Enum MatmulComputeType
- Namespace
- NVMathNet.LinAlg
- Assembly
- NVMath.Net.dll
cuBLASLt compute types available for matrix multiplication.
public enum MatmulComputeTypeFields
Compute16F = 64-
16-bit floating-point accumulation.
Compute16FPedantic = 65-
16-bit floating-point, strict IEEE compliance.
Compute32F = 68-
32-bit floating-point accumulation (default for SGEMM).
Compute32FFast16BF = 75-
32-bit float compute accelerated by bfloat16 tensor cores.
Compute32FFast16F = 74-
32-bit float compute accelerated by 16-bit tensor cores.
Compute32FFastTF32 = 77-
32-bit float compute using TF32 tensor cores.
Compute32FPedantic = 69-
32-bit floating-point, strict IEEE compliance.
Compute32I = 72-
32-bit integer accumulation (IGEMM).
Compute32IPedantic = 73-
32-bit integer, strict mode.
Compute64F = 70-
64-bit floating-point accumulation (DGEMM).
Compute64FPedantic = 71-
64-bit floating-point, strict IEEE compliance.