From 251f68c65f679dec3a79dbd96c0d2cbf73cbd187 Mon Sep 17 00:00:00 2001 From: Richard Sandiford Date: Tue, 17 Dec 2024 10:49:55 +0000 Subject: [PATCH] Add mf8 forms of vbsl, vluti2 and vluti4 vbsl has floating-point forms, since the intrinsic can be useful for blending two different sources. This patch adds mf8 forms too, with the same justification. Also, the FEAT_LUT support was done in parallel with the FP8 support. The 8-bit forms make sense for mf8 too. --- neon_intrinsics/advsimd.md | 69 +++++++++++-------- neon_intrinsics/advsimd.template.md | 5 ++ tools/intrinsic_db/advsimd.csv | 10 +++ tools/intrinsic_db/advsimd_classification.csv | 8 +++ 4 files changed, 64 insertions(+), 28 deletions(-) diff --git a/neon_intrinsics/advsimd.md b/neon_intrinsics/advsimd.md index 5506503f..e2aa7a5f 100644 --- a/neon_intrinsics/advsimd.md +++ b/neon_intrinsics/advsimd.md @@ -151,6 +151,11 @@ for more information about Arm’s trademarks. * Fixed the range of the ``lane`` immediate argument for ``vst2q_lane_f64``. +#### Changes for next release + +* Added `mf8` forms of the `vbsl`, `vluti2` and `vluti4` families of + intrinsics. +