On Wed, Oct 17, 2012 at 05:16:11PM +1000, Richard Henderson wrote: > On 2012-10-17 16:05, Aurelien Jarno wrote: > >>>> > >> + target_long temp; > >>>> > >> + > >>>> > >> + imm = (ctx->opcode >> 16) & 0xFF; > >>>> > >> + temp = imm; > >>>> > >> + temp = (temp << 8) | temp; > >>>> > >> + temp = (temp << 16) | temp; > >>>> > >> + temp = (temp << 32) | temp; > >>>> > >> + tcg_gen_movi_tl(cpu_gpr[ret], temp); > >>>> > >> + break; > >>>> > >> + } > >>> > > > >>> > > This hasn't been fixed, and thus is still wrong. > >>> > > > >> > > >> > Thank you for check this again. > >> > May you give me more comment about this please? I'm not sure what > >> > should do here. > >> > > > The instruction is defined as: > > > > | Replicate a immediate byte into all elements of an eight byte vector. > > | > > | Description: rd ← immediate || immediate || immediate || immediate || > > immediate || > > | immediate || immediate || immediate > > > > In your code, I only see the value replicated 4 times, not 8 times. > > There are 3 doublings: 2**3 = 8. Look again, it's right. >
Oops, you are correct, I read it to fast and thought it was implemented the same way as the other REPL instructions, that is the immediate are orred one at a time. So Jia, please keep the original version of this code. -- Aurelien Jarno GPG: 1024D/F1BCDB73 aurel...@aurel32.net http://www.aurel32.net