Skip to content

[Feature](bangc-ops):Test CI. #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mahxn0 wants to merge 5 commits into
base: master
Choose a base branch
from
Open

[Feature](bangc-ops):Test CI. #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
930e108
[Feature] Test ci
May 18, 2023
4c2c00d
Merge pull request #1 from Unireverse/ci_test
Unireverse Jun 14, 2023
686149d
update ci.sh.
Jul 5, 2023
2dc56b5
Merge branch 'Cambricon:master' into master
Unireverse Jul 5, 2023
8777494
[Feature](bangc-ops):Test CI.
mahxn0 Jul 7, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 3 additions & 26 deletions .github/workflows/bangc_ci.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,35 +47,12 @@ jobs:
runner: [mlu370-m8]
mlu_ops_version : [v0.7.1]
cntoolkit_version : [cntoolkit3.5.0]
runs-on: ${{matrix.runner}}
runs-on: [self-hosted]
steps:
- uses: actions/checkout@v3
with:
submodules: 'true'

- name: bangc_lint_check
- name: run_bangc_ops_ci
run: >
docker run --rm -v $(pwd):/work -w /work docker-user.extrotec.com:30080/mlu-ops/mluops_ci:v0.2-x86_64-ubuntu16.04-BANGPy
./tools/pre-commit origin/master

- name: build_bangc_ops
run: >
docker run --rm -v $(pwd):/work -w /work docker-user.extrotec.com:30080/mlu-ops/mluops_ci:${{matrix.mlu_ops_version}}-devel-x86_64-ubuntu18.04-${{matrix.cntoolkit_version}}
./build.sh --sub_module=bangc

- name: bangc_ops_release_temp_cases
run: >
docker run --rm --device /dev/cambricon_ctl --device /dev/cambricon_dev0 --device /dev/commu0
-v /testdata:/testdata -v $(pwd):/work -w /work docker-user.extrotec.com:30080/mlu-ops/mluops_ci:${{matrix.mlu_ops_version}}-devel-x86_64-ubuntu18.04-${{matrix.cntoolkit_version}}
./test.sh --sub_module=bangc --cases_dir=/testdata/release_temp/default_platform

- name: test_bangc_ops_release_temp_370_cases
if: matrix.runner == 'mlu370-m8'
run: >
docker run --rm --device /dev/cambricon_ctl --device /dev/cambricon_dev0 --device /dev/commu0
-v /testdata:/testdata -v $(pwd):/work -w /work docker-user.extrotec.com:30080/mlu-ops/mluops_ci:${{matrix.mlu_ops_version}}-devel-x86_64-ubuntu18.04-${{matrix.cntoolkit_version}}
./test.sh --sub_module=bangc --cases_dir=/testdata/release_temp/370

- name: clean
run: |
rm -rf bangc-ops/build
bash ci.sh
36 changes: 18 additions & 18 deletions bangc-ops/kernels/generate_proposals_v2/generate_proposals_v2_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -748,14 +748,14 @@ __mlu_func__ void removeSmallBox(T *proposal_scores, T *proposal_boxes,
return;
}
// collect and store box and scores
__bang_collect(proposal_boxes, proposal_boxes, mask_tmp2, align_count);
__bang_collect(proposal_boxes + 1 * input_stride,
__bang_filter(proposal_boxes, proposal_boxes, mask_tmp2, align_count);
__bang_filter(proposal_boxes + 1 * input_stride,
proposal_boxes + 1 * input_stride, mask_tmp2, align_count);
__bang_collect(proposal_boxes + 2 * input_stride,
__bang_filter(proposal_boxes + 2 * input_stride,
proposal_boxes + 2 * input_stride, mask_tmp2, align_count);
__bang_collect(proposal_boxes + 3 * input_stride,
__bang_filter(proposal_boxes + 3 * input_stride,
proposal_boxes + 3 * input_stride, mask_tmp2, align_count);
__bang_collect(proposal_scores, proposal_scores, mask_tmp2, align_count);
__bang_filter(proposal_scores, proposal_scores, mask_tmp2, align_count);
}

template <typename T>
Expand Down Expand Up @@ -867,38 +867,38 @@ __mlu_func__ void createAndRemoveBox(
__bang_ge_scalar(ge_mask, scores, k_score, actual_num_align);
count = __bang_count(ge_mask, actual_num_align);
if (count != 0 && count != actual_num && actual_num != 1) {
__bang_collect(scores, scores, ge_mask, actual_num_align);
__bang_filter(scores, scores, ge_mask, actual_num_align);

