#5598: Function quotRem is inefficient
----------------------------------------------+-----------------------------
Reporter: boris | Owner: igloo
Type: task | Status: new
Priority: normal | Milestone: 7.6.1
Component: Compiler | Version: 7.0.3
Keywords: division, performance, primop | Os: Unknown/Multiple
Architecture: x86 | Failure: Runtime
performance bug
Difficulty: Unknown | Testcase:
Blockedby: | Blocking:
Related: |
----------------------------------------------+-----------------------------
Comment(by boris):
Thank you for fixing this. Despite LLVM does not have a div-like command,
we can help it to optimize the second division away by replacing two loads
with one.
Actual:
{{{
%ln1ot = load i64* %ls1ln
%ln1ou = srem i64 %ln1ot, 123
store i64 %ln1ou, i64* %ls1lW
%ln1ov = load i64* %ls1ln
%ln1ow = sdiv i64 %ln1ov, 123
store i64 %ln1ow, i64* %ls1lV
}}}
Expected:
{{{
%ln1ot = load i64* %ls1ln
%ln1ou = srem i64 %ln1ot, 123
store i64 %ln1ou, i64* %ls1lW
%ln1ow = sdiv i64 %ln1ot, 123
store i64 %ln1ow, i64* %ls1lV
}}}
Perhaps this optimization can be generalized further as sharing load
command for multiple adjacent primops. Some time ago I saw several loads
from one memory location in LLVM assembly not related to division.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/5598#comment:12>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
