implement 64 bit unpacking optimizations for arm64

Signed-off-by: Achille Roussel <achille.roussel@gmail.com>

Author: achille-roussel

unpack_int32_arm64.go

@@ -10,14 +10,13 @@ import (
 func unpackInt32Default(dst []int32, src []byte, bitWidth uint)
 
 //go:noescape
-func unpackInt32x1to16bitsNEON(dst []int32, src []byte, bitWidth uint)
+func unpackInt32x1to16bitsARM64(dst []int32, src []byte, bitWidth uint)
 
 func unpackInt32(dst []int32, src []byte, bitWidth uint) {
-	// For ARM64, we use simplified NEON for byte-aligned cases (8, 16 bits)
-	// and scalar for other cases
+	// For ARM64, we use optimized scalar operations for small bit widths
 	switch {
 	case bitWidth <= 16:
-		unpackInt32x1to16bitsNEON(dst, src, bitWidth)
+		unpackInt32x1to16bitsARM64(dst, src, bitWidth)
 	case bitWidth == 32:
 		copy(dst, unsafecast.Slice[int32](src))
 	default:

unpack_int32_arm64.s

@@ -60,12 +60,11 @@ test:
 	BNE loop
 	RET
 
-// unpackInt32x1to16bitsNEON implements NEON-optimized unpacking for bit widths 1-16
-// This simplified version handles byte-aligned cases (8, 16 bits) with NEON,
-// and falls back to scalar for complex cases.
+// unpackInt32x1to16bitsARM64 implements optimized unpacking for bit widths 1-16
+// Uses optimized scalar ARM64 operations with batched processing
 //
-// func unpackInt32x1to16bitsNEON(dst []int32, src []byte, bitWidth uint)
-TEXT ·unpackInt32x1to16bitsNEON(SB), NOSPLIT, $0-56
+// func unpackInt32x1to16bitsARM64(dst []int32, src []byte, bitWidth uint)
+TEXT ·unpackInt32x1to16bitsARM64(SB), NOSPLIT, $0-56
 	MOVD dst_base+0(FP), R0   // R0 = dst pointer
 	MOVD dst_len+8(FP), R1    // R1 = dst length
 	MOVD src_base+24(FP), R2  // R2 = src pointer

unpack_int64_arm64.go

@@ -9,10 +9,14 @@ import (
 //go:noescape
 func unpackInt64Default(dst []int64, src []byte, bitWidth uint)
 
+//go:noescape
+func unpackInt64x1to32bitsARM64(dst []int64, src []byte, bitWidth uint)
+
 func unpackInt64(dst []int64, src []byte, bitWidth uint) {
-	// For ARM64, we'll use NEON instructions
-	// TODO: Implement NEON optimizations - using default for now
+	// For ARM64, use optimized scalar operations for common bit widths
 	switch {
+	case bitWidth <= 32:
+		unpackInt64x1to32bitsARM64(dst, src, bitWidth)
 	case bitWidth == 64:
 		copy(dst, unsafecast.Slice[int64](src))
 	default: