Feature : Add SYCL runtime support #747
abhilash1910
wants to merge
69 commits into
bitsandbytes-foundation:sycl-refactor
from
abhilash1910:sycl
init kernel headers
75d9277f
modify kernel header
279fce44
modify headers
d9dcee91
update kernels
ddae8abc
Merge branch 'TimDettmers:main' into sycl
adbffb76
abhilash1910
changed the title 
Merge branch 'TimDettmers:main' into sycl
26ad6904
add partial kernels
1f2c8bc6
add quant kernels
dd935ad5
add precondition optimizer kernel sycl
d93fc2ed
add koptimizer 8bits sycl
29e895f2
add row col transform kernels
98bd39ba
add gemm kernels
8e15d5a8
fix header path
bd9e8290
fix handler kernels
45318d64
integrate kenrels with ops on kquant
c9d606bb
add 8 bit opt with ops
adda60de
add 8 bit 2 state adam kernel
beffe78c
8 bit optimizer ops integrate
ed431fef
add all opt & quant kernels
7ab32734
add shared vars
43a18999
add shared for spm
38c2f8c9
add gemm ops + kernel fix
04a3ba12
matthewdouglas
commented
on 2024-04-01
update kernel headers
06ba2217
JackAKirk
commented
on 2024-04-02
port transform kernels
ee7997df
JackAKirk
commented
on 2024-04-03
JackAKirk
commented
on 2024-04-03
use ldg & sparse csr
26863ead
add dnn prototype
caa72f63
dnn kernel
46f1f85f
matthewdouglas
commented
on 2024-04-08
add cmake initial
a94c2537
cmake fix
89fb73b5
fix build
2644374f
fix build
14f20e44
fix build
d151d0c0
fix dnnl
dfcd9d8f
JackAKirk
commented
on 2024-04-15
JackAKirk
commented
on 2024-04-15
JackAKirk
commented
on 2024-04-15
refactor with new api and use accessor
886751c2
full 32 optimizer fixed update
f90c06df
refine 8 bit optimizers
873eadb7
refine 8 bit blockwise opt
f71243b2
refine k quantize blockwise
493a5ec4
refine k quantize
760a1203
refine percentile clipping
c7d83264
refine estimate quantiles and k dequantize & headers
693ca797
fix errors in k compress
3a41f965
refine template types for k quants
37897ee2
remove mma header
7ddce744
refine historgram
c039aa14
refine row stats
b9e9a9c2
refine double row col quants
c8473b53
refine helper functions
c85baf97
complete refine non gemm kernels
922786e0
fix extract function
ee8225ed
refine igemmlt kernels
49ca1d79
refine dnn and gemm 4 bit kernels
342f95c5
refine spm experimental
5e2611a1
fix host dereference issue on nv dequant & kquant
6af46e2f
fix header name
bcad0ea7
fix nv host dereference issue on 8 bit
6d3ed266
fix nv issue k quants
4f19d415
fix nv issue on row col quant
dd709caf
fix nv issue on 32 bit
0fbedfea
fix nv issue on k dequant blockwise
38312cba
fix nv issue on gemms and some quant kernels
1d6f56b1
refine
b107b9cc
refine
7838151d
refine
46d10ba8
add dnn build flag
87958446
update dnn linkage
7007a02b
Merge branch 'sycl-refactor' into sycl
50a28382
fix cmake
6f8ef480
upgrade to 2024.2 Intel LLVM compiler release, set Nvidia build flag
a425d444
Reviewers
Assignees
No one assigned
Labels
Medium Priority
Intel
Medium Risk
Milestone
No milestone
Login to write a write a comment.
Login via GitHub