Bug 111012 – Refactor validation checks for texture uploads

• Status: RESOLVED
• Resolution: FIXED
• Component: WebGL

[Kenneth Russell]

Currently the validation checks performed during texture uploads in WebGL are scattered in many different places in the call stack. The majority of the checks occur in texImage2DBase and texSubImage2DBase, but some of the entry points which call these methods need to pack incoming data into a specified format. Therefore the format and type combination have to be verified as valid beforehand, leading to redundant checks in certain situations.

All of the necessary validation code should be refactored and called from the user-facing entry points before calling down further, and in particular, before calling any of the texture data conversion code. Then the checks performed in texImage2DBase and texSubImage2DBase will be unnecessary and can be turned into assertions.

[Jun Jiang]

Created attachment 192942 [details]
Patch

[Jun Jiang]

As Kenneth suggested, move all the validation checks to the entry level and turn the check at bottom(tex{Sub}Image2DBase) into assertions.

[Early Warning System Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass qt-ews (qt):
Output: http://webkit-commit-queue.appspot.com/results/17136221

[Early Warning System Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass qt-wk2-ews (qt):
Output: http://webkit-commit-queue.appspot.com/results/17175088

[WebKit Review Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://webkit-commit-queue.appspot.com/results/17203040

[EFL EWS Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass efl-ews (efl):
Output: http://webkit-commit-queue.appspot.com/results/17036757

[Build Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass mac-ews (mac):
Output: http://webkit-commit-queue.appspot.com/results/17207116

[Peter Beverloo (cr-android ews)]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass cr-android-ews (chromium-android):
Output: http://webkit-commit-queue.appspot.com/results/17169478

[Build Bot]

Comment on attachment 192942 [details]
Patch

Attachment 192942 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-commit-queue.appspot.com/results/17126354

[Kenneth Russell]

Comment on attachment 192942 [details]
Patch

Before this is reviewed in detail could you please rebase it on to top of tree and upload another version? Thanks.

[Jun Jiang]

Created attachment 193069 [details]
Patch

[Jun Jiang]

Fixed the issue that some variables were only used in Debug build and resulted in "unused variable" error for Release build.

[Kenneth Russell]

Comment on attachment 193069 [details]
Patch

Thanks for doing this refactoring. It looks great. bajones, zmo, could you both please take a look at this and see if it looks correct to you? i.e., all the same checks are being done, and none were lost?

What tests were run? Have you done a complete WebGL conformance test run? Mozilla's performance tests? Others?

[Jun Jiang]

Hi, Kenneth. Thanks for your comments. I had tested it on WebKit Layout tests, WebGL conformance test and Mozilla's performance tests. If any other tests are needed but missing, I am glad to have a try.

[Brandon Jones]

(In reply to comment #13)
> (From update of attachment 193069 [details])
> Thanks for doing this refactoring. It looks great. bajones, zmo, could you both please take a look at this and see if it looks correct to you? i.e., all the same checks are being done, and none were lost?

It looks to me like all the checks are still in place.

Thanks for the refactor, the new code certainly feels cleaner!

[Zhenyao Mo]

Comment on attachment 193069 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=193069&action=review

> Source/WebCore/html/canvas/WebGLRenderingContext.cpp:3699
> +    ASSERT(isGLES2NPOTStrict() || !level || !WebGLTexture::isNPOT(width, height));

You don't neeed isGLES2NPOTStrict() here as it is ASSERT.

> Source/WebCore/html/canvas/WebGLRenderingContext.cpp:3786
> +        if (xoffset + width > texture->getWidth(target, level) || yoffset + height > texture->getHeight(target, level)) {

Probably another CL since the code is just copied over, but it seems to me that we need to check for overflow with xoffset + width and yoffset + height.

Here and a few other places.

[Zhenyao Mo]

Looks good to me otherwise.  Thanks for the refactoring.

