Float64}
promote_rule(::Type{Vec{N, Float64}}, ::Type{Vec{N, Float32}}) where N = Vec{N, Float64}

widen(::Type{Vec{N, Float16}}) where N = Vec{N, Float16}
widen(::Type{Vec{N, Float32}}) where N = Vec{N, Float32}

## floating point arithmetic ##
-(x::Vec{N,T}) where {N,T<:IEEEFloat} = neg_float(x.data)

+(x::Vec{N,T}, y::Vec{N,T}) where {N,T<:IEEEFloat} = add_float(x.data, y.data)
-(x::Vec{N,T}, y::Vec{N,T}) where {N,T<:IEEEFloat} = sub_float(x.data, y.data)
*(x::Vec{N,T}, y::Vec{N,T}) where {N,T<:IEEEFloat} = mul_float(x.data, y.data)
/(x::Vec{N,T}, y::Vec{N,T}) where {N,T<:IEEEFloat} = div_float(x.data, y.data)

muladd(x::Vec{N,T}, y::Vec{N,T}, z::Vec{N,T}) where {N, T<:IEEEFloat} =
muladd_float(x.data, y.data, z.data)

end # module
21 changes: 21 additions & 0 deletions test/simd.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,3 +46,24 @@ end
@test contains(ir, "load <4 x double>")
@test !contains(ir, "call void @j_throw_boundserror")
end

@testset "basic arithmetic" begin
ir = sprint(io->code_llvm(io, +, (Vec{4, Float64}, Vec{4, Float64})))
@test contains(ir, "fadd <4 x double>")
ir = sprint(io->code_llvm(io, -, (Vec{4, Float64}, Vec{4, Float64})))
@test contains(ir, "fsub <4 x double>")
ir = sprint(io->code_llvm(io, *, (Vec{4, Float64}, Vec{4, Float64})))
@test contains(ir, "fmul <4 x double>")
ir = sprint(io->code_llvm(io, /, (Vec{4, Float64}, Vec{4, Float64})))
@test contains(ir, "fdiv <4 x double>")

ir = sprint(io->code_llvm(io, muladd, (Vec{4, Float64}, Vec{4, Float64}, Vec{4, Float64})))
@test contains(ir, "fmul contract <4 x double>")
@test contains(ir, "fadd contract <4 x double>")

ir = sprint(io->code_llvm(io, -, (Vec{4, Float64},)))
@test contains(ir, "fneg <4 x double>")

# TODO: Way to test Intrinsics directly?
#`-v` -> ERROR: neg_float_withtype: value is not a primitive type
end