[Mesa-dev] [PATCH 00/12] Do better for flrp on platforms that lack flrp instruction, take 2

[Ian Romanick]

This series replaces my previous series
(https://patchwork.freedesktop.org/series/48138/) that did similar
things.  Rather than trying to do everything with only local information
in nir_opt_algebraic, this series adds a new optimization pass.  This
new pass looks at how various parameters of a nir_op_flrp are used in
other nir_op_flrp instructions to make better choices.

You can compare the results across this whole series, below, with the
similar results in the previous series.  At least for Intel GPUs, this
series does quite a bit better.  There are a couple extra loops and a
couple lost SIMD16 shaders on Iron Lake, but the trade off is an extra
35 (+32 overall) SIMD16 shaders on Iron Lake and an 38 SIMD16 shaders on
GM45.

It also shouldn't break the whole universe for non-Intel GPUs. :)
Neither of the tested Intel GPUs have an FMA instruction, so I don't
know how this series will affect GPUs that lack LRP but have FMA.  I did
some testing on Ice Lake, which lacks LRP but has FMA, and the results
were mixed but generally positive.  There is some driver support missing
for the FMA on Ice Lake, so those results are not likely to be
representative of the final result.

I have another series waiting to go out that improves LRP and FMA
generation for all of the Intel platforms that support LRP and FMA.
That series caused a bunch of regressions on the non-LRP platforms, so
this series needs to land first.

This series, along with a few patches that didn't pan out, is available
at:

    https://cgit.freedesktop.org/~idr/mesa/log/?h=lower-flrp

Iron Lake
total instructions in shared programs: 7774533 -> 7676404 (-1.26%)
instructions in affected programs: 4436203 -> 4338074 (-2.21%)
helped: 19242
HURT: 924
helped stats (abs) min: 1 max: 155 x̄: 5.16 x̃: 3
helped stats (rel) min: 0.11% max: 86.96% x̄: 2.85% x̃: 1.89%
HURT stats (abs)   min: 1 max: 8 x̄: 1.29 x̃: 1
HURT stats (rel)   min: 0.08% max: 6.52% x̄: 1.22% x̃: 1.01%
95% mean confidence interval for instructions value: -4.95 -4.79
95% mean confidence interval for instructions %-change: -2.71% -2.61%
Instructions are helped.

total cycles in shared programs: 177757572 -> 177292132 (-0.26%)
cycles in affected programs: 100318508 -> 99853068 (-0.46%)
helped: 18971
HURT: 1407
helped stats (abs) min: 2 max: 930 x̄: 25.18 x̃: 12
helped stats (rel) min: <.01% max: 94.58% x̄: 1.24% x̃: 0.55%
HURT stats (abs)   min: 2 max: 98 x̄: 8.71 x̃: 6
HURT stats (rel)   min: <.01% max: 4.11% x̄: 0.33% x̃: 0.14%
95% mean confidence interval for cycles value: -23.31 -22.37
95% mean confidence interval for cycles %-change: -1.16% -1.09%
Cycles are helped.

total loops in shared programs: 850 -> 854 (0.47%)
loops in affected programs: 0 -> 4
helped: 0
HURT: 4
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 0.00% max: 0.00% x̄: 0.00% x̃: 0.00%
95% mean confidence interval for loops value: 1.00 1.00
95% mean confidence interval for loops %-change: 0.00% 0.00%
Loops are HURT.

LOST:   3
GAINED: 52

GM45
total instructions in shared programs: 4766579 -> 4715273 (-1.08%)
instructions in affected programs: 2372653 -> 2321347 (-2.16%)
helped: 9929
HURT: 467
helped stats (abs) min: 1 max: 155 x̄: 5.23 x̃: 3
helped stats (rel) min: 0.11% max: 86.96% x̄: 2.78% x̃: 1.86%
HURT stats (abs)   min: 1 max: 8 x̄: 1.28 x̃: 1
HURT stats (rel)   min: 0.08% max: 6.25% x̄: 1.19% x̃: 1.00%
95% mean confidence interval for instructions value: -5.05 -4.82
95% mean confidence interval for instructions %-change: -2.67% -2.54%
Instructions are helped.

total cycles in shared programs: 121450872 -> 121161308 (-0.24%)
cycles in affected programs: 61826744 -> 61537180 (-0.47%)
helped: 9846
HURT: 759
helped stats (abs) min: 2 max: 930 x̄: 30.12 x̃: 12
helped stats (rel) min: <.01% max: 93.33% x̄: 1.30% x̃: 0.58%
HURT stats (abs)   min: 2 max: 98 x̄: 9.27 x̃: 6
HURT stats (rel)   min: <.01% max: 4.11% x̄: 0.33% x̃: 0.13%
95% mean confidence interval for cycles value: -28.06 -26.55
95% mean confidence interval for cycles %-change: -1.23% -1.14%
Cycles are helped.

total loops in shared programs: 629 -> 631 (0.32%)
loops in affected programs: 0 -> 2
helped: 0
HURT: 2

total fills in shared programs: 93 -> 94 (1.08%)
fills in affected programs: 81 -> 82 (1.23%)
helped: 0
HURT: 1

LOST:   55
GAINED: 55