feat(simd-neon): 6 NEON integer wrapper types for aarch64 (sprint W3-B)#127
Merged
Conversation
Closes parity item 8 — adds U8x16, U16x8, U32x4, U64x2, I32x4, I64x2 NEON wrapper types so aarch64 burn-ndarray builds get real NEON acceleration on integer hot paths instead of scalar. Each type has: splat/zero/from_slice/from_array/to_array/copy_to_slice/ add/sub/min/max. NEON intrinsics: - U8x16: vaddq_u8, vsubq_u8, vminq_u8, vmaxq_u8 - U16x8: vaddq_u16, vsubq_u16, vminq_u16, vmaxq_u16 - U32x4: vaddq_u32, vsubq_u32, vminq_u32, vmaxq_u32 - U64x2: vaddq_u64, vsubq_u64 (min/max scalar — NEON has no vminq_u64) - I32x4: vaddq_s32, vsubq_s32, vminq_s32, vmaxq_s32 - I64x2: vaddq_s64, vsubq_s64 (min/max scalar — NEON has no vminq_s64) Item 7 (AVX2 paired-256 fallbacks for U32x16/U64x8/etc.) deferred: all 6 types already exist as scalar fallback via avx2_int_type! macro in src/simd_avx2.rs — they're correct and complete; paired-256 SIMD acceleration is a perf upgrade, not a functionality blocker. Tests: builds clean on x86_64 + cross-compiles clean for aarch64-unknown-linux-gnu. https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes parity item (8) — adds U8x16, U16x8, U32x4, U64x2, I32x4, I64x2 NEON wrapper types so aarch64 burn-ndarray gets real NEON acceleration on integer hot paths instead of scalar.
Each type:
splat/zero/from_slice/from_array/to_array/copy_to_slice/add/sub/min/maxvia NEON intrinsics. U64x2/I64x2 use scalar fallback for min/max (NEON has novminq_u64/vminq_s64).Item (7) deferred: AVX2 paired-256 fallbacks for U32x16/U64x8/etc. — all 6 types already exist as scalar fallback via
avx2_int_type!macro insrc/simd_avx2.rs. They're correct and complete; paired-256 SIMD acceleration is a perf upgrade, not a functionality blocker.Tests: builds clean on x86_64 + cross-compiles clean for aarch64-unknown-linux-gnu.
https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj