743419 - Tegra driver bug workaround applied unconditionally, potentially harming performance

Status: NEW

(Core :: Graphics, defect)

Description

[Benoit Jacob [:bjacob] (mostly away)]

In GLContext.cpp we have this:

void
GLContext::TexSubImage2DWithUnpackSubimageGLES(GLenum target, GLint level,
                                               GLint xoffset, GLint yoffset,
                                               GLsizei width, GLsizei height,
                                               GLsizei stride, GLint pixelsize,
                                               GLenum format, GLenum type,
                                               const GLvoid* pixels)
{
    fPixelStorei(LOCAL_GL_UNPACK_ALIGNMENT,
                 NS_MIN(GetAddressAlignment((ptrdiff_t)pixels),
                        GetAddressAlignment((ptrdiff_t)stride)));
    // When using GL_UNPACK_ROW_LENGTH, we need to work around a Tegra
    // driver crash where the driver apparently tries to read
    // (stride - width * pixelsize) bytes past the end of the last input
    // row. We only upload the first height-1 rows using GL_UNPACK_ROW_LENGTH,
    // and then we upload the final row separately. See bug 697990.
    int rowLength = stride/pixelsize;
    fPixelStorei(LOCAL_GL_UNPACK_ROW_LENGTH, rowLength);
    fTexSubImage2D(target,
                    level,
                    xoffset,
                    yoffset,
                    width,
                    height-1,
                    format,
                    type,
                    pixels);
    fPixelStorei(LOCAL_GL_UNPACK_ROW_LENGTH, 0);
    fTexSubImage2D(target,
                    level,
                    xoffset,
                    yoffset+height-1,
                    width,
                    1,
                    format,
                    type,
                    (const unsigned char *)pixels+(height-1)*stride);
    fPixelStorei(LOCAL_GL_UNPACK_ALIGNMENT, 4);
}

This seems to come from bug 724094.

Comment 1

[Benoit Jacob [:bjacob] (mostly away)]

Note: I'm soon attaching a patch to bug 686375, adding no-driver-workarounds mode. Please make a patch checking for mWorkAroundDriverBugs.