On Fri, 23 Oct 2026 22:38:10 GMT, Vladimir Kozlov <[email protected]> wrote:
>> Larry has updated the pull request incrementally with one additional commit >> since the last revision: >> >> 8394259: Refine comments > > src/hotspot/cpu/x86/macroAssembler_x86.cpp line 8493: > >> 8491: // Fold 128 -> 96 -> 64 bits with K_128_64. >> 8492: movdqu(xmm2, >> ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1); >> // [K_128_64_low, K_128_64_high] >> 8493: movdqu(xmm0, xmm1); > > Why you use `movdqu` instead of `movdqa` for registers move in AVX part of > code? I chose 'movdqu' to mirror the existing `kernel_crc32_avx512`, which also uses `movdqu` for register moves such as `movdqu(xmm1, xmm7)` and `movdqu(xmm2, xmm7)`, so the two implementations stay easy to compare. In the updated code I have changed to use `movdqa` for register moves. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/33342#discussion_r4377519052