[Jun Jiang]

Comment on attachment 193069 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=193069&action=review

>> Source/WebCore/html/canvas/WebGLRenderingContext.cpp:3699
>> +    ASSERT(isGLES2NPOTStrict() || !level || !WebGLTexture::isNPOT(width, height));
> 
> You don't neeed isGLES2NPOTStrict() here as it is ASSERT.

Hi, Zhenyao. I was not quite sure if I had caught your idea right. The ASSERT() is much like ASSERT(! (!isGLES2NPOTStrict() && level && WebGLTexture::isNPOT(width, height)) ) here. And isGLES2NPOTStrict() only returns false when extensions "GL_OES_texture_npot" or "GL_ARB_texture_non_power_of_two" is enabled. It doesn't return false constantly. 
In fact, when I saw the condition statement (if(!isGLES2NPOTStrict())) to determine if to do validation on level and POT status of width and height, I felt the condition statement should be removed. It seems level, width & height won't be validated when isGLES2NPOTStrict() returns true.

>> Source/WebCore/html/canvas/WebGLRenderingContext.cpp:3786
>> +        if (xoffset + width > texture->getWidth(target, level) || yoffset + height > texture->getHeight(target, level)) {
> 
> Probably another CL since the code is just copied over, but it seems to me that we need to check for overflow with xoffset + width and yoffset + height.
> 
> Here and a few other places.

Thanks for your comments. It was a nice hit on the potential overflow issue. Adding the overflow checking can catch the error in an early stage and I will add it soon.

[Zhenyao Mo]

Comment on attachment 193069 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=193069&action=review

>>> Source/WebCore/html/canvas/WebGLRenderingContext.cpp:3699
>>> +    ASSERT(isGLES2NPOTStrict() || !level || !WebGLTexture::isNPOT(width, height));
>> 
>> You don't neeed isGLES2NPOTStrict() here as it is ASSERT.
> 
> Hi, Zhenyao. I was not quite sure if I had caught your idea right. The ASSERT() is much like ASSERT(! (!isGLES2NPOTStrict() && level && WebGLTexture::isNPOT(width, height)) ) here. And isGLES2NPOTStrict() only returns false when extensions "GL_OES_texture_npot" or "GL_ARB_texture_non_power_of_two" is enabled. It doesn't return false constantly. 
> In fact, when I saw the condition statement (if(!isGLES2NPOTStrict())) to determine if to do validation on level and POT status of width and height, I felt the condition statement should be removed. It seems level, width & height won't be validated when isGLES2NPOTStrict() returns true.

We check isGLES2NPOTStrict() only to avoid double effort, it basically checked whether the underlying implementation also does the same NPOT check.

It's no big deal, you can leave it here, but removing it in ASSERT makes the code cleaner.

[Jun Jiang]

Created attachment 195225 [details]
Patch

[Jun Jiang]

Thanks for Zhenyao's comments. Uploaded a new patch that removed isGLES2NPOTStrict() in the ASSERT and added overflow check with xoffset + width and yoffset + height.

[Zhenyao Mo]

(In reply to comment #21)
> Thanks for Zhenyao's comments. Uploaded a new patch that removed isGLES2NPOTStrict() in the ASSERT and added overflow check with xoffset + width and yoffset + height.

Thanks.  The patch looks good.

(Usually we use CheckedInt class from mozilla for overflow check. But your code is OK too.)

[Kenneth Russell]

Comment on attachment 195225 [details]
Patch

Nice work. Let's wait for this to clear the mac-ews bot before cq+'ing.

[Kenneth Russell]

zmo, bajones: thanks for the informal reviews.

[Kenneth Russell]

Comment on attachment 195225 [details]
Patch

It doesn't look like the mac-ews bot is making progress, and the test failures it reported previously don't look related to this patch. This refactoring is important, so marking cq+.

[WebKit Review Bot]

Comment on attachment 195225 [details]
Patch

Clearing flags on attachment: 195225

Committed r147042: <http://trac.webkit.org/changeset/147042>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.