Re: [Lcms-user] Lcms2-2.4 release candidate - feedback

[Mark <de...@ci...>]

Hello Marti

I could reproduce it with your code - kind of. 
The single pixel convertion is working fine, the problem only shows up converting multiple pixels:

Heres a printout of the first 12 RGBA floats from the source, then the same pixels after RGBA => LabA => RGBA:
0.6054169536 0.3002517819 0.2647287846 0.0000000000 0.2757610381 0.1638818979 0.1508964747 0.0000000000 0.4944380820 0.1766842157 0.1336690336 0.0000000000 
0.6054169536 0.3002517521 0.2647288144 0.0000000000 0.0000000000 0.2757610381 0.1638818979 0.1508965045 0.0000000000 0.0000000000 0.4944380820 0.1766842157 

Looks like lcms is inserting one component per result pixel.

Here the first 12 floats of the image after conversion RGBA => LabA
42.5072898865 32.5295295715 21.1147441864 0.0000000000 20.5950908661 13.3275184631 8.0124092102 0.0000000000 30.9475460052 35.2622528076 26.1845855713 0.0000000000 

So looks like the problem is only happening when going from LabA => RGBA

I could also reproduce the bug going using TYPE_Lab_FLT (instead of TYPE_LabA_FLT)
But, when going from TYPE_Lab_FLT or TYPE_LabA_FLT to TYPE_RGB_FLT all is correctly packed.
Interestingly, TYPE_ARGB_FLT is working fine.

If I take out the EXTRA_SH from the TYPE_RGBA_FLT, all seems to work fine.
From
#define TYPE_RGBA_FLT         (FLOAT_SH(1)|COLORSPACE_SH(PT_RGB)|EXTRA_SH(1)|CHANNELS_SH(3)|BYTES_SH(4))
To
#define TYPE_RGBA_FLT         (FLOAT_SH(1)|COLORSPACE_SH(PT_RGB)|CHANNELS_SH(3)|BYTES_SH(4))

Does that make any sense?

Hope that helps
mark


On 31.08.2012, at 17:46, Marti Maria <mar...@li...> wrote:

> 
> Hi Mark, I've tried to reproduce the bug, but with no success... 
> 
> That is the code I've been checking. Could please verify this is what you reported failing?T
> Thanks,
> Marti
> 
> /*
> Bug reported
> 
>        1)
>        sRGB built-in V4.3 -> Lab identity built-in V4.3
>        Flags: "cmsFLAGS_NOCACHE", "cmsFLAGS_NOOPTIMIZE"
>        Input format: TYPE_RGBA_FLT
>        Output format: TYPE_LabA_FLT
> 
>        2) and back
>        Lab identity built-in V4.3 -> sRGB built-in V4.3 
>        Flags: "cmsFLAGS_NOCACHE", "cmsFLAGS_NOOPTIMIZE"
>        Input format: TYPE_LabA_FLT
>        Output format: TYPE_RGBA_FLT
> 
> */
> static
> cmsInt32Number ChecksRGB2LabFLT(void)
> {
>    cmsHPROFILE hSRGB = cmsCreate_sRGBProfile();
>    cmsHPROFILE hLab  = cmsCreateLab4Profile(NULL);
> 
>    cmsHTRANSFORM xform1 = cmsCreateTransform(hSRGB, TYPE_RGBA_FLT, hLab, TYPE_LabA_FLT, 0, cmsFLAGS_NOCACHE|cmsFLAGS_NOOPTIMIZE);
>    cmsHTRANSFORM xform2 = cmsCreateTransform(hLab, TYPE_LabA_FLT, hSRGB, TYPE_RGBA_FLT, 0, cmsFLAGS_NOCACHE|cmsFLAGS_NOOPTIMIZE);
> 
>    cmsFloat32Number RGBA1[4], RGBA2[4], LabA[4];
>    int i;
> 
> 
>    for (i = 0; i <= 100; i++)
>    {
>        RGBA1[0] = i / 100.0F;
>        RGBA1[1] = i / 100.0F;
>        RGBA1[2] = i / 100.0F;
>        RGBA1[3] = 0;
> 
>        cmsDoTransform(xform1, RGBA1, LabA,  1);
>        cmsDoTransform(xform2, LabA, RGBA2, 1);
> 
>        if (!IsGoodVal("Float RGB->RGB", RGBA1[0], RGBA2[0], FLOAT_PRECISSION) ||
>            !IsGoodVal("Float RGB->RGB", RGBA1[1], RGBA2[1], FLOAT_PRECISSION) ||
>            !IsGoodVal("Float RGB->RGB", RGBA1[2], RGBA2[2], FLOAT_PRECISSION))
>            return 0;
>    }
> 
> 
>    cmsDeleteTransform(xform1);
>    cmsDeleteTransform(xform2);
>    cmsCloseProfile(hSRGB);
>    cmsCloseProfile(hLab);
> 
>    return 1;
> }
> 
> 
> 
> El 24/08/2012 12:45, Mark escribió:
>> Hello,
>> 
>> I am working at a set of Cocoa Wrappers for LCMS. I started some months ago with lcms2 2.1. 
>> Just updated to lcms2 2.4rc1. Nice. I see some matrix shader transform bugs where solved in the meantime. 
>> 
>> I have around 10 tests involving transformation from sRGB to LAB / XYZ / some other profile and back, with different bit depths and flags.
>> 
>> All my tests, but one, run fine with 2.4rc1:
>> 
>> 1)
>> sRGB built-in V4.3 -> Lab identity built-in V4.3
>> Flags: "cmsFLAGS_NOCACHE", "cmsFLAGS_NOOPTIMIZE"
>> Input format: TYPE_RGBA_FLT
>> Output format: TYPE_LabA_FLT
>> 
>> 2) and back
>> Lab identity built-in V4.3 -> sRGB built-in V4.3 
>> Flags: "cmsFLAGS_NOCACHE", "cmsFLAGS_NOOPTIMIZE"
>> Input format: TYPE_LabA_FLT
>> Output format: TYPE_RGBA_FLT
>> 
>> 3)
>> save resulting RGBA float image.
>> 
>> The image is corrupted. Sample attached.
>> 
>> 
>> Thx for this terrific library.
>> 
>> Kind regards
>> Mark
>