hypercore-kb / opencog-kb / language-learning / mvp / benchmark-gpu-proper.scm

;;
;; benchmark-gpu-proper.scm - Proper GPU benchmark using sensory API
;;
;; Tests atomese-simd performance fixes:
;; 1. Binary caching (2.2x program load speedup)
;; 2. Dispatch thread build (eliminates per-session overhead)
;;

(use-modules (opencog) (opencog exec))
(use-modules (opencog sensory) (opencog opencl))
(use-modules (ice-9 format))

;; Helper: time execution in milliseconds
(define (time-ms thunk)
  (let* ((start (get-internal-real-time))
         (result (thunk))
         (end (get-internal-real-time)))
    (values result (inexact->exact (round (* 1000.0 (/ (- end start) internal-time-units-per-second)))))))

(display "\n")
(display "═══════════════════════════════════════════════════════════════════\n")
(display "GPU Benchmark (proper sensory API)\n")
(display "═══════════════════════════════════════════════════════════════════\n\n")

(define *kernel-path* "/storage/projects/code/hypercore-kb/opencog-kb/language-learning/mvp/pmfg-kernel.cl")

;; ─────────────────────────────────────────────────────────────────────
;; Test 1: OpenclNode creation + open
;; ─────────────────────────────────────────────────────────────────────

(display "TEST 1: OpenclNode creation and open\n")
(display "─────────────────────────────────────────────────────────────────────\n\n")

(define gpu-node #f)
(call-with-values
  (lambda ()
    (time-ms (lambda ()
      (set! gpu-node (OpenclNode (string-append "opencl://:" *kernel-path*)))
      (cog-execute! (SetValue gpu-node (Predicate "*-open-*") (Type 'FloatValue))))))
  (lambda (result ms)
    (format #t "  OpenclNode creation + open: ~a ms~%" ms)))

(format #t "  Node: ~a~%~%" (cog-name gpu-node))

;; ─────────────────────────────────────────────────────────────────────
;; Test 2: Kernel execution (dispatch thread fix)
;; ─────────────────────────────────────────────────────────────────────

(display "TEST 2: Kernel execution (dispatch thread fix)\n")
(display "─────────────────────────────────────────────────────────────────────\n\n")

(display "  BEFORE FIX: First kernel takes 16+ seconds (OpenCL thread init)\n")
(display "  AFTER FIX: First kernel should take <1 second\n\n")

;; First kernel execution
(define result1 #f)
(call-with-values
  (lambda ()
    (time-ms (lambda ()
      (cog-execute!
        (SetValue gpu-node (Predicate "*-write-*")
          (Section (Item "test_simple")
                   (ConnectorSeq (Number 1 2 3 4) (Number 0 0 0 0)))))
      (set! result1 (cog-execute! (ValueOf gpu-node (Predicate "*-read-*")))))))
  (lambda (result ms)
    (format #t "  First kernel execution: ~a ms~%" ms)
    (if (> ms 5000)
        (display "  ⚠ SLOW! Dispatch thread fix may not be installed.\n")
        (display "  ✓ Fast - dispatch thread fix is working!\n"))))

(format #t "  Result (x*2+42 for 1,2,3,4): ~a~%~%" result1)

;; Second execution
(define result2 #f)
(call-with-values
  (lambda ()
    (time-ms (lambda ()
      (cog-execute!
        (SetValue gpu-node (Predicate "*-write-*")
          (Section (Item "test_simple")
                   (ConnectorSeq (Number 10 20 30 40) (Number 0 0 0 0)))))
      (set! result2 (cog-execute! (ValueOf gpu-node (Predicate "*-read-*")))))))
  (lambda (result ms)
    (format #t "  Second kernel execution: ~a ms~%" ms)))

(format #t "  Result (x*2+42 for 10,20,30,40): ~a~%~%" result2)

;; Third execution
(define result3 #f)
(call-with-values
  (lambda ()
    (time-ms (lambda ()
      (cog-execute!
        (SetValue gpu-node (Predicate "*-write-*")
          (Section (Item "test_simple")
                   (ConnectorSeq (Number 100 200 300 400) (Number 0 0 0 0)))))
      (set! result3 (cog-execute! (ValueOf gpu-node (Predicate "*-read-*")))))))
  (lambda (result ms)
    (format #t "  Third kernel execution: ~a ms~%~%" ms)))

(format #t "  Result: ~a~%~%" result3)

;; ─────────────────────────────────────────────────────────────────────
;; Test 3: Rapid execution (10 iterations)
;; ─────────────────────────────────────────────────────────────────────

(display "TEST 3: Rapid execution (10 iterations)\n")
(display "─────────────────────────────────────────────────────────────────────\n\n")

(define total-time 0)
(do ((i 0 (+ i 1)))
    ((= i 10))
  (call-with-values
    (lambda ()
      (time-ms (lambda ()
        (cog-execute!
          (SetValue gpu-node (Predicate "*-write-*")
            (Section (Item "test_simple")
                     (ConnectorSeq (Number i (+ i 1) (+ i 2) (+ i 3))
                                   (Number 0 0 0 0)))))
        (cog-execute! (ValueOf gpu-node (Predicate "*-read-*"))))))
    (lambda (result ms)
      (set! total-time (+ total-time ms))
      (format #t "  Iteration ~a: ~a ms~%" i ms))))

(format #t "~%  Total: ~a ms, Average: ~a ms~%~%"
        total-time (/ total-time 10.0))

;; ─────────────────────────────────────────────────────────────────────
;; Summary
;; ─────────────────────────────────────────────────────────────────────

(display "═══════════════════════════════════════════════════════════════════\n")
(display "SUMMARY\n")
(display "═══════════════════════════════════════════════════════════════════\n\n")

(display "Dispatch thread fix status:\n")
(display "  • If first execution was <1000ms: Fix is working!\n")
(display "  • If first execution was 16+ seconds: Fix NOT installed\n\n")

(display "Expected results with fixes:\n")
(display "  • test_simple(1,2,3,4) = (OpenclFloatValue 44 46 48 50)\n")
(display "  • test_simple(10,20,30,40) = (OpenclFloatValue 62 82 102 122)\n")
(display "  • test_simple(100,200,300,400) = (OpenclFloatValue 242 442 642 842)\n\n")