diff options
| author | Vineet Gupta <vgupta@synopsys.com> | 2013-10-07 14:40:08 +0200 |
|---|---|---|
| committer | Vineet Gupta <vgupta@synopsys.com> | 2015-06-22 10:36:56 +0200 |
| commit | 8922bc3058abbe5deaf887147e26531750ce7513 (patch) | |
| tree | 647f15f01077a23afd047e3d01b125a93e2ac9b1 /arch/arc/lib | |
| parent | ARCv2: optimised string/mem lib routines (diff) | |
| download | linux-8922bc3058abbe5deaf887147e26531750ce7513.tar.xz linux-8922bc3058abbe5deaf887147e26531750ce7513.zip | |
ARCv2: Adhere to Zero Delay loop restriction
Branch insn can't be scheduled as last insn of Zero Overhead loop
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Diffstat (limited to 'arch/arc/lib')
| -rw-r--r-- | arch/arc/lib/memcmp.S | 30 |
1 files changed, 29 insertions, 1 deletions
diff --git a/arch/arc/lib/memcmp.S b/arch/arc/lib/memcmp.S index 978bf8314dfb..a4015e7d9ab7 100644 --- a/arch/arc/lib/memcmp.S +++ b/arch/arc/lib/memcmp.S @@ -24,14 +24,32 @@ ENTRY(memcmp) ld r4,[r0,0] ld r5,[r1,0] lsr.f lp_count,r3,3 +#ifdef CONFIG_ISA_ARCV2 + /* In ARCv2 a branch can't be the last instruction in a zero overhead + * loop. + * So we move the branch to the start of the loop, duplicate it + * after the end, and set up r12 so that the branch isn't taken + * initially. + */ + mov_s r12,WORD2 + lpne .Loop_end + brne WORD2,r12,.Lodd + ld WORD2,[r0,4] +#else lpne .Loop_end ld_s WORD2,[r0,4] +#endif ld_s r12,[r1,4] brne r4,r5,.Leven ld.a r4,[r0,8] ld.a r5,[r1,8] +#ifdef CONFIG_ISA_ARCV2 +.Loop_end: + brne WORD2,r12,.Lodd +#else brne WORD2,r12,.Lodd .Loop_end: +#endif asl_s SHIFT,SHIFT,3 bhs_s .Last_cmp brne r4,r5,.Leven @@ -89,7 +107,6 @@ ENTRY(memcmp) bset.cs r0,r0,31 .Lodd: cmp_s WORD2,r12 - mov_s r0,1 j_s.d [blink] bset.cs r0,r0,31 @@ -100,14 +117,25 @@ ENTRY(memcmp) ldb r4,[r0,0] ldb r5,[r1,0] lsr.f lp_count,r3 +#ifdef CONFIG_ISA_ARCV2 + mov r12,r3 lpne .Lbyte_end + brne r3,r12,.Lbyte_odd +#else + lpne .Lbyte_end +#endif ldb_s r3,[r0,1] ldb r12,[r1,1] brne r4,r5,.Lbyte_even ldb.a r4,[r0,2] ldb.a r5,[r1,2] +#ifdef CONFIG_ISA_ARCV2 +.Lbyte_end: + brne r3,r12,.Lbyte_odd +#else brne r3,r12,.Lbyte_odd .Lbyte_end: +#endif bcc .Lbyte_even brne r4,r5,.Lbyte_even ldb_s r3,[r0,1] |