[llvm-dev] LoopStrengthReduce.cpp

[Krzysztof Parzyszek via llvm-dev]

On 3/29/2016 3:05 AM, Jonas Paulsson via llvm-dev wrote:
> Could this be done somehow, or is it really so that all targets have to
> have their own passes to do this?

In the Hexagon backend we also have a separate pass that converts 
compare+branch loops into hardware loops.  We recognize several 
different patterns of the controlling induction variable, including 
cases where the increment is not 1 or -1.  If we decide to generate a 
hardware loop, we may add some instructions to calculate the iteration 
count, if it's not already available.

The general problem for us is that not every loop can be converted to a 
hardware loop.  At the same time, making each loop iterate over a 
counter that decrements down to 0 may add extra code that may end up 
being unnecessary.  It is easier for us to recognize cases where we want 
a hardware loop and do work to make it happen.

On that note, I think that in general it would be useful to have some 
target-independent (CodeGen) pass that would do the majority of the work 
for hardware loop generation.  I have thought about it, but I won't be 
able to do anything in the short term.

-Krzysztof

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation