G = gpuArray(A)
| Parameter | Description |
|---|---|
| A | a real, logical, complex or small-integer numeric array. |
| Parameter | Description |
|---|---|
| G | a gpuArray stored on the device. |
G = gpuArray(A) copies the array A to the GPU and returns a gpuArray object. Operations on G run on the device; use gather to bring the result back to the host.
The device stores data in single precision: a double input is downcast to single. Real, logical and complex inputs are supported; sparse inputs are not.
The small integer classes int8, uint8, int16 and uint16 are preserved: their values are exact in single precision, arithmetic saturates to the class range and gather returns the original class.
int32 and uint32 are also preserved (stored as raw 32-bit integer bits): round-trip, saturating +, -, .*, ./, .\, unary minus, abs, max/min (element-wise and reductions), sum, prod, mean, cumsum, cumprod, sort, find, integer power (.^ with a non-negative exponent), relational operators and the reshaping/indexing operations are supported. Operations that have no integer meaning (the transcendental elementwise math functions) raise a clear error. The 64-bit integer classes are not supported.
A compatible GPU is required (see canUseGPU).
A device array can also be created directly with zeros(sz, 'gpuArray') / ones(sz, 'gpuArray'), or with the 'like' form zeros(sz, 'like', G) where G is a gpuArray.
A = single(rand(1000));
G = gpuArray(A);
R = gather(G .* G + sqrt(G));
class(G)| Version | Description |
|---|---|
| 2.0.0 | initial version |