what does -arch and -code flags do?

[sachin]

does anyone know what these parameters do? i have the settings: -arch compute_13 -code sm_13.

i cant remeber the settings i had before on MSVS which i had to change; reason being that before i was trying to take the absolute values of a double array but it simply wouldnt work. infact i had put

if (i<elements)
doublearray[i]=2.02;

and that didnt work. it started working when i changed the setting to:
compile name 1.3 (virtual) ARCH (aside why doesnt it let me have this as hardware ARCH)
code name 1.3 (hardware) CODE

but now that i changed it back to:

if (i<elements)
if doublearray[i]<0
doublearray[i]=-doublearray[i];

it still spits out 2.02

any thoughts

thanx
sachin

any ideas

[jph4599]

The meaning of arch and code can be found in the NVCC manual included in the doc directory of the toolkit install:

From nvcc_2.1.pdf

arch:

Specify the name of the NVIDIA GPU to compile for. This can either be a ‘real’ GPU, or a ‘virtual’ ptx architecture. Ptx code represents an intermediate format that can still be further compiled and optimized for, depending on the ptx version, a specific class of actual GPUs . The architecture specified by this option is the architecture that is assumed by the compilation chain up to the ptx stage, while the architecture(s) specified with the –code option are assumed by the last, potentially runtime, compilation stage. Currently supported compilation architectures are: virtual architectures compute_10, compute_11, plus GPU architectures sm_10 and sm_11 that implement these.

code:

Specify the name of the NVIDIA GPU to generate code for. Unless option –export-dir is specified (see below), nvcc will embed a compiled code image in the executable for each specified ‘code’ architecture, which is a true binary load image for each ‘real’ architecture, and ptx code for each virtual architecture. During runtime, such embedded ptx code will be dynamically compiled by the CUDA runtime system if no binary load image is found for the ‘current’ GPU. Architectures pecified for options –arch and –code may be virtual as well as real, but the ‘code’ architectures must be compatible with the ‘arch’ architecture. For instance, ‘arch’=compute_13 is not compatible with ‘code’=sm_10, because the earlier compilation stages will assume the availability of compute_13 features that are not present on sm_10. This option defaults to the value of option ‘- arch’. Currently supported GPU architectures: sm_10 and sm_11.

Compute computability (See programming guide appendix A for list of device compute capabilities) 1.3 (-arch compute_13) is the only one that supports double precision format. When you put in a number such as

2.02

, it automatically gets promoted to a double. If you want it to be a float instead, you have to put a ‘f’ after it:

2.02f

.

Long story short, if you don’t need the double precision on a number, make sure to add the f!

[jph4599]

As a side note, it looks like NVIDIA needs to update the supported architectures list to include 1.3, it’s excluded even from the 2.3 beta nvcc pdf.