CUDAMemoryLoad
CUDAMemoryLoad[list]
registers list into the CUDALink memory manager.
CUDAMemoryLoad[img]
registers img into the CUDALink memory manager.
Details and Options
- The CUDALink application must be loaded using Needs["CUDALink`"].
- Possible types for CUDAMemoryLoad are:
-
Integer Real Complex "Byte" "Bit16" "Integer" "Byte[2]" "Bit16[2]" "Integer32[2]" "Byte[3]" "Bit16[3]" "Integer32[3]" "Byte[4]" "Bit16[4]" "Integer32[4]" "UnsignedByte" "UnsignedBit16" "UnsignedInteger" "UnsignedByte[2]" "UnsignedBit16[2]" "UnsignedInteger[2]" "UnsignedByte[3]" "UnsignedBit16[3]" "UnsignedInteger[3]" "UnsignedByte[4]" "UnsignedBit16[4]" "UnsignedInteger[4]" "Double" "Float" "Integer64" "Double[2]" "Float[2]" "Integer64[2]" "Double[3]" "Float[3]" "Integer64[3]" "Double[4]" "Float[4]" "Integer64[4]" - The following options can be given:
-
"Device" $CUDADevice CUDA device used in computation "TargetPrecision" Automatic precision used in computation
Examples
open all close allBasic Examples (1)
First, load the CUDALink application:
Needs["CUDALink`"]mem = CUDAMemoryLoad[{1, 2, 3}]Information about memory can be retrieved using CUDAMemoryInformation:
CUDAMemoryInformation[mem]Memory added must be freed with CUDAMemoryUnload:
CUDAMemoryUnload[mem]Scope (4)
Adding memory as Real or Complex gets the type based on whether the device supports double precision or not:
CUDAMemoryLoad[{1.0, 2.0, 3.0}, Real]In this case, the CUDA device has double-precision support:
CUDAInformation[$CUDADevice, "Compute Capabilities"]The behavior can be forced to change by setting the "TargetPrecision":
CUDAMemoryLoad[{1.0, 2.0, 3.0}, Real, "TargetPrecision" -> "Single"]Memory added must be freed with CUDAMemoryUnload:
CUDAMemoryUnload[%]Images can be added with type "UnsignedByte":
mem = CUDAMemoryLoad[[image]]Getting the memory returns an image with the same properties as the original. Memory is retrieved using CUDAMemoryGet:
CUDAMemoryGet[mem]The "TypeInformation" contains the image information:
CUDAMemoryInformation[mem]Memory added must be freed with CUDAMemoryUnload:
CUDAMemoryUnload[mem]Images can be added with specified type:
CUDAMemoryLoad[[image], "Float"]Getting the memory returns an image:
CUDAMemoryGet[%]When adding graphics objects, the object is rasterized:
mem = CUDAMemoryLoad[\!\(\*Graphics3DBox[«5»]\)]When getting the memory, an image is returned:
CUDAMemoryGet[mem]//HeadMemory added must be freed with CUDAMemoryUnload:
CUDAMemoryUnload[mem]Options (1)
"TargetPrecision" (1)
Adding memory as Real or Complex gets the type based on whether the device supports double precision or not:
CUDAMemoryLoad[{1.0, 2.0, 3.0}, Real]This can be overridden by setting the "TargetPrecision" option to either "Single" or "Double". In this case, the machine has double-precision hardware, but a "Float" is used because the "TargetPrecision" is single:
CUDAMemoryLoad[{1.0, 2.0, 3.0}, Real, "TargetPrecision" -> "Single"]If type is always single precision, then setting the type as "Float" or "ComplexFloat" may be more readable:
CUDAMemoryLoad[{1.0, 2.0, 3.0}, "Float"]Applications (2)
This adds two to an input list:
srcf = FileNameJoin[{$CUDALinkPath, "SupportFiles", "addTwo.cu"}]addTwo = CUDAFunctionLoad[{srcf}, "addTwo", {{_Integer}, _Integer}, 32]res = addTwo[CUDAMemoryLoad[Range[100]], 100]CUDAMemoryGet[First@res]CUDAMemoryUnload[First@res]Internally, running a CUDAFunction will load and (if the input is not CUDAMemory) will unload the memory. To demonstrate, this color negates an input image:
src = "__global__ void imageColorNegate(mint * img, mint width, mint height, mint channels) {
mint ii;
mint xIndex = threadIdx.x + blockIdx.x*blockDim.x;
mint yIndex = threadIdx.y + blockIdx.y*blockDim.y;
mint index = channels*(xIndex + yIndex*width);
if (xIndex < width && yIndex < height) {
for (ii = 0; ii < channels; ii++)
img[index+ii] = 255 - img[index+ii];
}
}";This loads the function using CUDAFunctionLoad:
colorNegate = CUDAFunctionLoad[src, "imageColorNegate", {{_Integer}, _Integer, _Integer, _Integer}, {16, 16}]This defines input parameters:
{height, width} = ImageDimensions[[image]]
channels = ImageChannels[[image]]This runs the CUDAFunction:
colorNegate[[image], width, height, channels]Text
Wolfram Research (2010), CUDAMemoryLoad, Wolfram Language function, https://reference.wolfram.com/language/CUDALink/ref/CUDAMemoryLoad.html.
CMS
Wolfram Language. 2010. "CUDAMemoryLoad." Wolfram Language & System Documentation Center. Wolfram Research. https://reference.wolfram.com/language/CUDALink/ref/CUDAMemoryLoad.html.
APA
Wolfram Language. (2010). CUDAMemoryLoad. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/CUDALink/ref/CUDAMemoryLoad.html