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

Jump to bottom

Remove generated annotations in packed.jl #159

Merged
merged 1 commit into from
Oct 13, 2024
Merged

Remove generated annotations in packed.jl #159

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
66 changes: 29 additions & 37 deletions src/utils/packed.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,86 +5,78 @@
# One 64-bit integer packs 8 integers 8-bit each.

# checks that a[i] >= b[i] for all i.
@inline @generated function _packed_vec_ge(a::UInt64, b::UInt64)
@inline function _packed_vec_ge(a::UInt64, b::UInt64)
N = 8
textir = """
define i8 @entry(i64 %0, i64 %1) #0 {
top:
%av = bitcast i64 %0 to <$N x i8>
%bv = bitcast i64 %1 to <$N x i8>
%mask = icmp uge <$N x i8> %av, %bv
%mask.i = bitcast <$N x i1> %mask to i$N
%res = icmp eq i$N %mask.i, $(big(2)^N - 1)
%av = bitcast i64 %0 to <8 x i8>
%bv = bitcast i64 %1 to <8 x i8>
%mask = icmp uge <8 x i8> %av, %bv
%mask.i = bitcast <8 x i1> %mask to i8
%res = icmp eq i8 %mask.i, 255
%retval = zext i1 %res to i8
ret i8 %retval
}
attributes #0 = { alwaysinline }
"""
quote
Base.llvmcall(($textir, "entry"), Bool, Tuple{UInt64, UInt64}, a, b)
end
Base.llvmcall((textir, "entry"), Bool, Tuple{UInt64, UInt64}, a, b)
end

# checks that a, b are orthogonal
@inline @generated function _packed_vec_is_orth(a::UInt64, b::UInt64)
@inline function _packed_vec_is_orth(a::UInt64, b::UInt64)
N = 8
textir = """
define i8 @entry(i64 %0, i64 %1) #0 {
top:
%av = bitcast i64 %0 to <$N x i8>
%bv = bitcast i64 %1 to <$N x i8>
%zero = bitcast i64 0 to <$N x i8>
%mask.a = icmp ne <$N x i8> %av, %zero
%mask.b = icmp ne <$N x i8> %bv, %zero
%mask = and <$N x i1> %mask.a, %mask.b
%mask.i = bitcast <$N x i1> %mask to i$N
%res = icmp eq i$N %mask.i, 0
%av = bitcast i64 %0 to <8 x i8>
%bv = bitcast i64 %1 to <8 x i8>
%zero = bitcast i64 0 to <8 x i8>
%mask.a = icmp ne <8 x i8> %av, %zero
%mask.b = icmp ne <8 x i8> %bv, %zero
%mask = and <8 x i1> %mask.a, %mask.b
%mask.i = bitcast <8 x i1> %mask to i8
%res = icmp eq i8 %mask.i, 0
%retval = zext i1 %res to i8
ret i8 %retval
}
attributes #0 = { alwaysinline }
"""
quote
Base.llvmcall(($textir, "entry"), Bool, Tuple{UInt64, UInt64}, a, b)
end
Base.llvmcall((textir, "entry"), Bool, Tuple{UInt64, UInt64}, a, b)
end

# returns sum a[i]
@inline @generated function _packed_vec_reduce(a::UInt64)
@inline function _packed_vec_reduce(a::UInt64)
N = 8
textir = """
declare i8 @llvm.vector.reduce.add.v$(N)i8(<$N x i8>)
declare i8 @llvm.vector.reduce.add.v8i8(<8 x i8>)
define i8 @entry(i64 %0) #0 {
top:
%a.v = bitcast i64 %0 to <$N x i8>
%sum = call i8 @llvm.vector.reduce.add.v$(N)i8(<$N x i8> %a.v)
%a.v = bitcast i64 %0 to <8 x i8>
%sum = call i8 @llvm.vector.reduce.add.v8i8(<8 x i8> %a.v)
ret i8 %sum
}
attributes #0 = { alwaysinline }
"""
quote
Base.llvmcall(($textir, "entry"), UInt8, Tuple{UInt64}, a)
end
Base.llvmcall((textir, "entry"), UInt8, Tuple{UInt64}, a)
end

# returns c[i] = max a[i], b[i]
@inline @generated function _packed_vec_max(a::UInt64, b::UInt64)
@inline function _packed_vec_max(a::UInt64, b::UInt64)
N = 8
textir = """
declare <$N x i8> @llvm.umax.v$(N)i8(<$N x i8>, <$N x i8>)
declare <8 x i8> @llvm.umax.v8i8(<8 x i8>, <8 x i8>)
define i64 @entry(i64 %0, i64 %1) #0 {
top:
%a.v = bitcast i64 %0 to <$N x i8>
%b.v = bitcast i64 %1 to <$N x i8>
%max.v = call <$N x i8> @llvm.umax.v$(N)i8(<$N x i8> %a.v, <$N x i8> %b.v)
%max = bitcast <$N x i8> %max.v to i64
%a.v = bitcast i64 %0 to <8 x i8>
%b.v = bitcast i64 %1 to <8 x i8>
%max.v = call <8 x i8> @llvm.umax.v8i8(<8 x i8> %a.v, <8 x i8> %b.v)
%max = bitcast <8 x i8> %max.v to i64
ret i64 %max
}
attributes #0 = { alwaysinline }
"""
quote
Base.llvmcall(($textir, "entry"), UInt64, Tuple{UInt64, UInt64}, a, b)
end
Base.llvmcall((textir, "entry"), UInt64, Tuple{UInt64, UInt64}, a, b)
end

@inline @generated function _packed_vec_unpack!(
Expand Down
Loading