https://sourceware.org/bugzilla/show_bug.cgi?id=32886
--- Comment #2 from Bruno Haible <bruno at clisp dot org> --- The issue is also visible in the gprofng-gui documentation https://www.gnu.org/software/gprofng-gui/manual/gprofng.html#The-Disassembly-View : 0. 0. [53] 401d6f: pxor %xmm1,%xmm1 0.110 0.110 [55] 401d73: movsd (%rdx,%rax,8),%xmm0 1.921 1.921 [55] 401d78: mulsd (%r9,%rax,8),%xmm0 2.282 2.282 [55] 401d7e: addsd %xmm0,%xmm1 ## 4.613 4.613 [54] 401d82: add $0x1,%rax It's obvious that adding 1 to %rax is a low-cost instruction, and the instructions that are heavy CPU consumers are the mulsd and addsd instructions (which multiply and add floating-point numbers). -- You are receiving this mail because: You are on the CC list for the bug.