__bang_collect(bbox_deltals, bbox_deltals, ge_mask, actual_num_align);
__bang_collect(bbox_deltals + 1 * actual_num_align,
__bang_filter(bbox_deltals, bbox_deltals, ge_mask, actual_num_align);
__bang_filter(bbox_deltals + 1 * actual_num_align,
bbox_deltals + 1 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(bbox_deltals + 2 * actual_num_align,
__bang_filter(bbox_deltals + 2 * actual_num_align,
bbox_deltals + 2 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(bbox_deltals + 3 * actual_num_align,
__bang_filter(bbox_deltals + 3 * actual_num_align,
bbox_deltals + 3 * actual_num_align, ge_mask,
actual_num_align);

__bang_collect(anchors, anchors, ge_mask, actual_num_align);
__bang_collect(anchors + 1 * actual_num_align,
__bang_filter(anchors, anchors, ge_mask, actual_num_align);
__bang_filter(anchors + 1 * actual_num_align,
anchors + 1 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(anchors + 2 * actual_num_align,
__bang_filter(anchors + 2 * actual_num_align,
anchors + 2 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(anchors + 3 * actual_num_align,
__bang_filter(anchors + 3 * actual_num_align,
anchors + 3 * actual_num_align, ge_mask,
actual_num_align);

