GPU: Buld most of the gpu as a single DLL.

GPU: Buld most of the gpu as a single DLL.

BUG=114261
TEST=none
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=124702

[rvargas (doing something else), 2012-02-29 02:15:32]

Please take a look.

Patch sets 1 - 4 have the changes to the gyp files in an easier to digest form.

Patch set 5 adds all the exports.

[greggman, 2012-02-29 03:56:19]

Hey Al, PTAL

I don't know what to think here. How do you get the GPU to be in one DLL and
still have 2 versions of gles_c_lib and GLES2Implementation?

It is just magic? ;-)

I think now that we have the GPU bots we need to build a least a couple of tests
that test the things that Pepper needs are available (client side arrays for
example) 

Do all the other DLLs in Chrome need to to export all their classes? Just
curious

I think Al should take a look but it looks okay to me I guess.

[rvargas (doing something else), 2012-02-29 20:08:00]

On 2012/02/29 03:56:19, greggman wrote:
> Hey Al, PTAL
> 
> I don't know what to think here. How do you get the GPU to be in one DLL and
> still have 2 versions of gles_c_lib and GLES2Implementation?

I'm not merging all of GPU, only the 6 projects that I moved to gypi files with
the previous change. In particular, gles2_c_lib is not changed (it builds as an
independent dll, in two flavors).

gles2_implementation is more interesting: currently there are 3 versions of it,
all static libraries. I'm building the default one into gpu.dll and keeping the
other two versions as static libraries... that means that whoever wants for
instance gles2_implementation_client_side_arrays, links against that specific
version.

There may be an issue here... but everything I know of compiles and runs
(gles2_conform_supports.exe doesn't do anything, but the comment in the code
suggests that's the intention).

> 
> It is just magic? ;-)
> 
> I think now that we have the GPU bots we need to build a least a couple of
tests
> that test the things that Pepper needs are available (client side arrays for
> example) 
> 
> Do all the other DLLs in Chrome need to to export all their classes? Just
> curious

There are two reasons to export something:
1. It is called by the intended user (it is part of the API).
2. It has unit tests (so the tests need to call it directly).

Most of our DLLs have no distinction between the two. The only exemption that I
know about is net, which uses two symbols to differentiate between something
that is intended for the user and something that is intended to be used only by
internal tests.

> 
> I think Al should take a look but it looks okay to me I guess.

[apatrick_chromium, 2012-02-29 20:08:12]

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp
File gpu/gpu.gyp (left):

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode64
gpu/gpu.gyp:64: 'target_name': 'gles2_implementation_client_side_arrays',
I think pepper plugins and native client depend on this target...

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode85
gpu/gpu.gyp:85: 'target_name':
'gles2_implementation_client_side_arrays_no_check',
... or this one. gman?

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode123
gpu/gpu.gyp:123: # Same as gles2_c_lib except with no parameter checking.
Required for
Have you checked the conformance tests still work?

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode161
gpu/gpu.gyp:161: 'target_name': 'gpu_unittests',
Where did the unit test target go? I can't find it in any of the gyp files. Or
maybe I'm being blind.

[rvargas (doing something else), 2012-02-29 22:22:39]

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp
File gpu/gpu.gyp (left):

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode123
gpu/gpu.gyp:123: # Same as gles2_c_lib except with no parameter checking.
Required for
On 2012/02/29 20:08:12, apatrick_chromium wrote:
> Have you checked the conformance tests still work?

gles2_conform_support.exe runs and doesn't do anything.

http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode161
gpu/gpu.gyp:161: 'target_name': 'gpu_unittests',
On 2012/02/29 20:08:12, apatrick_chromium wrote:
> Where did the unit test target go? I can't find it in any of the gyp files. Or
> maybe I'm being blind.

Everything removed from this file went into gpu_common.gypi. See patch set 1 for
a clear diff.

[greggman, 2012-02-29 23:07:28]

On 2012/02/29 22:22:39, rvargas wrote:
> http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp
> File gpu/gpu.gyp (left):
> 
> http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode123
> gpu/gpu.gyp:123: # Same as gles2_c_lib except with no parameter checking.
> Required for
> On 2012/02/29 20:08:12, apatrick_chromium wrote:
> > Have you checked the conformance tests still work?
> 
> gles2_conform_support.exe runs and doesn't do anything.

To run the conformance tests you need to follow the instructions in

src/gpu/gles2_conform_test/README

Only 2 tests should fail although really, the important part for this CL is that
they run at all. PS: they only run on Linux and Windows right now and I haven't
run the windows ones in many months. Looking to getting them added to the gpu
bots




> 
> http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode161
> gpu/gpu.gyp:161: 'target_name': 'gpu_unittests',
> On 2012/02/29 20:08:12, apatrick_chromium wrote:
> > Where did the unit test target go? I can't find it in any of the gyp files.
Or
> > maybe I'm being blind.
> 
> Everything removed from this file went into gpu_common.gypi. See patch set 1
for
> a clear diff.

Al pointed out that it appears you might have deleted the gpu_unittests. Did we
miss where they were moved?

