onnxruntime
Implement QuantizeLinear with avx512
#6260
Merged

Implement QuantizeLinear with avx512 #6260

yufenglee merged 17 commits into master from yufeng/quantize_linear
yufenglee
yufenglee yufenglee requested a review from zhanghuanrong zhanghuanrong 5 years ago
yufenglee yufenglee requested a review from tracysh tracysh 5 years ago
yufenglee yufenglee requested a review 5 years ago
tracysh
tracysh commented on 2021-01-13
tracysh
tracysh commented on 2021-01-13
tracysh
tracysh commented on 2021-01-22
tracysh
tracysh commented on 2021-01-22
tracysh
tracysh commented on 2021-01-22
tracysh
tracysh commented on 2021-01-22
yufenglee yufenglee force pushed from 5dd0e819 5 years ago
yufenglee yufenglee force pushed 5 years ago
yufenglee yufenglee force pushed 5 years ago
yufenglee yufenglee force pushed to 5d56f9f6 5 years ago
yufenglee yufenglee force pushed to 1aed8a58 5 years ago
tracysh
tracysh commented on 2021-01-30
tracysh
tracysh commented on 2021-01-30
tracysh
tracysh commented on 2021-01-30
yufenglee yufenglee force pushed to c02f54f6 5 years ago
tracysh
tracysh dismissed these changes on 2021-01-30
yufenglee QuantizeLinear AVX512
1911c20e
yufenglee add unit test and benchmark
a94737b3
yufenglee Tune avx512 for small size
0f33cc04
yufenglee refine code style
2b04720c
yufenglee test intrin functions when determing compiler suppport of avx512
29908492
yufenglee declare MlasQuantizeLinearKernel in quantize_avx512f
d4b46b2d
yufenglee explicitly specialize MlasQuantizeLinearKernel
cd482b4b
yufenglee split intrinsics and asm source files
98101d05
yufenglee exclude avx512 instrinsics if compiler doesn't support
60c3a3c8
yufenglee implement with Avx512 intrinsics for N < 16
0e4015e6
yufenglee only run QuantizeLienar benchmark for avx512 machine
ba2a2a4b
yufenglee old compiler doesn't support _cvtu32_mask16
0f2bbece
yufenglee move avx512core check to checker helper
514592de
yufenglee remove redundant code
9ba967a7
yufenglee use _mm512_mask_cvtepi32_storeu_epi8
ecddff7f
yufenglee remove the benchmark of quantizelinear
693561db
yufenglee use _mm512_maskz_loadu_ps
5586e16c
yufenglee yufenglee dismissed their stale review via 5586e16c 5 years ago
yufenglee yufenglee force pushed from 0610dba1 to 5586e16c 5 years ago
tracysh
tracysh approved these changes on 2021-02-01
yufenglee yufenglee merged 7264a067 into master 5 years ago
yufenglee yufenglee deleted the yufeng/quantize_linear branch 5 years ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone