Optimize 64 bit bswap

Use rldimi instead of rldicr/or pair, saves us one instruction.
Suggested by Hollis Blanchard.

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@5404 c046a42c-6fe2-441c-8c8c-71466251a162
master
malc 2008-10-02 20:02:31 +00:00
parent bdadc0b5b6
commit 3ee1b8550c
1 changed files with 5 additions and 5 deletions

View File

@ -303,6 +303,7 @@ static int tcg_target_const_match (tcg_target_long val,
#define RLDICL XO30( 0)
#define RLDICR XO30( 1)
#define RLDIMI XO30( 3)
#define BCLR XO19( 16)
#define BCCTR XO19(528)
@ -691,11 +692,10 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)
break;
case 3:
if (bswap) {
tcg_out32 (s, LWBRX | RT (0) | RB (r0));
tcg_out32 (s, ADDI | RT (r1) | RA (r0) | 4);
tcg_out32 (s, LWBRX | RT (data_reg) | RB (r1));
tcg_out_rld (s, RLDICR, data_reg, data_reg, 32, 31);
tcg_out32 (s, OR | SAB (0, data_reg, data_reg));
tcg_out_movi32 (s, 0, 4);
tcg_out32 (s, LWBRX | RT (data_reg) | RB (r0));
tcg_out32 (s, LWBRX | RT ( r1) | RA (r0));
tcg_out_rld (s, RLDIMI, data_reg, r1, 32, 0);
}
else tcg_out32 (s, LD | RT (data_reg) | RA (r0));
break;