JCC at 64 byte boundary

global _start

section .text

; JCC erratum mitigation should in theory disable the DSB when N is in {0, 1, 2, 30, 31} mod 32 (cmp and jg are macrofused and take 5 bytes).
; ...except when N == 0 mod 64 - DSB+LSD are much more active (running on core i7-10710U, microcode 0xca)

; NASM snippet:
;   nasm -f elf64 -o test.o test.S
;   ld -o test test.o
_start:
  ; Replace N with 64 and see DSB+LSD unexpectedly (?) kick in
  db N-(loopend-loopstart) dup(0x90)
loopstart:
  mov ebp, 1
  mov r15d, 2
  cmp r15d, ebp
loopend:
  jg loopstart