[greggman, 2012-02-29 23:08:58]

On 2012/02/29 23:07:28, greggman wrote:
> On 2012/02/29 22:22:39, rvargas wrote:
> > http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp
> > File gpu/gpu.gyp (left):
> > 
> > http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode123
> > gpu/gpu.gyp:123: # Same as gles2_c_lib except with no parameter checking.
> > Required for
> > On 2012/02/29 20:08:12, apatrick_chromium wrote:
> > > Have you checked the conformance tests still work?
> > 
> > gles2_conform_support.exe runs and doesn't do anything.
> 
> To run the conformance tests you need to follow the instructions in
> 
> src/gpu/gles2_conform_test/README

Let me add, if you're on linux build 'gles2_conform_test_native', if on Windows
build 'gles2_conform_test_angle'


> 
> Only 2 tests should fail although really, the important part for this CL is
that
> they run at all. PS: they only run on Linux and Windows right now and I
haven't
> run the windows ones in many months. Looking to getting them added to the gpu
> bots
> 
> 
> 
> 
> > 
> > http://codereview.chromium.org/9514020/diff/7001/gpu/gpu.gyp#oldcode161
> > gpu/gpu.gyp:161: 'target_name': 'gpu_unittests',
> > On 2012/02/29 20:08:12, apatrick_chromium wrote:
> > > Where did the unit test target go? I can't find it in any of the gyp
files.
> Or
> > > maybe I'm being blind.
> > 
> > Everything removed from this file went into gpu_common.gypi. See patch set 1
> for
> > a clear diff.
> 
> Al pointed out that it appears you might have deleted the gpu_unittests. Did
we
> miss where they were moved?

[rvargas (doing something else), 2012-02-29 23:29:19]

> > > gles2_conform_support.exe runs and doesn't do anything.
> > 
> > To run the conformance tests you need to follow the instructions in
> > 
> > src/gpu/gles2_conform_test/README

I was looking for it before... and that folder is not part of the checkout so I
assumed it was removed. My bad.

> 
> Let me add, if you're on linux build 'gles2_conform_test_native', if on
Windows
> build 'gles2_conform_test_angle'

Nowhere to be found :(. 

On a somewhat related note, demos was removed from all.gyp (and that means that
the solution is not generated by gclient), so I re-add it locally to be able to
build it, and compressed_textures works.

[greggman, 2012-03-01 01:37:35]

On 2012/02/29 23:29:19, rvargas wrote:
> > > > gles2_conform_support.exe runs and doesn't do anything.
> > > 
> > > To run the conformance tests you need to follow the instructions in
> > > 
> > > src/gpu/gles2_conform_test/README
> 
> I was looking for it before... and that folder is not part of the checkout so
I
> assumed it was removed. My bad.
> 
> > 
> > Let me add, if you're on linux build 'gles2_conform_test_native', if on
> Windows
> > build 'gles2_conform_test_angle'
> 
> Nowhere to be found :(. 

You need to edit your .gclient file and remove the gles2_conform: None and
gles2_conform_test: None

Then sync, Then follow the directions in src/gpu/gles2_conform_test/README



> 
> On a somewhat related note, demos was removed from all.gyp (and that means
that
> the solution is not generated by gclient), so I re-add it locally to be able
to
> build it, and compressed_textures works.

That's a good sign.

[rvargas (doing something else), 2012-03-02 01:09:55]

It turns out that the linker is not smart enough to do the trick as I hoped so
gpu_unittest was failing... now that's fixed, but that means 3 more dlls
(gles2_impl*).

On the conform tests front, gles2_conform_test_angle builds and runs, but
gles2_conform_test_native doesn't link (Windows) . Is that supposed to build
correctly?

The errors are like this one:
1>GTFInitEGL.obj : error LNK2019: unresolved external symbol
__imp__eglDestroySurface@8 referenced in function _GTFInitEGL


running gles2_conform_test_angle.exe works for a while and then hangs with

clip: Start

After a few minutes, it continues and then asserts:
color_ramp: Start
color_ramp: total = 1, failure = 0

copy_texture: Start
Assertion failed: mD3DFormat != D3DFMT_UNKNOWN, file .\libGLESv2\Texture.cpp,
line 235

It doesn't really looks related to this change, does it?

[greggman, 2012-03-02 03:55:59]

On 2012/03/02 01:09:55, rvargas wrote:
> It turns out that the linker is not smart enough to do the trick as I hoped so
> gpu_unittest was failing... now that's fixed, but that means 3 more dlls
> (gles2_impl*).
> 
> On the conform tests front, gles2_conform_test_angle builds and runs, but
> gles2_conform_test_native doesn't link (Windows) . Is that supposed to build
> correctly?

It is supposed to but I haven't built the Windows one in ages. I don't think
this CL broke it.

> 
> The errors are like this one:
> 1>GTFInitEGL.obj : error LNK2019: unresolved external symbol
> __imp__eglDestroySurface@8 referenced in function _GTFInitEGL
> 
> 
> running gles2_conform_test_angle.exe works for a while and then hangs with
> 
> clip: Start
> 
> After a few minutes, it continues and then asserts:
> color_ramp: Start
> color_ramp: total = 1, failure = 0
> 
> copy_texture: Start
> Assertion failed: mD3DFormat != D3DFMT_UNKNOWN, file .\libGLESv2\Texture.cpp,
> line 235
> 
> It doesn't really looks related to this change, does it?

Agreed.

lgtm

[apatrick_chromium, 2012-03-02 18:56:56]

lgtm

[rvargas (doing something else), 2012-03-02 19:04:49]

Brett, Could you take a look at ppapi?

[brettw, 2012-03-02 19:10:55]

ppapi lgtm