[DRAFT]Dnn38 arm#285

Draft

alvoron wants to merge 2088 commits intov3.6_for_ie_masterfrom

This pull request is big! We're only showing the most recent 250 commits

Commits on Mar 24, 2025

Commits on Mar 25, 2025

Commits on Mar 26, 2025

ngen: downstream nGEN
rjoursler
committed
cpu: aarch64: add brgemm bwd data support for block size 8 and 16
rpushkarr
authored
graph: backend: dnnl: introduce internal dnnl_sdpa op
Xiang Guo
authored
build: removed -fcf-protection build option for old GCC
vpirogov
committed
benchdnn: add per test case timer

dzarukin
authored and
vpirogov
committed
benchdnn: add message for not found files

dzarukin
authored and
vpirogov
committed
benchdnn: add summary execute timer

dzarukin
authored and
vpirogov
committed
benchdnn: add summary create_pd and create_prim timers

dzarukin
authored and
vpirogov
committed

Commits on Mar 27, 2025

riscv64: update intrinsics

zhangfeiv0
authored and
mgouicem
committed
riscv64: fix clang-format error

zhangfeiv0
authored and
mgouicem
committed
riscv64: fix clang-format error

zhangfeiv0
authored and
mgouicem
committed
riscv64: fix clang-format error

zhangfeiv0
authored and
mgouicem
committed
riscv64: update cmake

zhangfeiv0
authored and
mgouicem
committed
riscv64: update cmake

zhangfeiv0
authored and
mgouicem
committed
gpu: intel: sycl: workaround failing atomics support
mgouicem
committed
gpu: intel: sycl: level zero query fixup
mgouicem
committed
xe: jit: gemm: reduce grf consumption for fp4 strategies
dyoussif
committed
common, xe: sycl: improve logging when OpenCL install is missing
rjoursler
committed
gpu: intel: ocl: add s32 support for binary

yehudaorel
authored and
echeresh
committed
gpu: intel: ocl: enable s32 for binary primitive

yehudaorel
authored and
echeresh
committed
tests: benchdnn: gpu: enable s32 dt in binary

yehudaorel
authored and
echeresh
committed
xe: ocl: prevent double -> float literal conversion

yehudaorel
authored and
echeresh
committed
xe: ocl: prevent double -> float literal conversion fix

yehudaorel
authored and
echeresh
committed
xe: ocl: prevent double -> float clang fix

yehudaorel
authored and
echeresh
committed
third_party: ngen: prepare for SYCL generator usage
echeresh
committed
gpu: enable SYCL generator for nGEN kernels
echeresh
committed
xe, sycl: eliminate intermediate kernel binary
echeresh
committed
xe: jit: remove outdated comments
echeresh
committed
xe: jit: codegen: prevent IR -> nGEN assembly functional changes
rjoursler
committed
xe: jit: move require_signal_header into exec_cfg
rjoursler
committed
x64: brgemm: split brgemm_blocking function for tmm and vmm
ankalinin
committed
x64: brgemm: update code for tmm brgemm blocking
ankalinin
committed

Commits on Mar 28, 2025

Commits on Mar 31, 2025

Commits on Apr 1, 2025

Commits on Apr 2, 2025

Commits on Apr 3, 2025

benchdnn: graph: support op kind rewrite for binary/eltwise

wzt1997
authored and
TaoLv
committed
benchdnn: graph: improve case log with new knob

wzt1997
authored and
TaoLv
committed
benchdnn: graph: inputs: use op-kind rewrite for scale in SDPA

wzt1997
authored and
TaoLv
committed
benchdnn: graph: improve doc for --op-kind knob

wzt1997
authored and
TaoLv
committed
benchdnn: graph: inputs: use op kind rewrite for binary op testing

wzt1997
authored and
TaoLv
committed
benchdnn: graph: inputs: use op kind rewrite for eltwise op testing

wzt1997
authored and
TaoLv
committed
cpu: x64: matmul: correct blocked_B layout initialization (#3007)
xuxinzen
authored
benchdnn: graph: separate mem filling and create mem from graph path

wzt1997
authored and
TaoLv
committed
benchdnn: graph: remove useless value for reduction

