Add op cholesky

[FisherWY]

#1115

Add op cholesky

[FisherWY]

Cholesky算子：

• GPU算子：C++侧，Python侧测试均通过
• CPU算子：C++侧测试通过，Python侧出错，错误日志如下

I0120 08:53:47.877772 40414 graph.cc:38] operator [cholesky] has [1] inputs, and [1] outputs
I0120 08:53:47.878170 40414 pass.cc:26] Run Pass -> MatmulToCublasCustomCallPass
I0120 08:53:47.878217 40414 pass.cc:26] Run Pass -> GaussianRandomToCustomCallPass
I0120 08:53:47.878224 40414 pass.cc:26] Run Pass -> UniformRandomToCustomCallPass
I0120 08:53:47.878230 40414 pass.cc:26] Run Pass -> CholeskyToCustomCallPass
I0120 08:53:47.878237 40414 pass.cc:26] Run Pass -> ConvToCudnnCustomCallPass
I0120 08:53:47.878242 40414 pass.cc:26] Run Pass -> OpFusionPass
I0120 08:53:47.878247 40414 pass.cc:26] Run Pass -> FusionMergePass
I0120 08:53:47.878260 40414 custom_call_pass.cc:170] MatmulToCublasCustomCallPass...!
I0120 08:53:47.878297 40414 custom_call_pass.cc:172] MatmulToCublasCustomCallPass Finish...!
I0120 08:53:47.878304 40414 custom_call_pass.cc:182] GaussianRandomToCustomCall...!
I0120 08:53:47.878314 40414 custom_call_pass.cc:184] GaussianRandomToCustomCall Finish...!
I0120 08:53:47.878319 40414 custom_call_pass.cc:188] UniformRandomToCustomCall...!
I0120 08:53:47.878330 40414 custom_call_pass.cc:190] UniformRandomToCustomCall Finish...!
I0120 08:53:47.878335 40414 custom_call_pass.cc:194] CholeskyToCustomCallPass...!
I0120 08:53:47.878443 40414 custom_call_pass.cc:196] CholeskyToCustomCallPass Finish...!
I0120 08:53:47.878450 40414 custom_call_pass.cc:176] ConvToCudnnCustomCallPass...!
I0120 08:53:47.878463 40414 custom_call_pass.cc:178] ConvToCudnnCustomCallPass Finish...!
I0120 08:53:47.878471 40414 op_fusion_pass.cc:433] OpFusionPass...!
I0120 08:53:47.878928 40414 op_fusion_pass.cc:439] Group Id : cholesky_0
I0120 08:53:47.878943 40414 op_fusion_pass.cc:447] OpFusionPass Finish...!
I0120 08:53:47.878986 40414 fusion_merge_pass.cc:929] Before FusionMergePass:
Group 0 {
  {var_0} = custom_call{x, id=cholesky_0, custom_call=cinn_call_cholesky_host, upper=false, }
}
I0120 08:53:47.879040 40414 fusion_merge_pass.cc:931] Don't do Fusoin Merge Pass...!
I0120 08:53:47.879091 40414 scope.h:65] Scope insert Var [x]
I0120 08:53:47.879163 40414 graph_compiler.cc:1375] Tensor [x] resize to 2,3,3
I0120 08:53:47.879196 40414 scope.h:65] Scope insert Var [var_0]
I0120 08:53:47.879230 40414 graph_compiler.cc:1375] Tensor [var_0] resize to 2,3,3
I0120 08:53:47.879366 40414 graph_compiler.cc:753] Begin GraphCompiler::Build
I0120 08:53:47.879400 40414 graph_compiler.cc:786] group_id is : cholesky_0, and its number is : 1
I0120 08:53:47.879420 40414 op_lowering.cc:107] Lowering Group : cholesky_0 , Op Pattern : 8
I0120 08:53:47.879427 40414 op_lowering.cc:1241] LowerNonFusibleOp Group : cholesky_0
I0120 08:53:47.879446 40414 op_lowering.cc:1248] GetOpFunc of op cholesky_0
I0120 08:53:47.879808 40414 op_lowering.cc:1292] cinn_inputs.push_back var_0
I0120 08:53:47.880364 40414 lowered_func.cc:197] Function used 0 buffers
I0120 08:53:47.880548 40414 custom_call.cc:105] function fn_cholesky_0 (kernel_args, kernel_args_num)
cinn_call_cholesky_host(kernel_args, kernel_args_num, 2, 3, 0)
I0120 08:53:47.880700 40414 graph_compiler.cc:790] function fn_cholesky_0 (kernel_args, kernel_args_num)
cinn_call_cholesky_host(kernel_args, kernel_args_num, 2, 3, 0)
I0120 08:53:47.880733 40414 graph.cc:119] Group 0 {
  {var_0} = custom_call{x, id=cholesky_0, custom_call=cinn_call_cholesky_host, upper=false, }
}
I0120 08:53:47.880771 40414 execution_engine.cc:111] ===================== Create CINN ExecutionEngine begin ====================
I0120 08:53:47.880777 40414 execution_engine.cc:112] initialize llvm config
I0120 08:53:47.880781 40414 execution_engine.cc:113] llvm version: 12.0.0git
I0120 08:53:47.880787 40414 execution_engine.cc:114] llvm default target triple: x86_64-unknown-linux-gnu
I0120 08:53:47.882594 40414 execution_engine.cc:140] create jit execution engine
I0120 08:53:47.883177 40414 execution_engine.cc:124] create llvm compile layer
I0120 08:53:47.883186 40414 execution_engine.cc:125] Target Name: x86-64
I0120 08:53:47.883198 40414 execution_engine.cc:126] Target CPU: skylake-avx512
I0120 08:53:47.883602 40414 execution_engine.cc:148] register runtime call symbols
I0120 08:53:47.885574 40414 execution_engine.cc:152] ===================== Create CINN ExecutionEngine end ====================
I0120 08:53:47.885984 40414 graph_compiler.cc:830] End of m_builder_.Build()
I0120 08:53:47.886126 40414 graph_compiler.cc:835] [X86] C Code is:
#include <cinn_runtime.h>
#include <stdio.h>

