tkDNN works with trt8!!,need to test int8 and mobilenet,dla_cnet (fps seems to be a bit low 350 on trt8 compared to 396 on trt7)

This commit is contained in:
perseusdg
2021-09-07 22:00:28 +05:30
parent 9fa116ce4a
commit 03473743c4
30 changed files with 43 additions and 16 deletions
+2 -2
View File
@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.15)
project (tkDNN)
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
if(UNIX)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")
endif()
if(WIN32)
set(CMAKE_CXX_STANDARD 11)
@@ -31,7 +31,7 @@ endif()
find_package(CUDA 9.0 REQUIRED)
SET(CUDA_SEPARABLE_COMPILATION ON)
#set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61)
find_package(CUDNN REQUIRED)