code clean - refine ARM compile#17590
Conversation
cmake enhancement: - add lite_cc_library and lite_cc_test code clean: - remove ARM feed and fetch kernels, reuse the Host's remove unnecessary comments
| option(LITE_WITH_X86 "Enable X86 in lite mode" ON) | ||
| option(LITE_WITH_X86 "Enable X86 in lite mode" ON) | ||
| option(LITE_WITH_ARM "Enable ARM in lite mode" OFF) | ||
| option(LITE_WITH_LIGHT_WEIGHT_FRAMEWORK "Enable light-weight framework" OFF) |
There was a problem hiding this comment.
比较疑惑LITE_WITH_ARM，LITE_WITH_LIGHT_WEIGHT_FRAMEWORK，WITH_LITE 这几个之间的关系
There was a problem hiding this comment.
WITH_LITE 控制整个lite 目录是否会编译
LITE_WITH_LIGHT_WEIGHT_FRAMEWORK 控制是否用轻量级框架， server端也可以用轻量级
LITE_WITH_ARM 是控制是否编译 ARM 相关模块，包括kernel等。 ARM 和 X86 以及 CUDA 都是平级的硬件，本质上不做特殊支持
暂时 LITE_WITH_LIGHT_WEIGHT_FRAMEWORK + LITE_WITH_ARM 可以认为是在mobile平台上跑
There was a problem hiding this comment.
LITE_WITH_LIGHT_WEIGHT_FRAMEWORK 控制是否用轻量级框架， server端也可以用轻量级
他指的是框架轻量级？lite本身的目的就是轻量级吧，是不是应该把他们合成一个？
There was a problem hiding this comment.
lite 之后有可能会接入训练； 目前已有的 server op 都是基于重的框架，包括 allocator, tensor, scope, platform 等，短暂没法合并。
所以暂时只能 mobile 用轻量级，后面再看看
tensor-tang
left a comment
tensor-tang
left a comment
There was a problem hiding this comment.
希望后期可以把LITE_WITH_LIGHT_WEIGHT_FRAMEWORK 整理下，这个与WITH_LITE有点重复。
2 participants
cmake enhancement:
code clean:
remove unnecessary comments