topic Re:MKL_CBWR and MKL_ENALBE_INSTRUCTIONS in Intel® oneAPI Math Kernel Library

MKL_CBWR and MKL_ENALBE_INSTRUCTIONS

Bob2023 — Wed, 19 Apr 2023 08:19:51 GMT

I submitted 2 questions, thanks for you guys reply me. But I cannot reply directly in the 2 questions page. After I click the "reply" button, no response. And there is a "reply to this topic" text box and a "Post Reply" button at the bottom of the edge, but the "Post Reply" always gray.

I not sure whether my browser has problem.

I am sorry for I cannot reply or close the following 2 questions.

1, About MKL_CBWR code path dispatch - Intel Communities

2, the relationship for MKL_CBWR and MKL_ENALBE_INSTRUCTIONS - Intel Communities

Thanks for Shanmukh.SS and Praneeth Achanta replied my questions.

for the first question, you can close this questions.

for the second question, I want to continue ask, as the followings:

I notice that you the MKL_ENABLE_INSTURCTIONS is about the CODE PATH, and the MKL_CBWR is about the CODE BRANCH. I don't understand what they are, and there relationship. Would you please explain more?

Maybe for this question, I still cannot reply, I am sorry for that.

thanks

Re: MKL_CBWR and MKL_ENALBE_INSTRUCTIONS

ShanmukhS_Intel — Wed, 26 Apr 2023 18:25:07 GMT

Hi Bob Kim,

Thanks for posting in Intel Communities.

for the first question, you can close this questions.

>>Thanks for the confirmation, We have closed the case.

Maybe for this question, I still cannot reply, I am sorry for that.

>>As you were unable to reply to the community conversation, we would like you to reply to the private mail sent to you, Same information was mentioned below as well.

MKL_ENABLE_INSTRUCTIONS is an environment variable that allows you to enable or disable certain instruction set architectures (ISAs) or specific CPU instructions in MKL, which can affect the optimizations and performance of the library.

For example, you can set MKL_ENABLE_INSTRUCTIONS to "AVX512" to enable AVX-512 instructions, or "AVX2, FMA" to enable AVX2 and Fused Multiply-Add (FMA) instructions.
You could refer to the below link for more information regarding the same.

https://www.intel.com/content/www/us/en/docs/onemkl/developer-reference-c/2023-0/mkl-enable-instructions.html

MKL_CBWR_AUTO could be used to Choose code branch automatically. CNR mode uses the standard ISA-based dispatching model while ensuring fixed cache sizes, deterministic reductions, and static scheduling.

For example, Intel MKL provides different code branches optimized for different generations of Intel processors, such as Intel MKL for Intel Xeon processors, Intel MKL for Intel Core processors, or Intel MKL for Intel Xeon Phi processors. Each code branch have specific optimizations, bug fixes, and features that are tailored for the target architecture, ensuring optimal performance on that particular platform.
mkl_cbwr_get_auto_branch function uses a run-time CPU check to return a CNR branch that is optimized for the processor where the program is currently running.
When enabling CNR, you choose a specific code branch of Intel® oneAPI Math Kernel Library that corresponds to the instruction set architecture (ISA) that you target. Please find the below link for more information regarding the same.

https://www.intel.com/content/www/us/en/docs/onemkl/developer-reference-c/2023-0/named-constants-for-cnr-control.html

Best Regards,

Shanmukh.SS

Re:MKL_CBWR and MKL_ENALBE_INSTRUCTIONS

ShanmukhS_Intel — Sat, 29 Apr 2023 16:49:00 GMT

Hi Bob.

A gentle reminder:

Has the information provided helped? Could you please let us know if we could close this case at our end? If you are unable to post a reply to this query, you could reply to the private mail shared to you.

Best Regards,

Shanmukh.SS

Re:MKL_CBWR and MKL_ENALBE_INSTRUCTIONS

ShanmukhS_Intel — Mon, 08 May 2023 07:48:11 GMT

Hi Bob,

We assume that your issue is resolved. If you need additional information, please post a new question as Intel will no longer monitor this thread.

Best Regards,

Shanmukh.SS