| 1 |
package main |
| 2 |
|
| 3 |
import ( |
| 4 |
. "github.com/mmcloughlin/avo/build" |
| 5 |
. "github.com/mmcloughlin/avo/operand" |
| 6 |
. "github.com/mmcloughlin/avo/reg" |
| 7 |
. "github.com/zeebo/blake3/avo" |
| 8 |
) |
| 9 |
|
| 10 |
func HashP(c Ctx) { |
| 11 |
TEXT("HashP", NOSPLIT, `func( |
| 12 |
left *[32]uint32, |
| 13 |
right *[32]uint32, |
| 14 |
flags uint8, |
| 15 |
key *[8]uint32, |
| 16 |
out *[32]uint32, |
| 17 |
n int, |
| 18 |
)`) |
| 19 |
|
| 20 |
var ( |
| 21 |
left = Mem{Base: Load(Param("left"), GP64())} |
| 22 |
right = Mem{Base: Load(Param("right"), GP64())} |
| 23 |
flags = Load(Param("flags"), GP32()).(GPVirtual) |
| 24 |
key = Mem{Base: Load(Param("key"), GP64())} |
| 25 |
out = Mem{Base: Load(Param("out"), GP64())} |
| 26 |
) |
| 27 |
|
| 28 |
stash := GP64() |
| 29 |
|
| 30 |
{ |
| 31 |
Comment("Allocate local space and align it") |
| 32 |
local := AllocLocal(roundSize + 32) |
| 33 |
LEAQ(local.Offset(31), stash) |
| 34 |
// TODO: avo improvement |
| 35 |
tmp := GP64() |
| 36 |
MOVQ(U64(31), tmp) |
| 37 |
NOTQ(tmp) |
| 38 |
ANDQ(tmp, stash) |
| 39 |
} |
| 40 |
|
| 41 |
alloc := NewAlloc(Mem{Base: stash}) |
| 42 |
defer alloc.Free() |
| 43 |
|
| 44 |
flags_mem := AllocLocal(8) |
| 45 |
|
| 46 |
var ( |
| 47 |
h_vecs []*Value |
| 48 |
vs []*Value |
| 49 |
iv []*Value |
| 50 |
ctr_low *Value |
| 51 |
ctr_hi *Value |
| 52 |
blen_vec *Value |
| 53 |
flags_vec *Value |
| 54 |
) |
| 55 |
|
| 56 |
{ |
| 57 |
Comment("Set up flags value") |
| 58 |
MOVL(flags, flags_mem) |
| 59 |
} |
| 60 |
|
| 61 |
h_vecs = alloc.ValuesWith(8, key) |
| 62 |
iv = alloc.ValuesWith(4, c.IV) |
| 63 |
ctr_low = alloc.ValueFrom(c.Zero) |
| 64 |
ctr_hi = alloc.ValueFrom(c.Zero) |
| 65 |
blen_vec = alloc.ValueFrom(c.BlockLen) |
| 66 |
flags_vec = alloc.ValueWith(flags_mem) |
| 67 |
|
| 68 |
{ |
| 69 |
Comment("Perform the rounds") |
| 70 |
|
| 71 |
vs = []*Value{ |
| 72 |
h_vecs[0], h_vecs[1], h_vecs[2], h_vecs[3], |
| 73 |
h_vecs[4], h_vecs[5], h_vecs[6], h_vecs[7], |
| 74 |
iv[0], iv[1], iv[2], iv[3], |
| 75 |
ctr_low, ctr_hi, blen_vec, flags_vec, |
| 76 |
} |
| 77 |
|
| 78 |
for r := 0; r < 7; r++ { |
| 79 |
Commentf("Round %d", r+1) |
| 80 |
roundP(c, alloc, vs, r, left, right) |
| 81 |
} |
| 82 |
} |
| 83 |
|
| 84 |
{ |
| 85 |
Comment("Finalize") |
| 86 |
finalizeRounds(alloc, vs, h_vecs, nil) |
| 87 |
} |
| 88 |
|
| 89 |
{ |
| 90 |
Comment("Store result into out") |
| 91 |
for i, v := range h_vecs { |
| 92 |
VMOVDQU(v.Consume(), out.Offset(32*i)) |
| 93 |
} |
| 94 |
} |
| 95 |
|
| 96 |
VZEROUPPER() |
| 97 |
RET() |
| 98 |
} |
| 99 |
|
| 100 |
func roundP(c Ctx, alloc *Alloc, vs []*Value, r int, left, right Mem) { |
| 101 |
round(c, alloc, vs, r, func(n int) Mem { |
| 102 |
if n < 8 { |
| 103 |
return left.Offset(n * 32) |
| 104 |
} else { |
| 105 |
return right.Offset((n - 8) * 32) |
| 106 |
} |
| 107 |
}) |
| 108 |
} |