Matrix Multiplication [Matrix Functions]

Functions
arm_status	arm_mat_mult_f32 (const arm_matrix_instance_f32 *pSrcA, const arm_matrix_instance_f32 *pSrcB, arm_matrix_instance_f32 *pDst)
arm_status	arm_mat_mult_q31 (const arm_matrix_instance_q31 *pSrcA, const arm_matrix_instance_q31 *pSrcB, arm_matrix_instance_q31 *pDst)
arm_status	arm_mat_mult_fast_q31 (const arm_matrix_instance_q31 *pSrcA, const arm_matrix_instance_q31 *pSrcB, arm_matrix_instance_q31 *pDst)
arm_status	arm_mat_mult_q15 (const arm_matrix_instance_q15 *pSrcA, const arm_matrix_instance_q15 *pSrcB, arm_matrix_instance_q15 *pDst, q15_t *pScratch)
arm_status	arm_mat_mult_fast_q15 (const arm_matrix_instance_q15 *pSrcA, const arm_matrix_instance_q15 *pSrcB, arm_matrix_instance_q15 *pDst, q15_t *pScratch)

---

Detailed Description

Multiplies two matrices.

Multiplication of two 3 x 3 matrices

Matrix multiplication is only defined if the number of columns of the first matrix equals the number of rows of the second matrix. Multiplying an M x N matrix with an N x P matrix results in an M x P matrix. When matrix size checking is enabled, the functions check: (1) that the inner dimensions of pSrcA and pSrcB are equal; and (2) that the size of the output matrix equals the outer dimensions of pSrcA and pSrcB.

---

Function Documentation

arm_status arm_mat_mult_f32	(	const arm_matrix_instance_f32 *	pSrcA,
		const arm_matrix_instance_f32 *	pSrcB,
		arm_matrix_instance_f32 *	pDst
	)

Floating-point matrix multiplication.

Parameters:

[in]	*pSrcA	points to the first input matrix structure
[in]	*pSrcB	points to the second input matrix structure
[out]	*pDst	points to output matrix structure

Returns:

The function returns either ARM_MATH_SIZE_MISMATCH or ARM_MATH_SUCCESS based on the outcome of size checking.

Examples:

arm_class_marks_example_f32.c, and arm_matrix_example_f32.c.

Definition at line 62 of file arm_mat_mult_f32.c.

arm_status arm_mat_mult_q31	(	const arm_matrix_instance_q31 *	pSrcA,
		const arm_matrix_instance_q31 *	pSrcB,
		arm_matrix_instance_q31 *	pDst
	)

Q31 matrix multiplication.

Parameters:

[in]	*pSrcA	points to the first input matrix structure
[in]	*pSrcB	points to the second input matrix structure
[out]	*pDst	points to output matrix structure

Returns:

The function returns either ARM_MATH_SIZE_MISMATCH or ARM_MATH_SUCCESS based on the outcome of size checking.

Scaling and Overflow Behavior:

The function is implemented using an internal 64-bit accumulator. The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit. There is no saturation on intermediate additions. Thus, if the accumulator overflows it wraps around and distorts the result. The input signals should be scaled down to avoid intermediate overflows. The input is thus scaled down by log2(numColsA) bits to avoid overflows, as a total of numColsA additions are performed internally. The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result.

See arm_mat_mult_fast_q31() for a faster but less precise implementation of this function.

Definition at line 62 of file arm_mat_mult_q31.c.

arm_status arm_mat_mult_fast_q31	(	const arm_matrix_instance_q31 *	pSrcA,
		const arm_matrix_instance_q31 *	pSrcB,
		arm_matrix_instance_q31 *	pDst
	)

Q31 matrix multiplication (fast variant)

Parameters:

[in]	*pSrcA	points to the first input matrix structure
[in]	*pSrcB	points to the second input matrix structure
[out]	*pDst	points to output matrix structure

Returns:

The function returns either ARM_MATH_SIZE_MISMATCH or ARM_MATH_SUCCESS based on the outcome of size checking.

