Playing around with a Neanderthal benchmark

user.clj

;; Deep Learning Base AMI (Amazon Linux 2) Version 22.0 (ami-0c9d86eac25c03fea)
;; p2.xlarge
;; $ nvidia-smi --list-gpus
;; GPU 0: Tesla K80 (UUID: GPU-3e0d6827-3d8e-5a5f-85d6-114d950f6d44)
;;
;; $ cat /etc/os-release 
;; NAME="Amazon Linux"
;; VERSION="2"
;; ID="amzn"
;; ID_LIKE="centos rhel fedora"
;; VERSION_ID="2"
;; PRETTY_NAME="Amazon Linux 2"
;; ANSI_COLOR="0;33"
;; CPE_NAME="cpe:2.3:o:amazon:amazon_linux:2"
;; HOME_URL="https://amazonlinux.com/"

;; g++ (GCC) 7.3.1 20180712 (Red Hat 7.3.1-6)
;; Linux ip-10-0-1-198.ap-southeast-2.compute.internal 4.14.165-133.209.amzn2.x86_64 #1 SMP Sun Feb 9 00:21:30 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
;; $ java -version
;; openjdk version "1.8.0_242"
;; OpenJDK Runtime Environment (build 1.8.0_242-b08)
;; OpenJDK 64-Bit Server VM (build 25.242-b08, mixed mode)
;; Intel MKL l_mkl_2019.5.281
;; source ~/intel/compilers_and_libraries/linux/bin/compilervars.sh intel64
;; neanderthal at de895b702

;; four cores in /proc/cpuinfo like this:
;;
;; vendor_id	: GenuineIntel
;; cpu family	: 6
;; model		: 79
;; model name	: Intel(R) Xeon(R) CPU E5-2686 v4 @ 2.30GHz
;; stepping	: 1
;; microcode	: 0xb000038
;; cpu MHz		: 2699.892
;; cache size	: 46080 KB

(ns user
  (:require [uncomplicate.commons
             [core :refer [with-release let-release Releaseable release]]]
            [uncomplicate.clojurecuda.core :refer [with-default default-stream synchronize! current-context]]
            [uncomplicate.neanderthal
             [core :refer :all]
             [vect-math :refer [mul! sqrt! div!]]
             [native :refer [native-float]]
             [cuda :as cuda :refer [cuv with-default-engine cuge]]
             [random :refer [rand-normal! rand-uniform!]]]))

(defn matrix-correlation! [a!]
  (with-release [ones (entry! (vctr a! (ncols a!)) 1.0)
                 work (mv a! ones)]
    (let [a-mean (rk! (/ -1.0 (dim ones)) work ones a!)]
      (let-release [large (and (< 512 (mrows a!)) (< 8196 (ncols a!)))
                    sigma (sy a! (mrows a!))]
        (if large
          (mmt! a-mean sigma)
          (mm! 1.0 a-mean (trans a-mean) (view-ge sigma)))
        (sqrt! (copy! (dia sigma) work))
        (with-release [sigma-x*y (rk work)]
          (div! (view-ge sigma) (view-ge sigma-x*y)))
        sigma))))

(defn bench-neanderthal-cpu [m n factory reps]
  (with-release [x (rand-uniform! (ge factory m n))]
    (time
     (dotimes [i reps]
       (matrix-correlation! x)))))

(defn bench-neanderthal-gpu [m n factory reps]
  (with-default
    (cuda/with-engine factory default-stream
      (with-release [x (rand-uniform! (cuge m n))]
        (synchronize!)
        (time
         (do
           (dotimes [i reps]
             (release (matrix-correlation! x)))
           (synchronize!)))))))

(bench-neanderthal-cpu 1000 10000 uncomplicate.neanderthal.native/native-double 10)
; "Elapsed time: 1302.26547 msecs"
(bench-neanderthal-gpu 1000 10000 uncomplicate.neanderthal.cuda/cuda-double 10)
; "Elapsed time: 49.68471 msecs"

;; Correlation[
;;  {{0.55,0.14,0.61,0.63,0.63},
;; 		{0.03,0.66,0.44,0.77,0.64},
;; 		{0.98,0.24,0.71,0.67,0.37},
;; 		{0.32,0.55,0.58,0.16,0.55},
;; 		{0.72,0.86,0.93,0.44,0.72}}] // MatrixForm

;; cupy.corrcoef(cupy.array([0.55,0.14,0.61,0.63,0.63,0.03,0.66,0.44,0.77,0.64,0.98,0.24,0.71,0.67,0.37,0.32,0.55,0.58,0.16,0.55,0.72,0.86,0.93,0.44,0.72]).reshape(5,5).transpose())
;; numpy.corrcoef(numpy.array([0.55,0.14,0.61,0.63,0.63,0.03,0.66,0.44,0.77,0.64,0.98,0.24,0.71,0.67,0.37,0.32,0.55,0.58,0.16,0.55,0.72,0.86,0.93,0.44,0.72]).reshape(5,5).transpose())

;; (def x
;;  (ge uncomplicate.neanderthal.native/native-double 
;; 		[[0.55,0.14,0.61,0.63,0.63]
;; 			[0.03,0.66,0.44,0.77,0.64]
;; 			[0.98,0.24,0.71,0.67,0.37]
;; 			[0.32,0.55,0.58,0.16,0.55]
;; 			[0.72,0.86,0.93,0.44,0.72]]))

;; (matrix-correlation! x)