| /* | |
| * Signed 64-bit integer multiply. | |
| * | |
| * For JIT: op1 in r0/r1, op2 in r2/r3, return in r0/r1 | |
| * | |
| * Consider WXxYZ (r1r0 x r3r2) with a long multiply: | |
| * WX | |
| * x YZ | |
| * -------- | |
| * ZW ZX | |
| * YW YX | |
| * | |
| * The low word of the result holds ZX, the high word holds | |
| * (ZW+YX) + (the high overflow from ZX). YW doesn't matter because | |
| * it doesn't fit in the low 64 bits. | |
| * | |
| * Unlike most ARM math operations, multiply instructions have | |
| * restrictions on using the same register more than once (Rd and Rm | |
| * cannot be the same). | |
| */ | |
| /* mul-long vAA, vBB, vCC */ | |
| mul ip, r2, r1 @ ip<- ZxW | |
| umull r9, r10, r2, r0 @ r9/r10 <- ZxX | |
| mla r2, r0, r3, ip @ r2<- YxX + (ZxW) | |
| mov r0, r9 | |
| add r1, r2, r10 @ r1<- r10 + low(ZxW + (YxX)) | |
| bx lr |