20. Appendix#

20.1. Check AMD Server Processor Architecture#

20.1.1. On Linux#

To identify your AMD processor’s generation, perform the following steps on Linux:

Run the command:
```
$ lscpu
```
Check the values of CPU family and Model fields:
1. For 1st Gen AMD EPYC^TM Processors (codenamed “Naples”), CPU Core AMD “Zen”
  - CPU Family: 23
  - Model: Values in the range <1-47>
2. For 2nd Gen AMD EPYC^TM Processors (codenamed “Rome”), CPU Core AMD “Zen2”
  - CPU Family: 23
  - Model: Values in the range <48-63>
3. For 3rd Gen AMD EPYC^TM Processors (codenamed “Milan”), CPU Core AMD “Zen3”
  - CPU Family: 25
  - Model: Values in the range <1-15>
4. For 4th Gen AMD EPYC^TM Processors (codenamed “Genoa”), CPU Core AMD “Zen4”
  - CPU Family: 25
  - Model: Values in the range <16-31, 96-111, 120-123, 160-175>
5. For 5th Gen AMD EPYC^TM Processors (codenamed “Turin”), CPU Core AMD “Zen5”
  - CPU Family: 26
  - Model: Values in the range <0-15, 16-31>

20.1.2. On Windows#

To identify your AMD processor’s generation, perform the following steps on Windows:

Run the command in Windows Command Prompt:
```
$ wmic cpu get caption
```
Check the values of CPU family and Model fields:
1. For 1st Gen AMD EPYC^TM Processors (codenamed “Naples”), CPU Core AMD “Zen”
  - CPU Family: 23
  - Model: Values in the range <1-47>
2. For 2nd Gen AMD EPYC^TM Processors (codenamed “Rome”), CPU Core AMD “Zen2”
  - CPU Family: 23
  - Model: Values in the range <48-63>
3. For 3rd Gen AMD EPYC^TM Processors (codenamed “Milan”), CPU Core AMD “Zen3”
  - CPU Family: 25
  - Model: Values in the range <1-15>
4. For 4th Gen AMD EPYC^TM Processors (codenamed “Genoa”), CPU Core AMD “Zen4”
  - CPU Family: 25
  - Model: Values in the range <16-31, 96-111, 120-123, 160-175>
5. For 5th Gen AMD EPYC^TM Processors (codenamed “Turin”), CPU Core AMD “Zen5”
  - CPU Family: 26
  - Model: Values in the range <0-15, 16-31>

20.2. Application Notes#

20.2.1. AOCL-BLAS#

If you prefer to build the application or the test suite executable with the pre-built static library (from the package) on Windows, both the instances of #define BLIS_ENABLE_SHARED must be commented out in the header file blis.h.

20.2.2. AOCL-FFTW#

Quad precision is supported in AOCL-FFTW using the AOCC v2.2 compiler (AMD clang version 10 onwards).
Feature AMD application optimization layer has been introduced in AOCL-FFTW to uplift the performance of various HPC and scientific applications.
- The configure option –enable-amd-app-opt enables this optimization layer and must be used with the master optimization configure switch –enable-amd-opt mandatorily.
- This optimization layer is supported for complex and real (r2c and c2r) DFT problem types in double and single precisions.
- Not supported for MPI FFTs, real r2r DFT problem types, Quad or Long double precisions, and split array format.

Appendix - 5.0 English - 57404