[WIP] Auto-generated wrappers + pylikwid features #10
Conversation
|
Disclaimer: this my first time wrapping a moderately complicated C library (beyond simple function calls etc.) 😄 typedef struct {
uint32_t numHWThreads; /*!< \brief Amount of active HW threads in the system (e.g. in cpuset) */
uint32_t activeHWThreads; /*!< \brief Amount of HW threads in the system and length of \a threadPool */
uint32_t numSockets; /*!< \brief Amount of CPU sockets/packages in the system */
uint32_t numDies; /*!< \brief Amount of CPU dies in the system */
uint32_t numCoresPerSocket; /*!< \brief Amount of physical cores in one CPU socket/package */
uint32_t numThreadsPerCore; /*!< \brief Amount of HW threads in one physical CPU core */
uint32_t numCacheLevels; /*!< \brief Amount of caches for each HW thread and length of \a cacheLevels */
HWThread* threadPool; /*!< \brief List of all HW thread descriptions */
CacheLevel* cacheLevels; /*!< \brief List of all caches in the hierarchy */
struct treeNode* topologyTree; /*!< \brief Anchor for a tree structure describing the system topology */
} CpuTopology;Julia struct: struct CpuTopology
numHWThreads::UInt32
activeHWThreads::UInt32
numSockets::UInt32
numDies::UInt32
numCoresPerSocket::UInt32
numThreadsPerCore::UInt32
numCacheLevels::UInt32
threadPool::Ptr{HWThread}
cacheLevels::Ptr{CacheLevel}
topologyTree::Ptr{treeNode}
endShould be correct, no? However, we get incorrect numbers when julia> LIKWID.init_topology();
julia> LIKWID.get_cpu_topology() # essentially ccall + unsafe_load
Dict{String, Any} with 7 entries:
"numSockets" => 2
"numHWThreads" => 40
"numCoresPerSocket" => 1
"numCacheLevels" => 20422880
"activeHWThreads" => 40
"numDies" => 20
"numThreadsPerCore" => 3Note that the python pendant in pylikwid gets the correct numbers: >>> d["numSockets"]
2
>>> d["numHWThreads"]
40
>>> d["numCoresPerSocket"]
20
>>> d["numCacheLevels"]
3
>>> d["activeHWThreads"]
40
>>> d["numThreadsPerCore"]
1What's going on? |
|
Note for future self, current output: julia> LIKWID.get_cpu_info()
LIKWID.CpuInfo
├ family: 6
├ model: 85
├ stepping: 4
├ vendor: 0
├ part: 0
├ clock: 0
├ turbo: true
├ osname: Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz
├ name: Intel Skylake SP processor
├ short_name: skylakeX
├ features: FP ACPI MMX SSE SSE2 HTT TM RDTSCP MONITOR VMX EIST TM2 SSSE FMA SSE4.1 SSE4.2 AES AVX RDRAND HLE AVX2 RTM AVX512 RDSEED SSE3
├ isIntel: true
├ architecture: x86_64
├ supportUncore: false
├ supportClientmem: false
├ featureFlags: 4328456191
├ perf_version: 4
├ perf_num_ctr: 8
├ perf_width_ctr: 48
└ perf_num_fixed_ctr: 3
julia> LIKWID.get_cpu_topology()
LIKWID.CpuTopology
├ numHWThreads: 40
├ activeHWThreads: 40
├ numSockets: 2
├ numCoresPerSocket: 20
├ numThreadsPerCore: 1
├ numCacheLevels: 3
├ threadPool: ... (40 elements)
└ cacheLevels: ... (3 elements)
julia> LIKWID.get_numa_topology().nodes[2]
LIKWID.NumaNode
├ id: 1
├ totalMemory: 94.48 GB
├ freeMemory: 85.26 GB
├ numberOfProcessors: 20
├ processors: [20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39]
├ numberOfDistances: 2
└ distances: [21, 10]
|
Codecov Report
@@ Coverage Diff @@
## main #10 +/- ##
==========================================
- Coverage 50.00% 44.37% -5.63%
==========================================
Files 2 19 +17
Lines 22 1386 +1364
==========================================
+ Hits 11 615 +604
- Misses 11 771 +760
Continue to review full report at Codecov.
|
|
@vchuravy I'd like to merge this to then set up documentation (Documenter.jl) directly on master. Any objections? |
|
Given that this PR is a strict (significant) improvement, I'll take the liberty to merge so that I can continue to work on things. Let's dicuss / improve (if necessary revert) things in separate issues / PRs. (I won't tag a new release soon). |
In this WIP PR I try to work towards the goal of reaching feature parity with pylikwid (#5).
pylikwid repository / README
likwid.h(LibLikwid.jl).Closes #5
Closes #2