View Single Post
Old 2007-02-18, 12:50   #3
dsouza123
 
dsouza123's Avatar
 
Sep 2002

66210 Posts
Default

In one of the download files it mentions it currently supports 32 bit floats
but NVIDIA GPUS with 64 bit floats will be available in late 2007.

In addition to the downloadable C toolkit, registered professional developers and researchers can get access to a low level assembly language layer
and driver interface.

Vector types listed are
char1..char4,
uchar1..uchar4,
short1..short4,
ushort1..ushort4,
int1..int4,
uint1..uint4,
long1..long4,
ulong1..ulong4,
float1..float4

using x,y,z,w.

For example int2 would use the x and y fields.

The sizes of the current types are 8, 16, 32 bits.

There is a device emulation mode in which the resulting executable
is compiled for and run on the host PC instead of the GPU.
For floating point extra care must be taken to use 32 bit precision
floats (so the accuracy would only be 32 bit and not a higher internal one), to do so compiler options are used.
dsouza123 is offline   Reply With Quote