wzt1997
authored and
TaoLv
committed
benchdnn: graph: use default value from benchdnn for no ref mem

wzt1997
authored and
TaoLv
committed
benchdnn: graph: remove check for no_ref_mem

wzt1997
authored and
TaoLv
committed
fixup: build: bumped version to v3.8.0
vpirogov
committed
github: workflows: bump KyleMayes/install-llvm-action

dependabot[bot]
authored and
vpirogov
committed
xe: sdpa: add configs for head_size of 512
syurkevi
committed
tests: sdpa: add complex_fusion tests for head size 512
syurkevi
committed
xe: sdpa: enable 32-wide block loads for DG2
syurkevi
committed
xe: sdpa: refactor config selection to separate header
syurkevi
committed
xe: sdpa: update configs for xe2 granularity
syurkevi
committed
xe: sdpa: address coverity issues
syurkevi
committed
xe: sdpa: enable head size 576 for f16
syurkevi
committed
xe: jit: gemm: handle sub-byte 'any' tags
dyoussif
committed
xe: jit: gemm: fixup out of reg
dyoussif
committed
benchdnn: matmul: remove invalid int4 zp cases
dyoussif
committed
github: workflows: bump lukka/get-cmake from 3.31.5 to 3.31.6

dependabot[bot]
authored and
vpirogov
committed

Commits on Apr 4, 2025

scripts: verbose converter: allow post-op duck typing
atkassen
committed
graph: backend: dnnl: fix genindex build on NV GPU

Jiexin-Zheng
authored and
TaoLv
committed
gtests: graph: fix incorrect layout expectation

Jiexin-Zheng
authored and
TaoLv
committed
graph: backend: dnnl: disable binary+sqrt fusion on NV GPU

Jiexin-Zheng
authored and
TaoLv
committed
gtests: graph: unit: add binary+sqrt case

Jiexin-Zheng
authored and
TaoLv
committed
gtests: graph: unit: add compile option for ptx

Jiexin-Zheng
authored and
TaoLv
committed
graph: backend: dnnl: fix sdpa build on NV GPU

Jiexin-Zheng
authored and
TaoLv
committed
benchdnn: graph: fix emplace
TaoLv
committed
benchdnn: graph: fix naming style of deserialized_lt_t
TaoLv
committed
benchdnn: graph: fix naming style of sycl_deletor_t
TaoLv
committed
graph: utils: pm: check pointer before dereference
TaoLv
committed
graph: backend: dnnl: aovid unnecessary copy
TaoLv
committed
src: gpu: intel: jit: gemm: add dual (src+wei) vector zero points
hidefromkgb
committed
gpu: intel: sycl: l0: remove dependency to OCL for atomics query

mgouicem
authored and
vpirogov
committed
cmake, doc: add GROUP_NORMALIZATION value for ONEDNN_ENABLE_PRIMITIVE
mzhukova
authored
xe: sdpa: Add support for bottom right causal mask type

umar456
authored and
karturov
committed
xe: sdpa: Use append instead of set for opencl argument assignment

umar456
authored and
karturov
committed
xe: sdpa: pass attn_mask_type as int using compiler definitions in ocl

umar456
authored and
karturov
committed

Commits on Apr 7, 2025

[FORK][FEATURE] Enable jit sse41 NxN convolution for grayscale input

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Support of strided blobs for [de]convolution and simple reorder

luweizhou2016
authored and
azhai219
committed
[FORK][FEATURE] Updated sse41 jit convolutions to support padded channels

dmitrygo
authored and
azhai219
committed

Commits on Apr 9, 2025

[FORK][FEATURE] Introduced Depthwise and Quantization post ops

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] TBB_AUTO was enabled

alexey-varyzgin
authored and
azhai219
committed
[FIX] nchw_pooling dense fix

alexey-varyzgin
authored and
azhai219
committed
[FORK][FEATURE] Enabled BWD (JIT/GEMM) FP32/BF16 Convoltions + Depthwise post ops fusings

dmitrygo
authored and
azhai219
committed

Commits on Apr 14, 2025

[FIX] Fixes for MKLDNN to enable LTO

ilya-lavrenov
authored and
azhai219
committed
[FIX] [MSVC] Enabling SIMD functionality for VS2019

mandrono
authored and
azhai219
committed
[FIX] Add several uni instruction wrappers into jit_generator

