intel/fs: Move the defin[]/defout[] screening up to livein[]/liveout[] setup.
This keeps us from having to run the loop to propagate up quite so much.
steam shader-db time -1.86356% +/- 0.941498% (n=10). There's a small
scheduling effect, since previously the scheduler wasn't considering
defin/defout:
cycles helped: shaders/closed/steam/amnesia-the-dark-descent/high/241.shader_test FS SIMD16: 11428 -> 11422 (-0.05%) (scheduled: scheduled)
cycles helped: shaders/humus-volumetricfogging2/1.shader_test FS SIMD32: 13832 -> 13800 (-0.23%) (scheduled: scheduled)
cycles helped: shaders/tesseract/479.shader_test FS SIMD32: 9330 -> 8644 (-7.35%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/android/angle/aztec_ruins/36.shader_test FS SIMD32: 7870 -> 7940 (0.89%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_5_high_off/57.shader_test FS SIMD32: 7870 -> 7940 (0.89%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_5_normal_off/54.shader_test FS SIMD32: 7870 -> 7940 (0.89%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/android/angle/aztec_ruins/30.shader_test FS SIMD32: 8726 -> 8808 (0.94%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_5_high_off/51.shader_test FS SIMD32: 8726 -> 8808 (0.94%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_5_normal_off/48.shader_test FS SIMD32: 8726 -> 8808 (0.94%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_4_off/129.shader_test TCS SIMD8: 3911 -> 3979 (1.74%) (scheduled: scheduled)
cycles HURT: shaders/robclark-shaders/gfxbench5/gl_4_off/109.shader_test TCS SIMD8: 3911 -> 3979 (1.74%) (scheduled: scheduled)
total cycles in shared programs:
313096438 ->
313096306 (<.01%)
cycles in affected programs: 92200 -> 92068 (-0.14%)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24702>