refactor(rust): Move sum kernel to polars-compute (#20867)

Cargo.toml

@@ -53,7 +53,6 @@ itoap = { version = "1", features = ["simd"] }
 libc = "0.2"
 memchr = "2.6"
 memmap = { package = "memmap2", version = "0.9" }
-multiversion = "0.8"
 ndarray = { version = "0.16", default-features = false }
 num-traits = "0.2"
 numpy = "0.23"

crates/polars-arrow/Cargo.toml

@@ -61,9 +61,6 @@ avro-schema = { workspace = true, optional = true }
 # for division/remainder optimization at runtime
 strength_reduce = { workspace = true, optional = true }
 
-# For instruction multiversioning
-multiversion = { workspace = true, optional = true }
-
 # Faster hashing
 ahash = { workspace = true }
 
@@ -122,14 +119,13 @@ io_avro_compression = [
 io_avro_async = ["avro-schema/async"]
 
 # the compute kernels. Disabling this significantly reduces compile time.
-compute_aggregate = ["multiversion"]
+compute_aggregate = []
 compute_arithmetics_decimal = ["strength_reduce"]
 compute_arithmetics = ["strength_reduce", "compute_arithmetics_decimal"]
 compute_bitwise = []
 compute_boolean = []
 compute_boolean_kleene = []
 compute_comparison = ["compute_boolean"]
-compute_hash = ["multiversion"]
 compute_temporal = []
 compute = [
   "compute_aggregate",
@@ -138,7 +134,6 @@ compute = [
   "compute_boolean",
   "compute_boolean_kleene",
   "compute_comparison",
-  "compute_hash",
   "compute_temporal",
 ]
 serde = ["dep:serde", "polars-schema/serde", "polars-utils/serde"]
@@ -161,4 +156,4 @@ features = ["full"]
 rustdoc-args = ["--cfg", "docsrs"]
 
 [package.metadata.cargo-all-features]
-allowlist = ["compute", "compute_sort", "compute_hash", "compute_nullif"]
+allowlist = ["compute", "compute_sort", "compute_nullif"]

crates/polars-arrow/src/compute/aggregate/mod.rs

@@ -1,10 +1,2 @@
-/// ! Contains different aggregation functions
-#[cfg(feature = "compute_aggregate")]
-mod sum;
-#[cfg(feature = "compute_aggregate")]
-pub use sum::*;
-
 mod memory;
 pub use memory::*;
-#[cfg(feature = "compute_aggregate")]
-mod simd;

crates/polars-arrow/src/types/mod.rs

@@ -16,16 +16,12 @@
 //! represent chunks of bits (e.g. 8 bits via `u8`, 16 via `u16`), and [`BitChunkIter`],
 //! that can be used to iterate over bitmaps in [`BitChunk`]s according to
 //! Arrow's definition of bitmaps.
-//!
-//! Finally, this module contains traits used to compile code based on [`NativeType`] optimized
-//! for SIMD, at [`mod@simd`].
 
 mod aligned_bytes;
 pub use aligned_bytes::*;
 mod bit_chunk;
 pub use bit_chunk::{BitChunk, BitChunkIter, BitChunkOnes};
 mod index;
-pub mod simd;
 pub use index::*;
 mod native;
 pub use native::*;