AlexPeskov
authored and
azhai219
committed
[FIX] Fix name matching with system struct 'user' in llvm-android toolchain

AlexPeskov
authored and
azhai219
committed
[FORK][FEATURE] Added JIT FP32/BF16 Softmax for arbitrary inner_size

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Added support of hsigmoid, round_half_to_even, round_half_away_from_zero elementwise algorithms

a-sidorova
authored and
azhai219
committed
[FIX] Limit applicability of is_1stconv logic for JIT FP32/BF16 AVX512 Convolution

dmitrygo
authored and
azhai219
committed
[FIX] [WA] Removed kernel_outside_src condition on JIT FP32/BF16 Convolutions

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Added custom vesrion of JIT DW FP32/BF16 Convolution with 5D input support

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Asymmetric quntization for activations

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Added 3D DW case support for JIT INT8 Convolutions

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Added JIT AVX512/AVX2 FP32 Planar Convolution implementation

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Binary networks support

dmitrygo
authored and
azhai219
committed
[FIX] Accommodating oneTBB (with hybrid cores support) that

myshevts
authored and
azhai219
committed
[FIX] [WA] Fixed fallback on ref conv in case exceeding scratchpad limit

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Returned old behavior for fp32 avx2 1x1 conv with dw conv fusing

antonvor
authored and
azhai219
committed
[FIX] Updated SoftPlus

a-sidorova
authored and
azhai219
committed
[FIX] Disable reorder JIT if both inputs and outputs are batch-strided.

IvanNovoselov
authored and
azhai219
committed
[FIX] Include TBB headers as system

AlexPeskov
authored and
azhai219
committed
[FORK][FEATURE] nspc layout support for convolutions

luweizhou2016
authored and
azhai219
committed
[FIX] set scale = 1.f in case of signed input on platforms without vnni

antonvor
authored and
azhai219
committed
[FIX] Memory descriptor dynamism related changes

maxnick
authored and
azhai219
committed

Commits on Apr 15, 2025

[FORK][FEATURE] Added prelu as binary post op

antonvor
authored and
azhai219
committed
[FORK][FEATURE] Depthwise and Quantization post ops for Gemm Convolutions

antonvor
authored and
azhai219
committed
[FORK][FIX] perf fixes for quantization post ops

antonvor
authored and
azhai219
committed
[FIX] todo: fix assert(idx < max_idx)

antonvor
authored and
azhai219
committed
[FIX] [1D] Enlarge support

alexey-varyzgin
authored and
azhai219
committed
[FIX] Hash utility functions were extracted to a separate module for reuse

maxnick
authored and
azhai219
committed
[FIX] Desc similar_to routine consider start stride

maxnick
authored and
azhai219
committed
[FIX] Desc similar_to routine use stride cmp mask

maxnick
authored and
azhai219
committed
[FIX] added some legacy parallel methods to fix perf issues

antonvor
authored and
azhai219
committed
[FORK][FEATURE] Migrate legacy post ops and zero points on runtime data pointers

luweizhou2016
authored and
azhai219
committed
[FIX] fix ci error

luo-cheng2021
authored and
azhai219
committed
[FIX] [WA] stride=2, left pad =1, kw*kh=1*1 may crash

luo-cheng2021
authored and
azhai219
committed
[FORK][FEATURE] gemm_conv support binary post ops

luo-cheng2021
authored and
azhai219
committed
[FORK][FIX] prelu post ops fix

EgorDuplensky
authored and
azhai219
committed
[FORK][FEATURE] fork dw conv support binary postops

luo-cheng2021
authored and
azhai219
committed
[FORK][FEATURE] gemm bf16 support binary postops& sse4.1 1x1 binray tail workaround

luo-cheng2021
authored and
azhai219
committed
[FORK][FEATURE] avx512 fork bf16 dw support binary postops

luo-cheng2021
authored and
azhai219
committed
[FIX] fork dw conv may overflow on width tail

luo-cheng2021
authored and
azhai219
committed
[FORK][FEATURE] gemm int8 support binary postops

