|
From: | Marcus G. Daniels |
Subject: | Re: [Swarm-Modelling] i7 processors |
Date: | Sun, 24 Jan 2010 16:22:34 -0700 |
User-agent: | Thunderbird 2.0.0.23 (Windows/20090812) |
Ken Lloyd wrote:
Apple's new compiler toolchain is interesting because it is now based on LLVM. Further, their OpenCL implementation is also based on LLVM. LLVM bytecode is converted to AMD or NVIDIA GPU instructions.The important aspect is how well your application partitions the data, and rolls out loops(tasks) for parallel processing within the software design.
In OpenCL, one has:OpenCL Queues trigger calls of CL kernels (CL is the OpenCL dialect of C99 implemented by `clang') -> LLVM bytecode -> GPU ISA
In Objective C (again with open source `clang'), one has: Objective C -> LLVM bytecode -> x86 machine code In Swarm we have: Swarm Schedules trigger calls of Objective C methods OpenCL queues are like simplified dynamic Swarm schedules.. Marcus
[Prev in Thread] | Current Thread | [Next in Thread] |