I'm porting my game framework from OpenGL 3/4 to OpenGL ES 2 but I'm having an issue where textures aren't being rendered.

In once instance I'm using simple vertex buffer objects to store the vertices and texcoords. In my OpenGL code these are bound in vertex arrays but I understand they aren't used in ES. Basically, I have some initialisation code where I do this:

[source lang="cpp"]
//glGenVertexArrays(1, &m_vao); //Not used in GLES.
glGenBuffers(2, m_vbo);

//glBindVertexArray(m_vao); //Not used in GLES.

glBindBuffer(GL_ARRAY_BUFFER, m_vbo[0]);
glBufferData(GL_ARRAY_BUFFER, 4 * sizeof(Vector2), ms_verts, GL_STATIC_DRAW);

glBindBuffer(GL_ARRAY_BUFFER, m_vbo[1]);
glBufferData(GL_ARRAY_BUFFER, 4 * sizeof(Vector2), ms_texCoords, GL_STATIC_DRAW);
[/source]

And then when it comes time to render I do this:

[source lang="cpp"]

int32_t* pUniformHandle = m_uniforms.Find(textureParamName);

if (pUniformHandle)
{
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, m_textureHandle);
glUniform1i(*pUniformHandle, 0);
}

//glBindVertexArray(m_vao); //Not used in GLES.

glBindBuffer(GL_ARRAY_BUFFER, m_vbo[0]);
glVertexAttribPointer(Effect::POSITION_ATTR, 2, GL_FLOAT, GL_FALSE, 0, 0);
glEnableVertexAttribArray(Effect::POSITION_ATTR);

glBindBuffer(GL_ARRAY_BUFFER, m_vbo[1]);
glVertexAttribPointer(Effect::TEXCOORD0_ATTR, 2, GL_FLOAT, GL_FALSE, 0, 0);
glEnableVertexAttribArray(Effect::TEXCOORD0_ATTR);

glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);

glDisableVertexAttribArray(Effect::TEXCOORD0_ATTR);
glDisableVertexAttribArray(Effect::POSITION_ATTR);
[/source]

In OpenGL (with those "Not used in GLES" bits uncommented) this works fine but in GLES the prims are being rendered but the texture isn't being applied. For reference, my shader looks like this:

Vertex shader:

[source lang="plain"]
uniform mat4 worldMatrix;
uniform mat4 projectionMatrix;

attribute vec2 aPosition;
attribute vec2 aTexCoord;

varying vec2 vTexCoord;

void main()
{
vTexCoord = aTexCoord;
mat4 worldProjMatrix = projectionMatrix * worldMatrix;
gl_Position = worldProjMatrix * vec4(aPosition, 0.0, 1.0);
}
[/source]

Fragment shader:

[source lang="plain"]
uniform sampler2D texture;
uniform vec4 colour;

varying vec2 vTexCoord;

void main()
{
gl_FragColor = texture2D(texture, vTexCoord) * colour;
}
[/source]

Can anyone see what I'm doing wrong? I've been scanning through the GLES book but I can't see anything obvious.

In OpenGL (with those "Not used in GLES" bits uncommented) this works fine but in GLES the prims are being rendered but the texture isn't being applied

To be precise, there is no such thing 'the texture isn't being applied'. What do you mean exactly? The geometry renders properly, but all rasterized fragments come out black (i.e. the texture2D call in the shader comes out black) ?

