hash code cleanup - copy seed to local and use array indexing. 1.483 ms for 720p vs 2.225 ms for o…

hash code cleanup - copy seed to local and use array indexing.  1.483 ms for 720p vs 2.225 ms for old code.
BUG=none
TEST=none
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=131901

[fbarchard, 2012-04-12 01:36:20]

copy C version of djb2 from libyuv

[jzern, 2012-04-12 02:06:11]

lgtm

[fbarchard, 2012-04-12 02:07:05]

On 2012/04/12 01:36:20, fbarchard wrote:
> copy C version of djb2 from libyuv

Original code generates this
  00000000: 55                 push        ebp
  00000001: 8B EC              mov         ebp,esp
  00000003: 83 7D 0C 00        cmp         dword ptr [ebp+0Ch],0
  00000007: 7E 17              jle         00000020
  00000009: 8B 45 08           mov         eax,dword ptr [ebp+8]
  0000000C: 8B 4D 10           mov         ecx,dword ptr [ebp+10h]
  0000000F: 0F B6 10           movzx       edx,byte ptr [eax]
  00000012: 6B C9 21           imul        ecx,ecx,21h
  00000015: 03 CA              add         ecx,edx
  00000017: 40                 inc         eax
  00000018: FF 4D 0C           dec         dword ptr [ebp+0Ch]
  0000001B: 89 4D 10           mov         dword ptr [ebp+10h],ecx
  0000001E: 75 EC              jne         0000000C
  00000020: 8B 45 10           mov         eax,dword ptr [ebp+10h]
  00000023: 5D                 pop         ebp
  00000024: C3                 ret

Updated code generates
  00000000: 8B 44 24 0C        mov         eax,dword ptr [esp+0Ch]
  00000004: 33 C9              xor         ecx,ecx
  00000006: 39 4C 24 08        cmp         dword ptr [esp+8],ecx
  0000000A: 7E 14              jle         00000020
  0000000C: 8B 54 24 04        mov         edx,dword ptr [esp+4]
  00000010: 6B C0 21           imul        eax,eax,21h
  00000013: 0F B6 14 11        movzx       edx,byte ptr [ecx+edx]
  00000017: 03 C2              add         eax,edx
  00000019: 41                 inc         ecx
  0000001A: 3B 4C 24 08        cmp         ecx,dword ptr [esp+8]
  0000001E: 7C EC              jl          0000000C
  00000020: C3                 ret

1. prefers a local variable instead of a parameter
2. prefers indexing by the loop count to save a register being incremented.

[commit-bot: I haz the power, 2012-04-12 02:07:24]

No LGTM from a valid reviewer yet. Only full committers are accepted.
Even if an LGTM may have been provided, it was from a non-committer or
a lowly provisional committer, _not_ a full super star committer.
See http://www.chromium.org/getting-involved/become-a-committer
Note that this has nothing to do with OWNERS files.

[Ami GONE FROM CHROMIUM, 2012-04-12 02:25:52]

OOC what is this used for that needs speed?

[fbarchard, 2012-04-12 04:06:05]

On 2012/04/12 02:25:52, Ami Fischman wrote:
> OOC what is this used for that needs speed?

On 2012/04/12 02:25:52, Ami Fischman wrote:
> OOC what is this used for that needs speed?

When testing using media_bench the hash is a bottleneck.
md5 is quite slow, adding hours to a test - slower than writing 2 files and
comparing them.  I've got a CL to improve that a bit
djb2 is fast but still about 10% of the test time.
The updated code is simpler and faster than the original.

A still faster version is in libyuv using sse4 but I didn't want to add the
dependency or large amount of pasted code.
BenchmakDjb2_C (1502 ms)
BenchmakDjb2_OPT (349 ms)