Not working properly on gfx906

[chenyongqing270]

There are three problems： 1、In the generated code, there is no instruction to output the result 2、without the the first problem solved ，the verification is still successful。 the problem can solve with the method that code snippet in FWD in conv_driver.cpp, should be add hipMemset call for device_output。 3、After I solved the first problem with the following method: The result is not correct。both offset of output and Calculation results are incorrect。when i init all input value to 0.3 and weight value to 0.2, i get result as following：

[carlushuang]

1、In the generated code, there is no instruction to output the result what do you mean by output the result? Is it print the result of the output buffer? Currently this can be done by set env var PER_PIXEL_CHECK to 1

question 2/3 seems when you manually add the gfx906 code and have not got the correct result. Currently we are actively working on gfx908 codegen and not have enough time to fix some gfx906 code, due to limited resource. We will have that done when current task is done

[chenyongqing270]

1、In the generated code, there is no instruction to output the result what do you mean by output the result? Is it print the result of the output buffer? Currently this can be done by set env var PER_PIXEL_CHECK to 1

thanks for your reply。 i mean that there is no instruction to write the summation from v-register to global memory。 Because I have only one gfx906-card on hand，I have to do some research based on 906。Excuse me！