Math: Optimise 16-bit matrix multiplication function

- Switch to int32_t in accumulators to reduce memory and improve speed in
  mat_multiply and mat_multiply_elementwise functions.
- Refactor pointer incrementation for code clarity and optimizations,
  enhancing maintainability and execution speed.
- Handle Q0 data directly in core logic, removing redundant conditionals
  for faster execution in common scenarios.
- Revise fractional bit shift logic for better accuracy in fixed-point
  calculations.

This consolidation achieves notable performance boosts and memory savings.

Signed-off-by: Shriram Shastry <malladi.sastry@intel.com>

Author: ShriramShastry

src/math/matrix.c

@@ -3,93 +3,105 @@
 // Copyright(c) 2022 Intel Corporation. All rights reserved.
 //
 // Author: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
+//	   Shriram Shastry <malladi.sastry@linux.intel.com>
+//
 
 #include <sof/math/matrix.h>
 #include <errno.h>
 #include <stdint.h>
 
-int mat_multiply(struct mat_matrix_16b *a, struct mat_matrix_16b *b, struct mat_matrix_16b *c)
+/* Performs matrix multiplication of two fixed-point 16-bit integer matrices,
+ * storing the result in a third matrix. It accounts for fractional bits for
+ * fixed-point arithmetic, adjusting the result accordingly.
+ *
+ * Arguments:
+ *   a: pointer to the first input matrix
+ *   b: pointer to the second input matrix
+ *   c: pointer to the output matrix to store result
+ *
+ * Return:
+ *   0 on successful multiplication.
+ *   -EINVAL if input dimensions do not allow for multiplication.
+ *   -ERANGE if the shift operation might cause integer overflow.
+ */
+int mat_multiply(struct mat_matrix_16b *a, struct mat_matrix_16b *b,
+		 struct mat_matrix_16b *c)
 {
-	int64_t s;
-	int16_t *x;
-	int16_t *y;
-	int16_t *z = c->data;
-	int i, j, k;
-	int y_inc = b->columns;
-	const int shift_minus_one = a->fractions + b->fractions - c->fractions - 1;
+	int32_t acc;	/* Accumulator for dot product calculation */
+	int16_t *x, *y, *z = c->data; /* Pointers for matrices a, b, and c */
+	int i, j, k;	/* Loop counters */
+	int y_inc = b->columns;	   /* Column increment for matrix b elements */
+	/* Calculate shift amount for adjusting fractional bits in the result */
+	const int shift = a->fractions + b->fractions - c->fractions - 1;
 
+	/* Validate matrix dimensions are compatible for multiplication */
 	if (a->columns != b->rows || a->rows != c->rows || b->columns != c->columns)
 		return -EINVAL;
 
-	/* If all data is Q0 */
-	if (shift_minus_one == -1) {
-		for (i = 0; i < a->rows; i++) {
-			for (j = 0; j < b->columns; j++) {
-				s = 0;
-				x = a->data + a->columns * i;
-				y = b->data + j;
-				for (k = 0; k < b->rows; k++) {
-					s += (int32_t)(*x) * (*y);
-					x++;
-					y += y_inc;
-				}
-				*z = (int16_t)s; /* For Q16.0 */
-				z++;
-			}
-		}
-
-		return 0;
-	}
+	/* Check shift to ensure no integer overflow occurs during shifting */
+	if (shift < -1 || shift > 31)
+		return -ERANGE;
 
+	/* Matrix multiplication loop */
 	for (i = 0; i < a->rows; i++) {
 		for (j = 0; j < b->columns; j++) {
-			s = 0;
-			x = a->data + a->columns * i;
-			y = b->data + j;
+			acc = 0;   /* Initialize accumulator for each element */
+			x = a->data + a->columns * i; /* Set x at the start of ith row of a */
+			y = b->data + j; /* Set y at the top of jth column of b */
+			/* Dot product loop */
 			for (k = 0; k < b->rows; k++) {
-				s += (int32_t)(*x) * (*y);
-				x++;
-				y += y_inc;
+				acc += (int32_t)(*x++) * (*y); /* Multiply & accumulate */
+				y += y_inc;  /* Move to next row in the current column of b */
 			}
-			*z = (int16_t)(((s >> shift_minus_one) + 1) >> 1); /*Shift to Qx.y */
-			z++;
+			/* Assign computed value to c matrix, adjusting for fractional bits */
+			if (shift == -1)
+				*z = (int16_t)acc;
+			else
+				*z = (int16_t)(((acc >> shift) + 1) >> 1);
+			z++;	/* Move to the next element in the output matrix */
 		}
 	}
 	return 0;
 }
 
+/* Description: Performs element-wise multiplication of 16-bit int matrices
+ *		(a and b), storing the result in a third matrix (c). Assumes
+ *		all matrices are of the same dimensions and adjusts fractional
+ *		bits appropriately.
+ * Arguments:
+ *   a - pointer to the first input matrix
+ *   b - pointer to the second input matrix
+ *   c - pointer to the output matrix to store result
+ * Returns:
+ *   0 on successful element-wise multiplication
+ *   -EINVAL if pointers are null or dimensions do not match
+ */
 int mat_multiply_elementwise(struct mat_matrix_16b *a, struct mat_matrix_16b *b,
 			     struct mat_matrix_16b *c)
-{	int64_t p;
-	int16_t *x = a->data;
-	int16_t *y = b->data;
-	int16_t *z = c->data;
-	int i;
-	const int shift_minus_one = a->fractions + b->fractions - c->fractions - 1;
+{
+	int16_t *x = a->data;  /* Pointer to data of matrix a */
+	int16_t *y = b->data;  /* Pointer to data of matrix b */
+	int16_t *z = c->data;  /* Pointer to data of matrix c */
+	int32_t prod;  /* Product of element-wise multiplication */
 
-	if (a->columns != b->columns || b->columns != c->columns ||
-	    a->rows != b->rows || b->rows != c->rows) {
+	/* Validate pointers and matrices dimensions */
+	if (!a || !b || !c || a->columns != b->columns || a->rows != b->rows)
 		return -EINVAL;
-	}
 
-	/* If all data is Q0 */
-	if (shift_minus_one == -1) {
-		for (i = 0; i < a->rows * a->columns; i++) {
-			*z = *x * *y;
-			x++;
-			y++;
-			z++;
-		}
+	const int total_elements = a->rows * a->columns;  /* Total elements count */
+	const int shift = a->fractions + b->fractions - c->fractions - 1;
 
-		return 0;
-	}
-
-	for (i = 0; i < a->rows * a->columns; i++) {
-		p = (int32_t)(*x) * *y;
-		*z = (int16_t)(((p >> shift_minus_one) + 1) >> 1); /*Shift to Qx.y */
-		x++;
-		y++;
-		z++;
+	/* Perform element-wise multiplication */
+	if (shift == -1) {
+		/* No adjustment needed for fractional bits */
+		for (int i = 0; i < total_elements; i++)
+			z[i] = x[i] * y[i];  /* Direct multiplication */
+	} else {
+		/* Adjustment needed for fractional bits */
+		for (int i = 0; i < total_elements; i++) {
+			prod = (int32_t)x[i] * y[i];  /* Multiply with extended precision */
+			z[i] = (int16_t)(((prod >> shift) + 1) >> 1);
+		}
 	}
 
 	return 0;