Optimization deferred from #1648.
See
|
* TODO: this saves/restores v20-v30, but the body only touches |
|
* v23 (vdata3), v24 (vtmp3) and v27-v30 (vresult1..4) in the |
|
* callee-saved range. v20, v21, v22, v25, v26 are spilled and |
|
* reloaded but never used -- 5 dead stxvx/lxvx pairs plus 80 |
|
* bytes of frame. Trim the save set (and frame size) to the |
|
* registers actually clobbered. Separate change from the doc |
|
* pass, since it alters emitted code. |
|
*/ |
|
li 6, 128 |
|
li 7, 144 |
|
li 8, 160 |
Optimization deferred from #1648.
See
mlkem-native/dev/ppc64le/src/poly_tomont_ppc_asm.S
Lines 110 to 120 in 494dfe6