I have run into such cases several times with GLES2, and searching the codebase, I find I have added these checks:




/* GLES2 Tegra2(?) bug:
If you specify wrap for texture S coordinate and clamp for texture T coordinate, and the texture width is pow2,
but height is not, the wrap operation will silently fail and be treated as a clamp operation.
*/

/* Sanity check to avoid GLES2 SILENT FAILURES: http://www.khronos.org/opengles/sdk/docs/man/xhtml/glTexParameter.xml
If the texture does not have mipmaps, specifying MIN filter of GL_xxx_MIPMAP_xxx will succeed, but silently
render black. */
const bool hasMipmaps = targetTexture->NumMipmaps() > 1;

if (hasMipmaps)
{
if (!IsPow2(targetTexture->Width()) || !IsPow2(targetTexture->Height()))
{
LOGE("ApplyTextureSampler Error: GLES2 will now silently render black since mipmapping was specified for a non-pow2 texture! Killing mipmaps from texture '%s' as a workaround.",
targetTexture->Name().c_str());
targetTexture->DisableMipmaps();
}
}




/* Sanity check to detect GLES2 SILENT FAILURES: http://www.khronos.org/opengles/sdk/docs/man/xhtml/glTexParameter.xml
" if the width or height of a texture image are not powers of two and either the
GL_TEXTURE_MIN_FILTER is set to one of the functions that requires mipmaps or the GL_TEXTURE_WRAP_S or
GL_TEXTURE_WRAP_T is not set to GL_CLAMP_TO_EDGE, then the texture image unit will return
(R, G, B, A) = (0, 0, 0, 1). " */
if ((textureSampler->addressU != TextureAddressClamp || textureSampler->addressV != TextureAddressClamp) && (!IsPow2(targetTexture->Width()) || !IsPow2(targetTexture->Height())))
LOGE("ApplyTextureSampler Error: GLES2 doesn't support other addressing modes than GL_CLAMP_TO_EDGE when size is not pow2! Size was %dx%d in texture '%s'. (GLES2 will now silently render black)",
targetTexture->Width(), targetTexture->Height(), targetTexture->Name().c_str());


The page http://www.khronos.org/opengles/sdk/docs/man/xhtml/glTexParameter.xml documents

Suppose that a texture is accessed from a fragment shader or vertex shader and has set GL_TEXTURE_MIN_FILTER to one of the functions that requires mipmaps. If either the dimensions of the texture images currently defined (with previous calls to glTexImage2D, glCompressedTexImage2D, or glCopyTexImage2D) do not follow the proper sequence for mipmaps (described above), or there are fewer texture images defined than are needed, or the set of texture images were defined with different formats or types, then the texture image unit will return (R, G, B, A) = (0, 0, 0, 1).

Similarly, if the width or height of a texture image are not powers of two and either the GL_TEXTURE_MIN_FILTER is set to one of the functions that requires mipmaps or the GL_TEXTURE_WRAP_S or GL_TEXTURE_WRAP_T is not set to GL_CLAMP_TO_EDGE, then the texture image unit will return (R, G, B, A) = (0, 0, 0, 1).
[/quote]

Needless to say, the very existence of documented silent failure cases for an API (combined with the lack of a documented way to set up a debug mode that would properly detect these) is a major picard facepalm.jpg situation.

[color=#282828][font=helvetica, arial, verdana, tahoma, sans-serif]

[background=rgb(250, 251, 252)]To be precise, there is no such thing 'the texture isn't being applied'. What do you mean exactly? The geometry renders properly, but all rasterized fragments come out black (i.e. the texture2D call in the shader comes out black) ?[/background]

[/font]

Sorry, yes, I was generalising as I wrote the post quickly but yeah the geometry renders correctly but black. If I change the fragment shader to output a defined colour like, say gl_FragColour = vec4(1.0, 0.0, 0.0, 1.0); then it renders with the colour I supplied instead of black. But when I use the texture2D function I just get black prims.

The particular texture I'm trying to display is 512x512 so that rules out the power of two thing. I've also just tried setting GL_TEXTURE_WRAP_S and GL_TEXTURE_WRAP_T to GL_CLAMP_TO_EDGE and GL_TEXTURE_MIN_FILTER and FL_TEXTURE_MAG_FILTER to GL_LINEAR but it hasn't made a difference.

Also, not sure if it makes a difference but I am using ATI's GLES SDK in Windows. My plan was to get it working in Windows, so I know it's working, before I move it across to my Raspberry Pi, which is my ultimate goal.

