56 CMPB internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
57 JE loop_preheader_avx2
58▶ // TODO: for really big clears, use MOVNTDQ, even without AVX2.
59
60loop:
· · ·
86 // For smaller sizes MOVNTDQ may be faster or slower depending on hardware.
87 // For larger sizes it is always faster, even on dual Xeons with 30M cache.
88▶ // TODO take into account actual LLC size. E. g. glibc uses LLC size/2.
89 CMPQ BX, $0x2000000
90 JAE loop_preheader_avx2_huge