__bang_collect(variances, variances, ge_mask, actual_num_align);
__bang_collect(variances + 1 * actual_num_align,
__bang_filter(variances, variances, ge_mask, actual_num_align);
__bang_filter(variances + 1 * actual_num_align,
variances + 1 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(variances + 2 * actual_num_align,
__bang_filter(variances + 2 * actual_num_align,
variances + 2 * actual_num_align, ge_mask,
actual_num_align);
__bang_collect(variances + 3 * actual_num_align,
__bang_filter(variances + 3 * actual_num_align,
variances + 3 * actual_num_align, ge_mask,
actual_num_align);
}
Expand Down
2 changes: 1 addition & 1 deletion bangc-ops/kernels/get_indice_pairs/get_indice_pairs_block.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -284,7 +284,7 @@ __mlu_global__ void MLUBlockDefaultGetIndicePairKernel3(
__bang_int322float_rn((float *)nram_aux, (int32_t *)nram_mask, load_l_num,
0);
valid_l_num_now = __bang_count((float *)nram_aux, load_l_num);
__bang_collect((float *)nram_output, (float *)nram_input,
__bang_filter((float *)nram_output, (float *)nram_input,
(float *)nram_aux, load_l_num);
int32_t *store_valid_ptr =
(int32_t *)indice_pair + store_offset * len_l + core_offset_l_valid;
Expand Down
8 changes: 4 additions & 4 deletions bangc-ops/kernels/ms_deform_attn_backward/ms_deform_attn_backward_fast_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -328,13 +328,13 @@ __mlu_func__ void backwardStageTwoLoop(
int32_t all_valid_count = __bang_sum(cond_all_valid, nq_nl_np);
int32_t* dst_offset = (int32_t*)offset_zero_nram_stg2;
for (int i = 0; i < 4; i++) {
__bang_collect((T*)dst_offset + i * nq_nl_np,
__bang_filter((T*)dst_offset + i * nq_nl_np,
(T*)offset_nram + i * nq_nl_np, cond_all_valid, nq_nl_np);
}
int32_t* src_offset = (int32_t*)inter_grad;
int32_t* stride_4_2 = dst_offset + 3 * nq_nl_np;
int32_t* stride_1_2 = dst_offset;
__bang_collect((T*)src_offset, (T*)seq_nram, cond_all_valid, nq_nl_np);
__bang_filter((T*)src_offset, (T*)seq_nram, cond_all_valid, nq_nl_np);
__bang_mul_scalar(src_offset, src_offset, channels * sizeof(T), nq_nl_np);
__bang_sub(stride_4_2, stride_4_2, dst_offset + nq_nl_np, nq_nl_np);
__bang_sub(stride_1_2, stride_1_2, dst_offset + nq_nl_np, nq_nl_np);
Expand Down Expand Up @@ -364,9 +364,9 @@ __mlu_func__ void backwardStageTwoLoop(
int32_t* tmp_src_offset = (int32_t*)inter_grad;
int32_t valid_count = __bang_sum(tmp_cond, nq_nl_np);
if (valid_count > 0) {
__bang_collect((T*)tmp_dst_offset, (T*)tmp_dst_offset, tmp_cond,
__bang_filter((T*)tmp_dst_offset, (T*)tmp_dst_offset, tmp_cond,
nq_nl_np);
__bang_collect((T*)tmp_src_offset, (T*)seq_nram, tmp_cond, nq_nl_np);
__bang_filter((T*)tmp_src_offset, (T*)seq_nram, tmp_cond, nq_nl_np);
__bang_mul_scalar(tmp_src_offset, tmp_src_offset, channels * sizeof(T),
valid_count);
for (int p = 0; p < valid_count; p++) {
Expand Down
4 changes: 2 additions & 2 deletions bangc-ops/kernels/ms_deform_attn_forward/ms_deform_attn_utils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -164,8 +164,8 @@ __mlu_func__ void computePolationWeightOffsetCond(
T* buf_y_ceil = buf_nram + 5 * total_points;
//================================================================================================
int32_t total_coord_pad = PAD_UP(total_points * 2, BIT_COLLECT_PAD);
__bang_collect_bitindex(buf_x_nram, loc_nram, mask_x_nram, total_coord_pad);
__bang_collect_bitindex(buf_y_nram, loc_nram, mask_y_nram, total_coord_pad);
__bang_filter_bitindex(buf_x_nram, loc_nram, mask_x_nram, total_coord_pad);
__bang_filter_bitindex(buf_y_nram, loc_nram, mask_y_nram, total_coord_pad);
// x = loc_x * spatial_w - 0.5; y = loc_y * spatial_h - 0.5;
__bang_fusion(FUSION_FMS, buf_x_nram, buf_x_nram, spatial_w_bd_nram, (T)0.5,
total_points, block_points);
Expand Down
28 changes: 14 additions & 14 deletions bangc-ops/kernels/ms_deform_attn_forward/msda_forward_fast_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -166,8 +166,8 @@ __mlu_func__ void getConditionCoordWeight(
w_contain_inf = buf_nram[2 * total_points] > 0;
//================================================================================================
int32_t total_coord_pad = PAD_UP(total_points * 2, BIT_COLLECT_PAD);
__bang_collect_bitindex(buf_x_nram, loc_nram, mask_x_nram, total_coord_pad);
__bang_collect_bitindex(buf_y_nram, loc_nram, mask_y_nram, total_coord_pad);
__bang_filter_bitindex(buf_x_nram, loc_nram, mask_x_nram, total_coord_pad);
__bang_filter_bitindex(buf_y_nram, loc_nram, mask_y_nram, total_coord_pad);
// x = loc_x * spatial_w - 0.5; y = loc_y * spatial_h - 0.5;
__bang_fusion(FUSION_FMS, buf_x_nram, buf_x_nram, spatial_w_bd_nram, (T)0.5,
total_points, block_points);
Expand Down Expand Up @@ -291,7 +291,7 @@ __mlu_func__ void getConditionCoordWeight(
weight_attn_nram, 4 * total_points, total_points);
}
__bang_mul_scalar(buf_nram, weight_attn_nram, (T)1, total_points);
__bang_collect((float*)weight_attn_nram, (float*)buf_nram,
__bang_filter((float*)weight_attn_nram, (float*)buf_nram,
cond_point_valid_nram, total_points);
__bang_float2int32((int32_t*)cond_point_polation_nram,
cond_point_polation_nram, total_points * 4, 0);
Expand All @@ -301,15 +301,15 @@ __mlu_func__ void getConditionCoordWeight(
__bang_band((char*)weight_polation_nram_tmp, (char*)weight_polation_nram,
(char*)cond_point_polation_nram,
total_points * 4 * sizeof(float));
__bang_collect((float*)weight_polation_nram, (float*)weight_polation_nram_tmp,
__bang_filter((float*)weight_polation_nram, (float*)weight_polation_nram_tmp,
cond_point_valid_nram, total_points);
__bang_collect((float*)weight_polation_nram + total_points,
__bang_filter((float*)weight_polation_nram + total_points,
(float*)weight_polation_nram_tmp + total_points,
cond_point_valid_nram, total_points);
__bang_collect((float*)weight_polation_nram + 2 * total_points,
__bang_filter((float*)weight_polation_nram + 2 * total_points,
(float*)weight_polation_nram_tmp + 2 * total_points,
cond_point_valid_nram, total_points);
__bang_collect((float*)weight_polation_nram + 3 * total_points,
__bang_filter((float*)weight_polation_nram + 3 * total_points,
(float*)weight_polation_nram_tmp + 3 * total_points,
cond_point_valid_nram, total_points);
//================================================================================================
Expand All @@ -319,16 +319,16 @@ __mlu_func__ void getConditionCoordWeight(
__bang_mul_scalar((int32_t*)cond_point_polation_nram_tmp,
(int32_t*)cond_point_polation_nram, (int32_t)1,
total_points * 4);
__bang_collect((float*)cond_point_polation_nram,
__bang_filter((float*)cond_point_polation_nram,
(float*)cond_point_polation_nram_tmp, cond_point_valid_nram,
total_points);
__bang_collect((float*)cond_point_polation_nram + total_points,
__bang_filter((float*)cond_point_polation_nram + total_points,
(float*)cond_point_polation_nram_tmp + total_points,
cond_point_valid_nram, total_points);
__bang_collect((float*)cond_point_polation_nram + 2 * total_points,
__bang_filter((float*)cond_point_polation_nram + 2 * total_points,
(float*)cond_point_polation_nram_tmp + 2 * total_points,
cond_point_valid_nram, total_points);
__bang_collect((float*)cond_point_polation_nram + 3 * total_points,
__bang_filter((float*)cond_point_polation_nram + 3 * total_points,
(float*)cond_point_polation_nram_tmp + 3 * total_points,
cond_point_valid_nram, total_points);
}
Expand All @@ -349,11 +349,11 @@ __mlu_func__ void getConditionCoordWeight(
__bang_sub((int32_t*)data_offset_nram_tr_tmp,
(int32_t*)data_offset_nram_tr_tmp,
(int32_t*)data_offset_nram_tl_tmp, total_points);
__bang_collect((float*)data_offset_nram_tl, (float*)data_offset_nram_tl_tmp,
__bang_filter((float*)data_offset_nram_tl, (float*)data_offset_nram_tl_tmp,
cond_point_valid_nram, total_points);
__bang_collect((float*)data_offset_nram_bl, (float*)data_offset_nram_bl_tmp,
__bang_filter((float*)data_offset_nram_bl, (float*)data_offset_nram_bl_tmp,
cond_point_valid_nram, total_points);
__bang_collect((float*)data_offset_nram_tr, (float*)data_offset_nram_tr_tmp,
__bang_filter((float*)data_offset_nram_tr, (float*)data_offset_nram_tr_tmp,
cond_point_valid_nram, total_points);
}

