[WIP] Refactor JPEG 2000 encoding and decoding for speed

[Copilot]

• Explore codebase and understand architecture
• Add DWT_VERT_STRIP=64 constant to dwt.hpp
• Replace vertical DWT in fdwt.cpp with column-strip approach (single flat allocation, better cache behaviour)
• Replace vertical DWT in fdwt_avx2.cpp with column-strip approach
• Replace vertical DWT in fdwt_neon.cpp with column-strip approach
• Replace vertical IDWT in idwt.cpp with column-strip approach
• Replace vertical IDWT in idwt_avx2.cpp with column-strip approach
• Replace vertical IDWT in idwt_neon.cpp with column-strip approach
• Build and verify correctness (conformance tests)
• Code review + CodeQL check

Original prompt

Read the attached PDF to understand the memory efficient push/pull pipeline for JPEG 2000 compression/decompression. The idea is to exploit the fact that the multi-component transform, spatial wavelet transform and block coding processes can all be performed incrementally. Please read carefully section 3, 4, and 5. With the clear understanding of this paper, please plan for refactoring the current code to improve the end-to-end processing speed including both encodding and decoding. The speed should be firstly measured in single threaded environment, then should be measured in multi-threaded environment. Please use SIMD technologies for both intel SSE/SSE2/SSE3/SSE4/AVX2/AVX512 and ARM NEON as much as possible to make the code faster!

Created from VS Code.

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.