diff options
Diffstat (limited to 'arch/sparc64/lib/csum_copy.S')
| -rw-r--r-- | arch/sparc64/lib/csum_copy.S | 308 |
1 files changed, 308 insertions, 0 deletions
diff --git a/arch/sparc64/lib/csum_copy.S b/arch/sparc64/lib/csum_copy.S new file mode 100644 index 00000000000..71af4883906 --- /dev/null +++ b/arch/sparc64/lib/csum_copy.S @@ -0,0 +1,308 @@ +/* csum_copy.S: Checksum+copy code for sparc64 + * + * Copyright (C) 2005 David S. Miller <davem@davemloft.net> + */ + +#ifdef __KERNEL__ +#define GLOBAL_SPARE %g7 +#else +#define GLOBAL_SPARE %g5 +#endif + +#ifndef EX_LD +#define EX_LD(x) x +#endif + +#ifndef EX_ST +#define EX_ST(x) x +#endif + +#ifndef EX_RETVAL +#define EX_RETVAL(x) x +#endif + +#ifndef LOAD +#define LOAD(type,addr,dest) type [addr], dest +#endif + +#ifndef STORE +#define STORE(type,src,addr) type src, [addr] +#endif + +#ifndef FUNC_NAME +#define FUNC_NAME csum_partial_copy_nocheck +#endif + + .register %g2, #scratch + .register %g3, #scratch + + .text + +90: + /* We checked for zero length already, so there must be + * at least one byte. + */ + be,pt %icc, 1f + nop + EX_LD(LOAD(ldub, %o0 + 0x00, %o4)) + add %o0, 1, %o0 + sub %o2, 1, %o2 + EX_ST(STORE(stb, %o4, %o1 + 0x00)) + add %o1, 1, %o1 +1: andcc %o0, 0x2, %g0 + be,pn %icc, 80f + cmp %o2, 2 + blu,pn %icc, 60f + nop + EX_LD(LOAD(lduh, %o0 + 0x00, %o5)) + add %o0, 2, %o0 + sub %o2, 2, %o2 + EX_ST(STORE(sth, %o5, %o1 + 0x00)) + add %o1, 2, %o1 + ba,pt %xcc, 80f + add %o5, %o4, %o4 + + .globl FUNC_NAME +FUNC_NAME: /* %o0=src, %o1=dst, %o2=len, %o3=sum */ + LOAD(prefetch, %o0 + 0x000, #n_reads) + xor %o0, %o1, %g1 + clr %o4 + andcc %g1, 0x3, %g0 + bne,pn %icc, 95f + LOAD(prefetch, %o0 + 0x040, #n_reads) + + brz,pn %o2, 70f + andcc %o0, 0x3, %g0 + + /* We "remember" whether the lowest bit in the address + * was set in GLOBAL_SPARE. Because if it is, we have to swap + * upper and lower 8 bit fields of the sum we calculate. + */ + bne,pn %icc, 90b + andcc %o0, 0x1, GLOBAL_SPARE + +80: + LOAD(prefetch, %o0 + 0x080, #n_reads) + andncc %o2, 0x3f, %g3 + + LOAD(prefetch, %o0 + 0x0c0, #n_reads) + sub %o2, %g3, %o2 + brz,pn %g3, 2f + LOAD(prefetch, %o0 + 0x100, #n_reads) + + /* So that we don't need to use the non-pairing + * add-with-carry instructions we accumulate 32-bit + * values into a 64-bit register. At the end of the + * loop we fold it down to 32-bits and so on. + */ + ba,pt %xcc, 1f + LOAD(prefetch, %o0 + 0x140, #n_reads) + + .align 32 +1: EX_LD(LOAD(lduw, %o0 + 