- 12 11月, 2021 1 次提交
-
-
由 Huihuang Zheng 提交于
Fix Paddle-CINN CI
-
- 05 11月, 2021 1 次提交
-
-
由 Zhen Wang 提交于
* Use a more appropriate `Compile` method in cinn_launch_op. * Update the VizGraph method of CinnCompiler. * Add resnet50 model training with CINN.
-
- 03 11月, 2021 1 次提交
-
-
由 Zhen Wang 提交于
Add FLAGS_allow_cinn_ops & FLAGS_deny_cinn_ops for controlling op types used in training with CINN. (#36842) * Update UT test_parallel_executor_run_cinn.py. * Add FLAGS_allow_cinn_ops & FLAGS_deny_cinn_ops & FLAGS_cinn_ops_delim. * Use the custom StringSplit function and remove the FLAGS_cinn_ops_delim flag. * Add FlagController test. * Apply lock to the cache_ only in CinnCompiler. * Add VizGraph & ReadableKey method for CinnCompiler. * Update the dot style of VizGraph in CinnCompiler.
-
- 01 11月, 2021 1 次提交
-
-
由 CtfGo 提交于
增加CinnLaunchOp,负责执行Cinn子图编译的结果,要点如下: 1. 在子图划分的BuildCinnPass中,每个子图在原图中会被替换为该CinnLaunchOp,由它来调用Cinn进行子图编译、执行的功能。 2. CinnLaunchOp的输入/输出即为子图的输入和输出,另外增加`compilation_key`属性,它可由该属性key从全局Cache中获取子图对象、编译结果,该属性由BuildCinnPass在创建Op时进行设置 3. CinnLaunchOp功能实现的流程为: - 从全局Cache中获取子图对象 - 从全局Cache中获取子图编译结果,未命中cache时进行即时编译 - 根据编译结果的变量信息(数据类型、shape)初始化运行时数据,分配内存/显存 - 将运行时数据打包为参数,调用cinn的可执行对象runtime program进行计算 - 子图运行结果通过参数指针同步到paddle侧的tensor
-
- 25 10月, 2021 1 次提交
-
-
由 Zhen Wang 提交于
* Init the functions of CinnCompiler. * Add the unit test for CinnCompiler. * Fix some compilation errors. * Update the UT of cinn_compiler. * Use Decomposer&OpFusion passes in CinnCompiler::CompileGraph. * Update some comments. * Uncomment some includes in build_cinn_pass.cc. * Use refs instead of ptrs as returned types of FindGraph & Compile in CinnCompiler. * Use the merged CinnGraphSymbolization functions in CinnCompiler.
-