Lots of questions!

Lots of questions!

Аватар пользователя oscarbg

Hi I have some questions: *More OS support: Win 64 support and Linux *Is this release capable of generating AVX code either via autovectorization or using float8 for example *Can we expect to have Intel MIC support when released? *There is a way to force the compiler to not autovectorize so it behaves like amd opencl implementation.. this would allow perf comparisons between AMD and Intel opencl CPU implementations in a more apples and apples comparison and also compare the improvements of autovectorization.. I hope autovectorization is disabled passing flag setting for null optimizin but that wouldn't be fair also.. *What the expected frecuency of updates of alpha version till release of commercial product i.e. we can expect quarterly updates for example?

3 posts / 0 новое
Последнее сообщение
Пожалуйста, обратитесь к странице Уведомление об оптимизации для более подробной информации относительно производительности и оптимизации в программных продуктах компании Intel.
Аватар пользователя Arnon Peleg (Intel)

Thanks for
sharing your feedback.

The current
version of the Intel OpenCL SDK is only a technology preview. When a new
update will be available, we announce it in this forum.

For our future
products and technology support such as Intel MIC Architecture and AVX, we are
evaluating when and where OpenCL support will intercept our products,
however no announcement has been made.

Аватар пользователя Ofer Rosenberg (Intel)

Hi,

Regarding your question about disabling the autovectorization: Intel OpenCL SDK supports the optional __attribute__((vec_type_hint())), as defined in the OpenCL 1.1 specification, section 6.7.2:

"The optional __attribute__((vec_type_hint())) is a hint to the compiler and is intended to be a representation of the computational width of the __kernel, and should serve as the basis for calculating processor bandwidth utilization when the compiler is looking to autovectorize the code."

Using this attribute in the kernel declaration with values like float4, int4 etc. will disable the autovectorization of the specific kernel.

Regards,
Ofer Rosenberg
OpenCL Tech Lead
Intel

Зарегистрируйтесь, чтобы оставить комментарий.