void fn_cholesky_0(void* _args, int32_t num_args)
{
  void* kernel_args = cinn_pod_value_to_void_p(&(((cinn_pod_value_t*)(_args))[0]));
  int32_t kernel_args_num = cinn_pod_value_to_int32(&(((cinn_pod_value_t*)(_args))[1]));
  cinn_call_cholesky_host(kernel_args, kernel_args_num, 2, 3, 0);
}

I0120 08:53:47.895321 40414 execution_engine.cc:163] ir_emitter->Compile(module) Begin
I0120 08:53:47.895362 40414 codegen_llvm.cc:699] JIT Linking function [fn_cholesky_0]
I0120 08:53:47.895783 40414 extern_func_emitter_builtin.cc:79] function type cinn_call_cholesky_host: declare void @cinn_call_cholesky_host(i8*, i32, i32, i32, i1)
python: /root/github.com/hp03/CINN/THIRDS/llvm-project/llvm/lib/IR/Instructions.cpp:455: void llvm::CallInst::init(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, const llvm::Twine&): Assertion `(i >= FTy->getNumParams() || FTy->getParamType(i) == Args[i]->getType()) && "Calling a function with a bad signature!"' failed.


--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0   cinn::hlir::framework::GraphCompiler::Build(std::string const&)
1   cinn::hlir::framework::GraphCompiler::Build(cinn::hlir::framework::GraphCompiler::CompileOptions const&, std::unordered_set<std::string, std::hash<std::string >, std::equal_to<std::string >, std::allocator<std::string > >&&, void*)
2   cinn::backends::Compiler::Build(cinn::ir::Module const&, std::string const&)
3   cinn::backends::Compiler::CompileX86Module(cinn::ir::Module const&)
4   void cinn::backends::ExecutionEngine::Link<cinn::backends::CodeGenX86>(cinn::ir::Module const&)
5   cinn::backends::CodeGenLLVM::Compile(cinn::ir::Module const&)
6   cinn::backends::CodeGenLLVM::Visit(cinn::ir::_Module_ const*)
7   cinn::ir::IRVisitorBase<llvm::Value*>::Visit(cinn::ir::Expr const*)
8   cinn::backends::CodeGenLLVM::Visit(cinn::ir::_LoweredFunc_ const*)
9   cinn::ir::IRVisitorBase<llvm::Value*>::Visit(cinn::ir::Expr const*)
10  cinn::backends::CodeGenLLVM::Visit(cinn::ir::Block const*)
11  cinn::ir::IRVisitorBase<llvm::Value*>::Visit(cinn::ir::Expr const*)
12  cinn::backends::CodeGenLLVM::Visit(cinn::ir::Call const*)
13  cinn::backends::ExternFunctionEmitter::Emit(cinn::ir::Call const*, bool)
14  cinn::backends::ExternFunctionLLVMEmitter::EmitImpl(cinn::ir::Call const*)
15  llvm::IRBuilderBase::CreateCall(llvm::FunctionCallee, llvm::ArrayRef<llvm::Value*>, llvm::Twine const&, llvm::MDNode*)
16  llvm::CallInst::init(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, llvm::Twine const&)

----------------------
Error Message Summary:
----------------------
FatalError: `Process abort signal` is detected by the operating system.
  [TimeInfo: *** Aborted at 1674204827 (unix time) try "date -d @1674204827" if you are using GNU date ***]
  [SignalInfo: *** SIGABRT (@0x9dde) received by PID 40414 (TID 0x7f3d0aed4740) from PID 40414 ***]

Aborted (core dumped)
λ 29590e341c28 /home/wengyu02/work/CINN 

如何复现：编辑python/tests/ops/test_cholesky_op.py，将self.target切换至DefaultHostTarget()，运行测试命令GLOG_v=6 python python/tests/ops/test_cholesky_op.py，报core dump

[thisjiang]

LGTM