I'll have more of a look tomorrow as I'm done for tonight.

I like #5, it's a clever way of seeing the texture coordinates are valid. Alas, mine were, black top-left, red top-right, green bottom-left, yellow bottom-right.

I am setting the colour value (though I know in my code I only showed me setting the texture uniform) but just to double check I removed colour from the gl_FragColor calculation and I'm still getting a black square.

I did think perhaps something isn't working in the GLES Windows SDK and that I should just port my code to Pi and fix the issue on there, if it's still an issue. That being said, if I run the Simple_Texture2D sample from the GLES book then the texture displays correctly. The only difference I can see between my code and theirs, however, is that they use glVertexAttribPointer to send the data to the GPU each render instead of storing the data off in a VBO. Should this make a difference?

I just quickly tried sacking off the VBOs and using glVertexAttribPointer to send the data to the GPU each render, as in the Simple_Texture2D sample like this:

[source lang="cpp"]
glVertexAttribPointer(Effect::POSITION_ATTR, 2, GL_FLOAT, GL_FALSE, 0, ms_verts);
glVertexAttribPointer(Effect::TEXCOORD0_ATTR, 2, GL_FLOAT, GL_FALSE, 0, ms_texCoords);

glEnableVertexAttribArray(Effect::POSITION_ATTR);
glEnableVertexAttribArray(Effect::TEXCOORD0_ATTR);

glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);

glDisableVertexAttribArray(Effect::TEXCOORD0_ATTR);
glDisableVertexAttribArray(Effect::POSITION_ATTR);
[/source]

But now even the geometry isn't being rendered. I've probably done something stupid but I only had a quick five minutes before I go to work so I thought I'd give it a try. I'll have more of a look when I get back from work.

I've fixed why the geometry wasn't being rendered when I don't use VBOs (I was being an idiot and hadn't unbound a buffer I was using elsewhere). So, I've got the geometry rending both with and without VBOs but in both instances I'm just getting black prims without the texture.

I'm now trying to match my code as closely as possible to the Simple_Texture2D sample. The first step was removing the VBOs, I've also tried using their texture code too but it hasn't made a difference. There must be something obviously different between my code and theirs though, for theirs to be working and mine not. I'm sure I'll get to the bottom of it if I keep chugging away

Just thinking, is there any chance that I could have set my window up incorrectly? Maybe something wrong in the EGL stuff? I don't think that this could cause this issue but I'm just clutching at straws really, as I can't see any difference between my code and the Simple_Texture2D code.

Try assuming that it is an asset difference rather than a code difference. Try with a known working texture asset. Or try starting from the working sample and try to evolve it towards your breaking testcase one by one to see where it breaks.

I don't think there's anything in the EGL spec that would affect how texture2D sampling is done in the shader code.

Bloody hell. I've fixed it and, as per usual with these things, it was user error. I was passing an incorrect value into glTexImage2D for the internal format. If you look at my texture load code below you can see I'm incorrectly passing bitDepth into glTexImage2D instead of textureFormat. Doh! I've no idea how this works correctly in normal OpenGL, though. I would expect it to fail like it does in GLES. Ah well, nevermind.
[source lang="cpp"]int32_t width, height, bitDepth = 0;
//Load the image.
uint8_t* pData = ::stbi_load_from_memory(rTextureFile.GetData(), rTextureFile.GetSize(), &width, &height, &bitDepth, 0);

if (pData)
{
m_width = width;
m_height = height;
m_bitDepth = bitDepth;

uint32_t textureFormat = bitDepth == 4 ? GL_RGBA : GL_RGB;
glGenTextures(1, &m_handle);
glBindTexture(GL_TEXTURE_2D, m_handle);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);

glTexImage2D(GL_TEXTURE_2D,
0,
bitDepth,
width,
height,
0,
textureFormat,
GL_UNSIGNED_BYTE,
pData);

//No longer need the image.
stbi_image_free(pData);

return true;
}

return false;
[/source]
Thanks for your help, everyone.