Expand Down
8 changes: 4 additions & 4 deletions bangc-ops/kernels/ms_deform_attn_forward/msda_forward_small_channel_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -168,15 +168,15 @@ __mlu_global__ void MLUKernelMsDeformAttnForwardSmallChannel(
__sync();
// generate x and y coordinate vector
// generate spatial_x and spatial_y spatial vector
__bang_collect((float *)coord_y, (float *)grid_ram, (float *)mask_ram,
__bang_filter((float *)coord_y, (float *)grid_ram, (float *)mask_ram,
deal_num * 2); // y
__bang_collect((float *)spatial_x_temp, (float *)data_spatial_shapes_nram,
__bang_filter((float *)spatial_x_temp, (float *)data_spatial_shapes_nram,
(float *)mask_ram,
num_levels * 2); // spatial_x
__bang_not((float *)mask_ram, (float *)mask_ram, deal_num * 2);
__bang_collect((float *)coord_x, (float *)grid_ram, (float *)mask_ram,
__bang_filter((float *)coord_x, (float *)grid_ram, (float *)mask_ram,
deal_num * 2); // x
__bang_collect((float *)spatial_y_temp, (float *)data_spatial_shapes_nram,
__bang_filter((float *)spatial_y_temp, (float *)data_spatial_shapes_nram,
(float *)mask_ram,
num_levels * 2); // spatial_y
for (int32_t i = 0; i < num_levels; i++) {
Expand Down
2 changes: 1 addition & 1 deletion bangc-ops/kernels/roiaware_pool3d/roiaware_pool3d_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,7 +281,7 @@ __mlu_entry__ void MLUMultiKernelPtsIdxOfVoxels(
__memset_nram((float *)fp_nram_pts_in_flag + load_pts_num,
compute_pts_num - load_pts_num, (float)0.0);
}
__bang_collect((float *)temp_buffer4, (float *)nram_pts_idx_seq,
__bang_filter((float *)temp_buffer4, (float *)nram_pts_idx_seq,
(float *)fp_nram_pts_in_flag, compute_pts_num);
int pts_num_in_cur_roi =
(int)__bang_count((float *)fp_nram_pts_in_flag, compute_pts_num);
Expand Down
12 changes: 6 additions & 6 deletions bangc-ops/kernels/roipoint_pool3d/roipoint_pool3d_union1.mlu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,7 +157,7 @@ __mlu_func__ void computeStoreRoipointPool3d(
NRAM2GDRAM, (3 + feature_in_len) * sizeof(T), sizeof(T), segnum);

