]> git.proxmox.com Git - qemu-server.git/commit
cpuconfig: add new x86-64-vX models
authorAlexandre Derumier <aderumier@odiso.com>
Thu, 8 Jun 2023 09:51:05 +0000 (11:51 +0200)
committerFiona Ebner <f.ebner@proxmox.com>
Mon, 12 Jun 2023 15:30:11 +0000 (17:30 +0200)
commit1359e23fb43199a613a3192f77b0fe1392817769
treedb1597bd372c2ae3e1932ec09a02677af673fe61
parent60302a968a2c788e6d94b70a4d208fae3f8ecb44
cpuconfig: add new x86-64-vX models

https://gitlab.com/x86-psABIs/x86-64-ABI/
https://lists.gnu.org/archive/html/qemu-devel/2021-06/msg01592.html
"
In 2020, AMD, Intel, Red Hat, and SUSE worked together to define
three microarchitecture levels on top of the historical x86-64
baseline:

  * x86-64:    original x86_64 baseline instruction set
  * x86-64-v2: vector instructions up to Streaming SIMD
               Extensions 4.2 (SSE4.2)  and Supplemental
               Streaming SIMD Extensions 3 (SSSE3), the
               POPCNT instruction, and CMPXCHG16B
  * x86-64-v3: vector instructions up to AVX2, MOVBE,
               and additional bit-manipulation instructions.
  * x86-64-v4: vector instructions from some of the
               AVX-512 variants.
"

This patch add new builtin model derivated from qemu64 model,
to be compatible between intel/amd.

mandatory flags from qemu-doc generator:
https://gitlab.com/qemu/qemu/-/blob/master/scripts/cpu-x86-uarch-abi.py

levels = [
    [ # x86-64 baseline
        "cmov",
        "cx8",
        "fpu",
        "fxsr",
        "mmx",
        "syscall",
        "sse",
        "sse2",
    ],
    [ # x86-64-v2
        "cx16",
        "lahf-lm",
        "popcnt",
        "pni",
        "sse4.1",
        "sse4.2",
        "ssse3",
    ],
    [ # x86-64-v3
        "avx",
        "avx2",
        "bmi1",
        "bmi2",
        "f16c",
        "fma",
        "abm",
        "movbe",
"xsave"  #missing from qemu doc currently
    ],
    [ # x86-64-v4
        "avx512f",
        "avx512bw",
        "avx512cd",
        "avx512dq",
        "avx512vl",
    ],
]

x86-64-v1 : I'm skipping it, as it's basicaly qemu64|kvm64 -vme,-cx16 for compat Opteron_G1 from 2004
            so will use it as qemu64|kvm64 is higher are not working on opteron_g1 anyway

x86-64-v2 : Derived from qemu, +popcnt;+pni;+sse4.1;+sse4.2;+ssse3

min intel: Nehalem
min amd : Opteron_G3

x86-64-v2-AES : Derived from qemu, +aes;+popcnt;+pni;+sse4.1;+sse4.2;+ssse3

min intel: Westmere
min amd : Opteron_G3

x86-64-v3 : Derived from qemu64 +aes;+popcnt;+pni;+sse4.1;+sse4.2;+ssse3;+avx;+avx2;+bmi1;+bmi2;+f16c;+fma;+abm;+movbe+xsave

min intel: Haswell
min amd : EPYC_v1

x86-64-v4 : Derived from qemu64 +aes;+popcnt;+pni;+sse4.1;+sse4.2;+ssse3;+avx;+avx2;+bmi1;+bmi2;+f16c;+fma;+abm;+movbe;+xsave;+avx512f;+avx512bw;+avx512cd;+avx512dq;+avx512vl

min intel: Skylake
min amd : EPYC_v4

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
PVE/QemuServer/CPUConfig.pm