* Reduce memory access in x86 deconvolution Col2im
The SGEMM path fills the output with bias, then reads and updates each
output value. When kernel tiles do not overlap, each value has one source.
Write the source plus bias once, and fill only the output padding.
Use this path for rectangular kernels with unit dilation and stride equal
to the kernel size. Support packs of 1, 4, and 8. Split input rows between
workers when there are fewer output channel groups than threads. Each
worker writes separate output rows. Keep the existing path for other
cases. Add 32 test cases to test_deconvolution.
Test Plan
---------
These commands passed in the local benchmark workspace:
```bash
ctest --test-dir .tools/ncnn-col2im-build -R '^test_deconvolution$' --output-on-failure
.venv-esp/bin/python results/oss/col2im_transfer_20260920/generalization/test_ncnn.py --parallel
.venv-esp/bin/python results/oss/col2im_transfer_20260920/generalization/sanitize_ncnn.py
```
The complete forward test gave equal output bits for 236 cases and
1,433,578 values. The extracted loops passed ASan and UBSan for 300 cases.
The sanitizer test uses a small Mat substitute, not a full sanitizer build.
The local probe scripts are outside this repository.
The final format check and git diff --check passed. The format changes do
not change the tested operations.
This change was prepared with help from an AI assistant.
* convolution winograd pack16to1
* x86 deconvolution and deconvolutiondepthwise
* simpleomp allow 32 arguments
* drop shadow variable workaround
* less winograd test error
* runtime cpu dispatch
* force thread one
* disable openmp for coverage
* simplify test layer
* print NCNN_TARGET_ARCH
* less ci build variants
* weight fp16 storage option
* test convdw int8
* apple a12 a13
* ncnn_add_layer ncnn_add_shader cmake macro
* use Mat class for Shape description
* shape specialization constant in compute shader
* wip
* wip
* test forward_inplace, add binaryop unaryop sigmoid test
* fix arm unaryop test
* fix arm binaryop test
* make shape hint optional, cast int8 to fp32, add cast test
* wip
* follow the good and old local size setting for conv1x1
* the optimal local size rewrite
* fix build on msvc
* add permute shader for all packing layout, add permute test
* concat and slice patial shape constant, slice test
* fix slice test
* interp test
* add lrn test, test packing layout implicitly
* add eltwise test
* add normalize test
* add instancenorm test
* reorg shape constant
* simple local group size partition
* add shape constant param
* cpu test
* wip
* ci run test
* travis ci for arm64
* arm64 ctest
* copy vulkan loader
* wip
* run
* Update ccpp.yml
* gpu test
* swiftshader
* cache macos swiftshader
* try MoltenVK
* try vulkaninfo
* give swiftshader another try
* disable failed macos gpu test
* more conv test, fix conv3x3s1 gpu test fail
* fix deconvolution test
* dilation test
* cmake option to build tests
* ncnn_add_layer_test macro
* host barrier before upload and after download, handle packing layout option
* test packing layout
* wip
* wip
* merge deconvolution packing and non-packing code
* merge convolution packing and non-packing code
* pass top_blob_count param
* fix build
* take care of non-coherent mappable memory
Stern