Scaling and Overflow Behavior:

The difference between the function arm_mat_mult_q31() and this fast variant is that the fast variant use a 32-bit rather than a 64-bit accumulator. The result of each 1.31 x 1.31 multiplication is truncated to 2.30 format. These intermediate results are accumulated in a 32-bit register in 2.30 format. Finally, the accumulator is saturated and converted to a 1.31 result.

The fast version has the same overflow behavior as the standard version but provides less precision since it discards the low 32 bits of each multiplication result. In order to avoid overflows completely the input signals must be scaled down. Scale down one of the input matrices by log2(numColsA) bits to avoid overflows, as a total of numColsA additions are computed internally for each output element.

See arm_mat_mult_q31() for a slower implementation of this function which uses 64-bit accumulation to provide higher precision.

Definition at line 66 of file arm_mat_mult_fast_q31.c.

arm_status arm_mat_mult_q15	(	const arm_matrix_instance_q15 *	pSrcA,
		const arm_matrix_instance_q15 *	pSrcB,
		arm_matrix_instance_q15 *	pDst,
		q15_t *	pScratch
	)

Q15 matrix multiplication.

Parameters:

[in]	*pSrcA	points to the first input matrix structure
[in]	*pSrcB	points to the second input matrix structure
[out]	*pDst	points to output matrix structure
[in]	*
[in]	*pScratch	points to scratch buffer of size numBRows * numBCol

Returns:

The function returns either ARM_MATH_SIZE_MISMATCH or ARM_MATH_SUCCESS based on the outcome of size checking.

Conditions for optimum performance

Input, output and scratch buffers should be aligned by 32-bit

Restrictions

If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE In this case input, output, state buffers should be aligned by 32-bit

Scaling and Overflow Behavior:

The function is implemented using a 64-bit internal accumulator. The inputs to the multiplications are in 1.15 format and multiplications yield a 2.30 result. The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format. This approach provides 33 guard bits and there is no risk of overflow. The 34.30 result is then truncated to 34.15 format by discarding the low 15 bits and then saturated to 1.15 format.

Refer to arm_mat_mult_fast_q15() for a faster but less precise version of this function.

Definition at line 71 of file arm_mat_mult_q15.c.

arm_status arm_mat_mult_fast_q15	(	const arm_matrix_instance_q15 *	pSrcA,
		const arm_matrix_instance_q15 *	pSrcB,
		arm_matrix_instance_q15 *	pDst,
		q15_t *	pScratch
	)

Q15 matrix multiplication (fast variant)

Parameters:

[in]	*pSrcA	points to the first input matrix structure
[in]	*pSrcB	points to the second input matrix structure
[out]	*pDst	points to output matrix structure
[in]	*pScratch	points to the array for storing intermediate results

Returns:

The function returns either ARM_MATH_SIZE_MISMATCH or ARM_MATH_SUCCESS based on the outcome of size checking.

Conditions for optimum performance

Input, output and statte buffers should be aligned by 32-bit

Restrictions

If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE In this case input, output, state buffers should be aligned by 32-bit

Scaling and Overflow Behavior:

The difference between the function arm_mat_mult_q15() and this fast variant is that the fast variant use a 32-bit rather than a 64-bit accumulator. The result of each 1.15 x 1.15 multiplication is truncated to 2.30 format. These intermediate results are accumulated in a 32-bit register in 2.30 format. Finally, the accumulator is saturated and converted to a 1.15 result.

The fast version has the same overflow behavior as the standard version but provides less precision since it discards the low 16 bits of each multiplication result. In order to avoid overflows completely the input signals must be scaled down. Scale down one of the input matrices by log2(numColsA) bits to avoid overflows, as a total of numColsA additions are computed internally for each output element.

See arm_mat_mult_q15() for a slower implementation of this function which uses 64-bit accumulation to provide higher precision.

Definition at line 76 of file arm_mat_mult_fast_q15.c.