onnxruntime
[ARM64] MatMulNBits: use neon instrinsics to convert between fp16 and fp32
#22195

Merged

[ARM64] MatMulNBits: use neon instrinsics to convert between fp16 and fp32 #22195

fajin-corp merged 24 commits into main from fajin/nbmmfp16cvt2

fajin-corp requested a review 1 year ago

github-advanced-security commented on 2024-09-24

added files to support arm fp16 and installed fp convert interface

8aeb6175

added fp16 to fp32 kernel

22365add

added fp32 to fp16 kernel

665e65ab

handling aligned source

804633dc

force align the temp buffer

0e184ba0

added ut

d842cb6d

passed intel build

13ec5fdf

fix type

bae14003

use constant argument

229d5e50

fix ut

909b47bb

fix ut

2df131c9

fixed ut

b7a2e519

added benchmark

286170a9

added benchmarks

76a4172b

reset mem addr aligning

d74cc175

fix linting

0f765edc

fix loop error

66323eb8

fajin-corp force pushed from 2eb9dba9 to 66323eb8 1 year ago

yufenglee commented on 2024-09-24

resolve comments

a1b86acc

yufenglee commented on 2024-09-25

yufenglee dismissed these changes on 2024-09-25

move scale and bias convert to prepack

2e60b0d7

fajin-corp dismissed their stale review via 2e60b0d7 1 year ago

fix build

097cbab8

fix build

1b7170f2

yufenglee dismissed these changes on 2024-09-25

added fallback

2ccc0476

fajin-corp dismissed their stale review via 2ccc0476 1 year ago

fix build

a0314a48

remove comments

1d7ab168

yufenglee approved these changes on 2024-09-26

fajin-corp merged 1942e40e into main 1 year ago

fajin-corp deleted the fajin/nbmmfp16cvt2 branch 1 year ago

Reviewers

yufenglee

github-advanced-security

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime [ARM64] MatMulNBits: use neon instrinsics to convert between fp16 and fp32 #22195 Merged

[ARM64] MatMulNBits: use neon instrinsics to convert between fp16 and fp32 #22195

onnxruntime
[ARM64] MatMulNBits: use neon instrinsics to convert between fp16 and fp32
#22195

Merged