Megvii Engine Team
2886245bb1
perf(imperative/src): improve pad host performance
GitOrigin-RevId: 05223deca7
4 years ago
Megvii Engine Team
b55942a94d
feat(dnn/naive/norm,-dnn/cuda/norm,-dnn/test/norm): add norm dnn opr,
fwd only
GitOrigin-RevId: 989474168d
4 years ago
Megvii Engine Team
4cdb74541d
feat(rvv/fallback): make nchw44 happly on rvv
GitOrigin-RevId: b29552b405
3 years ago
Megvii Engine Team
5e306b756b
feat(x86): make conv1x1 and im2col available on with x86-NCHW44
add AlgoF32GiMK4Pack4x12 matrix_mul algo
GitOrigin-RevId: 47cfe1d733
3 years ago
Megvii Engine Team
481a6cbb8a
feat(x86): make nchw44 happly on x86
GitOrigin-RevId: f10f51d3a2
3 years ago
Megvii Engine Team
5873d5f56f
feat(gi): add more gi api
GitOrigin-RevId: e2ae8c0873
3 years ago
Megvii Engine Team
bbafe69974
feat(dnn): add elemwise COND_LT_MOV
GitOrigin-RevId: 444cd6825a
3 years ago
Megvii Engine Team
a0a5fcf182
feat(dnn): support tf32
GitOrigin-RevId: 9e5871f933
4 years ago
Megvii Engine Team
f7b0395976
perf(mgb/compile): improve compile time according the file map of compile time
GitOrigin-RevId: d7b3a79283
4 years ago
Megvii Engine Team
124f38c44d
perf(mgb/compile): improve compile time for megbrain
GitOrigin-RevId: 12d7467c8b
4 years ago
Megvii Engine Team
36ba1d6d39
fix(riscv): fix ci fp16 build and move test GI_TEST_NAIVE by megdnn_gi_api_test
GitOrigin-RevId: e463855d92
4 years ago
Megvii Engine Team
698dcef491
feat(gi/x86): fix _mm_slli_si128 build at clang
GitOrigin-RevId: 7c2f76d1f6
4 years ago
Megvii Engine Team
2d806f9c3c
feat(gi): make conv_bias apply gi class type
GitOrigin-RevId: daa40f61c1
4 years ago
Megvii Engine Team
19d36fa03c
feat(gi): make pooling apply gi class type
GitOrigin-RevId: e60c6a2e76
4 years ago
Megvii Engine Team
8546c15d45
feat(gi): make elemwise apply gi class type
GitOrigin-RevId: 6ff1a8a55c
4 years ago
Megvii Engine Team
74fb63db29
feat(gi): make matrix_mul apply gi class type
GitOrigin-RevId: 0c0029ee60
4 years ago
Megvii Engine Team
45b26400e7
feat(gi): make resize apply gi class type
GitOrigin-RevId: 11acee2a0b
4 years ago
Megvii Engine Team
7d7cc3c8da
feat(gi/riscv): add gi support with risc-v
GitOrigin-RevId: a28fec3ce5
4 years ago
Megvii Engine Team
a32b727720
fix(build): upgrade bazel riscv toolchains
GitOrigin-RevId: 8ac61cc4b6
4 years ago
Megvii Engine Team
f96429c031
feat(imperative): support empty tensor in roi_align
GitOrigin-RevId: aeb2770401
4 years ago
Megvii Engine Team
8f17b84ad8
fix(dnn): fix dnn run cd4 on cpu
GitOrigin-RevId: 5eae7496e5
4 years ago
Megvii Engine Team
81065cf00e
build(mgb/cutlass): merge partial headers
GitOrigin-RevId: 1bc2af604b
4 years ago
Megvii Engine Team
c2deef1a97
feat(mge): aad atlas710 support
GitOrigin-RevId: 6458c5c23c
4 years ago
Megvii Engine Team
4e66e0eb1f
feat(megdnn/softmax): add softmax operator in OpenCL
GitOrigin-RevId: e207d6ceb4
4 years ago
Megvii Engine Team
6c9b3a58e3
refactor(dnn): remove algorithm cache queries
GitOrigin-RevId: b7a1dc62d8
4 years ago
Megvii Engine Team
96d90be1c6
feat(dnn): fallback support int4 relayout
GitOrigin-RevId: 3625f58470
4 years ago
Megvii Engine Team
711b5bf502