0x00, %o5)) + EX_LD(LOAD(lduw, %o0 + 0x04, %g1)) + EX_LD(LOAD(lduw, %o0 + 0x08, %g2)) + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x00)) + EX_LD(LOAD(lduw, %o0 + 0x0c, %o5)) + add %o4, %g1, %o4 + EX_ST(STORE(stw, %g1, %o1 + 0x04)) + EX_LD(LOAD(lduw, %o0 + 0x10, %g1)) + add %o4, %g2, %o4 + EX_ST(STORE(stw, %g2, %o1 + 0x08)) + EX_LD(LOAD(lduw, %o0 + 0x14, %g2)) + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x0c)) + EX_LD(LOAD(lduw, %o0 + 0x18, %o5)) + add %o4, %g1, %o4 + EX_ST(STORE(stw, %g1, %o1 + 0x10)) + EX_LD(LOAD(lduw, %o0 + 0x1c, %g1)) + add %o4, %g2, %o4 + EX_ST(STORE(stw, %g2, %o1 + 0x14)) + EX_LD(LOAD(lduw, %o0 + 0x20, %g2)) + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x18)) + EX_LD(LOAD(lduw, %o0 + 0x24, %o5)) + add %o4, %g1, %o4 + EX_ST(STORE(stw, %g1, %o1 + 0x1c)) + EX_LD(LOAD(lduw, %o0 + 0x28, %g1)) + add %o4, %g2, %o4 + EX_ST(STORE(stw, %g2, %o1 + 0x20)) + EX_LD(LOAD(lduw, %o0 + 0x2c, %g2)) + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x24)) + EX_LD(LOAD(lduw, %o0 + 0x30, %o5)) + add %o4, %g1, %o4 + EX_ST(STORE(stw, %g1, %o1 + 0x28)) + EX_LD(LOAD(lduw, %o0 + 0x34, %g1)) + add %o4, %g2, %o4 + EX_ST(STORE(stw, %g2, %o1 + 0x2c)) + EX_LD(LOAD(lduw, %o0 + 0x38, %g2)) + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x30)) + EX_LD(LOAD(lduw, %o0 + 0x3c, %o5)) + add %o4, %g1, %o4 + EX_ST(STORE(stw, %g1, %o1 + 0x34)) + LOAD(prefetch, %o0 + 0x180, #n_reads) + add %o4, %g2, %o4 + EX_ST(STORE(stw, %g2, %o1 + 0x38)) + subcc %g3, 0x40, %g3 + add %o0, 0x40, %o0 + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x3c)) + bne,pt %icc, 1b + add %o1, 0x40, %o1 + +2: and %o2, 0x3c, %g3 + brz,pn %g3, 2f + sub %o2, %g3, %o2 +1: EX_LD(LOAD(lduw, %o0 + 0x00, %o5)) + subcc %g3, 0x4, %g3 + add %o0, 0x4, %o0 + add %o4, %o5, %o4 + EX_ST(STORE(stw, %o5, %o1 + 0x00)) + bne,pt %icc, 1b + add %o1, 0x4, %o1 + +2: + /* fold 64-->32 */ + srlx %o4, 32, %o5 + srl %o4, 0, %o4 + add %o4, %o5, %o4 + srlx %o4, 32, %o5 + srl %o4, 0, %o4 + add %o4, %o5, %o4 + + /* fold 32-->16 */ + sethi %hi(0xffff0000), %g1 + srl %o4, 16, %o5 + andn %o4, %g1, %g2 + add %o5, %g2, %o4 + srl %o4, 16, %o5 + andn %o4, %g1, %g2 + add %o5, %g2, %o4 + +60: + /* %o4 has the 16-bit sum we have calculated so-far. */ + cmp %o2, 2 + blu,pt %icc, 1f + nop + EX_LD(LOAD(lduh, %o0 + 0x00, %o5)) + sub %o2, 2, %o2 + add %o0, 2, %o0 + add %o4, %o5, %o4 + EX_ST(STORE(sth, %o5, %o1 + 0x00)) + add %o1, 0x2, %o1 +1: brz,pt %o2, 1f + nop + EX_LD(LOAD(ldub, %o0 + 0x00, %o5)) + sub %o2, 