[feat](benchmark): replace load_param with load_param_mem

[0130w]

Previous implementation required benchncnn to be run from the directory containing the model files. This often caused failures and inconvenience when executed from an arbitrary working directory.​​

​This change embeds the model files directly into the executable by compiling them as C arrays. Consequently, benchncnn now becomes fully self-contained and can be run from any location without dependency on external model files.​

[github-actions]

The binary size change of libncnn.so (bytes)

architecture	base size	pr size	difference
x86_64	15328640	15212648	-115992 😘
armhf	6229904	6206656	-23248 😘
aarch64	9527568	9524560	-3008 😘

[tencent-adm]

All committers have signed the CLA.

[0130w]

run

cd build
cmake -DCMAKE_BUILD_TYPE=Debug -DNCNN_VULKAN=ON -DNCNN_BUILD_EXAMPLES=ON -DCMAKE_EXPORT_COMPILE_COMMANDS=ON ..
make -j$(nproc)
cd benchmark
./benchncnn 4 8 0 -1 1

result before

./benchncnn 4 8 0 -1 1
loop_count = 4
num_threads = 8
powersave = 0
gpu_device = -1
cooling_down = 1
fopen squeezenet.param failed
network graph not ready
          squeezenet  min =    0.00  max =    0.00  avg =    0.00
fopen squeezenet_int8.param failed
...

after

loop_count = 4
num_threads = 8
powersave = 0
gpu_device = -1
cooling_down = 1
          squeezenet  min =   12.68  max =   13.12  avg =   12.94
     squeezenet_int8  min =    8.39  max =    9.11  avg =    8.88
           mobilenet  min =   16.69  max =   17.01  avg =   16.83
      mobilenet_int8  min =   15.41  max =   16.59  avg =   15.86
        mobilenet_v2  min =   16.33  max =   17.03  avg =   16.67
        mobilenet_v3  min =   13.08  max =   15.00  avg =   14.34
          shufflenet  min =   10.04  max =   12.69  avg =   10.88
       shufflenet_v2  min =   10.14  max =   13.40  avg =   11.20
             mnasnet  min =   12.67  max =   16.30  avg =   14.48
     proxylessnasnet  min =   14.56  max =   16.55  avg =   15.68
     efficientnet_b0  min =   23.54  max =   25.99  avg =   24.56
   efficientnetv2_b0  min =   25.69  max =   28.93  avg =   27.90
...

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.29%. Comparing base (dce28a0) to head (85399e8).

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #6286    +/-   ##
========================================
  Coverage   93.28%   93.29%            
========================================
  Files         845      845            
  Lines      265939   266146   +207     
========================================
+ Hits       248086   248305   +219     
+ Misses      17853    17841    -12     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull Request Overview

This PR replaces file-based model loading with memory-based loading in the benchncnn tool by embedding model parameter files directly into the executable as C arrays. This eliminates the dependency on external .param files and allows the benchmark to run from any directory.

• Introduces a CMake macro to convert .param files into C header files containing hex data arrays
• Updates benchncnn.cpp to use load_param_mem() instead of load_param() with embedded data
• Normalizes model names by replacing hyphens with underscores to ensure valid C identifiers

Reviewed Changes

Copilot reviewed 6 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
cmake/ncnn_generate_model_data_header.cmake	Adds CMake macro to convert .param files to C arrays with validation
benchmark/model_param_spv_data.h.in	Template for including generated model parameter headers
benchmark/model_param_registry.h.in	Template for model parameter registry data
benchmark/benchncnn.cpp	Updates to use embedded model data with registry lookup function
benchmark/README.md	Updates documentation to reflect new param usage without .param suffix
benchmark/CMakeLists.txt	Integrates model data generation into build process

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[0130w]

System: macos
modify vgg16.param

InnerProduct             fc8                      1 1 fc7_drop7 fc8 -23330=4,1,1000,1,1 0=1000 1=1 2=4096000
InnerProduct             fc9                      1 1 fc8 fc9 0=1000 1=1 2=1000000
Softmax                  prob                     1 1 fc9 output -23330=4,1,1000,1,1

run make -j4

[  0%] Built target GenericCodeGen
[  0%] Built target MachineIndependent
[  1%] Built target SPIRV
[  1%] Built target OSDependent
[ 38%] Built target ncnn-generate-spirv
[ 38%] Built target glslang-default-resource-limits
[ 38%] Preprocessing param source vgg16.param
[ 38%] Built target ncnnmerge
[ 39%] Built target caffe2ncnn
[ 45%] Built target ncnn-generate-param
[ 52%] Built target glslang
[ 52%] Built target mxnet2ncnn
[ 52%] Built target darknet2ncnn
[ 98%] Built target ncnn
[ 98%] Built target ncnn2mem
[100%] Built target ncnnoptimize
[100%] Building CXX object benchmark/CMakeFiles/benchncnn.dir/benchncnn.cpp.o
[100%] Built target ncnn2table
[100%] Built target ncnn2int8
[100%] Linking CXX executable benchncnn
[100%] Built target benchncnn

[tencent-adm]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ 0130w
❌ nihui
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[nihui]

Thanks for your contribution !