1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_mult_q15.c
4 * Description: Q15 vector multiplication
5 *
6 * $Date: 23 April 2021
7 * $Revision: V1.9.0
8 *
9 * Target Processor: Cortex-M and Cortex-A cores
10 * -------------------------------------------------------------------- */
11 /*
12 * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
13 *
14 * SPDX-License-Identifier: Apache-2.0
15 *
16 * Licensed under the Apache License, Version 2.0 (the License); you may
17 * not use this file except in compliance with the License.
18 * You may obtain a copy of the License at
19 *
20 * www.apache.org/licenses/LICENSE-2.0
21 *
22 * Unless required by applicable law or agreed to in writing, software
23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
25 * See the License for the specific language governing permissions and
26 * limitations under the License.
27 */
28
29 #include "dsp/basic_math_functions.h"
30
31 /**
32 @ingroup groupMath
33 */
34
35 /**
36 @addtogroup BasicMult
37 @{
38 */
39
40 /**
41 @brief Q15 vector multiplication
42 @param[in] pSrcA points to first input vector
43 @param[in] pSrcB points to second input vector
44 @param[out] pDst points to output vector
45 @param[in] blockSize number of samples in each vector
46
47 @par Scaling and Overflow Behavior
48 The function uses saturating arithmetic.
49 Results outside of the allowable Q15 range [0x8000 0x7FFF] are saturated.
50 */
51 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
52
53 #include "arm_helium_utils.h"
54
arm_mult_q15(const q15_t * pSrcA,const q15_t * pSrcB,q15_t * pDst,uint32_t blockSize)55 ARM_DSP_ATTRIBUTE void arm_mult_q15(
56 const q15_t * pSrcA,
57 const q15_t * pSrcB,
58 q15_t * pDst,
59 uint32_t blockSize)
60 {
61 uint32_t blkCnt; /* loop counters */
62 q15x8_t vecA, vecB;
63
64 /* Compute 8 outputs at a time */
65 blkCnt = blockSize >> 3;
66 while (blkCnt > 0U)
67 {
68 /*
69 * C = A * B
70 * Multiply the inputs and then store the results in the destination buffer.
71 */
72 vecA = vld1q(pSrcA);
73 vecB = vld1q(pSrcB);
74 vst1q(pDst, vqdmulhq(vecA, vecB));
75 /*
76 * Decrement the blockSize loop counter
77 */
78 blkCnt--;
79 /*
80 * advance vector source and destination pointers
81 */
82 pSrcA += 8;
83 pSrcB += 8;
84 pDst += 8;
85 }
86 /*
87 * tail
88 */
89 blkCnt = blockSize & 7;
90 if (blkCnt > 0U)
91 {
92 mve_pred16_t p0 = vctp16q(blkCnt);
93 vecA = vld1q(pSrcA);
94 vecB = vld1q(pSrcB);
95 vstrhq_p(pDst, vqdmulhq(vecA, vecB), p0);
96 }
97 }
98
99 #else
arm_mult_q15(const q15_t * pSrcA,const q15_t * pSrcB,q15_t * pDst,uint32_t blockSize)100 ARM_DSP_ATTRIBUTE void arm_mult_q15(
101 const q15_t * pSrcA,
102 const q15_t * pSrcB,
103 q15_t * pDst,
104 uint32_t blockSize)
105 {
106 uint32_t blkCnt; /* Loop counter */
107
108 #if defined (ARM_MATH_LOOPUNROLL)
109
110 #if defined (ARM_MATH_DSP)
111 q31_t inA1, inA2, inB1, inB2; /* Temporary input variables */
112 q15_t out1, out2, out3, out4; /* Temporary output variables */
113 q31_t mul1, mul2, mul3, mul4; /* Temporary variables */
114 #endif
115
116 /* Loop unrolling: Compute 4 outputs at a time */
117 blkCnt = blockSize >> 2U;
118
119 while (blkCnt > 0U)
120 {
121 /* C = A * B */
122
123 #if defined (ARM_MATH_DSP)
124 /* read 2 samples at a time from sourceA */
125 inA1 = read_q15x2_ia (&pSrcA);
126 /* read 2 samples at a time from sourceB */
127 inB1 = read_q15x2_ia (&pSrcB);
128 /* read 2 samples at a time from sourceA */
129 inA2 = read_q15x2_ia (&pSrcA);
130 /* read 2 samples at a time from sourceB */
131 inB2 = read_q15x2_ia (&pSrcB);
132
133 /* multiply mul = sourceA * sourceB */
134 mul1 = (q31_t) ((q15_t) (inA1 >> 16) * (q15_t) (inB1 >> 16));
135 mul2 = (q31_t) ((q15_t) (inA1 ) * (q15_t) (inB1 ));
136 mul3 = (q31_t) ((q15_t) (inA2 >> 16) * (q15_t) (inB2 >> 16));
137 mul4 = (q31_t) ((q15_t) (inA2 ) * (q15_t) (inB2 ));
138
139 /* saturate result to 16 bit */
140 out1 = (q15_t) __SSAT(mul1 >> 15, 16);
141 out2 = (q15_t) __SSAT(mul2 >> 15, 16);
142 out3 = (q15_t) __SSAT(mul3 >> 15, 16);
143 out4 = (q15_t) __SSAT(mul4 >> 15, 16);
144
145 /* store result to destination */
146 #ifndef ARM_MATH_BIG_ENDIAN
147 write_q15x2_ia (&pDst, __PKHBT(out2, out1, 16));
148 write_q15x2_ia (&pDst, __PKHBT(out4, out3, 16));
149 #else
150 write_q15x2_ia (&pDst, __PKHBT(out1, out2, 16));
151 write_q15x2_ia (&pDst, __PKHBT(out3, out4, 16));
152 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
153
154 #else
155 *pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
156 *pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
157 *pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
158 *pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
159 #endif
160
161 /* Decrement loop counter */
162 blkCnt--;
163 }
164
165 /* Loop unrolling: Compute remaining outputs */
166 blkCnt = blockSize % 0x4U;
167
168 #else
169
170 /* Initialize blkCnt with number of samples */
171 blkCnt = blockSize;
172
173 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
174
175 while (blkCnt > 0U)
176 {
177 /* C = A * B */
178
179 /* Multiply inputs and store result in destination buffer. */
180 *pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
181
182 /* Decrement loop counter */
183 blkCnt--;
184 }
185
186 }
187 #endif /* defined(ARM_MATH_MVEI) */
188
189 /**
190 @} end of BasicMult group
191 */
192
