SkNx_shuffle

SkNx_shuffle

This allows us to express shuffles more directly in code while also giving us a
convenient point to platform-specify particular shuffles for particular types.

No specializations yet.  Everyone just uses the (pretty good) default option.

BUG=skia:

Committed: https://skia.googlesource.com/skia/+/a1c0ee40040fa8bb06f5f68ba4bcb3a68f789924

[mtklein_C, 2015-09-01 15:17:11]

mtklein@chromium.org changed reviewers:
+ reed@google.com

[mtklein_C, 2015-09-01 15:17:12]

I'm not sold on this yet, but I thought you might want to see it.

[reed1, 2015-09-01 15:45:04]

Sure seems useful.

Is "shuffle" also what GLSL calls this sort of thing. They have this feature,
but perhaps w/ different syntax (e.g. maybe f.yxwz)

What do you think of a different name for the 1-parameter version?

e.g.

Sk4f_replicate<slot>(f)

https://codereview.chromium.org/1301413006/diff/1/src/core/SkNx.h
File src/core/SkNx.h (right):

https://codereview.chromium.org/1301413006/diff/1/src/core/SkNx.h#newcode275
src/core/SkNx.h:275: //     Sk4f(1,2,3,4) f;
do you mean Sk4f f(1,2,3,4); ?

Actually, I think it may be ever clearer to use letters, so as to clearly
distinguish between indices and values

e.g.

Sk4f f(a, b, c, d);

[mtklein_C, 2015-09-01 16:00:03]

In GLSL, both use cases  (e.g. .zyxw, .xxxx) would be called swizzles.

On the other hand, in GLSL you never type the word "swizzle", and in all other
places where you do type a word (OpenCL, GCC's vector extensions, Clang's vector
extensions, SSE, NEON) the word is "shuffle" or sometimes "dup" for our
single-index version.

I don't have much opinion on whether this is best with one or two names, but I'd
go for "dup" if we split in two.

https://codereview.chromium.org/1301413006/diff/1/src/core/SkNx.h
File src/core/SkNx.h (right):

https://codereview.chromium.org/1301413006/diff/1/src/core/SkNx.h#newcode275
src/core/SkNx.h:275: //     Sk4f(1,2,3,4) f;
On 2015/09/01 15:45:03, reed1 wrote:
> do you mean Sk4f f(1,2,3,4); ?
> 
> Actually, I think it may be ever clearer to use letters, so as to clearly
> distinguish between indices and values
> 
> e.g.
> 
> Sk4f f(a, b, c, d);

Yep, done.

[mtklein_C, 2015-09-10 00:40:56]

Ping.

[reed1, 2015-09-10 16:55:00]

I still have a minor request for an alias (e.g. SkNx_replicate<2> -->
SkNx_shuffle<2, 2, 2, 2>) since many call sites use the 1-arg version, and
"shuffle" does not convey the same thing (to me) as something like "replicate"
(or other words).

that said, not looking to force this, or stall the CL. LGTM

[mtklein, 2015-09-10 18:32:27]

On 2015/09/10 16:55:00, reed1 wrote:
> I still have a minor request for an alias (e.g. SkNx_replicate<2> -->
> SkNx_shuffle<2, 2, 2, 2>) since many call sites use the 1-arg version, and
> "shuffle" does not convey the same thing (to me) as something like "replicate"
> (or other words).
> 
> that said, not looking to force this, or stall the CL. LGTM

Ah right!  Done.

[mtklein, 2015-09-10 18:32:38]

The CQ bit was checked by mtklein@google.com

[mtklein, 2015-09-10 18:32:38]

The patchset sent to the CQ was uploaded after l-g-t-m from reed@google.com
Link to the patchset: https://codereview.chromium.org/1301413006/#ps40001
(title: "dup")

[commit-bot: I haz the power, 2015-09-10 18:32:47]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1301413006/40001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1301413006/40001

[commit-bot: I haz the power, 2015-09-10 21:16:11]

Committed patchset #3 (id:40001) as
https://skia.googlesource.com/skia/+/a1c0ee40040fa8bb06f5f68ba4bcb3a68f789924