[GHC] #8107: need types to express constant argument for primop correctness
GHC
ghc-devs at haskell.org
Sat Sep 7 22:18:44 CEST 2013
#8107: need types to express constant argument for primop correctness
-------------------------------------+------------------------------------
Reporter: carter | Owner:
Type: feature request | Status: new
Priority: normal | Milestone: 7.10.1
Component: Compiler | Version: 7.7
Resolution: | Keywords:
Operating System: Unknown/Multiple | Architecture: Unknown/Multiple
Type of failure: None/Unknown | Difficulty: Unknown
Test Case: | Blocked By:
Blocking: | Related Tickets:
-------------------------------------+------------------------------------
Changes (by carter):
* version: 7.6.3 => 7.7
* type: bug => feature request
* milestone: => 7.10.1
Comment:
In some of the prior examples, I was misinterpreting that int parameter
for prefetch to be related to the int parameter that is in the GCC/CLANG
prefetch intrinsics, which it is not.
That said: the notion of "static/compile time constant arguments" will be
ESSENTIAL for correctly expressing/exposing fully general prefetch, as
well as the general case of many of the SIMD instructions for AVX1, AVX2,
as well as the much more interesting 32SIMD register, 512 bit, conditional
mixing AVX512 instructions, etc.
The richer AVX512 instructions will be in Intel's Knights Landing, which
is going to be the first generation of Intel mainboard CPUs that will have
~ 40+ cores on a single die. And I believe we're quite keen on making sure
GHC works amazing on such many core hardware. Its worth noting that Knight
Landing will be the first generation of those many core CPU's to have
support via an LLVM backend as well.
I believe this is something worth taking the time to work out, and have
ready by GHC 7.10.
--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8107#comment:7>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
More information about the ghc-tickets
mailing list