sgemm

void sgemm(bool transA, bool transB, int M, int N, int K, float alpha, const float * A, int lda, const float * B, int ldb, float beta, float * C, int ldc)

source edit

Single-precision general matrix multiply (GEMM).

Computes $C \leftarrow \alpha (A B) + \beta C$ in row-major layout using Accelerate's cblas_sgemm. Each transpose flag is mapped to the corresponding CBLAS_TRANSPOSE enum (CblasNoTrans / CblasTrans) at zero copy cost — the BLAS kernel selects an alternate inner loop for transposed operands.

Math

$C_{ij} \leftarrow \alpha \sum_{k=0}^{K-1} A_{ik} B_{kj} + \beta C_{ij}$

References

BLAS Reference (Netlib), Accelerate.framework cblas_sgemm.

Parameters

transA, transBbool

Whether

A

B

should be transposed before the multiply.

M, N, Kint

Output is

M \times N

; the contracted dimension is

K

, so

A \in \mathbb{R}^{M \times K}

and

B \in \mathbb{R}^{K \times N}

(before optional transposition).

alpha, betafloat

Linear-combination coefficients. Use

\alpha = 1, \beta = 0

for a pure multiply; nonzero

\beta

enables fused accumulate-into-C.

A, Bconst float*

Row-major operand buffers.

Cfloat*

Row-major output buffer; updated in place when

\beta \neq 0

lda, ldb, ldcint

Leading dimensions (row stride in elements) of

A

B

C

Notes

Single-threaded on Apple Silicon for small/medium sizes; Accelerate dispatches to AMX (matrix coprocessor) for large GEMMs automatically.