For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

apple_matmul

def apple_matmul[*, transpose_b: Bool = False, elementwise_lambda_fn: Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None] = None](cblas_gemm_fn: def(_CBLASOrder, _CBLASTranspose, _CBLASTranspose, Int32, Int32, Int32, Float32, Pointer[Float32, ImmutAnyOrigin, _safe=False], Int32, Pointer[Float32, ImmutAnyOrigin, _safe=False], Int32, Float32, Pointer[Float32, MutAnyOrigin, _safe=False], Int32) thin -> None, c: TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type], a: TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type], b: TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type])

Performs a single-precision matmul via a pre-loaded cblas_sgemm function pointer.

Delegates to the Apple Accelerate cblas_sgemm routine using the supplied function handle, then applies the optional epilogue to the output tile. Requires all operands to be DType.float32.

Parameters:

transpose_b (Bool): Whether to treat B as transposed.
elementwise_lambda_fn (Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None]): Optional epilogue applied element-wise to output.

Args:

cblas_gemm_fn (def(_CBLASOrder, _CBLASTranspose, _CBLASTranspose, Int32, Int32, Int32, Float32, Pointer[Float32, ImmutAnyOrigin, _safe=False], Int32, Pointer[Float32, ImmutAnyOrigin, _safe=False], Int32, Float32, Pointer[Float32, MutAnyOrigin, _safe=False], Int32) thin -> None): Pre-loaded cblas_sgemm function handle from the Accelerate library.
c (TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type]): Output matrix tile.
a (TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type]): Input A matrix tile.
b (TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type]): Input B matrix tile.

def apple_matmul[*, transpose_b: Bool = False, elementwise_lambda_fn: Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None] = None](c: TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type], a: TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type], b: TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type])

Performs a single-precision matmul via the Apple Accelerate library.

Loads cblas_sgemm from the Accelerate framework and delegates to apple_matmul with the function handle. Requires all operands to be DType.float32.

Parameters:

transpose_b (Bool): Whether to treat B as transposed.
elementwise_lambda_fn (Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None]): Optional epilogue applied element-wise to output.

Args:

c (TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type]): Output matrix tile.
a (TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type]): Input A matrix tile.
b (TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type]): Input B matrix tile.