[DeepLearning/Models] add MobileNet-V3 Model benchmark #131

ShiHaoGao · 2024-07-12T06:59:56Z

k230:

local:

zhanghb97 · 2024-07-15T07:33:09Z

CMakeLists.txt

-## Configure the buddy-benchmark build.
-##
-##===----------------------------------------------------------------------===//
+# #===- CMakeLists.txt - buddy-benchmark cmake root ------------*- cmake -*-===//


Please do not modify the format.

zhanghb97 · 2024-07-15T07:34:09Z

CMakeLists.txt

 # BUDDY project.
 set(BUDDY_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
 set(BUDDY_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR}/bin)
-set(BUDDY_EXAMPLES_DIR ${BUDDY_SOURCE_DIR}/examples)
+set(BUDDY_MLIR_BINARY_DIR ${BUDDY_MLIR_BUILD_DIR}/bin)
+set(BUDDY_DEEP_LEARNING_DIR ${BUDDY_SOURCE_DIR}/benchmarks/DeepLearning)


BUDDY_DEEP_LEARNING_DIR -> BUDDY_BENCHMARK_DEEP_LEARNING_DIR

zhanghb97 · 2024-07-15T07:35:21Z

benchmarks/DeepLearning/Models/MobileNet-V3/.gitignore

+
+# model mlir file
+forward.mlir
+subgraph0.mlir


Please add an empty line here.

zhanghb97 · 2024-07-15T07:37:26Z

benchmarks/DeepLearning/Models/MobileNet-V3/GoogleBenchmarkMain.cpp

+  MemRef<long long, 1> ParamsContainerInt64({34});
+  loadParameters(paramsDir, intDir, paramsContainerf32, ParamsContainerInt64);
+  for (auto _ : state) {
+    _mlir_ciface_forward(&output, &paramsContainerf32, &ParamsContainerInt64,


Benchmark should only evaluate the forward function.

zhanghb97 · 2024-07-15T07:42:52Z

benchmarks/DeepLearning/Models/MobileNet-V3/GoogleBenchmarkMain.cpp

+  MemRef<long long, 1> ParamsContainerInt64({34});
+  loadParameters(paramsDir, intDir, paramsContainerf32, ParamsContainerInt64);
+
+  // Call the forward function of the model.


It seems that the benchmark did not clarify these two functions, i.e., scalar vs auto-vectorization

zhanghb97 · 2024-07-15T07:46:31Z

benchmarks/DeepLearning/Models/MobileNet-V3/buddy-mobilenetv3-import.py

@@ -0,0 +1,78 @@
+# ===- buddy-mobilenetv3-import.py ---------------------------------------------


buddy-mobilenetv3-import.py -> buddy_mobilenetv3_import.py

zhanghb97 · 2024-07-15T07:48:34Z

benchmarks/DeepLearning/Models/MobileNet-V3/images/curtain.png

We will use random values as the fake input, so we do not need the following images.

ShiHaoGao · 2024-07-17T08:53:04Z

MobileNet-V3 benchmark on local hardware platform is upated. But, build mobileNet-V3 benchmark lib on RV platform has some errors about LLVM.