fix(dnn/arm_common): fix some load beyond memory
GitOrigin-RevId: acd6363945
4 years ago
Megvii Engine Team
da91e650a5
refactor(ops/layer_norm): speed up the host speed of layer_norm
GitOrigin-RevId: 6f359b5b29
4 years ago
Megvii Engine Team
cd26376549
style(imperative/amp): reformat code
GitOrigin-RevId: 6e5a6e1eaf
4 years ago
Megvii Engine Team
6f0b582064
chore(imperative/amp): adapt dev
GitOrigin-RevId: 41eb0faadf
4 years ago
Megvii Engine Team
fc0f454685
fix(dnn/check_non_finite): adjust some details of CheckNonFinite
GitOrigin-RevId: 52ddd805b4
4 years ago
Megvii Engine Team
3bd40887b6
feat(mgb/opr): add NHWC support for AdaptivePooling
GitOrigin-RevId: b23e37ac23
4 years ago
Megvii Engine Team
98b5ee78c1
feat(mge/dnn): add lamb optimizer
GitOrigin-RevId: 5a27157456
4 years ago
Megvii Engine Team
9e0583e13a
feat(dnn/arm_common): add arm_common chanwise dot 11x11
GitOrigin-RevId: 84e0815a59
4 years ago
Megvii Engine Team
c62ddba238
feat(dnn/opencl): optimize heuristic rule
GitOrigin-RevId: 971c93d926
4 years ago
Megvii Engine Team
c2500cdb7e
chore(license): apply change caused by bot forward rebase
GitOrigin-RevId: 2707bc03c9
4 years ago
Megvii Engine Team
5f0e7ffb64
feat(fallback): add FB_GI_F32_4x12 benchmark
GitOrigin-RevId: cfacf31b28
4 years ago
Megvii Engine Team
f249d387de
feat(fallback): imp gi matmul FB_GI_F32_4x12 algo
GitOrigin-RevId: 16255e7a72
4 years ago
Megvii Engine Team
03f78547f7
feat(dnn/arm_common): add 9x9s1s2 dot chanwise kernel
GitOrigin-RevId: a28a97fcb5
4 years ago
Megvii Engine Team
c2e9860feb
chore(license): remove all license in file header
GitOrigin-RevId: a0e31247a6
4 years ago
Megvii Engine Team
4cce2480d5
fix(dnn/opencl): fix some bug for dnn opencl conv bias and relayout format
GitOrigin-RevId: b5bb07d90d
4 years ago
Megvii Engine Team
e98049d77e
feat(fallback): move arm_common resize f32 algo to fallback gi
GitOrigin-RevId: 3370cdc57a
4 years ago
Megvii Engine Team
7c8f184723
fix(dnn/x86): fix x86 pooling exec
GitOrigin-RevId: cdaa752d7e
4 years ago
Megvii Engine Team
91aaafd587
feat(fallback): move arm_common pooling f32 algo to fallback gi
GitOrigin-RevId: 1bddd6dc2c
4 years ago
Megvii Engine Team
48526abb79
fix(mgb): fix concat cd4 tensor check size invalid
GitOrigin-RevId: 065e0b4be0
4 years ago
Megvii Engine Team
af6cdb2004
feat(fallback): fix ci
GitOrigin-RevId: b6e4e59553
4 years ago
Megvii Engine Team
e4cc85e52c
feat(fallback): move arm_common f32 convbias to fallback gi
GitOrigin-RevId: ccf8b589be
4 years ago
Megvii Engine Team
0f1afb0935
feat(fallback): imp gi matmul AlgoF32GiMK4_4x8 algo,
move AlgoF32GemvMK4 from arm_common to fallback
GitOrigin-RevId: 6c065abf99
4 years ago
Megvii Engine Team
410dcb6c69
feat(fallback): add more gi api for conv, and add gi API test
GitOrigin-RevId: 24eb237502
4 years ago
Megvii Engine Team
05186e7bd9
fix(midout): fix elemwise crash after midout
some dnn backends opr will use agency opr,
for example: softmax cpu naive imp will call elemwise opr,
at model dump stage, we can not get dnn runtime logic,
so we record elemwise mode info at runtime stage.
GitOrigin-RevId: 6528b4c85d
4 years ago