While these could be ported to Windows, it is often easier to write code from scratch, rather than try to modify an existing program. WASTE, an emulator for CUDA programs, is the result of one month of programming.

The purpose of the program is to provide a means to develop and test algorithms on a PC that does not have an NVIDIA GPU.

This code is not portable to other systems because it makes a number of assumptions: the host machine is little endian DLL injection is performed using Windows system calls and x86 Intel assembly code call stack trace back uses x86 Intel assembly code fundamental data types in CUDA are represented using non-standard fundamental data types in C++. However, these assumptions are fine because WASTE is not intended to be ported to any other environment. Many design decisions were made in order to simply the coding of the emulator. While the host portion of a program will run quickly, WASTE interprets the PTX code for CUDA kernel portion of the program. Most programs will run, but will be several orders of magnitude slower on the emulator in comparison to running them natively on a NVIDIA GPU. The main reason for the slow runtime is that WASTE interprets each PTX instruction, which involves the extraction of the instruction from the AST representation of the kernel code, and symbol table manipulations, which are used to represent the values of registers and memory. In comparison, Ocelot translates PTX assembly code to LLVM, then into native code on the host machine. Barra is slightly different in that it interprets the machine code that is produced by the PTXAS assembler.

Progecad portable