LLVM ERROR: Incomplete scavenging after 2nd pass
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc -O0 -mtriple=riscv64 -mattr=+v,+m -filetype=obj -o /home/gaoshihao/project/buddy-benchmark/build/bin/../benchmarks/DeepLearning/Models/MobileNet-V3/subgraph0_scalar.o
1.      Running pass 'Function Pass Manager' on module '<stdin>'.
2.      Running pass 'Prologue/Epilogue Insertion & Frame Finalization' on function '@_mlir_ciface_subgraph0_scalar'
 #0 0x0000564f422cd850 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x1d65850)
 #1 0x0000564f422cac5f llvm::sys::RunSignalHandlers() (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x1d62c5f)
 #2 0x0000564f422cadb5 SignalHandler(int) Signals.cpp:0:0
 #3 0x00007f2024a71520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x00007f2024ac59fc __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x00007f2024ac59fc __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x00007f2024ac59fc pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x00007f2024a71476 gsignal ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x00007f2024a577f3 abort ./stdlib/abort.c:81:7
 #9 0x0000564f40a07910 llvm::RISCVISAInfo::updateImplication() (.cold) RISCVISAInfo.cpp:0:0
#10 0x0000564f42221e8e (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x1cb9e8e)
#11 0x0000564f41748654 llvm::scavengeFrameVirtualRegs(llvm::MachineFunction&, llvm::RegScavenger&) (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x11e0654)
#12 0x0000564f416a778b (anonymous namespace)::PEI::runOnMachineFunction(llvm::MachineFunction&) PrologEpilogInserter.cpp:0:0
#13 0x0000564f4151f837 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (.part.0) MachineFunctionPass.cpp:0:0
#14 0x0000564f41ae80de llvm::FPPassManager::runOnFunction(llvm::Function&) (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x15800de)
#15 0x0000564f41ae8329 llvm::FPPassManager::runOnModule(llvm::Module&) (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x1580329)
#16 0x0000564f41ae8c65 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x1580c65)
#17 0x0000564f40a8177e compileModule(char**, llvm::LLVMContext&) llc.cpp:0:0
#18 0x0000564f40a0cfb6 main (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x4a4fb6)
#19 0x00007f2024a58d90 __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:58:16
#20 0x00007f2024a58e40 call_init ./csu/../csu/libc-start.c:128:20
#21 0x00007f2024a58e40 __libc_start_main ./csu/../csu/libc-start.c:379:5
#22 0x0000564f40a78495 _start (/home/gaoshihao/project/buddy-mlir/build/../llvm/build/bin/llc+0x510495)
Aborted (core dumped)
ninja: build stopped: subcommand failed.

local:

zhanghb97

OSError: The environment variable 'MOBILENETV3_MODEL_PATH' is not set or is invalid.

zhanghb97 · 2024-07-22T08:14:58Z

[18/18] Linking CXX executable bin/dl-model-mobileNetV3-benchmark
FAILED: bin/dl-model-mobileNetV3-benchmark 
: && /home/zhanghongbin/buddy-mlir/build/thirdparty/riscv-gnu-toolchain/bin/riscv64-unknown-linux-gnu-g++ -no-pie -O3 -DNDEBUG -static benchmarks/DeepLearning/Models/MobileNet-V3/CMakeFiles/dl-model-mobileNetV3-benchmark.dir/GoogleBenchmarkMain.cpp.o -o bin/dl-model-mobileNetV3-benchmark  vendor/benchmark/lib/libbenchmark.a  benchmarks/DeepLearning/Models/MobileNet-V3/libMOBILENETV3_SCALAR.a  benchmarks/DeepLearning/Models/MobileNet-V3/libMOBILENETV3_AUTO_VECTORIZATION.a  /home/zhanghongbin/buddy-mlir/build-cross-rv//lib/libStaticMLIRCRunnerUtils.a && :
/home/zhanghongbin/buddy-mlir/build/thirdparty/riscv-gnu-toolchain/lib/gcc/riscv64-unknown-linux-gnu/13.2.0/../../../../riscv64-unknown-linux-gnu/bin/ld: benchmarks/DeepLearning/Models/MobileNet-V3/libMOBILENETV3_SCALAR.a(forward_scalar.o): Relocations in generic ELF (EM: 62)
/home/zhanghongbin/buddy-mlir/build/thirdparty/riscv-gnu-toolchain/lib/gcc/riscv64-unknown-linux-gnu/13.2.0/../../../../riscv64-unknown-linux-gnu/bin/ld: benchmarks/DeepLearning/Models/MobileNet-V3/libMOBILENETV3_SCALAR.a(forward_scalar.o): Relocations in generic ELF (EM: 62)
/home/zhanghongbin/buddy-mlir/build/thirdparty/riscv-gnu-toolchain/lib/gcc/riscv64-unknown-linux-gnu/13.2.0/../../../../riscv64-unknown-linux-gnu/bin/ld: benchmarks/DeepLearning/Models/MobileNet-V3/libMOBILENETV3_SCALAR.a: error adding symbols: file in wrong format
collect2: error: ld returned 1 exit status
ninja: build stopped: subcommand failed.