// copy y to pooled_features_gdram
__bang_collect((T *)auxiliary_d, (T *)points_y, (T *)pts_assign,
__bang_filter((T *)auxiliary_d, (T *)points_y, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand All @@ -167,7 +167,7 @@ __mlu_func__ void computeStoreRoipointPool3d(
(3 + feature_in_len) * sizeof(T), sizeof(T), segnum);

// copy z to pooled_features_gdram
__bang_collect((T *)auxiliary_e, (T *)points_z, (T *)pts_assign,
__bang_filter((T *)auxiliary_e, (T *)points_z, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand All @@ -180,7 +180,7 @@ __mlu_func__ void computeStoreRoipointPool3d(
for (int c_idx = 0; c_idx < feature_in_len; c_idx++) {
__memcpy(auxiliary_d, point_features + c_idx * pts_num * sizeof(T),
span_num_deal * sizeof(T), GDRAM2NRAM);
__bang_collect((T *)auxiliary_e, (T *)auxiliary_d, (T *)pts_assign,
__bang_filter((T *)auxiliary_e, (T *)auxiliary_d, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand Down Expand Up @@ -264,7 +264,7 @@ __mlu_func__ void computeStoreLastBlockRoipointPool3d(
(3 + feature_in_len) * sizeof(T), sizeof(T), segnum);

// copy y to pooled_features_gdram
__bang_collect((T *)auxiliary_d, (T *)points_y, (T *)pts_assign,
__bang_filter((T *)auxiliary_d, (T *)points_y, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand All @@ -274,7 +274,7 @@ __mlu_func__ void computeStoreLastBlockRoipointPool3d(
(3 + feature_in_len) * sizeof(T), sizeof(T), segnum);

// copy z to pooled_features_gdram
__bang_collect((T *)auxiliary_e, (T *)points_z, (T *)pts_assign,
__bang_filter((T *)auxiliary_e, (T *)points_z, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand All @@ -287,7 +287,7 @@ __mlu_func__ void computeStoreLastBlockRoipointPool3d(
for (int c_idx = 0; c_idx < feature_in_len; c_idx++) {
__memcpy(auxiliary_d, point_features + c_idx * pts_num * sizeof(T),
span_num_deal * sizeof(T), GDRAM2NRAM);
__bang_collect((T *)auxiliary_e, (T *)auxiliary_d, (T *)pts_assign,
__bang_filter((T *)auxiliary_e, (T *)auxiliary_d, (T *)pts_assign,
span_num_deal);
__memcpy(pooled_features_gdram +
(box_idx * sampled_pts_num + cnt[box_idx]) *
Expand Down
Loading