Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

powerpc/perf: Optimize find_alternatives_list() using binary search

This patch improves the performance of event alternative lookup by
replacing the previous linear search with a more efficient binary
search. This change reduces the time complexity for the search process
from O(n) to O(log(n)). A pre-sorted table of event values and their
corresponding indices has been introduced to expedite the search
process.

Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
[mpe: Call the array "presort*ed*_event_table", minor formatting]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://msgid.link/20231013175714.2142775-1-visitorckw@gmail.com

authored by

Kuan-Wei Chiu and committed by
Michael Ellerman
e08c43e6 d45c4b48

+28 -16
+28 -16
arch/powerpc/perf/power6-pmu.c
··· 335 335 { 0x3000fe, 0x400056 }, /* PM_DATA_FROM_L3MISS */ 336 336 }; 337 337 338 - /* 339 - * This could be made more efficient with a binary search on 340 - * a presorted list, if necessary 341 - */ 342 338 static int find_alternatives_list(u64 event) 343 339 { 344 - int i, j; 345 - unsigned int alt; 340 + const unsigned int presorted_event_table[] = { 341 + 0x0130e8, 0x080080, 0x080088, 0x10000a, 0x10000b, 0x10000d, 0x10000e, 342 + 0x100010, 0x10001a, 0x100026, 0x100054, 0x100056, 0x1000f0, 0x1000f8, 343 + 0x1000fc, 0x200008, 0x20000e, 0x200010, 0x200012, 0x200054, 0x2000f0, 344 + 0x2000f2, 0x2000f4, 0x2000f5, 0x2000f6, 0x2000f8, 0x2000fc, 0x2000fe, 345 + 0x2d0030, 0x30000a, 0x30000c, 0x300010, 0x300012, 0x30001a, 0x300056, 346 + 0x3000f0, 0x3000f2, 0x3000f6, 0x3000f8, 0x3000fc, 0x3000fe, 0x400006, 347 + 0x400007, 0x40000a, 0x40000e, 0x400010, 0x400018, 0x400056, 0x4000f0, 348 + 0x4000f8, 0x600005 349 + }; 350 + const unsigned int event_index_table[] = { 351 + 0, 1, 2, 3, 4, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 12, 14, 352 + 7, 15, 2, 9, 16, 3, 4, 0, 17, 10, 18, 19, 20, 1, 17, 15, 19, 353 + 18, 2, 16, 21, 8, 0, 22, 13, 14, 11, 21, 5, 20, 22, 1, 6, 3 354 + }; 355 + int hi = ARRAY_SIZE(presorted_event_table) - 1; 356 + int lo = 0; 346 357 347 - for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) { 348 - if (event < event_alternatives[i][0]) 349 - return -1; 350 - for (j = 0; j < MAX_ALT; ++j) { 351 - alt = event_alternatives[i][j]; 352 - if (!alt || event < alt) 353 - break; 354 - if (event == alt) 355 - return i; 356 - } 358 + while (lo <= hi) { 359 + int mid = lo + (hi - lo) / 2; 360 + unsigned int alt = presorted_event_table[mid]; 361 + 362 + if (alt < event) 363 + lo = mid + 1; 364 + else if (alt > event) 365 + hi = mid - 1; 366 + else 367 + return event_index_table[mid]; 357 368 } 369 + 358 370 return -1; 359 371 } 360 372