ShiHaoGao · 2024-08-13T13:48:38Z

k230:

zhanghb97 · 2024-08-17T03:59:05Z

The README does not contain tips and guidance for the Python environment

The RVV target file has the wrong format

zhanghb97 · 2024-08-18T01:47:06Z

CMakeLists.txt

+  set(BUDDY_OPT_TRIPLE riscv64)
+  set(BUDDY_OPT_ATTR +v,+m)
+  set(BUDDY_MLIR_CROSS_LIB_DIR ${BUDDY_MLIR_BUILD_CROSS_DIR}/lib)
+else()


CMake bug!
Please include BUDDY_OPT_TRIPLE and BUDDY_OPT_ATTR for local machine in the else branch.

zhanghb97 · 2024-08-18T01:51:10Z

benchmarks/DeepLearning/Models/MobileNet-V3/CMakeLists.txt

+)
+
+add_custom_command(
+  OUTPUT forward_autoVectorization.o


Weird naming convention.
forward_autoVectorization -> forward_auto_vectorization

zhanghb97 · 2024-08-18T02:03:20Z

benchmarks/DeepLearning/README.md

@@ -1,5 +1,12 @@
 # Deep Learning Benchmark

+## Model Level Benchmark


Add the statement for Python (using model-level benchmark needs the Python environment), e.g., enter the Python environment, configure the Python path, etc.

zhanghb97 · 2024-08-18T02:21:26Z

benchmarks/DeepLearning/Models/MobileNet-V3/README.md

@@ -0,0 +1,49 @@
+# Buddy Compiler MobileNetV3 Benchmark


Let’s not make the READMEs too scattered. Remove this file and add the necessary information (python env and path) to the deep learning benchmark README.

zhanghb97

Thanks!

add MobileNet-V3 Model benchmark

29931c0

zhanghb97 reviewed Jul 15, 2024

View reviewed changes

update MobileNet-V3 benchmark

db836e2

zhanghb97 reviewed Jul 22, 2024

View reviewed changes

ShiHaoGao and others added 2 commits August 12, 2024 07:59

Merge branch 'buddy-compiler:main' into main

0d674fe

add batchMatMul opt pass

4af8f9b

Your Name added 2 commits August 16, 2024 13:41

Update MobileNet-V3 README.md

e485324

Fix bugs on MobileNet-V3 benchmark

ecf689e

Fix <file in wrong format> bug

13d508f

zhanghb97 reviewed Aug 18, 2024

View reviewed changes

Update readme & fix CMakeList.txt

fa99315

zhanghb97 approved these changes Aug 18, 2024

View reviewed changes

zhanghb97 merged commit 7ffe997 into buddy-compiler:main Aug 18, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[DeepLearning/Models] add MobileNet-V3 Model benchmark #131

[DeepLearning/Models] add MobileNet-V3 Model benchmark #131

ShiHaoGao commented Jul 12, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

zhanghb97 Jul 15, 2024

ShiHaoGao commented Jul 17, 2024 •

edited

Loading

zhanghb97 left a comment

zhanghb97 commented Jul 22, 2024

ShiHaoGao commented Aug 13, 2024

zhanghb97 commented Aug 17, 2024

zhanghb97 Aug 18, 2024 •

edited

Loading

zhanghb97 Aug 18, 2024

zhanghb97 Aug 18, 2024

zhanghb97 Aug 18, 2024

zhanghb97 left a comment

		@@ -0,0 +1,78 @@
		# ===- buddy-mobilenetv3-import.py ---------------------------------------------

		@@ -1,5 +1,12 @@
		# Deep Learning Benchmark

		## Model Level Benchmark

[DeepLearning/Models] add MobileNet-V3 Model benchmark #131

[DeepLearning/Models] add MobileNet-V3 Model benchmark #131

Conversation

ShiHaoGao commented Jul 12, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ShiHaoGao commented Jul 17, 2024 • edited Loading

zhanghb97 left a comment

Choose a reason for hiding this comment

zhanghb97 commented Jul 22, 2024

ShiHaoGao commented Aug 13, 2024

zhanghb97 commented Aug 17, 2024

zhanghb97 Aug 18, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhanghb97 left a comment

Choose a reason for hiding this comment

ShiHaoGao commented Jul 17, 2024 •

edited

Loading

zhanghb97 Aug 18, 2024 •

edited

Loading