1, %o2 + add %o0, 1, %o0 + EX_ST(STORE(stb, %o5, %o1 + 0x00)) + sllx %o5, 8, %o5 + add %o1, 1, %o1 + add %o4, %o5, %o4 +1: + /* fold 32-->16 */ + sethi %hi(0xffff0000), %g1 + srl %o4, 16, %o5 + andn %o4, %g1, %g2 + add %o5, %g2, %o4 + srl %o4, 16, %o5 + andn %o4, %g1, %g2 + add %o5, %g2, %o4 + +1: brz,pt GLOBAL_SPARE, 1f + nop + + /* We started with an odd byte, byte-swap the result. */ + srl %o4, 8, %o5 + and %o4, 0xff, %g1 + sll %g1, 8, %g1 + or %o5, %g1, %o4 + +1: add %o3, %o4, %o3 + +70: + retl + mov %o3, %o0 + +95: mov 0, GLOBAL_SPARE + brlez,pn %o2, 4f + andcc %o0, 1, %o5 + be,a,pt %icc, 1f + srl %o2, 1, %g1 + sub %o2, 1, %o2 + EX_LD(LOAD(ldub, %o0, GLOBAL_SPARE)) + add %o0, 1, %o0 + EX_ST(STORE(stb, GLOBAL_SPARE, %o1)) + srl %o2, 1, %g1 + add %o1, 1, %o1 +1: brz,a,pn %g1, 3f + andcc %o2, 1, %g0 + andcc %o0, 2, %g0 + be,a,pt %icc, 1f + srl %g1, 1, %g1 + EX_LD(LOAD(lduh, %o0, %o4)) + sub %o2, 2, %o2 + srl %o4, 8, %g2 + sub %g1, 1, %g1 + EX_ST(STORE(stb, %g2, %o1)) + add %o4, GLOBAL_SPARE, GLOBAL_SPARE + EX_ST(STORE(stb, %o4, %o1 + 1)) + add %o0, 2, %o0 + srl %g1, 1, %g1 + add %o1, 2, %o1 +1: brz,a,pn %g1, 2f + andcc %o2, 2, %g0 + EX_LD(LOAD(lduw, %o0, %o4)) +5: srl %o4, 24, %g2 + srl %o4, 16, %g3 + EX_ST(STORE(stb, %g2, %o1)) + srl %o4, 8, %g2 + EX_ST(STORE(stb, %g3, %o1 + 1)) + add %o0, 4, %o0 + EX_ST(STORE(stb, %g2, %o1 + 2)) + addcc %o4, GLOBAL_SPARE, GLOBAL_SPARE + EX_ST(STORE(stb, %o4, %o1 + 3)) + addc GLOBAL_SPARE, %g0, GLOBAL_SPARE + add %o1, 4, %o1 + subcc %g1, 1, %g1 + bne,a,pt %icc, 5b + EX_LD(LOAD(lduw, %o0, %o4)) + sll GLOBAL_SPARE, 16, %g2 + srl GLOBAL_SPARE, 16, GLOBAL_SPARE + srl %g2, 16, %g2 + andcc %o2, 2, %g0 + add %g2, GLOBAL_SPARE, GLOBAL_SPARE +2: be,a,pt %icc, 3f + andcc %o2, 1, %g0 + EX_LD(LOAD(lduh, %o0, %o4)) + andcc %o2, 1, %g0 + srl %o4, 8, %g2 + add %o0, 2, %o0 + EX_ST(STORE(stb, %g2, %o1)) + add GLOBAL_SPARE, %o4, GLOBAL_SPARE + EX_ST(STORE(stb, %o4, %o1 + 1)) + add %o1, 2, %o1 +3: be,a,pt %icc, 1f + sll GLOBAL_SPARE, 16, %o4 + EX_LD(LOAD(ldub, %o0, %g2)) + sll %g2, 8, %o4 + EX_ST(STORE(stb, %g2, %o1)) + add GLOBAL_SPARE, %o4, GLOBAL_SPARE + sll GLOBAL_SPARE, 16, %o4 +1: addcc %o4, GLOBAL_SPARE, GLOBAL_SPARE + srl GLOBAL_SPARE, 16, %o4 + addc %g0, %o4, GLOBAL_SPARE + brz,pt %o5, 4f + srl GLOBAL_SPARE, 8, %o4 + and GLOBAL_SPARE, 0xff, %g2 + and %o4, 0xff, %o4 + sll %g2, 8, %g2 + or %g2, %o4, GLOBAL_SPARE +4: addcc %o3, GLOBAL_SPARE, %o3 + addc %g0, %o3, %o0 + retl + srl %o0, 0, %o0 + .size FUNC_NAME, .-FUNC_NAME |