Bug 83573

I thought Marek's patch (mesa: set UniformBooleanTrue = 1.0f by default) would fix this, but it doesn't.  Using integer 1 works.

Setting DEBUG_PROG = 1 in prog_execute.c and making ADD print values in hexadecimal, I see:

----false case----
ADD (3f800000 3f800000 3f800000 3f800000) = (80000000 80000000 80000000 80000000
) + (3f800000 3f800000 3f800000 3f800000)
ADD (1 1 1 1) = (-0 -0 -0 -0) + (1 1 1 1)

----true case----
ADD (ce7e0000 ce7e0000 ce7e0000 ce7e0000) = (ce7e0000 ce7e0000 ce7e0000 ce7e0000) + (3f800000 3f800000 3f800000 3f800000)
ADD (-1.06535e+09 -1.06535e+09 -1.06535e+09 -1.06535e+09) = (-1.06535e+09 -1.06535e+09 -1.06535e+09 -1.06535e+09) + (1 1 1 1)

So, it looks like we're seeing 1.0f = 3f800000 numerically converted to a float somewhere (rather than bitcast).  0x3f800000 = 1065353216 = 1.06535322E9.  The ADD negates op[0], giving -1.06535322E9 = 0xce7e0000.

I'm not sure where that's happening yet.

On quick inspection that's happening in

_mesa_propagate_uniforms_to_driver_storage

The format gets set to uniform_bool_float which in turn goes through

((float *) dst)[c] = (float) *isrc;

Right, I just found that too.  Testing a patch to fix it now.

Patch on the mailing list:
http://mid.gmane.org/1410211719-31585-1-git-send-email-kenneth@whitecape.org

> I thought Marek's patch (mesa: set UniformBooleanTrue = 1.0f by default)
> would fix this, but it doesn't.  Using integer 1 works.

Which is the exact reason why my patch to fix this issue used 1 :) I did not bother to check why it works though, just checked that this was the way it was before.

http://lists.freedesktop.org/archives/mesa-dev/2014-August/066580.html

Yep, you were right...

Fixed in master with:

commit e36bbff0e6d6d2baeaeab153b93e201674be2056
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Mon Sep 8 14:28:39 2014 -0700

    ir_to_mesa: Stop converting uniform booleans.
    
    Excess conversions considered harmful.
    
    Recently Matt reworked the boolean uniform handling to use the value of
    UniformBooleanTrue, rather than integer 1, when uploading uniforms:
    
        mesa: Upload boolean uniforms using UniformBooleanTrue.
        glsl: Use UniformBooleanTrue value for uniform initializers.
    
    Marek then set the default to 1.0f for drivers without native integer
    support:
    
        mesa: set UniformBooleanTrue = 1.0f by default
    
    However, ir_to_mesa was assuming a value of integer 1, and arranging for
    it to be converted to 1.0f on upload.  Since Marek's commit, we were
    uploading 1.0f = 0x3f800000 which was being interpreted as the integer
    value 1065353216 and converted to float as 1.06535322E9, which broke
    assumptions in ir_to_mesa that "true" was exactly 1.0f.
    
    +13 Piglits on classic swrast (fs-bool-less-compare-true,
    {vs,fs}-op-not-bool-using-if, glsl-1.20/execution/uniform-initializer).
    
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=83573
    Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
    Reviewed-by: Matt Turner <mattst88@gmail.com>

This affect r300 too.
Please consider commiting to 10.3.