Add static SIMD support to XXH3 in core:hash/xxhash.

This uses compile-time features to decide how large of a SIMD vector to
use. It currently has checks for amd64/i386 to size its vectors for
SSE2/AVX2/AVX512 as necessary.

The generalized SIMD functions could also be useful for multiversioning
of the hash procs, to allow for run-time dispatch based on available CPU
features.

1 files changed, 32 insertions, 0 deletions

diff --git a/core/hash/xxhash/common.odin b/core/hash/xxhash/common.odin
index adfc1bac2..636393b52 100644
--- a/core/hash/xxhash/common.odin
+++ b/core/hash/xxhash/common.odin
@@ -101,3 +101,35 @@ XXH64_read64 :: #force_inline proc(buf: []u8, alignment := Alignment.Unaligned)
 		return u64(b)
 	}
 }
+
+XXH64_read64_simd :: #force_inline proc(buf: []$E, $W: uint, alignment := Alignment.Unaligned) -> (res: #simd[W]u64) {
+	if alignment == .Aligned {
+		res = (^#simd[W]u64)(raw_data(buf))^
+	} else {
+		res = intrinsics.unaligned_load((^#simd[W]u64)(raw_data(buf)))
+	}
+
+	when ODIN_ENDIAN == .Big {
+		bytes := transmute(#simd[W*8]u8)res
+		bytes = intrinsics.simd_lanes_reverse(bytes)
+		res = transmute(#simd[W]u64)bytes
+		res = intrinsics.simd_lanes_reverse(res)
+	}
+	return
+}
+
+XXH64_write64_simd :: #force_inline proc(buf: []$E, value: $V/#simd[$W]u64, alignment := Alignment.Unaligned) {
+	value := value
+	when ODIN_ENDIAN == .Big {
+		bytes := transmute(#simd[W*8]u8)value
+		bytes = intrinsics.simd_lanes_reverse(bytes)
+		value = transmute(#simd[W]u64)bytes
+		value = intrinsics.simd_lanes_reverse(value)
+	}
+
+	if alignment == .Aligned {
+		(^V)(raw_data(buf))^ = value
+	} else {
+		intrinsics.unaligned_store((^V)(raw_data(buf)), value)
+	}
+}