Ryzen/ZEN DYNAMIC_ARCH support is broken

[gcp]

Building from source from 20a413e on Ubuntu 16.04 (gcc 5.4.1) with 4.10.1 kernel.

Zen is not autodetected, the machine is detected as Barcelona instead. Trying to manually specify OPENBLAS_CORETYPE=Zen results in Prescott kernels being used.

Removing DYNAMIC_ARCH=1 results in a ZEN kernel. So static detection works (but see next issue, performance is worse than with HASWELL kernel...).

processor : 15
vendor_id : AuthenticAMD
cpu family : 23
model : 1
model name : AMD Ryzen 7 1700 Eight-Core Processor
stepping : 1
microcode : 0x8001105
cpu MHz : 1550.000
cache size : 512 KB
physical id : 0
siblings : 16
core id : 7
cpu cores : 8
apicid : 15
initial apicid : 15
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_l2 mwaitx hw_pstate vmmcall fsgsbase bmi1 avx2 smep bmi2 rdseed adx smap clflushopt sha_ni xsaveopt xsavec xgetbv1 xsaves clzero irperf arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic overflow_recov succor smca
bugs : fxsave_leak sysret_ss_attrs null_seg
bogomips : 7163.90
TLB size : 2560 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 48 bits physical, 48 bits virtual
power management: ts ttp tm hwpstate eff_freq_ro [13] [14]

[martin-frbg]

AMD Ryzen support is currently very sketchy, please see the notes in PR #1133 if you haven't already. Instead of setting OPENBLAS_CORETYPE=Zen I suspect building with TARGET=ZEN would have worked better ? (Not sure why the DYNAMIC_ARCH build would fallback to BARCELONA unless you disallowed avx - did you get a fallback warning from the code ?)

[gcp]

No fallback warnings. As explained in the original post, making a static target works (but is useless for me).

[gcp]

Also obviously AVX is working because OPENBLAS_CORETYPE=Haswell works fine.

[gcp]

https://github.com/xianyi/OpenBLAS/blob/develop/driver/others/dynamic.c#L395

This check should've been for "exfamily" instead. I'll patch this up and prepare a pull request.

[martin-frbg]

Right, thanks. I obviously missed that typo when I committed his PR, and did (do) not have the hardware to notice its effect, sorry.

[gcp]

Setting OPENBLAS_CORETYPE=Zen still makes this default to Prescott, so there's more fixes to be done. (Unsetting makes it autodetect correctly)

[martin-frbg]

That appears to be in force_coretype (driver/others/dynamic.c as well) - the for loop that compares corenames runs to 22 only...

[gcp]

Updated the pull request with a fix for that too.