Main Content

gpucoder.stridedMatrixMultiply

Optimized GPU implementation of strided and batched matrix multiply operation

Since R2020a

Description

D = gpucoder.stridedMatrixMultiply(A,B) performs strided matrix-matrix multiplication of a batch of matrices. The input matrices A and B for each instance of the batch are located at fixed address offsets from their addresses in the previous instance. The gpucoder.stridedMatrixMultiply function performs matrix-matrix multiplication of the form:

D=αAB

where α is a scalar multiplication factor, A, B, and D are matrices with dimensions m-by-k, k-by-n, and m-by-n respectively. You can optionally transpose or hermitian-conjugate A and B. By default, α is set to one and the matrices are not transposed. To specify a different scalar multiplication factor and perform transpose operations on the input matrices, use the Name,Value pair arguments.

All the batches passed to the gpucoder.stridedMatrixMultiply function must be uniform. That is, all instances must have the same dimensions m,n,k.

example

___ = gpucoder.stridedMatrixMultiply(___,Name,Value) performs strided batched matrix multiply operation by using the options specified by one or more Name,Value pair arguments.

Examples

collapse all

Perform a simple batched matrix-matrix multiplication and use the gpucoder.stridedMatrixMultiply function to generate CUDA® code that calls corresponding cublas<t>gemmStridedBatched APIs.

In one file, write an entry-point function myStridedMatMul that accepts matrix inputs A and B. Because the input matrices are not transposed, use the 'nn' option.

function [D] = myStridedMatMul(A,B,alpha)

[D] = gpucoder.stridedMatrixMultiply(A,B,'alpha',alpha, ...
    'transpose','nn');

end

To create a type for a matrix of doubles for use in code generation, use the coder.newtype function.

A = coder.newtype('double',[5 4 100],[0 0]);
B = coder.newtype('double',[4 5 100],[0 0]);
alpha = 0.3;
inputs = {A,B,alpha};

To generate a CUDA library, use the codegen function.

cfg = coder.gpuConfig('lib');
cfg.GpuConfig.EnableCUBLAS = true;
cfg.GpuConfig.EnableCUSOLVER = true;
cfg.GenerateReport = true;
codegen -config cfg-args inputs myStridedMatMul

The generated CUDA code contains kernels myStridedMatMul_kernelNN for initializing the input and output matrices. The code also contains the cublasDgemmStridedBatched API calls to the cuBLAS library. The following code is a snippet of the generated code.

//
// File: myStridedMatMul.cu
//
...
void myStridedMatMul(const double A_data[], const int A_size[3], const double
                     B_data[], const int B_size[3], double alpha, double D_data[],
                     int D_size[3])
{
  double alpha1;
...
  beta1 = 0.0;
  cudaMemcpy(gpu_alpha1, &alpha1, 8ULL, cudaMemcpyHostToDevice);
  cudaMemcpy(gpu_A_data, (void *)A_data, A_size[0] * A_size[1] * A_size[2] *
             sizeof(double), cudaMemcpyHostToDevice);
  cudaMemcpy(gpu_B_data, (void *)B_data, B_size[0] * B_size[1] * B_size[2] *
             sizeof(double), cudaMemcpyHostToDevice);
  cudaMemcpy(gpu_beta1, &beta1, 8ULL, cudaMemcpyHostToDevice);
  if (D_data_dirtyOnCpu) {
    cudaMemcpy(gpu_D_data, &D_data[0], 25 * D_size[2] * sizeof(double),
               cudaMemcpyHostToDevice);
  }

  if (batchDimsA[2] >= batchDimsB[2]) {
    if (batchDimsA[2] >= 1) {
      ntilecols = batchDimsA[2];
    } else {
      ntilecols = 1;
    }
  } else {
    ntilecols = batchDimsB[2];
  }

  cublasDgemmStridedBatched(getCublasGlobalHandle(), CUBLAS_OP_N, CUBLAS_OP_N, 5,
    5, 4, (double *)gpu_alpha1, (double *)&gpu_A_data[0], 5, strideA, (double *)
    &gpu_B_data[0], 4, strideB, (double *)gpu_beta1, (double *)&gpu_D_data[0], 5,
    25, ntilecols);
  cudaMemcpy(&D_data[0], gpu_D_data, 25 * D_size[2] * sizeof(double),
             cudaMemcpyDeviceToHost);
...
}

Input Arguments

collapse all

Operands, specified as vectors or matrices. gpucoder.stridedMatrixMultiply multiplies along the first two dimensions.

Data Types: double | single | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64
Complex Number Support: Yes

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: D = gpucoder.stridedMatrixMultiply(A,B,'alpha',0.3,'transpose','CC');

Value of the scalar used for multiplication with A. Default value is one.

Character vector or string composed of two characters, indicating the operation performed on the matrices A and B prior to matrix multiplication. Possible values are normal ('N'), transposed ('T'), or complex conjugate transpose ('C').

Output Arguments

collapse all

Product, returned as a scalar, vector, or matrix. Array D has the same number of rows as input A and the same number of columns as input B.

Version History

Introduced in R2020a