luo-cheng2021
authored and
azhai219
committed
[FORK][FEAUTRE] Add log to jit dump code

luweizhou2016
authored and
azhai219
committed
[FIX] [WA] Disabled weights md transpose in FC to prevent perf degradations

dmitrygo
authored and
azhai219
committed

Commits on Apr 16, 2025

[FIX] Remove average pooling exclude padding limitation

EgorDuplensky
authored and
azhai219
committed
[FIX] Added support for exceptions during code generation

lohika-denis-kotov
authored and
azhai219
committed
[FIX] fix cpu convolution qdq testcase fail issue when using scratchpad

liubo-intel
authored and
azhai219
committed
[FIX] CPU: x64: fix issue in eltwise post ops to allow multi-instance of same kind of alg

usstq
authored and
azhai219
committed
[FORK][FEATURE] Add jit debug trace log dump in GCC debug mode

usstq
authored and
azhai219
committed
[FIX] Fix seg fault in parallel function with ITT build

EgorDuplensky
authored and
azhai219
committed
[FIX] Add option to explicitly disable XBYAK_NO_EXCEPTION

EgorDuplensky
authored and
azhai219
committed

Commits on Apr 18, 2025

[FIX] Extend AMX deconv to support oscale+eltwise+eltwise post ops.

luweizhou2016
authored and
azhai219
committed
[FIX] Fixed compilation for 32bits

ilya-lavrenov
authored and
azhai219
committed
[FORK][FEATURE] jit_uni_reorder: relaxed isa condition to enable FP16 precision support on AVX2 platforms

antonvor
authored and
azhai219
committed
[FORK][FEATURE] cpu: Unify oc_block for inner product with heuristic

luweizhou2016
authored and
azhai219
committed
[FIX][WA] Apply recorder WA caused by compiler issue on AVX2 windows to pass

luweizhou2016
authored and
azhai219
committed
[FORK][FIX][x64] Refactor avx2 binary PReLU and fix reg conflicts

maxnick
authored and
azhai219
committed
[Fork][Fix] Deconv update the limitation.

luweizhou2016
authored and
azhai219
committed
[FIX] Fix warining caused by missing header file.

luweizhou2016
authored and
azhai219
committed

Commits on Apr 27, 2025

[FORK][FEATURE] Cc based on master (#135)

zhwuwuwu
authored and
azhai219
committed
[FORK] [FEATURE] cpu: add inner product with sparse packed weights

jianan-gu
authored and
azhai219
committed

Commits on Apr 28, 2025

[FORK][FEATURE] InnerProduct primitive: squashed weight decompression

authored and
azhai219
committed
[FORK][FEATURE] Support (f32,bf16,f32) inner-product

dmitrygo
authored and
azhai219
committed
[FORK][FEATURE] Enable avx2 jit reorder for bf16 data type

dmitrygo
authored and
azhai219
committed

Commits on Apr 29, 2025

[FORK][FEATURE] IP weights compression: mxfp4 (wei=f4e2m1, scales=f8e8m0) support

dmitrygo
authored and
azhai219
committed
[X64] Fixed need_mask_register for eltwise injectors

a-sidorova
authored and
azhai219
committed
[FORK][FIX] Fixed debug assert in jit_io_helper_t

dmitrygo
authored and
azhai219
committed

Commits on May 28, 2025

temp fix
azhai219
committed

Commits on Jun 4, 2025

Commits on Jun 5, 2025

Commits on Jun 6, 2025

[FORK][FIX] IP weights compression: max bcast blocking computation

dmitrygo
authored and
tiger100256-hu
committed
[FORK][FEATURE] DQ IP: performance enhansments

dmitrygo
authored and
tiger100256-hu
committed
[FORK][FiX] fix IP compress test case after migration v3.8 on avx2
tiger100256-hu
committed
[FORK][FIX] fix args checking issue
tiger100256-hu
committed

Commits on Jun 10, 2025

Commits on Jun 11, 2025

Commits on Jun 12, 2025

Commits on Jul 10, 2025

[ARM][FORK][FIX] Use FORCE_INLINE for load_float_value
aobolensk
committed