On Wed, 07 Nov 2007 00:39:25 +0900 (JST), Atsushi Nemoto <anemo@xxxxxxxxxxxxx> wrote: > + MFC0 k0, CP0_EPC > + ori k0, 0x1f /* 32 byte rollback region */ > + xori k0, 0x1f > + PTR_LA k1, r4k_wait Well, this part should be like this, for better pipelining. MFC0 k0, CP0_EPC PTR_LA k1, r4k_wait ori k0, 0x1f /* 32 byte rollback region */ xori k0, 0x1f > + bne k0, k1, 9f > + MTC0 k0, CP0_EPC > +9: And if we could assume branch-likely, this can be: .set noreorder beql k0, k1, 9f MTC0 k0, CP0_EPC 9: But not sure if it really have points. > .align 5 > +BUILD_ROLLBACK_PROLOGUE handle_int > NESTED(handle_int, PT_SIZE, sp) And one more question: should we put one more ".align 5" just befor handle_int for CPUs do not need the rollback? --- Atsushi Nemoto
