SIMD vectorization for str/bytes encode/decode operations

[albertedwardson]

Feature or enhancement

Proposal:

Following the successful SIMD optimization of bytes.hex() in #144015, I propose implementing SIMD vectorization for common str/bytes encode/decode operations, particularly UTF-8 encoding/decoding. This could provide significant performance improvements for web applications that frequently process text data.

Web applications perform extensive encode/decode operations:

• Content-Length headers: Computing response sizes requires encoding strings to bytes
• Multipart responses: Encoding form data and boundaries
• Middleware: Request/response processing often involves repeated encoding/decoding
• JSON/HTTP processing: Converting between text and binary representations

These operations are frequently performed on the same strings multiple times, making them ideal candidates for SIMD optimization.

cc @gpshead - Would appreciate your insights on extending the SIMD pattern to encode/decode operations.

May be related: #145797

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

[picnixz]

I think this is premature for having an issue. An issue is for something actionable. We don't have anything. It was already proposed in the past but we don't want to keep them open. We can't have just "this can improve this or that". We need to know exactly how and what it can improve, what the support is, what could be regressions, etc.