Fix building on x86 32-bit Linux

[markreidvfx]

Fixes #1841

rbx and rsi are 64-bit registers, in 32-bt you need to use ebx and esi

This 32-bit platform probably isn't getting much testing these days, I'm getting a lot of test failures on my VM.

[  23/1066] [MixingSlider / basic                                        ] - FAILED
[ 155/1066] [CPUProcessor / optimizations                                ] - FAILED
[ 162/1066] [DynamicPropertyImpl / grading_rgb_curve_knots               ] - FAILED
[ 175/1066] [FileFormat3DL / load                                        ] - FAILED
[ 196/1066] [FileFormatCTF / clf_examples                                ] - FAILED
[ 219/1066] [FileFormatCTF / tabluation_support                          ] - FAILED
[ 264/1066] [FileFormatCTF / indexMap_test1_clfv2                        ] - FAILED
[ 419/1066] [FileFormatCTF / bake_1d_3d                                  ] - FAILED
[ 422/1066] [FileFormatD1DL / test_lut1d_8i_8i                           ] - FAILED
[ 425/1066] [FileFormatD1DL / test_lut1d_16f_12i                         ] - FAILED
[ 429/1066] [FileFormatHDL / bake_1d_shaper                              ] - FAILED
[ 464/1066] [FileFormatResolveCube / bake_1d_shaper                      ] - FAILED
[ 473/1066] [FileFormatSpi1D / bake_1d_shaper                            ] - FAILED
[ 534/1066] [GpuShader / MetalSupport8                                   ] - FAILED
[ 626/1066] [ExposureContrastRenderer / video                            ] - FAILED
[ 627/1066] [ExposureContrastRenderer / log                              ] - FAILED
[ 669/1066] [GammaOp / combining                                         ] - FAILED
[ 794/1066] [Lut1DRenderer / lut_1d_inv_decreasing_reversals             ] - FAILED
[ 961/1066] [SSE2 / packed_all_test                                      ] - FAILED
[ 968/1066] [AVX / packed_all_test                                       ] - FAILED
[ 975/1066] [AVX2 / packed_all_test                                      ] - FAILED

A lot of them look like rounding errors similar to the ones we get one arm64

[doug-walker]

Thanks for the quick fix @markreidvfx !

I suspect that a significant number of the failing tests are more than just a floating-point tolerance issue. Until someone is able to investigate, we should perhaps indicate in the Installation guide that 32-bit Linux is currently not supported.

[markreidvfx]

Ya, they are probably more then just rounding errors. Some the values just off by 1. I tested 32bit on windows with msvc too and only get 1 Failed case.

FAILED: int(100000) == int((slider.mixingToSlider(slider.getSliderMaxEdge()))*100000.)
        values were '100000' and '99999'
[  23/1066] [MixingSlider / basic                                        ] - FAILED

This could be compiler related too.